The Economics of Football, 2nd Edition

This page intentionally left blank
The Economics of Football
The second edition of this popular book presents a detailed

economic analysis of professional football at club level, with
new material included to reflect the development of the eco-
nomics of professional football over the past ten years. Using
a combination of economic reasoning and statistical and
econometric analysis, the authors build upon the successes
and strengths of the first edition to guide readers through the
economic complexities and peculiarities of English club foot-
ball. The book uses a wide range of international comparisons
to help emphasise both the broader relevance as well as the
unique characteristics of the English experience. Topics cov-
ered include some of the most hotly debated issues currently
surrounding professional football, including player salaries,
the effects of management on team performance, betting on
football, racial discrimination and the performance of football
referees. This edition also includes new chapters on football
applications of game theory, club football around the world,
and the economics of the World Cup.
stephen d o b s o n is Professor of Economics at Hull

University Business School, Hull University.
john g o d d a r d is Professor of Financial Economics at

Bangor Business School, Bangor University.
The Economics of Football
Second edition
Stephen Dobson
and
John Goddard

cambridge university press

Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Tokyo, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press,

New York
www.cambridge.org
Information on this title: www.cambridge.org/9780521517140
© Stephen Dobson and John Goddard 2011
This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2011
Printed in the United Kingdom at the University Press, Cambridge
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data

Dobson, Stephen.
The economics of football / Stephen Dobson, John Goddard. – 2nd ed.
p. cm.
Includes bibliographical references and index.
ISBN 978-0-521-51714-0 (hardback)
1. Soccer–Economic aspects–Great Britain. 2. Soccer–Great Britain–
Finance. I. Goddard, John A. II. Title.
GV943.3.D63 2011
338.4′37963340941–dc22
2010040413
ISBN 978-0-521-51714-0 Hardback
Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate.

To: My daughters Imogen and Hannah

sd
To: My family and friends

jg
Contents
Preface page xi
Acknowledgements xiii
List of figures xiv
List of tables xv
1 Introduction 1
1.1 The economics of professional team sports: three seminal
contributions 2
1.2 Outline of this volume 8
2 The economic theory of professional sports leagues 12

2.1 The North American professional team sports model 14
2.2 A model of an n-team league 19
2.3 A model of a two-team league 23
2.4 Revenue sharing 27
2.5 Restraints on expenditure on players’ salaries 33
3 Competitive balance, uncertainty of outcome and home-field advantage 42

3.1 Measuring competitive balance and competitive inequality 42
3.2 Home-field advantage 50
3.3 Distributional properties of the goals scored by the home and
away teams 59
3.4 Good and poor sequences, and persistence in football match
results 67
4 Forecasting models for football match results 79

4.1 Previous literature on modelling and forecasting match results
in football 79
4.2 A goals-based forecasting model 82
4.3 Probabilistic forecasts for match results in ‘scores’ format 88
vii
viii Contents
4.4 A results-based forecasting model 95

4.5 Probabilistic forecasts for match results in ‘win-draw-lose’ format 98
4.6 Evaluation of the goals-based and results-based forecasting models 100
5 Game theory and football games 106

5.1 The penalty kick 107
5.2 A game-theoretic model of in-play strategic choice for football
teams 111
5.3 The timings of player dismissals and goals 119
5.4 An empirical model for the in-play arrival rates of player
dismissals and goals 122
5.5 Estimation results and interpretation 124
5.6 Stochastic simulations for in-play match result probabilities
conditional on the current state of the match 130
6 English professional football: historical development and commercial

structure 139
6.1 English professional football: competitive structure and team
performance 139
6.2 Match attendances 150
6.3 Financial structure of English football: overview of profit and loss
accounts 159
6.4 Gate revenues and admission prices 162
6.5 Broadcast revenues 171
6.6 Football’s labour market: players’ salaries and the transfer
system 179
6.7 Ownership, governance and finance 186
7 Determinants of professional footballers’ salaries 197

7.1 The compensation of professional footballers 198
7.2 The economics of superstars 201
7.3 Rank-order tournaments and intra-team earnings distributions 204
7.4 Determinants of players’ compensation: empirical evidence 209
8 Professional footballers: employment patterns and racial discrimination 216

8.1 Employment mobility, migration and career structure in English
football 217
8.2 International migration of professional footballers 227
8.3 Racial discrimination in professional team sports 236
9 The football manager 249

9.1 The role of the football manager 251
9.2 Measuring the managerial contribution: the production frontier
approach 254
Contents ix
9.3 Patterns of managerial change in English football 260

9.4 Determinants of managerial change 270
9.5 Estimation of hazard functions for managerial departure 276
9.6 The effect of managerial change on team performance 285
9.7 Managerial succession effects in English football 289
10 The football referee 295

10.1 The role of the football referee 296
10.2 Favouritism and referee behaviour 298
10.3 The incidence of disciplinary sanction in English Premier
League football 303
11 Spectator demand for football 321

11.1 Econometric analysis of football attendances 322
11.2 Modelling the demand for attendance at English league football,
1947–1997 336
11.3 Explaining base attendances, and the loyalty, league position,
price and goals scored coefficients 346
12 Gambling on football 352

12.1 Previous evidence on the informational efficiency of football and
other sports betting markets 353
12.2 A forecasting model for half-time/full-time match outcomes 365
12.3 Comparing the model’s probabilistic forecasts with betting odds 367
12.4 Testing the informational efficiency of the half-time/full-time
fixed-odds betting market 371
13 Football around the world: France, Germany, Brazil, Japan and China 380
13.1 France 380
13.2 Germany 385
13.3 Brazil 391
13.4 Japan 395
13.5 China 399
14 The economics of the World Cup 405

14.1 The World Cup 407
14.2 Costs and benefits of hosting a mega sporting event 409
14.3 Prospective economic impact studies 413
14.4 Retrospective economic impact studies 415
14.5 Intangible benefits of mega events 417
References 423
Index 447
Preface
Since the early 1990s, professional football in many countries has experienced
an astonishing transformation. Player salaries have risen exponentially, television
contracts yield revenues on a scale unimaginable only a few years ago, many foot-
ball stadia have been completely rebuilt, and the importance of commercial spon-
sorship and merchandising has increased beyond measure. Commercial aspects
of football feature regularly in the news headlines, and the media devote pages
to coverage of football finances. Football’s importance is not only economic, but
also social and cultural. Several million people attend matches each season, and
many millions more watch football on television and follow its fortunes through
coverage in the media. At the grassroots level, football’s popularity as a partici-
pant sport generates benefits for the health of the population. At the highest level,
international footballing success generates intangible benefits in the form of pres-
tige and goodwill.
Academic interest in the economic analysis of football has mirrored the growth
in the sport’s popularity. In the US, economists have written and published books
and scholarly articles on major league sports since the mid-1950s. Consequently,
the older academic literature on sports economics is dominated by studies of sports
such as baseball, basketball and (American) football. These writings shed light on
a wide range of issues, including the determinants of the compensation received
by sports professionals, the nature of joint production in team sports, competitive
balance, uncertainty of outcome and the distribution of playing talent in sports
leagues, and the contribution of the coach or manager to team performance. The
common thread linking research into all of these topics is the formulation and
testing of economic hypotheses using sports as a laboratory. A major attraction of
sports to empirical economists is that the availability of data permits investigation
of economic propositions that would be difficult to test in other areas, owing to a
lack of suitable data.
During the last decade of the old century and the first decade of the new, schol-
arly papers on the economics of football have been published with increasing regu-
larity in academic journals. Undergraduate and postgraduate students in many
universities study the economics of sports as part of their degree programmes. At
xi
xii Preface
the end of the 1990s, we felt that a monograph was needed to cover developments
in the subject, and present a unifying overview of this relatively new area of aca-
demic research. We therefore decided to write the first edition of The Economics of
Football, which was published by Cambridge University Press in September 2001.
Since publication, we have been encouraged by the responses to the first edition we
have received from scholars, students and other readers.
Since the appearance of the first edition, there has been a proliferation of schol-
arly work on the economics of football, in Europe and many other parts of the
world. Several new academic journals have been launched, dedicated to the study
of sports economics, finance and management. Football has received extensive
coverage in these field journals, and in mainstream journals in economics, finance
and management. Several books have been published, focusing on the economics
and business aspects of football.
When Cambridge University Press suggested that we should consider prepar-
ing a new edition of The Economics of Football, we agreed that the time was right
to revise and update the book. This second edition provides substantially revised
and updated coverage of all of the key topics from the first edition. In addition, a
broad range of new material reflects the growth in the scope and sophistication of
the literature over the past decade. The second edition includes new chapters on
football applications of game theory, football referees, football betting markets,
the economics of club football around the world, and the World Cup. As before,
our objective is to present a wide-ranging overview of the current state of theoret-
ical and empirical research on the economics of football.
Acknowledgements
We would like to thank colleagues and students at the Department of Economics,

Otago University; Bangor Business School, Bangor University; and the School
of Business and Economics, Swansea University for their direct and sometimes
unknowing help towards the development of this project. We would like to give
special thanks to Rick Audas, Tunde Buraimo, Andy Cooke, Juan Carlos Cuestas,
Peter Dawson, Neil Doncaster, David Forrest, Rodney Fort, Bill Gerrard, Leo
Kahane, Ruud Koning, Huw Lloyd Williams, Phil Molyneux, Dorian Owen, Phil
Quinn, Rob Simmons, Peter Sloane, Frank Stähler, Sarah Wale, Yizheng Wang,
Leighton Vaughan Williams and John Wilson.
We would also like to thank a number of staff at Cambridge University Press
for their assistance during the development of this project. In particular, we are
grateful to Chris Harrison (Publishing Director, Social Sciences) for his unfailing
support and encouragement, and Philip Good (Assistant Editor, Economics and
Business) for his advice and guidance. We are grateful to Matt Davies of Out of
House Publishing Solutions Ltd for his assistance during the process of produc-
tion, and Penny Harper (copy-editor) for her meticulous efforts in checking and
correcting the original manuscript.
xiii
Figures
2.1 Profit-maximisation and win-percent-maximisation

equilibria: two-team closed model page 25
2.2 Profit-maximisation and win-percent-maximisation
equilibria: two-team open model 26
2.3 Revenue sharing in a closed model: profit maximisation 28
2.4 Revenue sharing in a closed model: win-percent maximisation 29
2.5 Nash equilibrium and joint profit-maximisation equilibrium:
profit-maximisation model 31
2.6 Luxury tax: closed model 34
2.7 Luxury tax: open model 35
2.8 Payroll cap: closed model 36
2.9 Payroll cap: open model 37
2.10 G14 payroll cap: closed model 38
2.11 G14 payroll cap: open model 39
3.1 The Lorenz curve and Gini coefficient 46
6.1 Percentage shares in aggregate performance of clubs in Groups 1 to 5 150
6.2 Percentage shares in aggregate attendance of clubs in Groups 1 to 5 154
6.3 Percentage shares in aggregate gate revenue of clubs in Groups 1 to 5 168
7.1 Optimum investment in playing talent, rank-order tournament model 208
xiv
Tables
3.1 Percentages of home wins, draws and away wins, average numbers of
goals scored by the home and away teams, and win percentages per
season, English league, 1970–2009 seasons page 54
3.2 Home-team success percentages (2/1/0 points for W/D/L) by tier
(division) in five-season bands, English league, 1970–2009 seasons 56
3.3 Percentages of home wins, draws and away wins, and average numbers
of goals scored by home and away teams, international comparisons,
1973–2009 seasons 57
3.4 Joint and marginal percentage distributions of goals scored by home
and away teams, English league, 1970–1981 seasons 60
3.5 Joint and marginal percentage distributions of goals scored by home
and away teams, English league, 1982–2009 seasons 60
3.6 Percentage distribution of goals scored by the home team, conditional
on the number of goals scored by the away team, English league,
1982–2009 seasons 61
3.7 Percentage distribution of goals scored by the away team, conditional
on the number of goals scored by the home team, English league,
1982–2009 seasons 62
3.8 Parameter estimates for fitted double and bivariate Poisson and negative
binomial distributions, home- and away-team goals data, English
league, 1982–2009 seasons 66
3.9 Hypothesis tests for comparisons between fitted double and bivariate
Poisson and negative binomial distributions, home- and away-team
goals data, English league, 1982–2009 seasons 68
3.10 Longest runs of consecutive results, English league, 1970–2009 seasons 69
3.11 Empirical unconditional and conditional match result probabilities 70
3.12 League table, Premier League, 2009 season, and ordered probit team
quality parameter estimates 72
3.13 Simulated unconditional and conditional match result probabilities 74
3.14 Tests for persistence in sequences of consecutive match results 75
4.1 Goals-based forecasting model: estimated coefficients and p-values 85
xv
xvi List of tables
4.2 Data for calculation of average goals scored and conceded covariates,
Hull City vs Tottenham Hotspur fixture 89
4.3 Data for calculation of goals scored and conceded in recent matches
covariates, Hull City vs Tottenham Hotspur fixture 91
4.4 Estimated match result probabilities, Hull City vs Tottenham Hotspur
fixture 95
4.5 Results-based forecasting model: estimated coefficients and p-values 97
4.6 Fitted match result and selected score probabilities, Premier League,
weekend of 21–23 February 2009 102
4.7 Pseudo-R-square values for forecasting performance, 2009
season: goals-based forecasting model, and probabilities derived from
six betting firms’ prices 104
5.1 Observed proportions of penalty kicks, goalkeeper dives and goals 110
5.2 Hypothetical goal-scoring and player-dismissal probabilities for
numerical examples 114
5.3 Determination of the two teams’ optimal strategies for the 90th minute,
home team leading by one goal after 89 minutes 115
5.4 Determination of the two teams’ optimal strategies for the 90th minute,
scores level after 89 minutes 116
5.5 Hypothetical match result probabilities at the end of the first minute,
conditional on score and player dismissals, for numerical examples 118
5.6 Determination of the two teams’ optimal strategies for the first minute 118
5.7 Rates of player dismissal and goal scoring conditional on current
duration, English League, T1–T4, 2002–2009 seasons 119
5.8 Rates of player dismissal and goal scoring conditional on current
difference in scores, English League, T1–T4, 2002–2009 seasons 120
5.9 Rates of player dismissal and goal scoring conditional on numerical
disparity in players, English League, T1–T4, 2002–2009 seasons 120
5.10 Estimation results: player-dismissal hazard functions 125
5.11 Estimation results: goal-scoring hazard functions 128
5.12 Home-win probabilities, conditional on relative team strengths and the
state of the match at various durations 132
5.13 Draw probabilities, conditional on relative team strengths and the state
of the match at various durations 133
5.14 Away-win probabilities, conditional on relative team strengths and the
state of the match at various durations 134
5.15 In-play home-win/draw/away-win probabilities: illustration 136
6.1 Historical performance of top English teams in league and cup
competition 144
6.2 Incidence of giant-killings in FA Cup ties, 1974–2009 147
6.3 Group definitions 148
6.4 English league attendances, aggregate and by tier 151
6.5 Average revenue, wages and salaries and operating profit per season,
three-season periods, English League, 1994–2008 seasons 160
List of tables xvii
6.6 Financial data, leading English clubs and averages by tier, 2008
season, £m 161
6.7 TV and other revenue, leading English clubs and averages by tier,
2008 season, £m 162
6.8 English league average admission prices 164
6.9 English league gate revenues 166
6.10 Revenue from sale of broadcast rights, English Premier League,
average per season, 1993–2010 seasons, £m 174
6.11 Average net transfer expenditure within the Premier League/Football
League per season, selected three-season periods, £m 185
6.12 Average flows of transfer expenditure within the Premier League/
Football League per season, 2006–2008 seasons, £m 185
6.13 Gross transfer expenditure of Premier League and Football League
clubs per season, 2001–2008 seasons, £m 186
6.14 Average gross transfer expenditure and average revenue per season,
three-season periods, 1994–2008 seasons, £m 186
7.1 Average basic footballer’s salary by tier, English League,
2000 and 2006 seasons, £ 199
8.1 Employment totals for professional footballers by tier, Engish
League, 1986–2009 218
8.2 Percentage distribution of professional footballers in England by age
band and tier, English League, selected years 219
8.3 Percentage distribution of professional footballers in England by month
of birth, English League, selected years 220
8.4 Employment totals for professional footballers by birthplace (country),
English League, 1986–2009 221
8.5 Total numbers of professional footballers by birthplace (country) and
tier, English League, selected years 222
8.6 Percentage distribution of footballers by birthplace (region) and
location of club (region), English League, 1989 223
8.7 Percentage distribution of footballers by birthplace (region) and
location of club (region), English League, 1999 224
8.8 Percentage distribution of footballers in England by birthplace (region)
and location of club (region), English League, 2009 225
8.9 Four-year employment transition probabilities by tier, English League,
1989–1993 226
1993–1997 227
1997–2001 228
2001–2005 229
2005–2009 230
xviii List of tables
8.14 Cross-tabulation of squad players’ home countries and countries of

employment, Euro 2000 233
8.15 Cross-tabulation of squad players’ home countries and countries of
employment, Euro 2008 234
8.16 Estimation results: initial divisional status 243
8.17 Estimation results: retention 244
8.18 Estimation results: divisional transition 245
9.1 Managerial spells ranked by duration (matches), English League,
1973–2009 seasons 261
9.2 Managerial spells ranked by win ratio, English League, 1973–2009
seasons 263
9.3 Managerial departures, by season, English League, 1973–2009
seasons 267
9.4 Average number of managerial departures per season,
four-season bands, English League, 1974–2009 seasons 268
9.5 Managerial departures by month, English League, 1974–2009 seasons 269
9.6 Managerial departures by tier, four-season bands, English League,
1974–2009 seasons 269
9.7 Average duration (matches completed) of terminating managerial spells,
four-season bands, English League, 1974–2009 seasons 270
9.8 Average duration (matches completed) and average win ratio, in
terminating managerial spells by tier, English League, 1974–2009 seasons 271
9.9 Distribution of complete and right-censored managerial spells by
duration (in matches), crude estimates of the survivor, distribution,
hazard and probability functions, English League, 1973–2009 seasons 280
9.10 Involuntary and voluntary managerial job departure hazard
functions: estimation results 282
9.11 The managerial succession effect: estimation results 292
10.1 Numbers of yellow cards incurred by the home and away teams, English
Premier League, 1997–2009 seasons 303
10.2 Numbers of red cards incurred by the home and away teams, English
Premier League, 1997–2009 seasons 303
10.3 Sample frequency distribution for the bivariate disciplinary points
dependent variable, {Z1,j,Z2,j} 305
10.4 Unconditional model: fitted bivariate probabilities and observed
proportions for the numbers of disciplinary ‘points’ incurred by the
home and away teams 307
10.5 Average numbers of yellow and red cards awarded per match by team,
English Premier League, 1997–2009 seasons 311
10.6 Average numbers of yellow and red cards and disciplinary ‘points’
awarded per match by referee, English Premier League, 1997–2009 seasons 314
10.7 Rule changes and changes of interpretation, by season 316
List of tables xix
10.8 Average numbers of yellow cards, red cards and disciplinary ‘points’
awarded per match by season, English Premier League, 1997–2009 seasons 317
11.1 Stadium capacity, average attendance, capacity utilisation, 2009 season
T1 clubs, 1979, 1989, 1999 and 2009 325
11.2 Attendance model: first-stage estimation results 340
11.3 Attendance model: second-stage estimation results 347
12.1 Probabilities for half-time/full-time outcomes 369
12.2 Mean returns based on bookmakers’ ‘best odds’ 370
12.3 Illustration of ‘highest expected return’ betting strategy: English
Premier League, 11 May 2003 372
12.4 Returns from ‘highest expected return’ and indiscriminate betting
strategies 374
12.5 Average rate of return by tier (division): chronological analysis 376
12.6 Mean expected and actual returns: all possible bets ranked by expected
return 377
13.1 Historical performance of top French teams in league and cup
competition 382
13.2 Average league attendances, France Ligue 1 and German Bundesliga 1,
1981–2010 (’000) 385
13.3 Revenue, costs and profitability, aggregates and breakdown, France, all
Ligue 1 clubs, selected years (€m) 386
13.4 Historical performance of top German teams in league and cup
competition 388
13.5 Revenue, costs and profitability, aggregates, Germany, all Bundesliga 1
clubs, selected years (€m) 391
13.6 Historical performance of top Brazilian teams in league and cup
competition 393
13.7 Percentage breakdown of total revenue, twenty-one Brazilian clubs,
2004–2007 394
13.8 J. League winners and runners-up, Japan, 1993–2009 397
13.9 Average attendance, revenue and salary expenditure, Japan, J1 clubs,
1999–2008 (¥m) 399
13.10 Football League and Super League winners and runners-up, China,
1994–2009 400
14.1 History of the World Cup 408
1 Introduction
Academic interest in the economics of professional team sports dates back as far
as the mid-1950s. Since then, many books and journal articles have been written
on the subject. Much of the academic literature originates in the United States
(US). In common with trends that are evident throughout the subject discipline
of economics, empirical research on the economics of sport has become increas-
ingly sophisticated, both theoretically and in its use of econometric methodology.
Papers on the economics of sport now appear regularly in many of the leading
economics journals, and most economists would agree that in view of its social,
cultural and economic importance, professional sport is a legitimate area of inter-
est for both theoretical and empirical researchers. Indeed, many would argue that
the unique configurations of individual and team incentives, and the interactions
between cooperative and competitive modes of behaviour that professional team
sports generate, make this particularly fertile territory in which to explore the per-
ennial questions about incentives, effort, risk and reward that lie at the heart of all
areas of economic inquiry.
This volume makes a contribution to the burgeoning literature on the econom-
ics of team sports, by providing a comprehensive survey of research that is focused
on professional football. The spectacular recent increase in the size of football’s
audience is, of course, a strong motivating factor. Such a survey will recognise
and reflect not only football’s global popularity in the first part of the twenty-
first century, but also the special historical significance of England as the ori-
ginal birthplace of the sport. Club football played in the English Premier League
and Football League provides the laboratory for most of the original, empirical
research that is reported in this volume.
Each chapter of this volume concentrates on a particular aspect of the eco-
nomics of professional football. The previous theoretical and empirical literature
that is relevant to each topic is reviewed, and new and original empirical analyses
are presented. The review sections aim to convey an impression of the breadth
and depth of previous academic research into the economics of professional team
sports. Much has been written already about football, and much more has been
written about other sports, especially in the US where attention naturally tends to
1
2 Introduction
focus on the traditional major league sports of baseball, basketball, (American)

football and hockey. Though football is the main subject of this volume, due atten-
tion and emphasis is devoted to insights that have been obtained from research
into other sports, wherever these turn out to be of wider relevance.
As already emphasised, anyone who reads the academic literature on the eco-
nomics of sport cannot fail to be struck by the sheer volume of column inches that
have been devoted to this topic. Of all the articles that have been published, how-
ever, a few early contributions were particularly important in shaping the research
agenda for the economics of team sports in general, and for the economics of
football in particular. Many of the ideas contained in these articles are as rele-
vant to researchers today as they were when the articles were originally published,
many years ago. In Chapter 1, three such articles are highlighted and reviewed in
some detail. There is, of course, an element of subjectivity in selecting such a small
number of articles out of the many that have been published. Even so, a consensus
seems to have evolved that regards the articles by Simon Rottenberg (1956) and
Walter Neale (1964) as fundamental to the subsequent development of research
on team sports in general. Both articles address various economic implications
of the structural features of the markets within which professional sports teams
operate. The article by Peter Sloane (1971) has also had a major influence on the
developing research agenda, especially in respect of the economics of football. In
Section 1.1, each of these articles is reviewed in turn, and the subsequent develop-
ment of the economics of team sports as an academic discipline is outlined. The
aim of Section 1.1 is to place the research that is reported in the rest of this vol-
ume into its proper context. This is followed in Section 1.2 by a summary of the
contents of the remaining chapters of this volume.
1.1 The economics of professional team sports: three seminal contributions

Rottenberg: ‘The Baseball Players’ Labor Market’, Journal of Political Economy,
1956
Rottenberg is widely credited with writing the first academic analysis of the
economics of professional team sports. The paper was written at a time when US
professional baseball players’ contracts included a reserve clause. Once a player
signed his first one-year contract with a team in Major League Baseball (MLB),
he ceased to be a free agent. On expiry of his present contract, his team retained
the option to renew his contract for another year.1 This served to limit players’
freedom of movement, by binding them to their present employers. Effectively, the
reserve clause created a monopsony in the players’ labour market: each contracted
player could negotiate with only one potential buyer of his services. The baseball
authorities defended the reserve clause on the grounds that it was necessary to
ensure an equal distribution of playing talent among opposing teams. Without the
reserve clause, the rich teams (with the largest potential markets) would outbid the
poorer ones for the best available players. This would tend to reduce uncertainty of
The economics of professional team sports 3
outcome and spectator interest in the league competition as a whole, and depress
the attendances and revenues of all teams.
Rottenberg’s contribution was to argue that free agency in the players’ labour
market would not necessarily lead to a concentration of the best players in the
richest teams. In other words, a reserve clause was not a necessary condition to
ensure competitive balance. Professional team sports are intrinsically different
from other businesses, in which a firm is likely to prosper if it can eliminate com-
petition and establish a position as a monopoly supplier. In sports, it does not
pay a rich team to accumulate star players to the extent that (sporting) competi-
tion is greatly diminished, because of the joint nature of ‘production’ in sports.
Consequently, a team that attempts to accumulate all of the best available playing
talent will find at some stage that diminishing returns begin to set in.
In baseball no team can be successful unless its competitors also survive and prosper sufficiently
so that the differences in the quality of play among teams are not ‘too great.’ … At some point,
therefore, a first star player is worth more to poor team B than, say, a third star to rich team A.
At this point, B is in a position to bid players away from A in the market. (Rottenberg, 1956,
p254, 255)
If teams are rational profit maximisers, the distribution of playing talent among
the teams should be more or less equal. Neither a reserve clause nor explicit collu-
sion is necessary in order to bring about this result. It is in each team’s self-interest
to ensure that it does not become too strong relative to its competitors.
It follows that players will be distributed among teams so that they are put to their most ‘pro-
ductive’ use; each will play for the team that is able to get the highest return from his ser-
vices. But this is exactly the result which would be yielded by a free market. (Rottenberg, 1956,
p256)
A reserve clause will therefore deliver almost the same distribution of play-
ing talent between the competing teams as free agency. Whether players are free
agents or not, the distribution of playing talent is determined by the incentive to
maximise the capitalised value of the services supplied by individual players. If
there is another team for which this capitalised value would be higher than it is for
the player’s present team, then there is a price at which it is advantageous for both
teams to trade the player’s contract.
Rottenberg also discusses the implications of the reserve clause, and the mon-
opsony power it confers on teams as buyers of playing services, for players’ salar-
ies. Each player’s reservation wage (the minimum salary he would accept to play
baseball) is determined primarily by the next highest salary he could earn outside
baseball adjusted to reflect his valuation of the non-pecuniary costs and benefits
of playing baseball. Although theoretically the team has the contractual power to
impose the reservation wage on all players, Rottenberg notes that in practice this
does not seem to happen. Many players earn far more from baseball than they
could in alternative employment. This is attributed to the fact that players as well as
teams have bargaining power in salary negotiations: in an extreme case, a player can
4 Introduction
simply threaten to withdraw his services. If a player’s reservation wage is $10,000,

but he is worth $20,000 to his team, then a salary anywhere between $10,000 and
$20,000 is possible, depending on the ‘shrewdness and guile of the parties in devising
their bargaining strategies’ (Rottenberg, 1956, p253). Competition among sellers,
however, imposes limitations on players’ bargaining power. A star player worth
$40,000 to his team cannot extract a salary beyond $30,000 if a lesser player worth
$20,000 is willing to accept a salary of $10,000 to fill the same position.
The main effect of the reserve clause is that players receive salaries below their
value to the team that employs them. In other words, it tends to direct rents away
from players and towards teams. The reserve clause does not achieve its stated
aim of influencing the allocation of playing talent between teams. Rottenberg
concludes by considering several alternative regimes that might produce a more
or less equal distribution of playing talent between teams with free agency in the
players’ labour market. These include:
• Revenue sharing. If all revenues are shared equally, teams have no pecuniary
incentive to spend on players to enhance their own performance and revenue.
An equal distribution of mediocre playing talent is the most likely outcome.
• Imposition of a maximum salary. The effect on the distribution of talent depends
on the ability of the teams to circumvent the maximum by offering players non-
pecuniary rewards.
• Allocating multiple team franchises in large cities. If reasonable equality between
each team’s potential market size can be achieved, this is expected to create a
more equal distribution of playing talent.
Neale: ‘The Peculiar Economics of Professional Sports’, Quarterly Journal of
Economics, 1964
Neale’s analysis begins by emphasising the joint nature of production in pro-
fessional sports. Heavyweight boxing is used as an example to introduce what
Neale calls the ‘Louis Schmelling Paradox’. World champion Joe Louis’ earnings
were higher if there was an evenly matched contender available for him to fight
than if the nearest contender was relatively weak. The same point applies also in
baseball.
Suppose the Yankees used their wealth to buy up not only all the good players but also all of the
teams in the American League: no games, no receipts, no Yankees. When, for a brief period in
the late fifties, the Yankees lost the championship and opened the possibility of a non-Yankee
World Series they found themselves – anomalously – facing sporting disgrace and bigger crowds.
(Neale, 1964, p2)
Does this imply that professional sport is an industry in which monopoly is less
profitable than competition, contradicting what is taught to students and what
can be read in any Principles text? Neale addresses this paradox by distinguish-
ing between sporting and economic competition. Sporting competition is more
profitable than sporting monopoly for the reasons outlined above, but sporting
competition is not the same as economic competition. Similarly, although in law

the sports team is a firm (which may be motivated by profit), it is not a firm in
the economist’s sense. A single team cannot supply the entire market; if it did, it
would have no opposition to play against. Teams must cooperate with each other
to produce individual matches and a viable league competition, so there is joint
production. The league’s organising body exerts strict controls over a wide range
of matters including competition rules and schedules, player mobility and the
entry and exit of clubs. In short, the league rather than the individual team is the
‘firm’ in the economist’s sense. A sports league should be regarded as analogous to
a multi-plant firm, in which the individual teams are ‘plants’, subject to decisions
which are taken and implemented collectively at league level.
If the sports league is the ‘firm’ in the economic sense, this raises the question
as to why it is unusual to observe direct competition between rival leagues operat-
ing within the same sport. Although the National League and American League
do operate simultaneously in baseball, analytically they should be regarded as
one larger ‘multi-league’ firm, since they come together at the end of each season
to produce the World Series. Geographical division is a more common form of
segmentation, though according to Neale, one that is inherently unstable. Where
direct competition is precluded by geography, profit incentives tend to pro-
mote enlargement and the elimination of geographical boundaries. Competition
between different sports is more common than competition between rival leagues
within the same sport, though segmentation based on nation, region, season of
the year or even social class is still common.2 All such forms of segmentation tend
to inhibit direct head-to-head competition.
Neale suggests that the general lack of competition between sports leagues
arises because the cost and demand characteristics of the market for professional
team sports tend to create conditions of natural monopoly, making it efficient for
a single league to supply the entire market. On the cost side, Neale suggests that
the long-run average cost curve is probably horizontal. Although an increase in
the scale of production might entail the use of less efficient playing inputs, raising
average costs, this tends to be offset by an ‘enthusiasm effect’. If the sport operates
on a larger scale, public enthusiasm encourages more people to take up playing,
eventually raising the supply of players at the highest level. To some extent, the
enthusiasm effect makes supply and demand interdependent: if more people play
the sport, more will also want to attend matches at the professional level. Finally,
the existence of rival leagues would effectively break the monopsony power of
teams as buyers of playing services, enhancing players’ bargaining power in salary
negotiations. This would tend to make costs higher than they are when one league
operates as a monopoly supplier.
On the demand side, Neale suggests that baseball teams produce a number of
streams of utility: directly for spectators who buy tickets for seats in the stadium
and for television viewers who watch the match at home; and indirectly for any-
one who enjoys following the championship race as it unfolds. The closer the
6 Introduction
competition, the greater the indirect effect. For newspapers and television com-
panies in particular, the indirect effect is a marketable commodity that helps sell
more of their product. The size of the indirect effect depends on the scale and
universality of the championship, and is therefore maximised when the league is a
monopoly supplier. Overall, ‘it is clear that professional sports are a natural mon-
opoly, marked by definite peculiarities both in the structure and in the functioning
of their markets’ (Neale, 1964, p14).
An important implication is that the peculiar economic characteristics of pro-
fessional sports leagues and their constituent teams should be recognised by
legislatures, by the courts and by the general public, whenever practices such as
collective decision-making or other (apparently) anticompetitive types of behav-
iour come under scrutiny.
Sloane: ‘The Economics of Professional Football: The Football Club as a Utility
Maximiser’, Scottish Journal of Political Economy, 1971
Sloane’s paper questions Neale’s conclusion that the league rather than the indi-
vidual team or club is the relevant ‘firm’ (or decision-making unit) in professional
team sports. In the case of English football, for example, the sport’s governing
bodies merely set the rules within which clubs can freely operate. Most economic
decisions, such as how much money to spend on stadium development and how
many players to employ, are made by the clubs. Although the total quantity of
‘output’ (the number of matches played by each team) is regulated, this clearly
reflects the clubs’ common interest. In cartels, it is not unusual for firms to reach
joint decisions concerning price or production, but this does not imply that the
cartel should be elevated to the theoretical status of a ‘firm’. In short, Sloane
suggests that Neale’s argument tends to overemphasise mutual interdependence.
‘The fact that clubs together produce a joint product is neither a necessary nor a
sufficient condition for analysing the industry as though the league was a firm’
(Sloane, 1971, p128).
Having argued that the club is the relevant economic decision-maker, Sloane
goes on to raise a number of key questions concerning the objectives of sports
clubs. Implicit in the reasoning of both Rottenberg and Neale is an assumption
of profit maximisation. Despite the ‘peculiarities’ of sports economics elucidated
by Neale, the behaviour of professional sports teams is analysed within a very
conventional analytical framework. While this may be reasonable in the case of
US professional team sports, where many teams do have an established track
record of profitability, Sloane suggests that it may not be universally applicable.
Throughout the history of English football, profit-making clubs have been the
exception and not the rule. Most chairmen and directors of football clubs are
individuals who have achieved success in business in other fields. Their motives
for investing may include a desire for power or prestige, or simple sporting enthu-
siasm: a wish to see the local club succeed on the field of play. In many cases,
profit or pecuniary gain seems unlikely to be a significant motivating factor. If
so, it may be sensible to view the objective of the football club as one of utility
maximisation subject to a financial solvency constraint. The financial solvency

constraint recognises that the benevolence of any chairman or director must
reach its limit at some point.
Non-profit-maximising models of the firm had received considerable attention
in the economics literature during the decade prior to the publication of Sloane’s
article.
A major drawback to the general introduction of the utility maximisation assumption in the
theory of the firm is that it may be rationalised so that it is consistent with almost any type of
behaviour and therefore tends to lack operational significance. (Sloane, 1971, p133)
In the case of football clubs, however, it is not too difficult to identify several
plausible and easily quantifiable objectives. Sloane suggests the following:
• Profit. The expectation that profit is not the sole or even the most important
objective does not preclude its inclusion as one of several arguments in the util-
ity function.
• Security. Simple survival may be a major objective for many clubs. Decisions
(concerning, for example, sales of players) may aim more at ensuring security
than at maximising playing success.
• Attendance or revenue. A capacity crowd enhances atmosphere and a sense of
occasion, and may in itself be seen as a measure of success. Recently, an increas-
ing willingness to charge whatever ticket prices the market will bear suggests
that revenue (or profit) carry a heavier weight in the utility function than in
earlier periods, when it was usual to charge the same price for all matches, irre-
spective of the level of demand.
• Playing success. This is probably the most important objective of all, and one to
which chairmen, directors, managers, players and spectators can all subscribe.
• Health of the league. This enters the utility function in recognition of clubs’
mutual interdependence.
Formally, the club’s objective is to maximise:
U = u( P,A,X,π R − π 0 − T ) subject to π R ≥ π 0 + T [1.1]
where P = playing success; A = average attendance; X = health of the league; πR =
recorded profit; π0 = minimum acceptable after-tax profit; and T = taxes.
It is important to note that the utility maximisation model has implications that
are very different to those that follow from the profit-maximising assumptions of
Rottenberg and Neale. In particular, if the weighting of P in the utility function is
heavy relative to that of X and πR, the argument that diminishing returns would
prevent the accumulation of playing talent in the hands of a small number of rich
clubs does not necessarily hold, unless there are binding financial constraints pre-
venting expenditure on new players. The notion that profit incentives should help
maintain a reasonably even allocation of playing talent between richer and poorer
teams breaks down. The case for regulation to override the ‘free market’ outcome,
whether in the form of a reserve clause, revenue sharing or the taxation of transfer
8 Introduction
fees, therefore seems to be enhanced if clubs are pursuing non-profit rather than
profit objectives.
1.2 Outline of this volume

Since the appearance of the pioneering work of Rottenberg, Neale and Sloane,
there has been a proliferation of published academic research on the economics
of team sports, in journal articles and in books. Many of these contributions are
reviewed in this volume. By way of an introduction, this section contains a brief
outline of the contents of each of the following chapters.
Rottenberg’s argument – that market mechanisms can be relied upon to main-
tain a reasonable degree of competitive equality among the member teams of a
sports league – provides the motivation for a number of theoretical contribu-
tions to the sports economics literature, examining the implications for resource
allocation within a professional sports league of several regulatory mechanisms,
including the reserve clause, payroll caps, the reverse-order-of-finish draft, rev-
enue sharing, and rules governing the award of franchises and league member-
ship. According to the theoretical analysis, most of these devices promote the
survival of small-market teams; but in many cases they fail to reduce competitive
inequality because they do not create the necessary profit incentives. Chapter 2
reviews some of the key findings from this literature.
The theoretical models of the economics of sports leagues are concerned with
the degree of competitive balance or competitive inequality between the member
teams of a league competition, which in turn determines the extent of uncertainty
of outcome for individual match results and for the destination of the league cham-
pionship. In Chapter 3, the emphasis shifts away from this theoretical analysis at
the level of the league championship, towards the measurement of competitive
inequality and uncertainty of outcome for individual match results. Certain other
empirical patterns and regularities in football match results data are investigated,
including the phenomenon of home-field advantage, and ‘streaks’ or persistence
effects in sequences of consecutive match results.
Uncertainty of outcome seems to be an essential ingredient for any competitive
sport. If the match result were completely deterministic, what would be the point
for either competitors or spectators? The observation that the result of any indi-
vidual match is uncertain, however, does not immediately consign any forecasting
exercise to irrelevance. If prediction is interpreted as assessment of the probabil-
ities for home-win, draw and away-win outcomes, the results of individual football
matches are highly predictable, as evidenced by the wide variation from one match
to the next in the betting odds quoted by bookmakers and online betting firms for
these outcomes. Chapter 4 examines the specification and performance of econo-
metric forecasting models for match results in football, which generate probabil-
istic forecasts of the results of forthcoming matches, by number-crunching large
volumes of past match results data and other relevant information.
Outline of this volume 9
Game theory is defined as the study of decision-making in situations of conflict

and interdependence. Some economists have argued that sports such as football
offer promising opportunities for the investigation of propositions of game theory,
concerning the formulation by the players of optimising strategies. For example,
the high-pressure ‘game’ played out between the kicker and the goalkeeper when-
ever a penalty kick is awarded mirrors closely the textbook conditions for a two-
person non-cooperative zero-sum simultaneous game. The kicker decides in which
direction to shoot and the goalkeeper decides simultaneously in which direction
to dive; and either the kicker scores, or a goal is averted because the goalkeeper
saves or the kicker misses the target. Viewed more broadly, the football match in
its entirety has many of the characteristics of a strategic and dynamic ‘game’. The
two teams are pitched into direct opposition, and the team managers select and
adjust continuously their styles of play as the match evolves. Chapter 5 reviews the
theoretical and empirical literature, and presents some new empirical evidence on
applications of game theory in football.
The sporting attributes of professional football and its characteristics as a busi-
ness have always been closely connected. Chapter 6 presents an overview of the
historical development of English club football as a business, and an analysis of its
contemporary economic, financial and commercial structure. The analysis covers
English club football’s competitive structure, trends in spectator demand reflected
in match attendances and gate revenues, the market for TV broadcast rights, the
development of the labour market for playing talent, and the ownership, govern-
ance and financing of English football clubs.
The level of the top professional footballers’ remuneration has raised concerns
over the effect on the competitive structure and finances of football, and over
the morality of such disproportionately high financial rewards accruing to foot-
ballers relative to other occupations whose contribution to society’s well-being
is, arguably, much greater. Chapter 7 examines theoretical explanations that have
been put forward for the exceptionally high remuneration of the leading stars in
modern-day professional football, and other professional sports. The theoretical
analysis of the economics of superstars notes that, especially since the arrival of
pay-TV, sports stars are capable of servicing very large paying audiences simultan-
eously, incurring little or no incremental cost as the audience size increases. Other
occupational groups, such as nurses and teachers, service strictly finite number of
users. The rank-order tournament model views the remuneration of the highest-
paid stars as a means of providing all professionals with incentives to invest in the
development of their skills, in the hope of reaching the top of their profession and
being rewarded accordingly.
Chapter 8 discusses several other topics concerned with the economics of the
professional footballers’ labour market. Patterns of migration, employment mobil-
ity and career development among the professional footballers employed by the
ninety-two member-clubs of the Premier League and Football League are exam-
ined, including the impact of the arrival of large numbers of overseas players on
10 Introduction
the career prospects of locally born footballers. The factors influencing patterns of
international migration by footballers are examined. Empirical evidence on racial
discrimination in English football suggests that a form of hiring discrimination
affects the opportunities for indigenous black players to progress to professional
status. It seems likely, however, that any such effect has diminished over time.
The job description of the football team manager includes the selection, super-
vision and coaching of players, and the formulation of tactics and strategies. Many
football managers, especially in professional football’s lower tiers, are also respon-
sible for the buying and selling of players, wage negotiations, and various admin-
istrative duties. Chapter 9 examines various aspects of the role and contribution
to team performance of the football manager. In principle, the football manager’s
contribution subdivides into a direct and an indirect component. The direct contri-
bution is to maximise performance through astute team selection, superior tactics
and powers of motivation; while the indirect contribution is to coach players so as
to enhance their skills, and to strengthen the team through effective dealings in the
transfer market. The indivisibility of the team effort which ultimately determines
performance poses a major challenge for any researcher seeking to isolate and
measure the manager’s contribution. The modern-day football manager’s position
is renowned for its chronic insecurity, which raises several interesting questions for
researchers concerning the relationship between managerial turnover and team
performance.
While it is the football manager who usually pays the ultimate price for perceived
underachievement on the part of the players under his direction, the football ref-
eree often serves as a convenient scapegoat during the immediate aftermath of a
poor result or performance. Football referees, who are the subject of Chapter 10,
are routinely criticised by managers, players, journalists and spectators for being
incompetent, inconsistent and biased. Several recent academic studies have exam-
ined whether there is any substance to claims of favouritism and bias on the part
of football referees. Due to technological advances in broadcasting, the actions of
referees have never been more intensely scrutinised than they are today. Split-second
decisions taken by referees can have enormous financial consequences, due to the
fine line between spectacular success and catastrophic failure that exists in football.
Social history and sociology provide many useful insights into the causes of fluc-
tuating football attendances. Econometric modelling of variations in the attend-
ances of individual clubs, both season-by-season and match-by-match, has been
a subject of attention for sports economists since the 1970s. Chapter 11 provides
a non-technical review of the empirical literature, focusing on the issues of vari-
able definition, model specification, estimation and interpretation that are faced
by researchers in this area, and presents an analysis of the variation in the average
attendances per season of English football clubs during the post-Second World
War period.
Research concerning the relationship between the market prices for bets on
the outcomes of sporting contests, and the probabilities associated with these
Outline of this volume 11
outcomes, forms a subfield within the literature on financial market efficiency.

Chapter 12 examines the economics of sports betting, with particular emphasis
on the markets for betting on the outcomes of football matches. The growth of
internet betting during the 1990s and 2000s has presented tremendous new oppor-
tunities for betting firms and their customers, based on the rapid information pro-
cessing and transmission capabilities of internet technology.
Much of the descriptive and empirical analysis of the football industry that is
presented in this volume is focused on English club football. Chapter 13 widens the
perspective, by examining the historical development and present-day competitive
and commercial structure of football in five other countries: France, Germany,
Brazil, Japan and China. France and Germany both fail to match England in
terms of the income that is generated by professional football at the highest level.
Over several decades, however, both countries have outshone England in the inter-
national football arena. Brazil, also renowned for the excellence of its national
football team, has traditionally exported many of its star footballers to wealthy
clubs in western Europe, despite a strong indigenous football culture. In Japan and
China, by contrast, there is little or no grassroots football tradition; and recent
attempts to establish commercially vibrant professional football leagues in these
countries have been driven from the top down.
In the arena of international football, countries compete fiercely to host the
World Cup, as well as other mega sporting events. Many World Cup bids in recent
times have attempted to muster public support through extravagant claims con-
cerning the economic benefits that will flow from a successful bid. Academic econ-
omists, however, tend to view such claims with extreme caution. Some economists
believe that the net economic impact is either negligible, or perhaps even negative.
The final chapter of this volume, Chapter 14, examines the economic evidence as
to whether hosting a World Cup can be expected to yield economic benefits to the
host nation, either tangible or intangible, that are capable of justifying the massive
financial outlays.
Notes
1 This was subject only to a rule that prevented the salary from being cut by more than 25 per
cent in any one year.
2 Traditionally, English cricket and rugby union were upper- or middle-class sports, while
football and rugby league were working-class sports. More recently, class divisions may
have become more blurred, but they have not disappeared altogether. Cricket is played only
in summer; rugby league switched from a winter to a summer schedule in the late 1990s;
football and rugby union are played in winter. Rugby league still has a strong regional iden-
tity centred on Lancashire and Yorkshire. Cricket is popular throughout England, but is a
minority sport in Scotland.
2 The economic theory of
professional sports leagues
Introduction
As seen in Chapter 1, the idea that market mechanisms can be relied upon to
maintain a reasonable degree of competitive equality among the member teams
of a sports league, without the need for extensive regulation of player compensa-
tion or mobility by the sport’s governing body, was first articulated by Rottenberg
(1956). In discussing the economic structure and characteristics of the North
American baseball players’ labour market, Rottenberg considers the case of two
teams located in different towns, one of which has a larger population (or poten-
tial market) than the other. Other things being equal the marginal revenue prod-
uct of a player of a given level of ability is greater with the large-market team than
it is with the small-market team. Since the marginal revenue function declines as
the quantity of playing talent already held increases, however, it does not pay the
large-market team to accumulate the most talented players to the point where
complete competitive dominance is achieved.
This argument does not depend upon contractual arrangements or the struc-
ture of player compensation. It is valid under free agency, in which case the player
may be in a strong position to secure most or all of his marginal revenue product
in salary negotiations, since his reservation wage is the salary he could command
by signing for another team. It is also valid under a reserve clause, in which case
teams may have the opportunity to drive player compensation down, towards the
highest salary the player could command in employment outside the sport.
Subsequently, El-Hodiri and Quirk (1971) developed a mathematical model of
an n-team professional sports league that captures these and other insights. They
demonstrate that perfect competitive balance, with all teams having equal playing
strengths, is consistent with an assumption of profit-maximising behaviour only
if there is no buying and selling of players’ contracts, or if the revenue functions
of all teams are the same. In the type of model presented below, the latter would
require that all teams are located in towns of equal (or similar) population size, so
that each team’s potential market is the same. Since neither of these requirements
is likely to be met in practice, ‘(i)t is not surprising, then, that casual empiricism
12
Introduction 13
indicates that in no professional team sport under current rules of operation is

there a tendency toward equal playing strengths’ (El-Hodiri and Quirk, 1971,
p1313).
Variants of the El-Hodiri and Quirk (1971) model have been used to inves-
tigate a range of policy issues. Key results from this literature are presented by
Fort and Quirk (1995) and Vrooman (1995, 2007, 2009), who examine the eco-
nomic and incentive effects of a number of regulatory mechanisms, including the
reserve clause, payroll caps, the reverse-order-of-finish draft, revenue sharing, and
rules governing the award of franchises and league membership. According to the
theoretical analysis, most of these devices promote the survival of small-market
teams; but in many cases they fail to reduce competitive inequality, because they
do not create profit incentives for team owners to adjust in the direction of greater
competitive equality. Some of the devices may even create perverse incentives and
tendencies towards greater competitive inequality. Késenne (2007b) provides a
comprehensive survey of this literature.
Chapter 2 reviews the main findings of this body of work, using a modified
version of the models described by Fort and Quirk (1995), Vrooman (1995) and
others. Most of the earliest studies that are reviewed above were developed with
North American major league sports in mind. Section 2.1 describes some of the
key characteristics of the North American model for the organisation of profes-
sional team sports leagues.
One key feature of the North American model is that the labour market for
playing talent is far less open than in the case of European football. Accordingly,
much of the US literature on the economics of professional team sports leagues
has considered models in which the total stock of playing talent is assumed to
be fixed, and any reallocation of talent between the league member teams is a
zero-sum activity. This ‘closed’ labour market assumption is reasonable for North
American sports, which draw principally on home-grown talent and do not fre-
quently trade players with teams in other countries. A closed model is less appro-
priate for football, especially following the liberalisation of the rules governing
the European football players’ labour market in the mid-1990s, since when the
leading clubs in particular have recruited large numbers of foreign players. In the
case of football, therefore, an ‘open’ labour market assumption is appropriate.
In an open model, the total stock of playing talent at the disposal of the league
member teams is variable, and can be increased by hiring from an external market
in playing talent.
The next two sections of this chapter develop a theoretical model of resource
allocation in a professional team sports league, under both closed and open labour
market conditions. The implications for resource allocation and competitive
inequality of profit-maximising behaviour and win-maximising behaviour on the
part of team owners are considered. Section 2.2 examines the general case of an
n-team league, and Section 2.3 examines the stylised case of a two-team league.
For the n-team model, the explicit solutions for the amounts of talent hired by the
14 The economic theory of professional sports leagues
teams are algebraically complex. In the sports economics literature, it is common

practice to obtain insights by considering the special case n = 2.
The final two sections use the theoretical model to examine the effectiveness
of rules that might be implemented by the sport’s governing body in an attempt
to reduce competitive inequality between large-market and small-market teams,
focusing specifically on revenue sharing and various forms of payroll cap. Do
such arrangements help reduce competitive inequality, and perhaps make the
league a more exciting and watchable proposition for spectators? Alternatively,
might their effects be either neutral, or even counterproductive in the sense that
they help entrench competitive inequalities that already exist? Revenue-sharing
arrangements are discussed in Section 2.4; and payroll caps are examined in
Section 2.5.
2.1 The North American professional team sports model

The theoretical literature on the determinants of the degree of competitive
inequality in sports leagues was developed by US sports economists, with North
American team sports primarily in mind. Naturally the development of this litera-
ture reflects the characteristics of the North American model for the organisation
and regulation of team sports. This differs from the European model in several
key aspects. In reviewing this literature, therefore, an awareness of the key fea-
tures of the North American model is useful. This subsection outlines the organ-
isational structure of three North American professional team sports: baseball,
basketball and (American) football.1 Comparisons between the North American
and European models for the organisation of sports leagues are drawn by Noll
(2002, 2003) and Szymanski (2003).
Perhaps the most fundamental difference between the North American and
European models concerns the highly restricted nature of supply in North America.
In all three sports leagues membership is controlled through the award of fran-
chises by the leagues’ organising bodies. A franchise grants a territorial monopoly
to the team’s owner: no other team can operate within the owner’s territory, and
no owner can relocate to another town without the approval of 75 per cent of
all team owners. In Major League Baseball (MLB), sixteen teams compete in the
National League (NL) and fourteen in the American League (AL). In (American)
football thirty-two teams make up the National Football League (NFL). In bas-
ketball there are thirty teams in the National Basketball Association (NBA). In
each case, league competition takes place within a number of regionalised sections
(divisions), and end-of-season play-offs determine championship outcomes.
To appreciate the highly restrictive nature of the franchise system, it is
worth noting that the mid-2009 estimate of the US population was approxi-
mately 307 million. The top divisions of the big five European football
leagues each support 18 or 20 teams, with estimated national populations of
approximately 53 million (England and Wales), 65 million (France), 82 million
The North American professional team sports model 15
(Germany), 60 million (Italy) and 46 million (Spain). Even more important,
league membership in North America is determined exclusively through the
franchise system; there is no hierarchical divisional structure and no promo-
tion and relegation. Membership of the top tiers of European football is open
to a much larger number of teams operating in the lower tiers of the respective
leagues. Promotion and relegation between the tiers depends solely on competi-
tive prowess.
Profit maximisation is the prime objective of North American leagues and team
owners, so profitability is the main factor influencing decisions concerning the
award of franchises and relocation. There is also intense competition between
municipalities to attract and retain franchise holders. Inducements take the form
of offers to construct lavish new stadiums at public expense: a significant source of
public subsidy for leagues and teams that are already rich in cash (see Chapter 14).
The leagues have been able to maintain their monopoly positions largely because
the US courts, acknowledging the peculiarities of the economics of team sports
(see Chapter 1), have tended to accept the legitimacy of restrictions on the produc-
tion of sporting events that would be inadmissible elsewhere. In a landmark ruling
in 1922, the Supreme Court granted MLB exemption from the Sherman antitrust
laws. In 1961 the Sports Broadcasting Act was passed by Congress, entitling the
leagues to sell broadcasting rights collectively on behalf of their member teams
(Quirk and Fort, 1992; Scully, 1995).
The main role of the leagues within this framework is to implement rules aimed
at furthering the collective interest of the teams in achieving joint profit maximisa-
tion. The leagues therefore impose restrictions on the behaviour of the teams in both
the product and labour markets, which seek to prevent any individual team from
achieving a level of competitive dominance that would be damaging to the interest
that all teams share in maintaining a reasonable degree of competitive balance. In
the early history of all three sports, the most important restriction of this kind was
the reserve clause, described previously in Chapter 1. Having signed a contract as a
professional, the player’s team retained the option to renew his contract, effectively
binding the player to his present employer. The reserve clause was justified on the
grounds that it was necessary to prevent the richest teams from outbidding the rest
for the services of the top players; but its tendency to depress players’ compensation
was a highly convenient side-effect from the team owners’ perspective.
Baseball’s reserve clause, effective since 1880, was challenged successfully in the
Supreme Court in 1976, on the grounds that it should be interpreted only as a
one-year option clause. The Court ruled that by playing for a year without signing
a new contract, a player could satisfy his obligations under the reserve clause, and
subsequently become a free agent. Under the 1976 settlement, all MLB players
became free agents after completing a minimum number of years of major league
service. At the time of writing, players qualify as free agents after six years.
In basketball, the one-year interpretation had always applied, but teams that
signed a free agent had to pay compensation to his former employer. This system
was also abandoned in 1976, when moves towards the full implementation of free
agency were initiated. Players became free agents after four years’ NBA service or
upon expiry of their second contract if this happened sooner. At the time of writ-
ing, a player becomes a free agent at the end of his first contract (often four years),
although the first team has the right to retain the player if it matches the highest
offer received from another team.
In (American) football, free agency was not introduced until 1993, when a
one-year option clause (similar to MLB) was abandoned. The delay is attributed
by Quirk and Fort (1999) to the relatively weak bargaining position of the play-
ers’ union, explained in turn by the short average duration of playing careers
and the dominant role played in the sport by coaches (who tend to be even more
influential than star players). From 1993 NFL players became free agents after a
five-year qualifying period. At the time of writing, the qualifying period for unre-
stricted free agency is four years; players also qualify for restricted free agency
rights after three years.
In all three sports, trades involving in-contract players normally take the form
of player-exchange deals, or swaps of players for draft picks (see below). Trading
players for cash is unusual, though not completely unknown, and there is no North
American equivalent of European football’s multi-million pound transfer system.
Other restrictions on the free play of market forces in North American sports
include the draft system, payroll caps, luxury taxes and revenue sharing. The
reverse-order-of-finish draft system allows the weakest teams from the previ-
ous season the first pick of rookie players moving from college (or school in the
case of baseball) to professional level for the first time. The NFL introduced
the draft in 1936; the NBA followed suit soon after its formation in 1949; and
MLB adopted the draft in 1965 (though the draft does not apply to new players
arriving from overseas). Drafts are permissible under US antitrust law because
they are included in collective bargaining agreements between leagues and player
unions. These agreements also require minimum salaries for newly drafted play-
ers. The detail of how the draft works varies among the sports. There are seven
rounds of the draft in the NFL (each team has seven selections). In MLB there
are two drafts per year, in June and December.
In the 1984–5 season the NBA became the first organising body to operate a
payroll cap, in an attempt to offset the potentially damaging implications of free
agency for competitive inequality. The total salary expenditure of each team was
subject to a limit equivalent to 53 per cent of gross revenues divided by the number
of teams in the league. In the 2005 collective bargaining agreement, the cap was
set at 51 per cent of gross league revenues for seasons 2006–7 to 2011–12. A min-
imum level of salary expenditure is also defined, at 75 per cent of the salary cap.
Any team failing to achieve the minimum level of expenditure is surcharged at the
end of the season. The NBA cap is ‘soft’, because there are exceptions allowing
teams to exceed the cap in order to retain players whose contract renewal would
otherwise take the team’s salary expenditure above the cap limit.
The North American professional team sports model 17
In theory, by equalising expenditure on players’ compensation throughout

the league, the payroll cap should enable the small-market teams to compete
for championship success on equal terms with the large-market teams. In prac-
tice, however, the NBA salary cap has not succeeded in equalising expenditures,
mainly because the exemption clause is invoked routinely by the large-market
teams. Consequently total salaries regularly exceed 60 per cent of revenues, and
the high-spending teams’ salary bills are typically more than twice those of the
low-spending teams. Championship success in the NBA became more, rather than
less, concentrated following the introduction of the cap (Quirk and Fort, 1999).
However, competitive inequality in the NBA has been reduced since the 2002–3
season, due in part to the evolution of the luxury tax first introduced in 1999 (see
below) into a hard cap after 2002–3.
In 1994 the NFL adopted a ‘hard’ payroll cap, at 64 per cent of defined gross
revenues. The NFL cap cannot be exceeded in the long run. While pro-rated
upfront bonus payments can be made over the length of a player’s contract, the
actual payroll in future years must be reduced to account for the forward bonus.
In the 2006 collective bargaining agreement, the cap was set at 57.5 per cent of
total revenues for the 2008–9 season, and 58 per cent subsequently. Although the
percentage is lower than previously, a change in the definition of gross revenue
(to include revenue from the sale of luxury seats) has made the size of the revenue
pool larger. There is also a minimum payroll requirement, set in the range of
84–90 per cent of the cap.
An attempt to introduce a payroll cap in MLB resulted in a players’ strike in 1994
and 1995, and was eventually abandoned. A luxury tax, rather than a payroll cap,
has been part of each subsequent collective bargaining agreement. The luxury tax is
a tax on salary expenditure exceeding a set threshold. Team payrolls are permitted
to exceed the threshold, but with a liability to pay tax on the excess expenditure. The
tax starts at 22.5 per cent, increasing to 30 per cent and 40 per cent on the second and
third occasions the threshold is breached. During the 2000s, however, the thresh-
old has been set sufficiently high that only the largest-market team, the New York
Yankees, has been taxed regularly and in large amounts. According to figures quoted
by Vrooman (2009), the Yankees contributed $121.6 billion of the total luxury tax
revenue of $136.4 billion paid by all MLB teams between 2003 and 2007.
The NBA luxury tax, which was introduced in 1999, is more onerous than the
MLB version. Teams whose payroll exceeds a predetermined level pay one dollar
for each dollar their payroll exceeds the tax level. The tax level is computed by tak-
ing 61 per cent of projected basketball-related income, subtracting projected ben-
efits, and adjusting for whether the previous season’s basketball-related income
was above or below the projection. On several occasions following its introduction
the tax was not triggered; but the 2005 collective bargaining agreement guarantees
that the tax is effective in every season. Five teams paid $55 million in 2006–7;
eight teams paid $92.4 million in 2007–8; and seven teams paid $87.3 million in
2008–9 (Coon, 2010).
In all three sports revenue sharing plays an important part in offsetting inequal-
ities in drawing power between teams. Broadcast rights are sold in separate pack-
ages to national and local broadcasters. Practices concerning the distribution of
gate revenues vary between sports. Home and away teams in the NFL share gate
revenues on a 66:34 split. Other venue revenues are not shared, while local broad-
cast revenue is small, and is not shared. In MLB, 31 per cent of local revenues
(including gate, venue and broadcast revenues) are shared. In the NBA, local rev-
enues are not shared.
As a general principle, national broadcast revenues are shared equally between
all major league member teams. Since the mid-1990s, there has been exponen-
tial growth in the value of rights, and a steady migration of the rights away from
the traditional free-to-air broadcasters ABC, CBS and NBC. The beneficiaries
include Fox, the group owned by Rupert Murdoch’s News Corporation, and sev-
eral satellite and cable pay-TV broadcasters. The NFL deal for 2006–11, valued at
$22.1 billion, assigns the primary role to ESPN (a cable broadcaster affiliated with
ABC) and DirecTV (satellite). The MLB deal for 2007–13, valued at $5.6 billion,
divides the broadcast rights between Fox, ESPN (cable), TBS (cable, owned by
Time Warner) and DirecTV (satellite). Since 2002, the broadcast rights for NBA
have been held by the cable providers ESPN and TBS. The 2008–16 deal is valued
at $7.4 billion.
With the leagues maintaining a stranglehold over the supply of professional
team sports, it is unsurprising that the histories of the three sports are littered
with attempts on the part of outsiders to set up rival competitions. Following
the creation of the NL in 1876, there were several attempts to establish rival
leagues, including the AL which was set up in 1901. The current NL–AL mon-
opoly, achieved by agreement between the two leagues in 1903, has survived sub-
sequent challenges from the Federal League (1914), the Mexican League (in the
1940s), and a threatened Pacific Coast League (which was averted in 1958, partly
as a result of the moves of the Brooklyn Dodgers and New York Giants to Los
Angeles and San Francisco, respectively).
In basketball the NBA was created in 1949 from the remains of two rival leagues.
Its most powerful challenger since has been the American Basketball Association
(ABA), set up in 1967. Mounting financial difficulties eventually led to the demise
of the ABA, with four of its surviving teams being incorporated into the NBA in
1976. Since the creation of the American Professional Football Association (the
forerunner of the NFL) in 1920, there have been several attempts to establish rival
leagues in (American) football; the most successful being the fourth American
Football League (AFL), set up after the NFL had refused to award a number
of expansion franchises in the late 1950s. The AFL operated alongside the NFL
between 1960 and 1969, before a merger incorporating the AFL teams into the
NFL realised the original aims expressed by the AFL’s founders ten years earlier.
Although there is a prolific and colourful history of attempts to create rival
leagues, only the AL has endured for more than a few years, and then only by
A model of an n-team league 19
operating together with the NL as a de facto joint monopoly supplier. In some

instances, however, rival league proposals (whether implemented or only threat-
ened) have been successful in forcing the established leagues to make policy
changes. Clearly, there is a strong imperative in any sport to operate a competi-
tive structure which delivers a single champion. But while the established leagues
remain effectively closed to new entrants, excluding a number of cities capable
of supporting potentially valuable franchises, the incentives that have previously
brought so many rival league proposals to (temporary) fruition will undoubtedly
remain in place.
2.2 A model of an n-team league

This section develops a theoretical model of resource allocation in a professional
team sports league, for the general case where the number of competing teams,
denoted n, is unspecified. In the model, it is assumed that the member teams of a
sports league hire units of playing talent that are embodied in players, and that
each player’s remuneration is proportional to the number of units of talent he (or
she) embodies. In principle there is no limit to the quantity of talent that any team
is able to deploy. While football teams are restricted to deploying only eleven
players at any time, they can always deploy more talent by hiring more highly tal-
ented players. Accordingly, the model is developed by examining the team own-
er’s decision on the number of units of talent hired and deployed.
The following notation is used to develop the model:
ti = talent hired and deployed by team i
T = total quantity of talent hired by all league member teams
n = number of league member teams
t 8 = T/n = average quantity of talent hired by all league member teams
τi = ti / t 8 = nti / T = ratio of talent hired by team i to the average quantity of
talent hired by all league member teams
wi = team i’s win ratio or win percent
mi = market size of team i
c = cost per unit of talent
Ri = total revenue of team i
MRi = marginal revenue of team i
ARi = average revenue of team i
In order to develop the model, it will be useful to begin by obtaining some general
expressions for the effect of a change in ti on t 8 and on τi.
First, note that t 8 = T/n, where T = t1 + t2 + … + tn. The partial derivative with
respect to ti is
∂ t / ∂t i = (1 / n )[1 + ∑ ( ∂t j / ∂t i )] [2.1]
j≠ i
Second, note that τi = ti / t .8 The partial derivative with respect to ti is

∂τ i / ∂t i = ( −t i / t 2 )( ∂ t / ∂t i ) + 1 / t = (1 / t 2 )[ t − t i ( ∂ t / ∂t i )] [2.2]
According to [2.1] and [2.2], the effect of a change in ti on τi depends upon the
implications of the change in ti for tj, for all j≠i. In other words, the effect on τi of a
change in the talent hired by team i depends upon the implications of this change
for the amounts of talent that are hired by all of the other league member teams.
In the model, it is assumed that team i’s revenue is dependent on three factors:
(i) Team i’s market size, represented by mi;
(ii) The overall quality of playing standards in the league as a whole, measured
by t 8; and
(iii) The quality of team i relative to the average quality of all league member
teams, measured by τi.
The first of these factors allows for league membership to comprise teams with
differing ‘base’ levels of support. These could be due to differences in the popu-
lation sizes of the teams’ home cities or towns, or other historical factors that
have led to the emergence of such differences. The second factor allows for each
team’s spectator demand to depend on the absolute quantity of the teams across
the league as a whole. The third factor allows each team’s spectator demand to
depend on its own quality relative to the league average. The specification of the
total revenue function is
Ri = ηm i ( t − θ t 2 )( τ i − ξτ 2i ) [2.3]
The inclusion in [2.3] of τi as a determinant of team i’s revenue captures the notion
that supporters prefer a winning team, and are expected to attend home matches
in larger numbers when the home team is successful. In the literature, it is standard
practice to assume that team i’s win ratio or win percent2 in matches played against
team j is determined by the relative quantities of talent hired by the two teams, or
wi = ti/(ti + tj). For an n-team league, this assumption leads to a rather complex
expression for team i’s win ratio over all of its league matches, which takes the
form ∑ [t i /(t i + t j )] /( n − 1) . The inclusion of τi in [2.3], rather than wi, makes the
j≠ i
revenue function easier to manipulate. For a two-team league τi = 2wi, and the
choice between τi and wi is trivial, affecting only the parameterisation of the rev-
enue function.
It is important to note that the adoption of the revenue function defined by
[2.3] imposes some restrictions on the generality of the model. In particular, the
assumption that the parameters η, θ and ξ are identical for all teams, and that
market size is the only source of intra-team differences in revenue functions, is
restrictive. If a restriction of this kind is not imposed, however, it is extremely dif-
ficult to obtain generalisations from the model concerning the effects on competi-
tive inequality of variations in the objectives of team owners, or policy measures
like revenue sharing or payroll caps (Fort, 2000; Fort and Quirk, 2004).
A model of an n-team league 21
Team i’s marginal revenue is defined as the effect on total revenue of a change
in ti. A general expression for team i’s marginal revenue is obtained by partially
differentiating [2.3] with respect to ti. Team i’s average revenue is the average
revenue generated by each unit of talent hired by team i. A general expression for
team i’s average revenue is obtained by dividing both sides of [2.3] by ti.
MRi = ηm i [(1 − 2θ t )( τ i − ξτ 2i )( ∂ t / ∂t i ) + ( t − θ t 2 )(1 − 2ξτ i )( ∂τ i / ∂t i )]

ARi = ηm i ( t − θ t 2 )( τ i − ξτ 2i ) / t i [2.4]
For simplicity, it is assumed that each team’s total cost is linear and propor-
tional to the quantity of talent it hires, and that the teams incur no fixed costs.
These assumptions produce marginal and average cost functions for the two
teams that are constant, and equivalent to the market price for each unit of
talent.
C i = ct i
MC i = AC i = c for all i [2.5]
In most variants of the theoretical model, the equilibrium allocation of talent

between the league’s member teams usually depends upon the objective function
of the team owners. Below, two alternatives are considered: profit maximisation
and win-percent maximisation subject to a zero-profit constraint. For convenience
the latter is usually known simply as win-percent maximisation. With reference to
European football, in particular, it has been argued that the latter objective func-
tion provides a more accurate characterisation of the reality than profit maximisa-
tion (Sloane, 1971; Késenne, 1996, 2007a, 2007b; Vrooman, 2000, 2007; Fort and
Quirk, 2004; Dietl, Lang and Werner, 2009); see also Chapter 6, Section 6.7.
The n-team closed model

In a closed model, a change in the talent hired by team i has direct implications
for the quantities of talent hired by the other teams, because it is assumed that the
total quantity of talent is fixed. For convenience and without any loss of general-
ity, the units of measurement of talent can be normalised such that T = 1 and t 8 =
1/n. Equations [2.1] and [2.2] have the following implications:
∑ ( ∂t j / ∂t i ) = −1 ∂ t / ∂t i = 0 ∂τ i / ∂t i = 1 / t = n / T = n [2.6]
j≠ i
Using [2.6], the equations for total revenue, marginal revenue and average revenue
[2.3] and [2.4] can be rewritten as functions of mi, ti and τi ( = nti):
Ri = ηm i [(1 / n ) − (θ / n2 )]( τ i − ξτ 2i )
MRi = ηm i n[(1 / n ) − (θ / n2 )](1 − 2ξτ i )
ARi = ηm i [(1 / n ) − (θ / n2 )]( τ i − ξτ 2i ) / t i [2.7]
With no external market in playing talent, it is reasonable to assume that the sal-
ary per unit of talent is determined endogenously, within the model, by the total
level of demand for talent among all of the league’s member teams. The salary per
unit of talent adjusts to ensure that each team is able to hire as much talent as it
wishes at that salary, and that all of the fixed stock of talent is employed by the
league’s member teams.
For profit maximisation in the closed model, each team hires talent to the point
at which the marginal revenue generated by the last unit of talent equals the mar-
ginal cost incurred by hiring the last unit of talent, or MRi = MCi. For win-percent
maximisation in the closed model, each team hires talent to the point at which the
zero-profit constraint is attained, or ARi = ACi. Each of these conditions gives rise
to a system of n linear equations in ti. These equations, together with the add-
ing-up constraint T = ∑ t j = 1, can be used to obtain the solutions for ti and c,
j
under both the profit-maximisation assumption and the win-percent-maximisation
assumption.
The n-team open model

In an open model in which the teams make their hiring decisions independently,
a change in the quantity of the talent hired by one team has no direct implica-
tions for the quantities of talent that are hired by all of the other teams, unless
the teams adjust their own decisions as to how much talent to hire in response
to changes in other teams’ decisions. Equations [2.1] and [2.2] have the following
implications:
∑ ( ∂t j / ∂t i ) = 0 ∂ t / ∂t i = 1 / n
j≠ i
∂τ i / ∂t i = (1 / t 2 )( t − t i / n ) = ( n2 / T 2 )[( T − t i ) / n] = n∑ t j / T 2 [2.8]
j≠ i
In an open model, playing talent is purchased from an international market at a

price per unit of talent which, for simplicity, is assumed to be determined by the
interaction between supply and demand in the international market. The salary
per unit of talent is therefore predetermined, or determined exogenously, from the
perspective of the teams that purchase talent in this market. The elasticity of the
supply of talent with respect to salary is infinite in the open model, and zero in the
closed model. The implications of a talent supply function that lies between these
two extremes are explored by Cyrenne (2009) and Fort and Winfree (2009).
The total quantity of talent held by all of the domestic league’s member teams is
variable in an open model with an infinitely elastic supply of talent. When one team
acquires more talent, there is no direct impact on the quantities of talent held by
the other teams. It is assumed that profit-maximising or win-percent-maximising
team owners take their decisions on how much talent to acquire on the basis of an
assumption that other teams will stick with the amounts of talent they currently
hold. In other words, each team owner takes his decision on how much talent to
A model of a two-team league 23
hold on the basis of an assumption that other team owners will not react to this
decision by adjusting the amounts of talent they hold. This is known as a zero con-
jectural variations assumption. It gives rise to a type of equilibrium solution known
as a Nash equilibrium, at which each team owner hires the quantity of talent that
maximises his own objective function, while treating the quantities of talent hired
by his team’s rivals as if they were fixed and beyond his control.
In the profit-maximisation case, each team hires the quantity of talent that max-
imises its profit, conditional on the quantities of talent hired by every other team
(that maximise their profits) remaining unchanged at their current levels. Since
every team maximises its own profit, the quantities of talent hired are mutually
consistent. The equilibrium is stable unless, for some reason, one or more of the
teams decide to abandon the zero conjectural variations mode of behaviour. The
term ‘Nash equilibrium’ is commonly used by economists to refer to a profit-max-
imisation equilibrium of this kind. However, an analogue of the Nash equilibrium
also exists for the win-percent-maximisation model. Each team selects the quan-
tity of talent that maximises its win percent (subject to the zero-profit constraint),
conditional on the quantities of talent hired by every other team (that maximise
their win percents) remaining unchanged at their current levels.
Using [2.8], the equations for total revenue, marginal revenue and average
revenue [2.3] and [2.4] can be rewritten as functions of mi, T, ti, tj for j≠i, and
τi ( = nti/T):
Ri = ηm i [(1 / n )T − (θ / n2 )T 2 ]( τ i − ξτ 2i )
MRi = ηm i {[(1 / n ) − ( 2θ / n2 )T ]( τ i − ξτ 2i )
+ n[(1 / n )T − (θ / n2 )T 2 ](1 − 2ξτ i )∑ t j / T}
j≠ i
ARi = ηm i [(1 / n )T − (θ / n )T ]( τ i − ξτ ) / t i
2 2 2
i [2.9]
For profit maximisation each team hires talent to the point at which the marginal
revenue generated by the last unit of talent equals the marginal cost incurred by
hiring the last unit of talent, or MRi = MCi. This condition gives rise to a system
of n non-linear simultaneous equations in ti. The solution to this system of equa-
tions is the Nash equilibrium solution to the profit-maximisation model. For win-
percent maximisation, each team hires talent to the point at which the zero-profit
constraint is attained, or ARi = ACi. This condition also gives rise to a system of n
non-linear simultaneous equations in ti, which can be solved to find the analogue
of the Nash equilibrium solution for the win-percent-maximisation model.
2.3 A model of a two-team league

The previous section defined the structure of a very general model for an n-team
league. For the n-team model, however, the explicit solutions for the amounts of
talent hired by the teams are algebraically complex. In the sports economics lit-
erature, the standard (almost universal) practice is to present solutions for the
special case where n = 2. The notion of a two-team league is obviously counter-

factual, and may therefore seem unappealing when it is encountered for the first
time. However, it turns out that most of the key insights that can be obtained from
the two-team model remain applicable, in an appropriately modified form, in the
more realistic but less tractable case where n>2. Therefore the use of a theoretical
model for a stylised two-team league should be viewed as a convenient simplifica-
tion, enabling useful results to be obtained that are widely applicable.
In the two-team model, team i’s win percent is assumed to be equivalent to the
ratio of its own talent to the total quantity of talent hired by both teams, or wi =
ti / (t1 + t2). In Section 2.2, the revenue functions [2.3], [2.4], [2.7] and [2.9] are
expressed in terms of t ,8 τi, ti and tj (for j≠i). In Section 2.3, it is convenient to repa-
rameterise the revenue functions, so as to obtain expressions in terms of T, wi and
wj. In order to do so, note that T = t1 + t2, τi = nti / T = 2ti / T = 2ti / (t1 + t2), and t 8 =
T/n = T/2. Accordingly, τi = 2wi, and T = 2t .8
Finally, and without any loss of generality, it is assumed that there is a dispar-
ity between the market sizes of teams 1 and 2 in the two-team model, such that
m1>m2. Below, team 1 is sometimes referred to as the large-market team, and team
2 as the small-market team.
The two-team closed model

In the two-team closed model, T = t1 + t2 = 1 and wi = ti / (t1 + t2) = ti. Using [2.7],
the resulting expressions for total revenue, marginal revenue and average revenue
are
Ri = ψm i ( w i − φw 2i )
MRi = ψm i (1 − 2φw i )
ARi = ψm i (1 − φw i ) [2.10]
where ψ = 2η[(1/n) − (θ/n2)] and φ = 2ξ. For a team owner’s objective of profit maxi-
misation, the solutions to the model, obtained by solving the equations MR1 = MC,
MR2 = MC and t1 + t2 = 1 for t1, t2 and c, are
m1 − m 2 + 2φm 2 m 2 − m1 + 2φm1
t1π = ; t 2π =
2φ(m1 + m 2 ) 2φ(m1 + m 2 )
ψm1m 2 (2 − 2φ ) t1 m1 − m 2 + 2φm 2
π
cπ = ; =
m1 + m 2 t 2π m 2 − m1 + 2φm1 [2.11]
For a team owner’s objective of win-percent maximisation, the solutions, obtained
by solving the equations AR1 = AC, AR2 = AC and t1 + t2 = 1 for t1, t2 and c, are
m1 − m 2 + φm 2 m 2 − m1 + φm1
t1w = ; t 2w =
φ(m1 + m 2 ) φ(m1 + m 2 )
ψm1m 2 (2 − φ ) t1w m1 − m 2 + φm 2
cw = ; =
m1 + m 2 t 2w m 2 − m1 + φm1 [2.12]
A model of a two-team league 25
cw •
AR1
cπ •
AR2
MR2 MR1
t1π
t1 = 0, t2 = 1 w t1 = 1, t2 = 0
t1
Figure 2.1 Profit-maximisation and win-percent-maximisation equilibria: two-team closed
model
If the team owner’s objective is win-percent maximisation, the equilibrium salary

per unit of talent is higher than it is if the owner’s objective is profit maximisation
(cw > cπ); and competitive inequality is higher than it is if the owner’s objective is
profit maximisation (tw1 /tw2 > tπ1/tπ2).
Figure 2.1 compares the profit-maximisation and win-percent-maximisation
equilibria in the two-team closed model. The downward-sloping MR1 shows team
1’s marginal revenue as a function of t1 represented on the horizontal axis. MR2
shows team 2’s marginal revenue. Since t1 = 1 – t2, MR2 plotted against t1 is the
inverse of the same function plotted against t2, and is therefore upward-sloping.
The profit-maximisation equilibrium is located at the intersection of MR1 and
MR2. Team 1 hires t1π units of talent, and team 2 hires t2π = 1 – t1π, where t1π > 0.5
and t2π < 0.5. Similarly, the win-maximisation equilibrium {t1w,t2w} is located at the
intersection of the downward-sloping AR1 and upward-sloping AR2 functions.
In an extension of the profit-maximisation version of the closed model, Hoehn
and Szymanski (1999) investigate the implications for competitive balance in the
domestic league of involvement of large-market teams in European competition.
While team 2’s revenue depends only on its win percent against team 1 in domes-
tic competition, team 1’s total revenues are obtained from both domestic and
European competition. In the latter, team 1 competes against a European team
with a similar (large) home-town population. Team 1’s involvement in European
competition provides an incentive to invest more heavily in talent than it would
if it were not involved, in order to achieve a reasonable win percent in Europe
and maximise profits from both competitions. The effect is to make competition
more unequal in the domestic league. The higher the proportion of team 1’s total
c c
MC AC
AR1( t2w )
MR1( t2π ) AR2( t1w )
MR2( t1π )
t2π t1π t1, t2 t2w t1w t1, t2
Figure 2.2 Profit-maximisation and win-percent-maximisation equilibria: two-team open
model
r evenue that is derived from Europe, the greater is the degree of domestic com-
petitive inequality.
The two-team open model

In the two-team open model, T = t1 + t2 and wi = ti/(t1 + t2) = τi/2. Using [2.9], a
simplified expression for total revenue is
Ri = ηm i (T − γT 2 )(w i − φw 2i ) [2.13]
where γ = θ/2 and φ = 2ξ (as before). Partially differentiating Ri with respect to ti
yields
MRi = ηm i [(1 − 2 γT )( w i − φw 2i )( ∂T / ∂t i ) + ( T − γT 2 )(1 − 2φw i )( ∂w i / ∂t i ) [2.14]
Note that ∂T/∂ti = 1 and ∂wi/∂ti = tj/(t1 + t2)2 = wj / (t1 + t2) = wj / T. Substituting
these results into [2.14] yields
MRi = ηm i [(1 − 2γT)(w i − φw 2i )+(1 − γT)(1 − 2φw i )w j ] [2.15]
Using [2.13], team i’s average revenue is
ARi = ηm i (T − γT 2 )(w i − φw 2i )/t i [2.16]
For a team owner’s objective of profit maximisation, the solutions to the model,
obtained by solving the equations MR1 = MC, MR2 = MC for t1 and t2, are alge-
braically complex, but simply represented diagrammatically. The left-hand dia-
gram in Figure 2.2 shows team 1’s marginal revenue function, drawn conditional
on team 2 hiring tπ2 units of talent, intersecting with the (exogenously determined)
horizontal marginal cost function at tπ1; and likewise, team 2’s marginal revenue
function, drawn conditional on team 1 hiring tπ1 units of talent, intersecting with
the horizontal marginal cost function at tπ2. For a team owner’s objective of win-
percent maximisation, the solutions, obtained by solving the equations AR1 = AC
and AR2 = AC for t1 and t2, are also algebraically complex. The solution is illus-
trated in the right-hand diagram of Figure 2.2, which has a construction similar
to the left-hand diagram.
Revenue sharing 27
2.4 Revenue sharing

In this section, the theoretical model developed above is used to examine the
impact of revenue sharing on competitive inequality in the two-team model.
Would the redistribution of some of the revenues of the league’s stronger member
teams to their weaker counterparts help to reduce competitive inequality, perhaps
making the league a more exciting and watchable proposition for spectators? Or
would redistribution actually make no difference to competitive inequality? Or
might redistribution, as some economists have argued, even have perverse effects
of increasing, rather than reducing, competitive inequality?
The findings of the previous theoretical literature on this question are mixed.
According to the benchmark analyses of Quirk and El Hodiri (1974) and Fort
and Quirk (1995), based on a closed model with a team owner’s objective of profit
maximisation, revenue sharing does not influence competitive balance, but it does
tend to depress players’ salaries. If the team owner’s objective is win-percent maxi-
misation, however, or if team owners optimise using a utility function containing
arguments other than profits, then revenue sharing might assist in reducing com-
petitive inequality (Atkinson, Stanley and Tschirhart, 1988; Késenne, 1996, 2000,
2007b, 2009; Rascher, 1997). In an analysis based on an open model, Szymanski
and Késenne (2004) suggest that a revenue-sharing arrangement might have the
perverse and counterintuitive effect of increasing competitive inequality. We exam-
ine this proposition in detail below. Dietl and Lang (2008) draw a similar con-
clusion; see also Dietl, Franck and Lang (2008), Dietl, Lang and Werner (2009)
and Grossman, Dietl and Lang (2010). Easton and Rockerbie (2005) examine the
conditions for a revenue-sharing arrangement to achieve majority support among
the league’s member teams. The teams’ willingness to support such a measure
depends upon the nature of conjectures about the responses of competitors to any
team’s decision to acquire more talent.
Let NRi denote the net revenue of team i after redistribution of revenues under
a revenue-sharing arrangement, represented by the parameter 0 ≤ α ≤ 1. α = 1 rep-
resents no revenue sharing, and α = 0 represents equal revenue sharing, such that
each team receives one-half of the total revenue. The expressions for NRi are
NRi = αRi + [(1 − α )/2](R i + R j ) = [(1 + α )/2]R i + [(1 − α )/2]R j [2.17]
In the profit-maximisation model, the case in which there is equal revenue sharing
(α = 0) can be interpreted as equivalent to the case in which quantities of talent
are assigned to the two teams in a manner that maximises their joint profits. Since
each team has the same net revenue function when α = 0, maximisation of this net
revenue function is equivalent to joint profit maximisation.
Revenue sharing in the closed model

As before, if the team owners pursue an objective of profit maximisation, each
team hires talent to the point at which MNRi = MC, where MNRi denotes team
i’s marginal net revenue.
MR1(α=1)
MR1(α=.5) MR2(α=1)
MR1(α=0) MR2(α=.5)
π
c (α = 1) •
MR2(α=0)
π
c (α = .5) •
c π ( α = 0) •π
t1=0, t2=1 t1 t1=1, t2=0
Figure 2.3 Revenue sharing in a closed model: profit maximisation
Using t2 = 1 – t1 in the closed model, note that ∂Rj/∂ti = −∂Rj/∂tj and ∂R1/∂t2 =
−∂R1/∂t1. Using [2.10] and [2.17], the expressions for MNRi are
MNR1 = ∂NR1 / ∂t1 = ψ{m1 [(1 + α ) / 2](1 − 2φt1 ) − m 2 [(1 − α ) / 2](1 − 2φt 2 )}
MNR2 = ∂NR2 / ∂t 2 = ψ{m 2 [(1 + α ) / 2 ](1 − 2φt 2 ) − m1 [(1 − α ) / 2](1 − 2φt1 )} [2.18]
It is straightforward to show that the solutions for t1 and t2, obtained by solving
the equations MNR1 = MC and MNR2 = MC, are the same as in [2.11]. The solu-
tion for c, however, differs from [2.11]:
αψm1m 2 (2 − 2φ ) [2.19]
cπ =
m1 + m 2
Figure 2.3 illustrates the impact of revenue sharing in a closed model with team owners
in pursuit of profit maximisation. As the parameter α decreases within the range 0
≤ α ≤ 1, and the proportion of the home team’s revenue that is shared increases, the
marginal net revenue functions of both teams shift downwards by identical amounts.
The gain in (net) revenue accruing to either team as a result of hiring an additional
unit of talent is reduced by revenue sharing, because part of the gain in (gross) rev-
enue accrues to the other team. Consequently neither the allocation of playing talent
between the two teams, nor the level of competitive inequality, is affected by revenue
sharing. However, the equilibrium salary per unit of talent is reduced as α decreases;
and in the limiting case α = 0, the equilibrium salary falls to zero.
Identifying the limiting case α = 0 with joint profit maximisation, the alloca-
tion of playing talent is shown to be the same if the team owners maximise their
Revenue sharing 29
c
AR1(α=0) AR2(α=0) AR2(α=.5)
AR1(α=.5)
AR1(α=1)
AR2(α=1)
t1=0, t2=1 t1w = .5 t1w t1w t1=1, t2=0

(α=0) (α=.5) (α=1)
Figure 2.4 Revenue sharing in a closed model: win-percent maximisation
profits either individually or jointly (collectively). This result is an application of

the Coasian invariance proposition. However, individual action on the part of
the team owners affords the players bargaining power in their salary negotiations,
enabling them to achieve a salary equivalent to the teams’ marginal revenues.
Collective action on the part of the team owners eliminates the players’ bargain-
ing power, forcing the equilibrium salary down to zero.
If the team owners pursue an objective of win-percent maximisation, each team
hires talent to the point at which ANRi = AC, where ANRi denotes team i’s aver-
age net revenue, obtained by dividing [2.17] by ti. In this case the solutions for t1
and t2 are algebraically complex. The sensitivity of competitive balance to vari-
ation in the revenue-sharing parameter is explored in Figure 2.4. As α decreases
from α = 1 towards α = 0, and the proportion of revenues that are shared increases,
in the vicinity of the solution the large-market team’s average net revenue func-
tion shifts downwards, and the small-market team’s average net revenue function
shifts upwards. Consequently, there is a decrease in competitive inequality. For
equal revenue sharing with α = 0, the two teams’ net average revenue functions
are identical, and the solution is located at tw1 = tw2 = 0.5, with perfect competitive
equality.
Revenue sharing in the open model

To establish the solution to the open model in the case of profit maximisation, it
is necessary to identify the marginal net revenue functions, by taking the partial
derivatives of each of R1 and R2 in [2.13] with respect to t1 and t2.
∂R1 / ∂t1 = ηm1 [(1 − 2 γT )( w1 − φw12 ) + (1 − γT )(1 − 2φw1 )w 2 ]

∂R2 / ∂t1 = ηm 2 [(1 − 2 γT )( w 2 − φw 22 ) − (1 − γT )(1 − 2φw 2 )w1 ]
∂R1 / ∂t 2 = ηm1 [(1 − 2 γT )( w1 − φw12 ) − (1 − γT )(1 − 2φw1 )w 2 ]
∂R2 / ∂t 2 = ηm 2 [(1 − 2 γT )( w 2 − φw 22 ) + (1 − γT )(1 − 2φw 2 )w1 ] [2.20]
The expressions for MNRi are

MNR1 = [(1 + α ) / 2]( ∂R1 / ∂t1 ) + [(1 − α ) / 2 ]( ∂R2 / ∂t1 )
MNR2 = [(1 + α ) / 2]( ∂R2 / ∂t 2 ) + [(1 − α ) / 2 ]( ∂R1 / ∂t 2 ) [2.21]
The solutions for t1 and t2 are obtained by solving the simultaneous equations
MNR1 = MC and MNR2 = MC. A relatively simple measure of competitive
inequality is obtained by substituting [2.20] into [2.21], and using the condition
MNR1 = MNR2 at the profit-maximising-equilibrium, yielding
m1 (1 − 2 γT )( w1 − φw12 ) + m1 (1 − γT )(1 − 2φw1 )w 2

= m 2 (1 − 2 γT )( w 2 − φw 22 ) + m 2 (1 − γT )(1 − 2φw 2 )w1
Rearranging
(1 − 2φw 2 )w1 m1  1 − 2 γT   ( m1 / m 2 )( w1 − φw12 ) − ( w 2 − φw 22 )  [2.22]
= + α
(1 − 2φw1 )w 2 m 2  1 − γT   (1 − 2φw1 )w 2 
The left-hand side of [2.22] can be interpreted as a measure of competitive

inequality. Note that as w1 increases and w2 decreases, both terms in the numer-
ator of the left-hand side increase, and both terms in the denominator decrease.
Unambiguously, therefore, the left-hand side increases (decreases) as competi-
tive inequality increases (decreases). On the right-hand side of [2.22], (1 – 2γT)/
(1 – γT)>0, and (m1/m2)(w1 – φ w21) – (w2 – φ w22) > 0. All terms on the right-hand side
are positive. Competitive inequality is therefore increasing in the revenue-sharing
parameter α. Competition is more equal with equal revenue sharing (α = 0) than
it is with no revenue sharing (α = 1). As α decreases within the range 0 ≤ α ≤ 1, the
right-hand side of [2.22] decreases, and competitive inequality decreases.
The equilibrium conditions for win-percent maximisation are ANR1 = AC and
ANR2 = AC, where
ANRi = [(1 + α ) / 2 ]Ri /t i + [(1 − α ) / 2 ]R j / t i [2.23]
As before, the solutions for t1 and t2 are algebraically complex. Again, however, it
is possible to derive a relatively simple measure of competitive inequality using the
condition ANR1 = ANR2 at the win-percent maximisation equilibrium, yielding
w1 (1 + α )( m1 / m 2 )( w1 − φw12 ) + (1 − α )( w 2 − φw 22 ) [2.24]
=
w12 ) + (1 + α )( w 2 − φw 22 )
w 2 (1 − α )( m1 / m 2 )( w1 − φw
Revenue sharing 31
t2
α=0
•
α=0.5
•
BR1 win-percent
profit maximisation
maximisation
N
t π2 •
t 2w • α=1
α=1
• BR2
J α=0.5
•
α=0
t1π t1w t1
Figure 2.5 Nash equilibrium and joint profit-maximisation equilibrium: profit-maximisation
model
Note that for α = 1 (no revenue sharing), w1/w2 = (m1/m2)(w1 − φw21)/(w2 − φw22) > 1;
and for α = 0 (equal revenue sharing), w1/w2 = 1. As in the case of profit maximisa-
tion, competition is more equal with equal revenue sharing (α = 0) than it is with
no revenue sharing (α = 1). With equal revenue sharing, both teams have the same
net revenue function, and both take the same decision concerning the quantities of
talent they hire. Accordingly w1 = w2, and there is perfect competitive equality.
Figure 2.5 tracks the progression from the Nash equilibrium to the joint profit-
maximisation equilibrium in the profit-maximisation model. The quantities of tal-
ent hired by the two teams, t1 and t2, are represented on the horizontal and vertical
axes. For the case of no revenue sharing (α = 1), BR1 and BR2 are the best-response
functions of teams 1 and 2. BR1 is read clockwise from the vertical to the hori-
zontal axis, and shows, for each possible value of t2, the profit-maximising value
of t1. Similarly BR2 is read anticlockwise from the horizontal to the vertical axis,
and shows, for each possible value of t1, the profit-maximising value of t2. The
profit-maximising solution for both teams under the zero conjectural variations
assumption is (tπ1,tπ2), represented by point N and located at the intersection of BR1
and BR2.
If a revenue-sharing arrangement is introduced, the best-response functions
shift towards the bottom-left-hand corner of Figure 2.5, and also tend to tilt. The
shifts from α = 0 through α = 0.5 to α = 1 are such that the points of intersection
also move towards the bottom-left-hand corner but, in accordance with the inter-
pretation of [2.22], in a manner that implies a reduction in competitive inequality
(represented by the rays from the origin to the intersection of the best-response
functions becoming steeper). The joint profit-maximisation equilibrium, at which
α = 0, is represented by point J. Figure 2.5 also tracks the progression between the
counterparts of these two equilibria in the win-percent maximisation model, over

variations in the revenue-sharing parameter from α = 0 through α = 0.5 to α = 1.
The Szymanski-Késenne revenue-sharing analysis

The finding that revenue sharing reduces competitive inequality in the profit-
maximisation version of the open model is contrary to an analysis presented by
Szymanski and Késenne (2004), henceforth SK, who draw the opposite conclu-
sion. SK find that competitive inequality is higher in the case of equal revenue
sharing (α = 0) than it is in the case of no revenue sharing (α = 1). In other words,
and perhaps counterintuitively, the introduction of a revenue-sharing arrange-
ment increases competitive inequality. SK explain this result through a tendency
for revenue sharing to offset the negative external effect of team strengthening
by either team on the other team’s revenue. The negative external effect of team
strengthening by team 2 on team 1’s revenue is larger than the negative external
effect in the reverse direction. When these effects are partially neutralised by rev-
enue sharing, the reduction in team 2’s marginal revenue is larger than the reduc-
tion in team 1’s marginal revenue. Accordingly, the solution shifts in the direction
of greater competitive inequality.
A key component of the SK analysis is an assumption that revenue depends upon
win percent, but not upon the quantities of talent hired by either or both teams in
absolute terms. SK use a revenue function that is similar to [2.10]. Within the frame-
work that is developed in this chapter, [2.10] is appropriate for use in a closed model,
where the total quantity of talent hired by the two teams combined is assumed fixed,
and is therefore omitted from the revenue function. In an open model, however, the
assumption that revenue depends upon win percent only seems counterintuitive or
counterfactual. Subsequently Cyrenne (2009) and Feess and Stahler (2009) develop
models in which team revenues depend upon both win percent and the absolute
quantities of talent. Both interpret the SK analysis as a special case, applicable only
when a restricted formulation of the revenue function is adopted.
If a revenue function similar to [2.10] is used in an open model, an implica-
tion is that if the quantities of talent (expressed in absolute terms) are reduced
towards zero, but in a manner such that the ratio of the quantities of talent hired
by the two teams is maintained at a constant value, then revenue is unaffected.
This would imply that a fixture in English football’s Premier League for which
the team managers (hypothetically) selected teams comprising park footballers
would command the same market value as the same fixture starring talented
international superstars. This seems a highly dubious proposition.3
Clearly, both absolute and relative talent are determinants of spectators’ util-
ity, and the revenue function in an open model should include both. As has been
shown in this section, when the revenue function is specified accordingly, revenue
sharing always reduces competitive inequality. The negative external effect of team
Restraints on expenditure on players’ salaries 33
strengthening by either team on the other team’s revenue through the change in
win percent is offset by a positive external effect emanating from the increase in the
combined quantity of talent hired by both teams. When the net (overall) external
effect is partially neutralised by revenue sharing, the reduction in team 2’s mar-
ginal revenue is always smaller than the reduction in team 1’s marginal revenue.
Consequently the solution shifts in the direction of greater competitive equality.
2.5 Restraints on expenditure on players’ salaries

This section examines the impact on competitive inequality of three forms of
restraint on expenditure on players’ salaries: first, a tax on expenditure on salar-
ies above a specific threshold, known as a luxury tax; second, a limit on the total
amount each team is permitted to spend on salaries as a proportion of the revenue
of the average team, known as a payroll cap; and third, a limit on the ratio of each
team’s expenditure to its own revenue, known as the G14 payroll cap. In each
case, the implications for competitive inequality in the two-team closed and open
models are developed, for the case of profit maximisation. Summary results are
also provided for the case of win-percent maximisation, but for reasons of space
the full analysis is not presented.
Luxury tax: a tax on salary expenditure over a set threshold

The luxury tax is a relatively recent innovation in North American major league
sports, whereby the league taxes any team’s expenditures on players’ salaries in
excess of a specific tax threshold (Gustafson, 2006). The luxury tax is a more flex-
ible mechanism for restraining salary expenditure than the payroll cap (see below),
which imposes an absolute limit on salary expenditure. Gustafson and Hadley
(1996) and Marburger (1997) analyse the impact of the luxury tax on salary and
competitive inequality; and Van de Burg and Prinz (2005) examine a variant in
which a progressive tax rate is applied.
Let θ denote the threshold for salary expenditure above which the luxury tax
is payable, and let τ denote the rate at which tax is levied on expenditure above θ.
Let {t1π , t 2π } denote the solution when there is no luxury tax, obtained by solving
MR1 = c and MR2 = c. Suppose initially the tax threshold is set at a level that makes
only team 1 (the large-market team) liable to pay the tax, so that ct1π > θ . In this case,
team 1’s marginal cost function increases from c to c + τ. Let {t1π , t2π } denote the solu-
tion when team 1 is liable to pay the tax but team 2 (the small-market team) is not
liable, obtained by solving MR1 = c + τ and MR2 = c. Then suppose instead the tax
threshold is set at a level that also makes team 2 liable to pay the tax, so that ct2π > θ .
In this case, team 2’s marginal cost function also increases to c + τ. If team 1 and team
2 are both liable to pay the tax, the solution, denoted {t1π , t2π } , is obtained by solving
MR1 = c + τ and MR2 = c + τ.
~c π+ τ
cπ = c π + τ MR2
•
~c π •
cπ MR1
θb θa ~t π t1π = t1π t1
1
cπ ~c π
Figure 2.6 Luxury tax: closed model
Figure 2.6 illustrates the analysis for the closed model, in which c is determined
endogenously and t2 = 1 – t1. When there is no luxury tax, the solution is {tπ1, cπ},
located at the point where MR1 = MR2 = cπ. If the luxury tax threshold is θa, this
solution is invalid because cπtπ1 > θa. The solution switches to {t1π , c π } , such that
MR1 = c π + τ and MR2 = c π . At {t1π , c π }, c π t2π = c π (1 − t1π ) < θa If only the large-
market team is liable to pay the luxury tax, therefore, competitive inequality is
reduced and the equilibrium salary per unit of talent is reduced. If a lower tax
threshold of θb is set, this solution becomes invalid, because c π t2π > θb . The solu-
tion switches to { t1π , c π } , such that MR1 = MR2 = c π + τ . Note that t1π = t1π , but
c π < c π . If both teams are liable to pay the luxury tax, therefore, competitive
inequality is the same as it is when there is no luxury tax, but the equilibrium salary
per unit of talent is reduced by an amount equivalent to the marginal tax rate.
Figure 2.7 illustrates the analysis for the open model, in which c is exogenous and
there are no constraints on the values of t1 and t2. The solution when there is no
luxury tax, {tπ1, tπ2}, is located at the intersection of BR1(c) and BR2(c). If the luxury
tax threshold is θa, this solution is invalid because ctπ1 > θa. The solution switches to
{t1π , t2π } , located at the intersection of BR1(c + τ) and BR2(c). At {t1π , t2π } , ct2π < θa .
If only the large-market team is liable to pay the luxury tax, therefore, competitive
inequality is reduced, indicated by the steeper slope of the ray at {t1π , t2π } . If a lower
tax threshold of θb is set, however, this solution is also invalid because ct2π > θb .
The solution switches to { t1π , t2π }, located at the intersection of BR1(c + τ) and
BR2(c + τ). { t1π , t2π } is located on the same ray from the origin as {tπ1, tπ2}. If both
teams are liable to pay the luxury tax, therefore, competitive inequality is the same
as it is when there is no luxury tax.
t2
θa/c
{~t , ~t2 }
π π
•1
{t1π, t π2}
•
• BR2(c)
π π
{t1, t 2 }
BR2(c+τ)
θb/c
BR1(c)
BR1(c+τ)
θb/c θa/c t1
Figure 2.7 Luxury tax: open model
In both the closed and the open model, the luxury tax reduces competitive
inequality in the case where the large-market team is taxed while the small-market
team is untaxed, but the luxury tax has no effect on competitive inequality in the
case where both teams are taxed.
The solution when the team owners’ objective is win-percent maximisation (not
illustrated in Figures 2.6 and 2.7) is established in a similar manner. In the case
where team 1 is taxed but team 2 is not taxed, the solutions for t1 and t2 are obtained
by solving AR1 = {θ + (c + τ)}(t1 – θ/c)}/t1 and AR2 = c. Competitive inequality is
less than it is when there is no luxury tax. In the case where both teams are taxed,
the solutions for t1 and t2 are obtained by solving AR1 = {θ + (c + τ)}(t1 – θ/c)}/
t1 and AR2 = {θ + (c + τ)}(t2 – θ/c)}/t2. Competitive inequality is the same as it is
when there is no luxury tax.
Payroll cap: a limit on each team’s total salary expenditure as a percentage

of the average team’s revenue
The implications of a payroll cap for competitive inequality and player salaries
are examined by Quirk and Fort (1992), Staudohar (1999) and Késenne (2000,
2007b). Recently, Dietl et al. (2008) and Dietl, Lang and Rathke (2009) have exam-
ined the social welfare effects of payroll or salary caps.
Let φ denote the proportion of the average team’s revenue that defines the upper
limit for each team’s salary expenditure. As before, let {tπ1, tπ2} denote the solution
when there is no payroll cap, obtained by solving MR1 = c and MR2 = c. Suppose
initially the payroll cap is set at a level that imposes a constraint on team 1 only,
so that ctπ1 > φ(Rπ1 + Rπ2)/2, where {Rπ1, Rπ2} are the revenues associated with the
unconstrained solutions {tπ1, tπ2} (with other revenue expressions similarly defined).
{t1π , t2π } , the solution when team 1’s salary expenditure is constrained but team 2 is
c
~ ~ ~ ~
ct1 = φ a (R 1π + R π2 ) / 2 ct 2 = φ a (R 1π + R π2 ) / 2
ct1 = φ b (R 1π + R π2 ) / 2 ct 2 = φ b (R 1π + R π2 ) / 2
MR2
cπ •
~c π •
cπ • MR1
t1π = 0.5 ~t1π t1π t1

Figure 2.8 Payroll cap: closed model
unconstrained, is obtained by solving φ(R1 + R2)/(2t1) = c and MR2 = c. Suppose

instead the payroll cap is set at a level that also imposes a constraint on team 2’s
salary expenditure, so that ct 2π > φ( R  1π + R  2π ) / 2 . If team 1 and team 2 are both
constrained, the solutions for t1 and t2, denoted {t18 , t28 }, are obtained by solving
φ(R1 + R2)/(2t1) = c and φ(R1 + R2)/(2t2) = c.
Figure 2.8 illustrates the analysis for the closed model. When there is no payroll
cap, the solution is {tπ1, cπ}, located at the point where MR1 = MR2 = cπ. If the
payroll cap is φa, this solution is invalid because cπtπ1 > φa(Rπ1 + Rπ2)/2. The solution
switches to {t1π , c π } , such that φa ( R  1π + R  2π ) /( 2t1π ) = c π and MR2 = c π . At
{t1 , c }, c t2 = c (1 − t1 ) < θa . If only the large-market team is capped, therefore,
π π π π π π
competitive inequality is reduced and the equilibrium salary per unit of talent is
reduced. If a lower payroll cap of φb is set, however, this solution is also invalid
because c π t2π > φb ( R π +R  π ) / 2 . The solution switches to { t1π , c π } , such that
1 2
φb( R1 + R2 ) /( 2 t1 ) = c π and φb( R1π + R2π ) /( 2 t2π ) = c π . Since t2π = 1 − t1π , the solu-
π π π
tion is t1π = t2π = 0.5 . If both teams are capped, therefore, competitive inequality
is eliminated altogether and there is perfect competitive balance.
Figure 2.9 illustrates the analysis for the open model. The solution when there is
no payroll cap, {tπ1, tπ2}, is located at the intersection of BR1 and BR2. If the payroll
cap is φa, this solution is invalid because ctπ1 > φa(Rπ1 + Rπ2)/2. Team 1 must reduce its
salary expenditure by hiring less talent. Team 2, which is unconstrained, reacts by
increasing its quantity of talent hired. There is a leftward shift along BR2, towards
the solution {t1π , t2π } , at which ct1π = φa( R  1π + R  2π ) / 2 . If only the large-market team
is capped, therefore, competitive inequality is reduced, indicated by the steeper
slope of the ray at {t1π , t2π } . If a lower payroll cap of φb is set, however, this solution
becomes invalid because ct2π > φb( R  1π + R  2π ) / 2 . The solution switches to { t1 , t2 } ,
~ ~
t2 φ b (R 1π + R π2 ) /( 2c) φ a (R 1π + R π2 ) /( 2c)
~ ~
φ b (R 1π + R π2 ) /( 2c) φ a (R 1π + R π2 ) /( 2c)
φ a (R 1π + R π2 ) /( 2c)
~ ~
φ a (R 1π + R π2 ) /( 2c)
{~t1π , ~t2π} •
{t1π , t π2} ~ ~
• φ b (R 1π + R π2 ) /( 2c)
{t1π , t π2} φ b (R 1π + R π2 ) /( 2c)
•
BR2
BR1
t1
Figure 2.9 Payroll cap: open model
such that φb( R1π + R2π ) /( 2 t1π ) = c and φb( R1π + R2π ) /( 2 t2π ) = c . This implies t1π = t2π .
As in the closed model, if both teams are capped, competitive inequality is elimi-
nated altogether and there is perfect competitive balance.
In both the closed model and the open model, the payroll cap reduces competi-
tive inequality in the case where the large-market team is capped while the small-
market team is uncapped, and eliminates competitive inequality altogether in the
case where both teams are capped.
The solution when the team owners’ objective is win-percent maximisation is
established in a similar manner. If team 1’s salary expenditure is constrained while
team 2 is unconstrained, competitive inequality is reduced in comparison with the
case of no payroll cap. If both teams’ salary expenditures are constrained, com-
petitive inequality is eliminated altogether, and the solution is the same as in the
profit-maximisation model.
Vrooman (1995) dissents from the proposition that payroll caps are helpful in
reducing competitive inequality, interpreting most capping arrangements as a col-
lusive attempt to control costs, rather than to redistribute playing talent among the
league’s member teams. According to the invariance proposition, the allocation
of playing talent that maximises total league profits is unaffected by any restric-
tions on salary expenditure. If such restrictions are imposed, profit-maximising
team owners are expected to invent mechanisms to circumvent them and restore
the optimal allocation of playing talent for profit maximisation. Such mechanisms
might include player loan arrangements, whereby the team borrowing the player
pays part of the loanee’s salary, enabling the parent team to retain the property
rights in the player’s services while maintaining its salary expenditure within the
limit imposed by the cap (Vrooman, 1995; Marburger, 2006).
AR1
MR1
AR2
MR2
ρaAR 2
cπ •
~c π •
ρaAR1
cπ • ρ bAR 2
ρ bAR1
~π
t1π t1 t1π = t1w t1
Figure 2.10 G14 payroll cap: closed model
G14 payroll cap: constraint on the ratio of each team’s salary expenditure

to its own revenue
The G14 was a lobby group comprising leading football clubs from seven
European countries, to provide a unified voice in negotiations with the sport’s
international governing bodies. Originally formed in 2000 by fourteen clubs, the
membership of the G14 was increased to eighteen clubs in 2002, but the original
name was retained. In 2002 the G14 announced a voluntary agreement that its
members would adhere to a limit on salary expenditure of 70 per cent of each
club’s revenue, to take effect from the 2006 season. The G14 disbanded in 2008.
Késenne (2003, 2007b) examines the implications of the G14 payroll cap for sal-
aries and competitive inequality, and Dietl, Franck and Nuesch (2006) investigate
the conditions under which a voluntary cap is self-enforcing.
Let ρ denote the proportion of each team’s revenue that defines the upper limit
for its own salary expenditure under the G14 payroll cap. As before, let {tπ1, tπ2}
denote the solution when there is no G14 payroll cap, obtained by solving MR1 =
c and MR2 = c. At {tπ1, tπ2}, team 2 (the small-market team) has a higher ratio of
salary expenditure to total revenue than team 1 (the large-market team). Suppose
initially the G14 payroll cap is set at a level that imposes a constraint on team
2 only, so that ctπ2 > ρRπ2. Let {t1π , t2π } denote the solution when team 2’s salary
expenditure is capped but team 1 is uncapped, obtained by solving MR1 = c and
ρR2/t2 = c. Suppose instead the G14 payroll cap is set at a level that also imposes
a constraint on team 1, so that ct1π > ρR  1π . If team 1 and team 2 are both capped,
the solutions for t1 and t2, denoted { t1 , t 2} , are obtained by solving ρR1/t1 = c and
ρR2/t2 = c.
Figure 2.10 illustrates the analysis for the closed model. When there is no G14 pay-
roll cap, the solution is {tπ1, cπ}, located at the point where MR1 = MR2 = cπ. If the G14
t2 ρ bR1π / c ρaR 1π / c
{t1π, t π2}
• ρaR π2 / c
~
• ρaR π2 / c
{~t π1 , ~t 2π} BR2
• ρ bR π2 / c
{t1π, t π2}
BR1
t1
Figure 2.11 G14 payroll cap: open model
payroll cap parameter is ρa, this solution is invalid because cπtπ2 > ρRπ2. The solution
 1π or
switches to {t1π , c π } , such that MR1 = c π and ρa AR2 = c π . At {t1π ,c π }, c π t1π < ρa R
c < AR1 . If only the small-market team is capped, therefore, competitive inequality is
π
increased and the equilibrium salary per unit of talent is reduced. If a lower G14 pay-
roll cap parameter of ρb is set, however, this solution is invalid because c π t1π > ρ b R  1π .
The solution switches to { t1π , t2π } , which satisfies the constraints ρb R1π / t1π = c π and
ρb R2π / t2π = c π . If both teams are capped, therefore, competitive inequality is further
increased, and the equilibrium salary per unit of talent is further reduced. At { t1π , t2π } ,
it can be shown that the level of competitive inequality is the same as it is when the team
owners’ objective is win-percent maximisation and there is no cap.
Figure 2.11 illustrates the analysis for the open model. The solution when there
is no G14 payroll cap, {tπ1, tπ2}, is located at the intersection of BR1 and BR2. If the
G14 payroll cap parameter is ρa, this solution is invalid because ctπ2 > ρaRπ2. Team
2 must reduce its salary expenditure by hiring less talent. Team 1, which is uncon-
strained, reacts by hiring more talent. There is a rightward shift along BR1, towards
 2π . If only the small-market team is capped,
the solution {t1π , t2π } , at which ct2π = ρa R
therefore, competitive inequality is increased, indicated by the shallower slope of
the ray at {t1π , t2π } . If a lower G14 payroll cap parameter of ρb is set, however, this
solution is also invalid because ct1π > ρb R  1π . The solution { t1π , t2π } satisfies the con-
straints ρb R1 / t1 = c and ρb R2 / t2 = c . If both teams are capped, therefore, com-
π π π π
petitive inequality is further increased, indicated by the shallower slope of the ray at
{ t1π , t2π } . As in the closed model, if both teams are capped, the level of competitive
inequality is the same as it is when the team owners’ objective is win-percent maxi-
misation and there is no cap.
In both the closed model and the open model, therefore, the G14 payroll cap
succeeds in its stated objective of reducing team owners’ total salary expenditure,
but does so at the expense of an increase in competitive inequality. When both
teams’ salary expenditures are constrained, competitive inequality with a team

owners’ objective of profit maximisation is the same as it is when there are no
restraints on salary expenditure and the team owners’ objective is win-percent
maximisation.
When the team owners’ objective is win-percent maximisation, there is a range
of values for ρ such that team 2’s salary expenditure is constrained while team 1 is
unconstrained. Within this range, competitive inequality is higher than it is when
there is no cap. There is a (lower) range of values for ρ such that both teams’ salary
expenditures are constrained. Competitive inequality is the same as it is as when
there is no cap.
Conclusion
The notion that market mechanisms can be relied upon to maintain a reasonable
degree of competitive equality among the member teams of a sports league, with-
out the need for intervention by the sports governing body to redistribute revenue,
restrict expenditure on players’ salaries, or constrain players’ employment mobility,
has a long tradition in the theoretical sports economics literature. Economists tend
to take a sceptical view of the effectiveness of interventions designed to produce
outcomes different from those that would be delivered by free markets. Although
regulatory intervention may be helpful in lowering players’ salaries, thereby boost-
ing profitability or promoting the survival of smaller-market teams, this does not
imply that intervention will always create the incentives required for adjustment in
the direction of reduced competitive inequality. Some forms of intervention may
even create perverse incentives and tendencies in the opposite direction.
Much of the early literature on the economics of team sports leagues was writ-
ten from a North American perspective, and develops models in which the total
stock of playing talent is assumed fixed, so that any reallocation of talent between
the league’s member teams is a zero-sum activity. In contrast with North American
major league sports, which draw principally on home-grown talent, English and
European football operates with a more open labour market. The total stock of
playing talent at the disposal of the league’s member teams is variable, and can
be increased by hiring from an external market in playing talent. One of the main
objectives of this chapter has been to review and consolidate a growing body of
literature that examines the adaptation of theoretical models of resource alloca-
tion in professional sports leagues originally developed based on closed labour
market assumptions, to the open labour market case.
Key conclusions that emerge from the analysis of revenue sharing and several
forms of payroll cap are as follows. The effect of revenue sharing on competitive
inequality is neutral in a closed model. In an open model, in which team revenues
depend upon both the absolute and relative qualities of playing talent fielded by
the teams, revenue sharing reduces competitive inequality. The effect of a luxury
tax or a payroll cap on competitive inequality depends upon how the measure
is pitched: does it seek to constrain the salary expenditure of all of the league’s

member teams, or is it targeted more selectively? A luxury tax which targets only
the largest spenders should reduce competitive inequality, but if such a tax is
applied indiscriminately its effect on competitive inequality is neutral. A payroll
cap which imposes a specific limit on each team’s total salary expenditure should
reduce competitive inequality. In contrast, a payroll cap of the kind advocated by
the G14, which limits each team’s total salary expenditure to a specific proportion
of its own revenue, is likely to increase competitive inequality, by entrenching dif-
ferences that already exist between the drawing power of the league’s large- and
small-market member teams.
Notes
1 For the early history, see Quirk and Fort (1992, 1999) and Scully (1995). For recent history,
see Coon (2010) for the NBA and Vrooman (2009). This section draws upon all of these
sources.
2 For sports in which every match produces a winner and a loser, the terms ‘win ratio’ and
‘win percent’, which have been used interchangeably in the academic literature, refer to the
ratio of matches won to matches played. For sports such as football, in which some matches
are drawn or tied, an analogue of the win ratio or win percent can be constructed by taking
the ratio of total ‘points’ to matches played, where ‘points’ are awarded on a scale of 1 for a
win and 0.5 for a draw.
3 If revenues depend upon relative team quality only, the joint profit-maximisation solution
(α = 0) is degenerate. By reducing their absolute quantities of talent hired towards zero, the
teams always increase their joint profits. Revenues are unaltered as the absolute quantities
of talent are reduced, because revenues depend upon relative talent only; but costs decrease
because costs are proportional to the absolute quantities of talent. As α is progressively
lowered from α = 1 to α = 0, the SK revenue-sharing analysis envisages the teams adjusting
from the Nash equilibrium towards the degenerate joint profit-maximising equilibrium, pro-
gressively divesting themselves of all of their playing talent.
3 Competitive balance, uncertainty
of outcome and home-field
advantage
Introduction
The theoretical models of the economics of sports leagues reviewed in Chapter 2
are concerned with the degree of competitive balance or competitive inequality
between the member teams of a league competition, which in turn determines the
extent of uncertainty of outcome for individual match results and for the destin-
ation of the league championship. In Chapter 3, the emphasis shifts away from
this theoretical analysis at the level of the league championship, towards the meas-
urement and empirical investigation of competitive inequality and uncertainty of
outcome for individual match results in football.
The measurement of competitive balance or competitive inequality in a sports
league has been the subject of a large number of articles in the academic sports
economics literature in recent years. Section 3.1 reviews this literature, with par-
ticular emphasis on studies focusing on English or European football. Section 3.2
examines the nature of home-field advantage in professional team sports. Section
3.3 investigates the statistical properties of football match results data, and exam-
ines the accuracy with which several variants of the Poisson distribution and the
negative binomial distribution are able to describe the distributional properties
of match results data presented in scores format. Finally, Section 3.4 presents an
empirical investigation of persistence in sequences of consecutive match results.
Does a winning streak help build a team’s confidence, making it more likely that
further matches will also be won? Or does it lead to complacency, increasing the
likelihood that the next match will be drawn or lost? A Monte Carlo simulation
exercise is used to address these questions.
3.1 Measuring competitive balance and competitive inequality

In the North American sports economics literature, the measurement of com-
petitive inequality often focuses upon the dispersion in win ratio or win percent
between the member teams of a sports league. In sports where every match pro-
duces a winner and a loser, and there are no draws or ties, the cross-sectional
42
Measuring competitive balance and competitive inequality 43
dispersion in win ratios over an entire season can be measured using the standard
deviation:
n
∑ (w − 0.5)
2
i
σ= i =1 [3.1]
n
In [3.1], wi is team i’s win ratio (number of wins divided by number of matches
played), and n is the number of teams. The higher is the value of σ, the greater
is the level of competitive inequality. One difficulty with this measure, however,
is that n and the number of matches played are not always constant in the same
league over time, and are usually not the same for different sports. Accordingly,
observed values of σ measured for the same sport over time, or measured for
different sports, are not always comparable. To deal with this difficulty, σ can be
benchmarked against the standard deviation that would describe dispersion for a
league with perfect competitive equality, in which (ignoring home-field advantage)
every team has a probability of 0.5 of winning every match (Noll, 1988; Scully,
1989; Quirk and Fort, 1992; Fort and Quirk, 1995). The ideal standard deviation
is 0.5/√m, where m is the number of matches played by each team. The relative
competitive inequality measure is:
[3.2]
Relative σ = σ /( 0.5 / m )
For a league of perfect competitive equality, the relative σ measure should not
be significantly different from one. The higher is relative σ, the greater is the level
of competitive inequality. This measure is used to measure competitive balance or
competitive inequality by Vrooman (1995), Quirk and Fort (1992), Humphreys
(2002), Zimbalist (2002) and Schmidt and Berri (2003), among others. Fort (2006)
reviews the application of the relative σ measure to win-ratio data for MLB, the
NFL, NBA and National Hockey League (NHL).
Although the relative σ measure is computationally straightforward and simple
to interpret, it has been subject to criticism. Since teams playing at home tend to
win more often than they lose, two teams of equal playing strength are not equally
likely to win (Trandel and Maxcy, 2009). This introduces an upward bias into the
value of the ‘ideal’ standard deviation that is used in the denominator of [3.2],
causing the relative σ measure to understate the degree of competitive inequal-
ity. Owen (2010) shows that the relative σ measure is more sensitive to changes
in the number of teams and the number of matches played than is the unadjusted
standard deviation, [3.1]. This may lead to misleading comparisons of competitive
inequality across leagues or over time, if the numbers of teams and/or matches
played are not constant.
By extension, Eckard (1998) examines the effect on competitive inequality in
American college football of a range of restrictions on player recruitment, eligi-
bility and compensation that were introduced by the NCAA (National Collegiate
44 Competitive balance and home-field advantage
Athletic Association) in 1952. Pooled cross-section time-series data are used to

decompose the overall variance in win ratios into a time component for each team
and a cumulative component across teams:
∑ ∑ (w i,t − 0.5)2 / nT = ∑ ∑ (w i,t − w i )2 / nT + ∑ (w i − 0.5)2 / n [3.3]

i t i t i
where wi,t is team i’s win ratio in season t; w i = ∑ w i,t / T ; n is the number of teams;
t
and T is the number of seasons. Post-1952, there was a reduction in the first term and
an increase in the second term on the right-hand side of [3.3]. This suggests that the
regulations made competition more unequal: differences between strong and weak
teams became more entrenched, resulting in a reduction in the variation of each
team’s performance over time, and an increase in the variation in average perform-
ance between teams. Eckard (2001) applies similar methods to MLB data, finding an
increase in competitive balance following the introduction of free agency in 1976.
Humphreys (2002) uses the competitive balance ratio (CBR) measure to iden-
tify trends in competitive inequality over time, taking account of both the time-
series variation in each team’s win ratio calculated over a number of seasons, and
the cross-sectional variation in the teams’ win ratios within each season.
The standard deviation of team i’s win ratio measured over T seasons is:

T T
σ i ,T = ∑ ( w i,t − w i )2 / T , where w i = ∑ w i,t / T
t =1 t =1
The standard deviation of the win ratios measured over n teams in season t is:
n

σ n,t = ∑ ( w i,t − 0.5 )2 / n
i =1
The CBR is:

N T
CBR = σ T / σ n , where σ T = ∑ σ i,T , σ n = ∑ σ n,t [3.4]
i =1 t =1
An increase in the CBR indicates a reduction in competitive inequality, driven

by either an increase in ‘churning’ in the teams’ win ratios from season to season
(increase in σT), or a reduction in the dispersion of the teams’ win ratios within
each season (decrease in σn). For MLB, Humphreys finds CBR has increased over
time.
An issue that arises when applying the relative σ measure to football is how
to deal with drawn matches (Cain and Haddock, 2006; Fort, 2007). A simple
approach is to record a draw as half of a win, and calculate the win ratio on this
basis. For English football, Szymanski (2001) reports annual results from 1977 to
1999. Buzzacchi, Szymanski and Valletti (2003) report ten-year averages for the top
tier of English football, which show an increase in competitive inequality to the
end of the 1980s, followed by a reduction during the 1990s. Bourg (2004) reports
five-year averages for the ‘Big 5’ European leagues (England, France, Germany,
Italy and Spain). Between the 1980s and 2000 there was an increase in competitive
inequality in France and Germany, and a reduction in England, Spain and Italy.
Kringstad and Gerrard (2007) report ten-year averages for the period 1966–2006.
Competitive inequality was lower in the first ten-year period compared to the most
recent period in all cases except Italy. Lenten (2008) examines the implications for
competitive inequality of the introduction of an unbalanced playing schedule in
the Scottish Premier League in the 2001 season. An unbalanced schedule creates
biases in the teams’ win ratios, and so has implications for the measurement of
competitive inequality.
Koning (2000) approaches the measurement of competitive inequality by using
a match-level measure of uncertainty of outcome. An ordered probit model is
fitted to match results data for the top tier in Dutch football, in order to estimate
a team quality coefficient for each team. The cross-sectional standard deviation
of these coefficients in each season is interpreted as a measure of competitive
inequality. Marques (2002) adopts a similar approach in a study of the trend in
competitive inequality in Portuguese football. It appears that this method should
produce results very similar to those obtained from an analysis of the trend in the
dispersion of win ratios.
Goossens (2006) suggests complete predictability might represent a better
benchmark for the standardisation of [3.1] than perfect competitive equality. The
National Measure of Seasonal Imbalance (NAMSI) takes values between zero
and one, with a zero value representing the case where every team achieves a win
percentage of 0.5, and a value of one representing the case of complete predict-
ability, when the strongest team wins all of its matches, the second-strongest team
wins all of its matches except those against the strongest team, and so on:
n
∑ (w − 0.5)
2
i
NAMSI = i =1
n
∑ (w − 0.5)
2
i,max [3.5]
i =1
where wi,max is the win ratio achieved by team i when there is complete predictabil-
ity. Between 1963 and 2005, the NAMSI measure indicates no significant change
in competitive inequality for Germany and France, and a small but significant
increase for Belgium and England. For several other countries, including Spain,
Italy, the Netherlands and Denmark, there is no clear trend.
Several alternative measures of dispersion have also been used to measure com-
petitive inequality. Several US studies use the Lorenz curve and Gini coefficient
(Fort and Quirk, 1995; Schmidt, 2001; Schmidt and Berri, 2001; Larsen, Fenn and
Spenner, 2006). The Lorenz curve shows the variation in the cumulative win ratio
of the j most successful teams, as j varies from 1 to n (where n is the total number
of teams). Figure 3.1 illustrates. The teams are represented horizontally, from the
Cumulative
win ratio up to A
team j
O D
Teams (j=1,...,n) arranged from highest to lowest win ratio
Figure 3.1 The Lorenz curve and Gini coefficient
most to the least successful (reading from left to right) along the horizontal axis.
The vertical axis shows the cumulative win ratio (the sum of the win ratios of all
teams from team 1 to team j, as a function of j). If all teams have the same win
ratio of 0.5, the Lorenz curve is the 45-degree line OCA. If the distribution of win
ratios is skewed, the Lorenz curve is the concave curve OBA. With reference to
Figure 3.1, the Gini coefficient is defined as follows:
n j
area of the crescent between OBA and OCA

∑∑w i
j=1 i=1
G= = n
[3.6]
area of the triangle ODA
0.5(n +1)∑ w i
i=1
The minimum value G = 0 describes the case in which all teams have the same win
ratio of 0.5. Since no team can win all of the league’s matches, however, the max-
imum theoretical value of G = 1 is unattainable. Reporting G in unadjusted form
tends to understate the level of competitive inequality (Utt and Fort, 2002).
In reporting results for English League football throughout this volume, we
refer to football seasons by their end-year. For example, the 2008–2009 season
is referred to as the 2009 season. The official nomenclature at the time of writ-
ing (during the 2010 season) for the four tiers (divisions) of English football’s
Premier League and Football League is ‘Premiership’ (the Premier League) and
‘Championship’, ‘League One’ and ‘League Two’ (the three tiers of the Football
League). In the rest of this chapter, and throughout this volume, these four tiers
(divisions) are known as T1, T2, T3 and T4 (tiers one to four).
Michie and Oughton (2004) plot Lorenz curves for T1 for selected seasons
between 1950 and 2004. Goossens (2006) reports Gini coefficients for several
European leagues for the period 1963–2005. The least unequal were Denmark,
Sweden, Germany and Italy, and the most unequal were Portugal, the Netherlands,
Greece and Spain. Intermediate cases include France, Belgium and England.
The Herfindahl index (HI), defined as the sum of the squared win ratios, can
also be used as a measure of competitive inequality:
n
HI = ∑ w 2i
i =1
[3.7]
An increase in HI signifies an increase in competitive inequality. For England,

Michie and Oughton (2004) find the HI for T1 to have remained roughly unchanged
between 1947 and the 1980s, and to have risen thereafter. For the period as a
whole, the HI increased by around 13 per cent. However, part of the increase was
due to a reduction in the number of teams from twenty-two until the 1989 season,
to twenty from the 1996 season onwards (with several variations between twenty
and twenty-two between the 1990 and 1995 seasons).
Depken (1999) notes that HI is bounded from below by 1/n, and proposes an
adjusted Herfindahl index, dHI = HI – 1/n. Variations over time in n affects both
the lower bound of HI and the upper bound. Using the dHI measure, Depken
finds that competitive inequality in MLB over the period 1920–96 became more
stable over time. Michie and Oughton (2004) standardise HI using SHI = 100
[HI/(1/n)]. In a perfectly balanced league, SHI = 100. For the top tier of English
football, SHI increased by 21 per cent between 1947 and 2004, with most of this
increase having occurred after the late 1980s. For the Scottish Premier League,
the patterns in SHI and the Gini coefficient for seasons 1997 to 2006 are similar
(Lenten, 2008).
Applied to win ratio data, the m-team concentration ratio provides a meas-
ure of competitive inequality that focuses on the performance of the top teams.
Michie and Oughton (2004) report a five-team concentration ratio, denoted C5,
and a standardised version, denoted C5B, in which the benchmark is a league in
which all teams have an identical win ratio of 0.5:
5
C5 = ∑ w j /( 0.5n ); C5B = C5 /(5 / n ) [3.8]
j=1
In a similar vein, Goossens (2006) tracks the trend in the numbers of different
teams achieving a top three position over five consecutive seasons for several
European countries. For England, Curran, Jennings and Sedgwick (2009) report
the ratio of the number of top four positions occupied by the four most successful
teams over a ten-year period to the maximum number of top four positions that
could be occupied by four teams (= 40), for the period 1949–2008. This measure
reflects the ascendancy of Arsenal, Chelsea, Liverpool and Manchester United
during the 2000s (see also Chapter 6, Section 6.1).
Groot (2008) proposes a ‘surprise index’ measure of competitive inequality, based

on the realised number of surprise points and the maximum number of surprise
points available if teams are perfectly balanced (every team wins its home matches
or every game ends in a draw). Surprise points are earned when a lower-ranked
team in the final league table either wins (two surprise points) or draws (one point)
a match against a higher-ranked team. Surprise points are multiplied by the diffe-
rence in rankings and aggregated. In a league with maximum competitive inequality
and no surprises, the surprise index is zero: the top team wins all its matches; the
second-placed team wins all its matches except those against the top team; and so
on. For a highly balanced league, the surprise index approaches its maximum value
of one. For the top tier in English football, for example, there is a clear upward
trend in competitive inequality over the twentieth century as a whole.
Measures of concentration or inequality that are used widely in industrial eco-
nomics, such as the concentration ratio, the Herfindahl index and Gini coefficient,
can also be used to measure the concentration of success in terms of the number
of championship wins per team recorded over a number of seasons. Using the HI,
Kringstad and Gerrard (2007) report an increase in concentration in England,
Germany and Italy between 1966 and 2006, and a decrease in France and Spain.
Szymanski and Kuypers (1999) fit Lorenz curves to data on the number of cham-
pionship wins over the period 1946–98 for five countries.
Lenten (2009a) uses a dynamic measure of competitive inequality, which com-
pares the win ratios of each team in consecutive seasons, in order to produce a
one-period metric of change that takes the form of a mobility gain function. This
approach offers a bridge between the within-season and between-season analyses.
The method is applied to data from the Australian Football League (AFL) and
the National Rugby League (NRL). There was no significant change in competi-
tive inequality in the AFL, but there was a decline in competitive inequality in the
NRL during the 1950s.
Hadley, Ciecka and Krautmann (2005) focus on the dynamics of final league
standings, rather than win ratios, by designating MLB teams as ‘winners’ (quali-
fiers for the play-offs) or ‘losers’ (non-qualifiers), and estimating transition prob-
abilities that, from one season to the next, a winner will remain a winner; a winner
will become a loser; a loser will become a winner; and a loser will remain a loser.
After 1994 (during the post-strike era) the likelihood of a winner remaining a
winner in the following season increased by nearly threefold (while the number of
teams advancing to the play-offs doubled). This suggests an increase in competi-
tive inequality measured by transition probabilities between winning and losing
states.
Mizak, Neral and Stair (2007) use an index for measuring competitive inequal-
ity based on changes in team standings over time. The year-to-year average team
movement in the standings is:
n
C t = ∑ |ri,t − ri,t −1 |/n [3.9]
i =1
where Ct is the churn in team standings for year t, |ri,t – ri,t-1| is the absolute value of
team i’s change in rank from season t–1 to season t, and n is the number of teams.
The maximum possible value of Ct depends on n. Since n can vary over time, the
churn index is divided by its maximum value to produce a standardised measure.
The standardised churn varies between zero (no change in league standings from
one year to the next) and one (maximum possible change in league standings from
one year to the next). According to the adjusted churn index, there has been a
decline in competitive inequality in MLB since the 1990s, especially in the AL.
For football’s top tiers in England, France, Germany, Italy and Spain, Kringstad
and Gerrard (2007) report Spearman rank correlation coefficients (SRC) between
the teams’ final league standings in consecutive seasons. If there is no churning
and the teams finish in the same order, SRC is one; if there is no relationship
between league standings in consecutive seasons, SRC is close to zero. Kendall’s
tau statistic, an alternative measure of rank correlation, is used in Groot’s (2008)
long-term analysis of competitive inequality in European football. For the top
tier in English football, Kendall’s tau increased from 0.17 in 1888 to 0.38 in 2005.
Most of the decline in churning occurred after 1947; between 1888 and 1939 there
was a weak trend in the opposite direction.
Several authors apply the tools of time-series econometrics, such as unit root
tests and structural break tests, to indicators of competitive inequality in sports
leagues (Schmidt, 2001; Schmidt and Berri, 2001, 2003; Lee and Fort, 2005;
Brandes and Franck, 2007; Fort and Lee, 2007; Lenten 2009b). For example, Lee
and Fort (2009) search for structural breaks in regressions that describe the trend
in the relative σ measure, and in an m-team concentration ratio, for English top-
tier football. Both competitive inequality indicators are based on league points
(rather than win ratios). The explanatory variables used in the regressions are
a constant and a time trend. The analysis identifies break points in 1901, 1906,
1937, and 1995 or 1997 (or both). The analysis is consistent with the notion that
competitive inequality increased following the formation of the Premier League
at the start of the 1993 season.
Dobson and Goddard (2004) are critical of the direct application of competitive
inequality measures, such as the standard deviation of win ratios, which were devel-
oped with US major league sports in mind, to data from the top tiers of European
football. Such measures fail to take into account the tiered competitive structure of
the European leagues. Intra-tier variation in win ratios might say something about
competitive inequality within an entire league; but such measures seem unlikely to
provide a powerful indication. Most of the action is inter-rather than intra-tier, but
inter-tier inequality is not considered. In England, however, all Premier League and
Football League teams take part in a competition that does permit direct comparisons
between the playing strengths of teams from different tiers. The FA Cup (Football
Association Challenge Cup) is a sudden-death knockout tournament involving both
league and non-league teams. For spectators, a major attraction is the cup’s propen-
sity to produce shock results, such as the elimination of a Premier League team by an
opponent from the lower reaches of the Football League or from non-league.
Szymanski (2001) uses FA Cup match attendance data in an effort to identify

trends in competitive inequality in English football. Analysing cup attendances
in matches where a corresponding league fixture (between the same teams in the
same season) took place, Szymanski finds that cup attendances declined relative to
league attendances. In the demand for sports literature, one version of the uncer-
tainty of outcome hypothesis asserts an inverse relationship between match-level
uncertainty and match attendance. Accordingly Szymanski interprets declining
cup attendances as evidence of increasing inter-tier competitive inequality. League
attendances are less affected by a rise in competitive inequality, which is primarily
an inter- rather than an intra-tier phenomenon.
Dobson and Goddard (2004) measure inter-tier competitive inequality directly
using FA Cup match results data, rather than indirectly via a hypothesised and
unsubstantiated relationship with match attendance. The empirical evidence for
any relationship between match-level uncertainty of outcome and match attend-
ance is widely recognised as being rather weak and unconvincing (see Chapter 11,
Section 11.1). A statistical analysis of win probabilities in FA Cup results suggests
there was an increase in competitive inequality during the 1920s and 1930s, and a
further increase during the mid to late 1970s, 1980s and 1990s. Chapter 6, Section
6.1 presents some of the more recent data on ‘giant-killings’, identifying the fre-
quency with which lower-ranked teams overcame higher-ranked opponents (from
higher tiers) in FA Cup ties played between the 1970s and 2000s.
3.2 Home-field advantage

Home-field advantage, a pervasive feature of professional team sports, has been
the subject of extensive academic research, reviewed previously by Courneya
and Carron (1992), Nevill and Holder (1999) and Carron, Loughhead and Bray
(2005). The sources of home-field advantage in any sport are difficult to iden-
tify or quantify empirically with any precision, but fall generally into four main
categories: familiarity (or unfamiliarity) with the home team’s stadium and
facilities; disruption of the away team’s preparation due to the need to travel;
various types of effect of the crowd on the home team (encouragement), the
away team (intimidation) or match officials (pressure leading, either consciously
or subconsciously, to biased decisions favouring the home team); and in cer-
tain sports, rules that may favour the home team.1 Summarising a number of
studies, Nevill and Holder (1999) find that the weight of evidence suggests that
the crowd is the most important source of home-field advantage, and that the
influence of the crowd on the match result operates mainly through the match
officials, rather than through any direct effect on the performance of the home-
or away-team players.
Several aggregate measures of home-field advantage are employed in the aca-
demic literature. Nevill, Newell and Gale (1996) and Attrill et al. (2008) use
percentage of matches won by the home and away teams; and Jacklin (2005)
Home-field advantage 51
uses the ratio of home wins to away wins. Pollard and Pollard (2005) use the
ratio of league points won to total points available to compare long-term trends
in home-field advantage in several North American professional team sports
with English football. Home-field advantage in most of the sports examined
was greatest in the earliest years of the sport’s existence. Home-field advantage
shows little evidence of any long-term trend in MLB and the NFL, but may
have increased slightly in the latter in recent years. Home-field advantage in
baseball is consistently lower than in the other sports examined. Home-field
advantage in the NBA and in hockey (the NHL), and in English football, has
declined in recent decades.2
Using cross-sectional data on seventy-two countries and six football seasons
from 1999 to 2004 inclusive, Pollard (2006) compares the size of the home-field
advantage effect by country. In Europe, home-field advantage is above average
in several countries in the Balkans, and lower than average throughout most
of northern Europe, including the Baltic, Scandinavia and the British Isles. In
Latin America, home-field advantage is above average in the Andean countries
Bolivia, Peru, Ecuador and Colombia, and below average in the southern coun-
tries Argentina, Paraguay and Uruguay. Pollard and Gomez (2009) report a more
detailed analysis for south-west Europe (France, Italy, Portugal and Spain).
Page and Page (2007) examine home-field advantage in two-leg ties played on a
knockout basis in European football club competition. Using data extending back
to the inception of European club competition in the mid-1950s, teams drawn to
play the away leg first and the home leg second are more likely to be successful in
the tie. The magnitude of this effect has diminished over time. Explanations for
this effect relating to competition rules (in some cases stronger teams were seeded
and guaranteed to play the second leg at home; and European ties that are level
at the end of 90 minutes’ play in the second leg are extended for an extra 30 min-
utes, conferring an additional advantage on the home team in the second leg) are
investigated, but discounted as explanations for the entire discrepancy between
the performance of the teams playing at home and away in the second leg. Poulter
(2009) examines home-field advantage in the Champions League over the period
2001–7.
Brown et al. (2002) analyse 3,914 match results played by the 32 international
teams that participated in the 1998 World Cup in France. The matches were played
between January 1987 and the end of the World Cup on 12 July 1998. The impact
on home-field advantage of three factors is examined: familiarity with the play-
ing facility, the importance of the match and the distance travelled by the away
team. Home-field advantage is stronger when the home team played in a familiar
stadium. Home-field advantage is positively related to distance travelled by the
away team.
For English club football, Pollard and Pollard (2005) note that home-field
advantage was several percentage points higher before the Second World War
than it was when football resumed afterwards, following a seven-year wartime
break. This shift is interpreted as circumstantial evidence that familiarity may

have been an important source of home-field advantage (players new to the
sport when football resumed would have been equally unfamiliar with their own
and their opponents’ stadia), and that crowd effects may have been unimportant
(English football matches attracted record attendances in the immediate post-
war seasons).
Among other studies with a more specific focus, Barnett and Hilditch (1993)
report that four English clubs that used artificial playing surfaces during the 1980s
and early 1990s (Luton Town, Oldham Athletic, Preston North End and Queen’s
Park Rangers) enjoyed a moderately enhanced home-field advantage. Clarke
and Norman (1995) analyse home-field advantage at club level using 1980s data.
Home-field advantage does not vary significantly by tier, but it does vary over
time. London clubs had a below-average home-field advantage, and home-field
advantage effects had greater leverage on winning than on goal margins. Pollard
(2002) finds home advantage is somewhat reduced when a team moves to a new
stadium.
Using data from the 1993 season, Nevill, Newell and Gale (1996) find that
home-field advantage in English and Scottish football was associated with the
average attendance in each tier. More penalties were awarded to the home team
and fewer home-team players were sent off in the higher tiers with large attend-
ances. Boyko, Boyko and Boyko (2007) and Johnston (2008) report conflicting
findings concerning the impact of the crowd size on home-field bias. Each of these
studies also considers whether referee bias contributes to home-field advantage.
Page and Page (2010) report that the magnitude of home-field advantage depends
upon the referee, and that this relationship is moderated by the size of the crowd.
Some referees are more prone to be influenced by the crowd than others (see also
Chapter 10).
Travel may contribute to home-field advantage, if players become tired after
travelling long distances, often on the day of the match for lower-tier teams in par-
ticular. Anecdotally, Koyama and Reade (2010) note that Carlisle United, one of
the most remotely located English clubs, won sixteen consecutive home matches
during the 2008 season. However, when T3 and T4 were organised on a regional
basis (Division 3 North and South, seasons 1921 to 1958 inclusive), home-field
advantage was greater than it became subsequently, even though the distances
travelled by away teams were smaller. Several other studies suggest distance exerts
only a weak effect on home-field advantage (Pace and Carron, 1992; Nevill and
Holder, 1999; Carron, Loughhead and Bray, 2005; Pollard, 2006).
Other factors that have been proposed as causes of home advantage include
territoriality and psychology. Defending the home ground may evoke sentiments
associated with territoriality. Neave and Wolfson (2003), for example, reported
higher concentrations of testosterone in players before a home match than before
an away match. Several researchers have sought explanations for home-field
advantage in player psychology. For example, Waters and Lovell (2002) report
Home-field advantage 53
that players expressed greater confidence when playing at home, and were more
optimistic about the likely outcome.
Table 3.1 presents an analysis of home-field advantage in the English league
over a forty-year period from the 1970 season to the 2009 season. The tabulation
shows the proportions of matches in all four tiers that resulted in home wins,
draws and away wins, and the average numbers of goals scored per match by
the home and away teams and in total. The final two columns report summary
measures of home-field advantage, in the form of home-team success percentages.
The penultimate column reports the percentage of league points that were gained
by home teams, calculated using a scale of 2/1/0 (two points for a win, one for a
draw and zero for a loss). This was the scoring system for league points until the
end of the 1981 season. Subsequently, three league points have been awarded for
a win; and from the 1982 season onwards the final column reports the equivalent
home-team success percentages calculated using a scale of 3/1/0.
The change in the points scoring system results in an increase of about one
per cent in the home-field advantage measure. A continuous series based on a
common scoring system therefore appears preferable to one that switches mid-
way from 2/1/0 to 3/1/0. Because the 2/1/0 series gives equal weighting to every
match (whereas the 3/1/0 series gives more weight to matches that are won or lost
than it gives to draws); and because the 2/1/0 series is directly comparable to a
win ratio or win-percent measure for American sports which do not permit draws
(ties), the 2/1/0 series is our preferred home-field advantage measure. On the 2/1/0
measure, average home-field advantage for the period 1980–2003 is 61.1 per cent.
Subsequently there has been a slight reduction, to an average of 58.0 per cent for
the 2004–9 period. For the period 1980–2003, the corresponding figures quoted by
Pollard and Pollard (2005) for North American major league sports are 53.7 per
cent for MLB, 57.2 per cent for NHL, 58.0 per cent for NFL and 62.6 per cent for
NBA.
Table 3.2 reports average values of the 2/1/0 home-field advantage measure by
tier (division), calculated over five-season intervals over the same period. In all
tiers, there has been a marked improvement in the average performance of away
teams, and a corresponding deterioration in the performance of home teams. The
reduction in home-field advantage has been more pronounced in the lower tiers
than in the higher tiers. In 1970–4, T1 had the lowest home-field advantage and T4
the highest; but by 2005–9 these positions had been reversed.
A particularly striking feature of Table 3.1 is the large jump in the average
number of goals scored by away teams, which coincided with the introduction
of the award of three league points (rather than two) for a win, from the 1982
season onwards. The improvement in the performance of away teams since the
early 1980s was initially (in the 1980s) due to an improvement in their offensive
or goal-scoring capability, but has been sustained subsequently (in the 1990s
and 2000s) by an improvement in their defensive capability to avoid conceding
goals.
Table 3.1 Percentages of home wins, draws and away wins, average numbers of goals scored by the home and
away teams, and win percentages per season, English league, 1970–2009 seasons
Average Average Home-team Home-team

goals per goals per win per cent win per
Home wins Draws Away wins match: match: (2/1/0 pts) cent
Season (per cent) (per cent) (per cent) home team away team (3/1/0 pts)
1970 49.9 28.6 21.5 1.64 1.00 64.2 —

1971 48.8 28.3 22.9 1.54 0.99 63.0 —
1972 51.4 27.6 21.0 1.62 0.97 65.2 —
1973 51.7 28.4 19.9 1.57 0.92 65.9 —
1974 48.9 30.2 21.0 1.51 0.92 64.0 —
1975 52.4 27.9 19.7 1.59 0.92 66.4 —
1976 50.2 27.7 22.1 1.59 0.99 64.1 —
1977 51.9 27.9 20.2 1.66 0.99 65.9 —
1978 50.1 30.9 19.0 1.62 0.98 65.5 —
1979 48.0 29.6 22.4 1.55 1.01 62.8 —
1980 49.9 27.9 22.2 1.58 0.98 63.8 —
1981 49.6 28.0 22.4 1.53 0.94 63.6 —
1982 47.1 27.4 25.5 1.53 1.07 60.8 61.9
1983 52.0 26.3 21.6 1.72 1.06 65.2 66.6
1984 50.8 26.0 23.2 1.69 1.06 63.8 65.1
1985 50.1 24.3 25.6 1.62 1.10 62.3 63.4
1986 50.1 24.6 25.3 1.69 1.11 62.4 63.5
1987 48.9 27.4 23.7 1.55 1.05 62.6 63.9
1988 46.1 27.3 26.6 1.55 1.08 59.8 60.7
1989 46.8 28.6 24.6 1.58 1.08 61.1 62.3
1990 46.8 27.3 25.9 1.54 1.09 60.4 61.5
1991 48.4 27.6 24.1 1.56 1.06 62.2 63.4
1992 47.5 26.9 25.5 1.52 1.08 61.0 62.1
1993 46.5 26.4 27.1 1.58 1.12 59.7 60.7
1994 45.6 27.4 27.0 1.53 1.14 59.3 60.2
1995 46.4 27.5 26.1 1.51 1.08 60.1 61.2
1996 45.0 29.6 25.3 1.48 1.06 59.8 60.9
1997 46.4 28.1 25.5 1.47 1.03 60.4 61.5
1998 47.9 27.8 24.3 1.51 1.05 61.8 63.0
1999 45.0 28.0 27.0 1.46 1.08 59.0 59.9
2000 45.5 27.1 27.4 1.46 1.07 59.1 60.0
2001 46.5 27.4 26.2 1.50 1.07 60.1 61.2
2002 46.7 26.5 26.8 1.54 1.09 59.9 60.9
2003 44.3 27.0 28.7 1.48 1.15 57.8 58.5
2004 45.2 27.9 26.9 1.50 1.10 59.1 60.1
2005 44.3 28.6 27.1 1.46 1.10 58.6 59.5
2006 43.3 29.5 27.2 1.42 1.07 58.1 59.0
2007 46.4 24.7 29.0 1.43 1.08 58.7 59.5
2008 43.4 26.5 30.2 1.41 1.12 56.6 57.2
2009 43.0 27.5 29.5 1.43 1.11 56.8 57.5
Source: Rothmans/Sky Sports Football Yearbook

Table 3.2 Home-team success percentages (2/1/0 points for W/D/L) by tier
(division) in five-season bands, English league, 1970–2009 seasons
Seasons T1 T2 T3 T4 All tiers
1970–1974 62.5 64.9 63.7 66.4 64.4

1975–1979 63.5 64.2 66.0 65.6 64.9
1980–1984 63.5 63.2 63.9 63.2 63.4
1985–1989 61.1 62.3 61.3 61.8 61.6
1990–1994 60.1 61.1 60.9 59.8 60.5
1995–1999 59.9 60.8 60.2 60.0 60.2
2000–2004 59.8 59.0 58.4 59.9 59.2
2005–2009 59.8 58.1 57.6 56.1 57.7
Inspection of Table 3.1 raises the question whether the long-term erosion of the
importance of home-field advantage is attributable solely to changes in incentives
arising from the introduction of three rather than two league points for a win
(Jacklin, 2005; Dilger and Geyer, 2009; Moschini, 2010), or whether it reflects
changes in the underlying ‘technology’ defining the processes which generate
match results. Such changes might include improvements in training methods
enabling players to perform more consistently away from home; improvements
in the quality of transportation making away travel less onerous; or improve-
ments in professionalism, making footballing ability a more decisive influence
on match results, or making psychological effects on players or officials deriving
from the attitude of the crowd less decisive (Smith, 2003). Arguing along similar
lines, Koyama and Reade (2010) suggest that an increase in the opportunity for
supporters to scrutinise players’ performances away from home, resulting from
the increasing scope of TV coverage, has raised incentives for players to perform
to their maximum ability in all matches.
Because England’s adoption of three-points-for-a-win pre-dated the
implementation of the same change in a number of other leading football
nations by more than a decade, some international comparisons of home- and
away-team win ratios and goal-scoring records should shed some light on this
matter. Table 3.3 summarises the relevant data for teams in the top tiers of
England, France, Germany, Scotland and Spain, for seasons 1973 to 2009
inclusive. For England the data are aggregated into three nine-season and two
five-season periods (seasons 1973–1981, 1982–1990, 1991–1999, 2000–2004
and 2005–2009, inclusive). For the other four countries, the data are similarly
aggregated, except for the 1991–9 period, which is split into two subperiods at
the point of introduction of three-points-for-a-win: the 1995 season in France
and Scotland, and the 1996 season in Germany and Spain.
Table 3.3 Percentages of home wins, draws and away wins, and average numbers of goals scored by home and away
teams, international comparisons, 1973–2009 seasons
Average Average
goals per goals per
Home wins Draws Away wins match: match: Home-team
Seasons (per cent) (per cent) (per cent) home team away team win ratio
England
1973–1981 50.0 28.0 21.9 1.58 1.00 0.641
1982–1990* 48.6 26.1 25.3 1.59 1.06 0.617
1991–1999* 45.6 28.6 25.8 1.51 1.09 0.599
2000–2004* 46.8 25.9 27.3 1.54 1.13 0.598
2005–2009* 47.2 25.4 27.5 1.47 1.05 0.598
France
1973–1981 57.0 25.1 18.0 1.92 1.01 0.696
1982–1990 55.1 27.4 17.4 1.67 0.87 0.688
1991–1994 51.7 31.7 16.6 1.45 0.76 0.676
1995–1999* 49.2 31.0 19.8 1.47 0.87 0.647
2000–2004* 50.1 27.1 22.8 1.46 0.93 0.636
2005–2009* 45.3 31.3 23.4 1.32 0.90 0.609
Germany
1973–1981 54.3 25.5 20.1 2.29 1.32 0.671
1982–1990 50.9 27.7 21.5 2.18 1.26 0.648
1991–1995 45.4 31.2 23.4 1.76 1.19 0.610
1996–1999* 44.9 29.3 25.8 1.73 1.22 0.596
2000–2004* 50.1 24.4 25.5 1.72 1.16 0.623
2005–2009* 46.0 25.6 28.4 1.62 1.22 0.588
Table 3.3 (cont.)
Average Average
goals per goals per
Home wins Draws Away wins match: match: Home-team
Seasons (per cent) (per cent) (per cent) home team away team win ratio
Scotland
1973–1981 46.7 24.7 28.5 1.62 1.21 0.591
1982–1990 45.4 25.3 29.3 1.53 1.14 0.581
1991–1994 40.7 29.3 30.0 1.36 1.11 0.554
1995–1999* 43.4 27.3 29.2 1.51 1.16 0.571
2000–2004* 45.0 22.5 32.5 1.57 1.23 0.562
2005–2009* 44.6 23.3 32.0 1.46 1.15 0.563
Spain
1973–1981 61.6 25.1 13.3 1.74 0.79 0.742
1982–1990 53.6 26.9 19.5 1.57 0.88 0.671
1991–1995 50.4 28.4 21.2 1.51 0.94 0.646
1996–1999* 47.9 27.1 25.0 1.60 1.09 0.615
2000–2004* 48.2 26.9 24.9 1.58 1.10 0.616
2005–2009* 46.9 24.9 28.2 1.49 1.13 0.594
Note: * denotes three league points awarded for a win.

Distributional properties of goals scored 59
The results for France, Germany and Spain indicate that long-term decline in the
size of home-field advantage is an international phenomenon, and has progressed
further in each of these three countries than in England. This partly reflects the
fact that home-field advantage counted for more in these three countries than it
did in England in the 1970s. Since then a process of international convergence has
been underway. In France and Spain, the short-term impact of the introduction
of three-points-for-a-win in the mid-1990s was the same as in England at the start
of the 1980s: an immediate increase in the average number of goals scored by the
away team, and a corresponding increase in the proportion of matches finishing
in away wins. In Germany most of the erosion of the importance of home-field
advantage appears to have occurred during the 1970s and 1980s, before the intro-
duction of three-points-for-a-win.
The proportion of away wins was higher in Scotland than in any of the other
four countries throughout the period covered by Table 3.3. Although there has
been some erosion of home-field advantage and a corresponding improvement
in the performance of away teams, the shift has been smaller than in any of the
other four countries. The relatively strong performance of away teams may reflect
a high level of competitive imbalance in the Scottish league: home advantage is
less important in matches between teams that are highly unequal. The Scottish
data may also be affected by a series of changes to the league and divisional struc-
ture during the 1970s, 1980s and 1990s, which had implications for competitive
balance. The introduction of three-points-for-a-win in the 1995 season coincided
with a reduction in the number of teams in the top division from twelve to ten.
Other things being equal this would be expected to reduce competitive inequality,
and increase the importance of home-field advantage. This latter effect appears to
have dominated the three-points-for-a-win effect, since the home-team win ratio
was higher in 1995–9 than it was in 1991–4.
3.3 Distributional properties of the goals scored by the home and away teams
Section 3.3 investigates the statistical properties of football match results data,
and examines the use of several variants of the Poisson distribution and the nega-
tive binomial distribution that are able to describe the distributional properties of
match results data presented in scores format. The analysis begins by reporting, in
Tables 3.4 and 3.5, the percentage distributions of the numbers of goals scored in
each match by the home and away teams based on league match results between
the 1970 and 2009 seasons (inclusive). Table 3.4 reports the results for the twelve
seasons prior to the introduction of three-points-for-a-win, from 1970 to 1981,
inclusive; and Table 3.5 reports the results for the twenty-eight seasons subse-
quently, from 1982 to 2009, inclusive.
For example, Table 3.5 indicates that 8.1 per cent of matches during the 1982–
2009 period finished as 0–0 draws, 10.9 per cent finished as 1–0 wins for the home
team and 7.8 per cent finished as 0–1 wins for the away team. A 1–1 draw was the
Table 3.4 Joint and marginal percentage distributions of goals scored by home and
away teams, English league, 1970–1981 seasons
Goals scored by the away team

0 1 2 3 4 5 6+ Total
Goals scored by 0 9.3 6.8 3.3 1.2 0.3 0.1 0.0 20.8
the home team 1 11.8 13.2 5.1 1.9 0.5 0.2 0.0 32.6
2 9.2 9.3 5.1 1.4 0.3 0.0 0.0 25.5
3 5.0 4.7 2.4 0.9 0.2 0.0 0.0 13.2
4 2.1 1.8 1.0 0.3 0.1 0.0 0.0 5.3
5 0.7 0.6 0.4 0.1 0.0 0.0 0.0 1.9
6+ 0.3 0.2 0.2 0.0 0.0 0.0 0.0 0.7
Total 38.3 36.6 17.4 5.9 1.4 0.3 0.1 100.0
Source: Rothmans Football Yearbook
Table 3.5 Joint and marginal percentage distributions of goals scored by home and
away teams, English league, 1982–2009 seasons

0 1 2 3 4 5 6+ Total
Goals scored by 0 8.1 7.8 4.1 1.5 0.5 0.1 0.0 22.1
the home team 1 10.9 12.8 6.4 2.3 0.7 0.2 0.0 33.2
2 8.1 9.4 5.1 1.7 0.4 0.1 0.0 25.0
3 4.2 4.4 2.5 1.0 0.3 0.0 0.0 12.5
4 1.7 1.7 0.9 0.4 0.1 0.0 0.0 4.9
5 0.6 0.6 0.3 0.1 0.0 0.0 0.0 1.6
6+ 0.2 0.2 0.1 0.0 0.0 0.0 0.0 0.6
Total 33.9 36.9 19.4 7.1 2.0 0.5 0.1 100.0
most common outcome, in 12.8 per cent of all matches. The comparison between
Tables 3.4 and 3.5 suggests that the change from two to three-points-for-a-win
had the desired effect of encouraging attacking play, particularly by away teams.
The proportion of matches that finished 0–0 was higher in 1971–82 (9.3 per cent)
than in 1982–2009. Similarly among other low-scoring results, the proportions of
1–0 home wins and 1–1 draws were higher in 1970–81 (11.8 per cent and 13.2 per
cent) than in 1982–2009. There was, however, a smaller proportion of 0–1 away
wins in 1970–81 (6.8 per cent) than in 1982–2009.
Table 3.6 Percentage distribution of goals scored by the home team, conditional on
the number of goals scored by the away team, English league, 1982–2009 seasons

0 1 2 3 4 5 6+
Goals scored by 0 24.0 21.0 20.8 21.4 23.5 25.3 25.9

the home team 1 32.2 34.6 32.8 32.5 33.3 33.9 33.3
2 23.9 25.5 26.5 24.2 21.9 26.1 22.2
3 12.3 12.0 13.1 14.0 13.3 8.9 9.9
4 5.0 4.7 4.7 6.0 5.5 3.9 8.6
5 1.8 1.6 1.6 1.5 1.9 1.2 0.0
6+ 0.7 0.7 0.5 0.4 0.7 0.8 0.0
Average 1.51 1.52 1.55 1.57 1.53 1.39 1.42
The final columns and rows of Tables 3.4 and 3.5 show the marginal probability
distributions of goals per match scored by the home and away teams respectively,
calculated by summing horizontally across the other columns and vertically down
the other rows of each table. For example, Table 3.5 indicates that the home team
failed to score in 22.1 per cent of all matches played in 1982–2009, and the home
team scored once in 33.2 per cent of all matches. Similarly, the away team failed
to score in 33.9 per cent of all matches, and scored once in 36.9 per cent of these
matches. For away teams in particular, there has been a marked change in the
shape of the marginal probability distribution of goals scored. The modal (most
frequent) number of away goals scored changed from zero in 1970–81, to one in
1982–2009.
Tables 3.6 and 3.7 show the conditional distributions for the numbers of goals
scored by the home and away team. To save space, only the results for the 1982–
2009 period are included. The conditioning is on the number of goals scored by
the opposing team. The conditional distributions can be used to identify whether
there is interdependence between the numbers of goals scored by the home and
away teams in each match, or whether the numbers of goals scored by the two
teams can be considered as (approximately) independent of each other.
For example, the first column of Table 3.6 shows, for matches in which the away
team failed to score, the proportions of occasions on which the home team scored
0, 1, 2, 3 goals, and so on. In 24.0 per cent of matches in which the away team and
so on failed to score, therefore, the home team also failed to score; in 32.2 per cent
of matches in which the away team failed to score, the home team scored once;
and so on. The second column shows the same for matches in which the away team
scored once. In Table 3.7, the conditioning is reversed. The first row shows, for
Table 3.7 Percentage distribution of goals scored by the away team, conditional on
the number of goals scored by the home team, English league, 1982–2009 seasons

0 1 2 3 4 5 6+ Average
Goals scored by 0 36.8 35.1 18.3 6.9 2.1 0.5 0.2 1.05
the home team 1 32.8 38.5 19.2 7.0 2.0 0.5 0.1 1.09
2 32.4 37.7 20.6 6.9 1.7 0.5 0.1 1.10
3 33.5 35.4 20.4 8.0 2.1 0.3 0.1 1.11
4 34.5 35.3 18.7 8.6 2.2 0.4 0.2 1.11
5 37.2 35.1 18.8 6.4 2.2 0.3 0.0 1.02
6+ 38.4 38.9 15.1 4.9 2.2 0.5 0.0 0.95
matches in which the home team failed to score, the proportions of occasions on
which the away team scored 0, 1, 2, 3 goals, and so on. The second row shows the
same for matches in which the home team scored once.
If the numbers of goals scored by the home team and the away team are
independent, then the number of away goals should not affect the conditional
probabilities for the number of home goals. This implies the probability dis-
tributions in each column of Table 3.6 should look similar to each other (and
similar to the marginal distribution for the number of home goals in the final
column of Table 3.5). Likewise, the number of home goals should not affect
the probabilities for the number of away goals, so the probability distributions
in each row of Table 3.7 should also look similar to one another (and similar
to the marginal distribution for the number of away goals in the final row of
Table 3.5).
In fact, the probabilities do not satisfy this condition for independence. In Table
3.7, for example, when the home team scored 0, the most frequent score for the
away team was 0 (36.8 per cent of matches when the home team scored 0). When
the home team scored 0, the away team scored 1 in 35.1 per cent of matches. When
the home team scored 1, however, the most frequent score for the away team was
also 1 (38.5 per cent of matches when the home team scored 1). When the home
team scored 1, the away team scored 0 in only 32.8 per cent of matches. The con-
ditional mean scores reported in the final column of Table 3.7 indicate that the
(conditional) average number of away goals was an increasing function of the
number of home goals over the range 0 to 3 home goals. Similarly, according to
the bottom row of Table 3.6, the (conditional) average number of home goals was
an increasing function of the number of away goals over the range 0 to 3 away
goals.
In common with the results presented by Dixon and Coles (1997), these tabula-
tions suggest that the greatest degree of interdependence occurs in low-scoring
games, since the largest discrepancies between the probabilities in adjacent rows or
columns are towards the top-left-hand corners of Tables 3.6 and 3.7. Nevertheless,
it is apparent that the shapes of the conditional probability distributions do vary
to some extent over the full range of values for the conditioning variable, and
that the numbers of goals scored by the home and away teams are therefore
interdependent.
This section concludes by presenting an analysis of the statistical properties of
the goals data for seasons 1982–2009 that are summarised in Tables 3.5–3.7. This
analysis involves using theoretical probability distributions to describe the data on
the goals scored by the home and away teams. The two candidate distributions are
the Poisson distribution and the negative binomial distribution. For each of these
two distributions, we consider three variants, as follows.
First, by treating the goals scored by the home and away teams in each match
as independent, the data can be described by fitting two separate univariate distri-
butions to the data for home teams and away teams individually. This approach,
which produces a double Poisson distribution or double negative binomial dis-
tribution, does not take account of any possible dependence between the goals
scored by the two teams.
Second, zero-inflated variants of both the Poisson distribution and the nega-
tive binomial distributions are available, to allow for departure from the basic
shapes of these theoretical distributions in the case of the ‘zero’ outcome, which
in the bivariate case is the 0–0 draw. In a wide range of applications, it has been
found that the theoretical distributions can have a tendency to generate probabil-
ities that understate the observed proportion of zero outcomes. A zero-inflated
adjustment involves increasing the theoretical probability for the zero outcome
by an amount sufficient to bring this probability into line with the observed pro-
portion. The theoretical probabilities for all other outcomes are deflated pro rata
by the amounts required to maintain the condition that the probabilities add up
to one. By extension, inflated adjustments can also be applied to the probabilities
for other outcomes if required. In the case of goals data, a case has been made
for applying diagonal-inflated adjustments to the probabilities for the most com-
monly observed drawn results of 0–0, 1–1 and 2–2 (Karlis and Ntzoufras, 2003).
Third, by relaxing the independence assumption, the data can be described
by fitting bivariate distributions to the data for the home teams and away teams
jointly. An explicit formula exists for the joint probability function of the bivariate
Poisson distribution. No such formula exists in the case of the bivariate negative
binomial distribution; but it is possible to create a ‘synthetic’ bivariate distribution
from two univariate distributions, using a copula function. The two sets of uni-
variate probabilities are treated as inputs to the copula function, which generates
the bivariate probabilities with one or more additional parameters introduced to
describe the dependence. The Frank copula requires a single additional parameter.
This procedure for creating a bivariate distribution can be applied to a wide range
of univariate distributions, including both the Poisson and the negative binomial.
In order to facilitate direct comparisons between these two distributions as can-
didates for describing the goals data, a ‘synthetic’ bivariate Poisson distribution
generated via a copula function is used below, rather than the theoretical bivariate
Poisson distribution.3
The technical details are as follows. Let S1 and S2 denote the numbers of goals
scored by the home and away teams expressed as random variables, and let s1 =
0,1,2, … and s2 = 0,1,2, … denote the realised values of S1 and S2. f1(s1) = P(S1 =
s1) denotes the probability that the home team scores s1 goals, for s1 = 0,1,2, … ,
known as the marginal probability function for S1. Similarly, f2(s2) = P(S2 = s2)
denotes the probability that the away team scores s2 goals, known as the marginal
probability function for S2.
The formulae for fi(si) in each case, together with the corresponding expressions
for the expected value of Si and the variance of Si, are as follows:
Poisson distribution: f i (s i ) = exp( − λ i,j )λ si,ji /s i!
Expected value and variance: E(Si ) = λ i var(Si ) = λ i [3.10]
Negative binomial distribution:

f i (s i ) = [Γ (ρi + s i )/{s i!Γ (ρi )}]{ρi /(λ i + ρi )}ρi {λ i /(λ i + ρi )}si
Expected value and variance:
E(Si ) = λ i var(Si ) = λ i (1 + κ i λ i ) where κ i = 1/ρi [3.11]
where exp( ) denotes the exponential function and Γ( ) denotes the gamma
function.
The distinction between the Poisson distribution and the negative binomial
distribution is found in the ancillary parameter ρi≥0, which allows for greater
flexibility in the dispersion of Si across matches than is allowed by the Poisson
distribution. Note that for ρi = 0, var(Si) = λi. In this limiting case, the negative
binomial distribution is the same as the Poisson distribution. For ρi>0, however,
var(Si)>λi. In this general case, the negative binomial distribution allows for over-
dispersion. In the sample data, the degree of overdispersion is small but positive.
The sample mean values of S1 and S2 are 1.5259 and 1.0838, and the sample vari-
ances are 1.5839 and 1.1038. The sample correlation coefficient between S1 and S2
of 0.0104 is significantly different from zero at the 0.01 level.
As indicated above, dependence between S1 and S2 is addressed by creating a
‘synthetic’ bivariate distribution from the two univariate distributions. Let Fi(si)
denote the univariate distribution functions for Si corresponding to fi(si). In other
words, F1(s1) = P(S1≤s1) expresses the probability that the home team scores s1 or
fewer goals, for all possible values of s1 ( = 0,1,2, …). Similarly, F2(s2) = P(S2≤s2)
is the probability that the away team scores s2 or fewer goals. The bivariate joint
distribution function is defined as follows:
G[F1 (s1 ),F2 (s 2 )] = P(S1 ≤ s1 ,S2 ≤ s 2 )

1  {exp[ϕF1 (s1 )] − 1}{exp[ϕF2 (s 2 )] − 1}
= ln 1 + 
ϕ  exp(ϕ ) − 1
[3.12]
The bivariate joint distribution function G[F1(s1),F2(s2)] expresses the probability

that the home team scores s1 or fewer goals and the away team scores s2 or fewer
goals. The ancillary parameter ϕ determines the nature of any correlation between
S1 and S2: for ϕ<0 the correlation between S1 and S2 is positive, and for ϕ>0 the
correlation is negative. G[F1(s1),F2(s2)] is undefined for ϕ = 0, but in this case it is
conventional to write G[F1(s1),F2(s2)] = F1(s1)F2(s2). The bivariate joint probability
function corresponding to G[F1(s1),F2(s2)] expresses the probability that the home
team scores (exactly) s1 goals and the away team scores (exactly) s2 goals. The
bivariate joint probability function f(s1,s2) = P(S1=s1,S2=s2) is obtained iteratively,
as follows:
f(0,0) = G[F1 (0),F2 (0)]

f(s1 ,0) = G[F1 (s1 ),F2 (0)] − G[F1 (s1 − 1),F2 (0)] for s1 = 1,2,
f(0,s 2 ) = G[F1 (0),F2 (s2 )] − G[F1 (0),F2 (s2 − 1)] for s 2 = 1,2,
f(s1 ,s 2 ) = G[F1 (s1 ),F2 (s 2 )] − G[F1 (s1 − 1),F2 (s 2 )]
− G[F1 (s1 ),F2 (s 2 − 1)] + G[F1 (s1 − 1),F2 (s 2 − 1)] for s1 ,s 2 =1,2, [3.13]

Finally, the diagonal-inflated adjustment for 0–0, 1–1 and 2–2 draws is applied by
introducing three additional parameters, π0, π1 and π2, adjusting the joint prob-
abilities as follows:

f(0,0) 
= P(S = 0,S = 0) = (1 − π )P(S = 0,S = 0) + π (1 − π − π )
1 2 0 1 2 0 1 2

f(1,1 )  1 = 1,S2 = 1) = (1 − π 0 )P(S1 = 1,S2 = 1) + π 0 π1
= P(S
f(2,2)  1 = 2,S2 = 2) = (1 − π 0 )P(S1 = 2,S2 = 2) + π 0 π 2
= P(S
 ,s ) = P(S
f(s  1 = s1 ,S2 = s 2 ) = (1 − π 0 )P(S1 = s1 ,S2 = s 2 )
1 2
for{s1 ,s 2 } ≠ {0,0},{1,1}or{2,2} [3.14]
Table 3.8 reports the parameter estimates obtained by fitting 16 alternative speci-
fications to the home and away teams’ goals data for the 59,922 English league
matches played during the 28 seasons from 1982 to 2009 (inclusive). The four
specifications used for the probability functions are double Poisson, bivariate
Poisson, double negative binomial and bivariate negative binomial. Four variants
are reported in each of these four cases: (i) unadjusted; (ii) zero-inflated adjust-
ment for 0–0 draws only; (iii) diagonal-inflated adjustments for 0–0 and 1–1 draws;
and (iv) diagonal-inflated adjustments for 0–0, 1–1 and 2–2 draws. As a descriptive
measure of the quality of each fitted model, the final column of Table 3.8 reports
pseudo R-square, the geometric mean of the probabilities assigned by the fitted
Table 3.8 Parameter estimates for fitted double and bivariate Poisson and negative binomial distributions, home- and away-team
goals data, English league, 1982–2009 seasons
Specification and code number λ1 λ2 π0 π1 π2 ϕ κ1 κ2 Pseudo R-square
1. double Poisson 1.5259 1.0838 .054276

2. double Poisson 0–0 1.5424 1.0955 .0106 .054306
3. double Poisson 0–0,1–1 1.5504 1.0982 .0222 .4502 .054324
4. double Poisson 0–0,1–1,2–2 1.5498 1.0969 .0238 .4268 .0608 .054325
5. bivariate Poisson 1.5258 1.0839 −.1287 .054286
6. bivariate Poisson 0–0 1.5418 1.0951 .0103 −.0151 .054306
7. bivariate Poisson 0–0,1–1 1.5507 1.0984 .0224 .4474 .0076 .054324
8. bivariate Poisson 0–0,1–1,2–2 1.5504 1.0972 .0244 .4201 .0630 .0167 .054326
9. double nb 1.5259 1.0838 .0244 .0166 .054299
10. double nb 0–0 1.5388 1.0929 .0084 .0159 .0082 .054314
11. double nb 0–0,1–1 1.5467 1.0952 .0208 .5469 .0183 .0168 .054337
12. double nb 0–0,1–1,2–2 1.5445 1.0915 .0241 .4997 .1376 .0219 .0215 .054342
13. bivariate nb 1.5258 1.0838 −.1305 .0244 .0166 .054309
14. bivariate nb 0–0 1.5360 1.0910 .0065 −.0573 .0177 .0100 .054315
15. bivariate nb 0–0,1–1 1.5446 1.0938 .0194 .5803 −.0410 .0197 .0180 .054337
16. bivariate nb 0–0,1–1,2–2 1.5430 1.0905 .0230 .5197 .1405 −.0302 .0228 .0223 .054342
Note: nb denotes negative binomial distribution.

Sequences and persistence in match results 67
model to the actual outcome of every match. A higher pseudo R-square indicates
a closer fit.
Table 3.9 reports the results of likelihood ratio tests for various comparisons
between the sixteen specifications. Panels 1a/b compare the double Poisson and
double negative binomial with their inflated-adjusted variants, by testing the
contribution to the model of the parameters π0, π1 and π2. The inflated adjust-
ments for 0–0, 1–1 and 2–2 draws are added and tested sequentially. The inflated
adjustment for 2–2 draws does not improve the quality of the double Poisson
specification inflated-adjusted for 0–0 and 1–1 draws significantly; but the same
adjustment does improve the corresponding double negative binomial specifica-
tion significantly. Panels 2a/b repeat these comparisons for the bivariate Poisson
and bivariate negative binomial specifications respectively, with similar results.
Panels 3a/b compare the double Poisson and double negative binomial speci-
fications with the corresponding bivariate specifications, by testing the contribu-
tion to the model of the parameter ϕ. In the absence of the inflated adjustments,
ϕ makes a significant contribution; but if the inflated-adjustment parameters are
included, ϕ becomes insignificant. The contributions to the model of ϕ on the one
hand, and π0, π1 and π2 on the other hand, are similar in the sense that the former
operates primarily to increase the probabilities for closely contested matches (and
reduce the probabilities for large differences in scores), while the latter operates
specifically upon the draw probabilities. The results reported in Table 3.9 suggest
the specific adjustments to the draw probabilities are more effective in improving
the quality of the fitted model than the adjustment that allows for dependence by
means of the copula function.
Finally, Panels 4a/b present direct comparisons of the corresponding Poisson
and negative binomial specifications, by testing the contribution to the model of
the overdispersion parameters κ1 and κ2 (or, equivalently, ρ1 and ρ2). These param-
eters make a significant contribution in every case, indicating that the negative
binomial specification outperforms the Poisson. Based on these hypothesis test
results, specification 12, the double negative binomial with inflated-adjusted prob-
abilities for 0–0, 1–1 and 2–2 draws, provides the best representation of the goals
data.
3.4 Good and poor sequences, and persistence in football match results
A question of enduring fascination to sports fans concerns the nature of per-
sistence in sequences of consecutive match results. Does a sequence of wins tend
to build a team’s confidence and morale, increasing the probability that the next
match will also be won? Or does it tend to create pressures or breed complacency,
increasing the likelihood that the next match will be drawn or lost? Does a sequence
of losses tend to sap confidence or morale, increasing the probability of a further
loss in the next match? Or does it tend to inspire greater effort, increasing the like-
lihood that the next match will be won or drawn? These questions are examined by
Table 3.9 Hypothesis tests for comparisons between fitted double and bivariate Poisson and negative binomial distributions,
home- and away-team goals data, English league, 1982–2009 seasons
Specifications under null and Specifications under null and

alternative hypotheses alternative hypotheses
Degrees of Degrees of
Null Alternative Chi-square freedom p-value Null Alternative Chi-square freedom p-value
Panel 1a Panel 1b
1 2 63.2 1 .000 9 10 31.3 1 .000
2 3 38.5 1 .000 10 11 47.9 1 .000
3 4 2.1 1 .145 11 12 10.3 1 .001
Panel 2a Panel 2b
5 6 41.4 1 .000 13 14 11.8 1 .001
6 7 38.4 1 .000 14 15 46.6 1 .000
7 8 2.3 1 .128 15 16 9.7 1 .002
Panel 3a Panel 3b
1 5 22.1 1 .000 9 13 22.2 1 .000
2 6 0.2 1 .655 10 14 2.7 1 .100
3 7 0.1 1 .806 11 15 1.4 1 .237
4 8 0.3 1 .610 12 16 0.8 1 .383
Panel 4a Panel 4b
1 9 48.1 2 .000 5 13 48.2 2 .000
2 10 16.2 2 .000 6 14 18.7 2 .000
3 11 25.5 2 .000 7 15 26.9 2 .000
4 12 33.7 2 .000 8 16 34.2 2 .000
Note: The model specifications under the null and alternative hypotheses that are compared in each test are identified by their code numbers,
as listed in Table 3.8. Panels 1a and 2a report tests for the significance of the zero-inflated parameters in the double Poisson and bivariate
Poisson specifications. Panels 3a and 3b report tests for the interdependence parameter in the bivariate Poisson and bivariate negative
binomial specifications. Panels 4a and 4b report tests for the significance of the overdispersion parameter in the double and bivariate binomial
specifications.
Table 3.10 Longest runs of consecutive results, English league, 1970–2009 seasons
Matches without a loss End-month Consecutive wins End-month
Arsenal 49 Oct-04 Arsenal 14 Aug-02

Nottm Forest 42 Nov-78 Newcastle Utd 13 Oct-92
Chelsea 40 Oct-05 Reading 13 Oct-85
Reading 33 Feb-06 Charlton Athletic 12 Mar-00
Bristol Rovers 32 Jan-74 Fulham 12 Oct-00
Liverpool 31 Mar-88 Liverpool 12 Oct-90
Arsenal 30 Oct-02 Luton Town 12 Apr-02
Leeds Utd 30 Feb-74 Manchester Utd 12 Aug-00
Consecutive losses End-month Matches without a win End-month

Sunderland 17 Aug-03 Derby County 36 Aug-08
Walsall 15 Feb-89 Cambridge Utd 31 Apr-84
Brighton & Hove Albion 12 Jan-73 Hull City 27 Nov-89
Brighton & Hove Albion 12 Oct-02 Oxford Utd 27 Aug-88
Carlisle Utd 12 Dec-03 Newport County 25 Jan-71
Barnet 11 Oct-93 Rochdale 25 Aug-74
MK Dons 11 Mar-04
Stoke City 11 Aug-85
West Bromwich Albion 11 Dec-95
Dobson and Goddard (2003), using English league match results data for seasons
1970–1999 (inclusive). This section presents an update of this earlier study, using
data to the end of the 2009 season.
Table 3.10 reports the longest sequences of consecutive results in the league
match results data set between the 1970 and 2009 seasons inclusive, based on four
criteria: (i) matches without a win; (ii) matches without a loss; (iii) consecutive
wins; and (iv) consecutive losses. For the purposes of counting sequences of con-
secutive results, breaks between seasons are ignored.
Table 3.11 reports empirical unconditional and conditional match result prob-
abilities, where conditioning is on the duration of various preceding sequences of
consecutive similar results. The first row reports the unconditional home and away
probabilities for a win (columns (1) and (2)), a win or draw ((3) and (4)), a loss ((5)
and (6)) and a loss or draw ((7) and (8)). Of the 81,258 matches in the data set,
38,775 were home wins, 22,426 were draws and 20,057 were away wins. Therefore
0.477 = 38775/81258 is the unconditional home win probability (column (1)), and
so on.
Subsequent rows of Table 3.11 report the conditional probabilities that each
result represents a ‘reversal’ of a previous sequence of consecutive ‘identical’
Table 3.11 Empirical unconditional and conditional match result probabilities
Probability of a Probability of a Probability of a Probability of a

win, conditional win or draw, loss, conditional loss or draw,
on n = number of conditional on on n = number conditional on
previous n = number of of previous n = number
consecutive previous consecutive of previous
matches consecutive matches consecutive
without a win losses without a loss wins
Home Away Home Away Home Away Home Away
n (1) (2) (3) (4) (5) (6) (7) (8)
0 0.477 0.247 0.753 0.523 0.247 0.477 0.523 0.753

1 0.465 0.236 0.731 0.494 0.232 0.464 0.496 0.738
2 0.451 0.228 0.715 0.474 0.222 0.449 0.478 0.716
3 0.438 0.219 0.695 0.456 0.210 0.438 0.460 0.693
4 0.433 0.216 0.686 0.442 0.199 0.423 0.451 0.663
5 0.422 0.212 0.670 0.415 0.191 0.406 0.424 0.646
7 0.409 0.202 0.611 0.413 0.168 0.387 0.382 0.583
10 0.396 0.190 — — 0.132 0.344 — —
15 0.344 0.177 — — 0.114 0.303 — —
20 0.346 0.163 — — 0.086 0.294 — —
results, conditioning on the type and duration of the sequence. Two types of
reversal are considered. First, a sequence of wins and draws is reversed by a
loss; and a sequence of losses is reversed by a win or draw. Second, a sequence
of wins is reversed by a draw or loss; and a sequence of draws and losses is
reversed by a win. These two types of reversal are ‘WD|L reversals’ and ‘W|DL
reversals’, respectively. Columns (3) to (6) of Table 3.11 show the conditional
probabilities for WD|L reversals; and columns (1), (2), (7) and (8) show the
conditional probabilities for W|DL reversals. For the purpose of calculating
these probabilities, the results of any cup, European or friendly matches played
within a sequence of league matches are ignored. So too are the venues (home
or away) of the matches comprising the sequence of prior matches. The condi-
tional probabilities themselves, however, are specific to the venue of match in
question.
For example, the conditioning for the home win probabilities in column (1) is on
the number of previous consecutive matches without a win (the number of previ-
ous matches drawn or lost). The home team had failed to win its four most recent
matches in 15,260 of the 81,258 matches in the data set. This figure includes cases
in which the sequence without a win was longer than four matches. In 6,610 of
these 15,260 matches, the match result was a home win, implying a W|DL rever-
sal. Therefore 0.433 = 6,610/15,260 is the home win probability conditional on the
home team having played at least four consecutive matches without a win, prior
to the match in question.4
Table 3.11 shows that the conditional probabilities of a good result (however
defined) tend to decline with the duration of an unsuccessful spell, and the condi-
tional probabilities of a poor result decline with the duration of a successful spell.
Without further investigation, however, it would be incorrect to attribute this
pattern to a positive persistence effect. The pattern in the conditional probabil-
ities might be explained by the variation between teams in their relative quality
or playing strength. The calculation of the win probability conditional on a long
spell without a win, for example, is based mainly on the experience of weaker
teams, whose win probability is below average because they are weak, but not
specifically because they have not won recently. In other words, the pattern in the
conditional probabilities reported in Table 3.11 might be explained by a team het-
erogeneity effect. Therefore any test for persistence in sequences of match results
needs to control for heterogeneous team strengths.
Below, a Monte Carlo analysis is used to test for persistence effects. In the
absence of any persistence effects, it is assumed that the following statistical model
would accurately represent the distribution of match results in each season in each
tier. According to this model, the result of the match between home team i and
away team j is generated as follows:
Home win(k = 2) if µ 2 < y*i,j + ε i,j

Draw (k = 1) if µ1 < y*i,j + ε i,j < µ 2
Away win (k = 0) if y*i,j + ε i,j < µ1 [3.15]

where y*i,j = αi – αj; αi and αj are parameters reflecting the quality or playing
strengths of team i and team j; μ1 and μ2 are additional parameters, known as
‘cut-off parameters’; and εi,j ~ N(0,1) is a random disturbance term, which follows
a standard Normal distribution (with zero mean and variance of one). The dis-
turbance term represents the unsystematic or random element in the result of the
match between teams i and j.
Table 3.12 illustrates the correspondence between the final league table and the
estimated parameters of [3.15] for T1 (the Premier League) in the 2009 season.
The ordering of the teams reflects the final league table, obtained by awarding
three league points for a win and one for a draw. Table 3.12 also reports each
team’s win percent, obtained by awarding 1 for each win and 0.5 for each draw,
and dividing the total by 38 (the number of matches played by each team). Table
3.12 also reports each team’s α ˆ i for the 2009 season, with the parameter for the
bottom-placed team, West Bromwich Albion, set to zero.
Table 3.12 League table, Premier League, 2009 season, and ordered probit team
quality parameter estimates
Won Drawn Lost League Win αˆ i

points ratio
Manchester United 28 6 4 90 .8158 1.6534

Liverpool 25 11 2 86 .8026 1.4863
Chelsea 25 8 5 83 .7632 1.3943
Arsenal 20 12 6 72 .6842 1.1348
Everton 17 12 9 63 .6053 .8941
Aston Villa 17 11 10 62 .5921 .8372
Fulham 14 11 13 53 .5132 .6379
Tottenham Hotspur 14 9 15 51 .4868 .5685
West Ham United 14 9 15 51 .4868 .5622
Manchester City 15 5 18 50 .4605 .4771
Wigan Athletic 12 9 17 45 .4342 .3757
Stoke City 12 9 17 45 .4342 .3944
Bolton Wanderers 11 8 19 41 .3947 .2242
Portsmouth 10 11 17 41 .4079 .3004
Blackburn Rovers 10 11 17 41 .4079 .2744
Sunderland 9 9 20 36 .3553 .1556
Hull City 8 11 19 35 .3553 .1808
Newcastle United 7 13 18 34 .3553 .1809
Middlesbrough 7 11 20 32 .3289 .0685
West Bromwich Albion 8 8 22 32 .3158 .0000
Cut-off parameters: ˆμ1 = –.7119 μˆ 2 = .0250
Illustrative fitted match result probabilities
Home win Draw Away win
Liverpool v Middlesbrough 0.673 0.209 0.118

Middlesbrough v Liverpool 0.309 0.285 0.406
Aston Villa v Blackburn Rovers 0.563 0.252 0.185
Blackburn Rovers v Aston Villa 0.417 0.284 0.299
Manchester City v Wigan Athletic 0.501 0.269 0.230
Wigan Athletic v Manchester City 0.480 0.274 0.246
All matches (average) 0.455 0.255 0.290
The estimates of μ1 and μ2 are shown at the foot of Table 3.12, together with
illustrative fitted match result probabilities. The latter are calculated using:
ˆ 2 − ŷ i,j )
P(home win) = 1 − Φ(µ
*
* *
P(draw) = Φ( µ
ˆ 2 − yˆ i,j ) − Φ(µ
ˆ 1 − yˆ i,j )
P(away win) = Φ(µ
ˆ 1 − yˆ i,j )*
[3.16]
where Φ is the distribution function for the standard Normal distribution; and
*
yî,j = α
î − α
ˆ j . The values of the cut-off parameters μˆ 1 and μˆ 2 allow for home-field
advantage. The examples illustrate the implications of variations in αˆ i and αˆ j for
the home win, draw and away win probabilities.
The Monte Carlo simulations enable comparisons to be drawn between the
observed numbers of reversals in the data set (as defined above), and the numbers
of reversals that should be obtained if [3.15] is the statistical model that describes
correctly the distribution of match results if there is no persistence effect. Two test
statistics are used to test for persistence effects: the first is based on the number
of WD|L reversals; and the second is based on the number of W|DL reversals. In
each case, the test statistic is τ = total number of match results divided by total
number of reversals. If the observed value of τ is similar to its expected value
obtained from the Monte Carlo simulations, the null hypothesis of no persistence
cannot be rejected. If the observed τ is significantly higher than its expected value,
reversals occur less frequently than they should occur if the null hypothesis is true.
In this case the null hypothesis is rejected in favour of an alternative hypothesis
of positive persistence. Conversely, if the observed τ is significantly lower than its
expected value, reversals occur more frequently than they should when the null
hypothesis is true. In this case the null hypothesis is rejected in favour of an alter-
native hypothesis of negative persistence.
The approach is similar in principle to the well-known runs test (Mood,
1940), which investigates the randomness of sequences of positive or negative
increments to a time series. Mood used analytic methods to derive asymptotic
sampling distributions for the numbers of ‘runs’ (sequences of consecutive posi-
tive or negative increments) expected under the randomness assumption, based
on the binomial distribution. In the present case, simulation rather than ana-
lytic methods are used, because the expected numbers of reversals depend on
the degree of inequality in the team strength parameters within each tier, and
therefore vary between tiers and between seasons. The greater is the degree of
inequality, the easier it is for a strong team to sustain a sequence of good results
(and the harder it is for a weak team to break a sequence of poor results), and
so the smaller is the expected number of reversals. Fort and Rosenman (1999)
apply the runs test directly to sequences of match results in MLB to test for the
presence of ‘streaks’. Since there are no draws, match results are binary and the
runs test is directly applicable. Each team is tested separately. The test controls
for the quality of the team being tested (via the overall win percentage), but in
contrast to the present analysis there is no control for variation in the quality of
the opposition team.
To generate the expected mean durations of sequences of consecutive results
under the null hypothesis of zero persistence, 160 sets of ordered probit estimates
of the parameters of [3.15] are obtained. Using the actual fixture calendars as
originally completed, a computer program then generates a complete set of simu-
lated match results for the full 40-season period, under the assumption of zero
Table 3.13 Simulated unconditional and conditional match result probabilities
Probability of a Probability of a Probability of a Probability of a

win, conditional win or draw, loss, conditional loss or draw,
on n = number of conditional on on n = number conditional on
previous n = number of of previous n = number
consecutive previous consecutive of previous
matches consecutive matches consecutive
without a win losses without a loss wins
n Home Away Home Away Home Away Home Away
(1) (2) (3) (4) (5) (6) (7) (8)
0 0.477 0.247 0.753 0.523 0.247 0.477 0.523 0.753

1 0.461 0.230 0.731 0.489 0.231 0.462 0.488 0.730
2 0.444 0.217 0.707 0.462 0.218 0.444 0.455 0.701
3 0.428 0.206 0.684 0.439 0.206 0.427 0.425 0.671
4 0.415 0.197 0.663 0.417 0.195 0.411 0.397 0.644
5 0.403 0.189 0.642 0.396 0.185 0.396 0.373 0.618
7 0.382 0.175 0.598 0.357 0.167 0.368 0.333 0.573
10 0.356 0.158 0.537 0.303 0.145 0.333 0.288 0.518
15 0.317 0.135 — — 0.120 0.290 — —
20 0.284 0.116 — — 0.102 0.258 — —
persistence, by substituting randomly drawn values of εi,j ~ N(0,1) into [3.15]. This
exercise is repeated 5,000 times, in order to generate 5,000 sets of simulated match
results each of which covers the entire 40-season period.
Table 3.13 reports match result probabilities conditional on each result rep-
resenting a reversal of a previous sequence of consecutive results, calculated
from the Monte Carlo simulations based on an assumption of no persistence. A
comparison between these simulated conditional probabilities and the observed
conditional probabilities reported in Table 3.11 (and allowing for occasional
random variation in the latter) confirms that the actual probability of a rever-
sal occurring is higher than the simulated probability under assumptions of no
persistence.
In order to test the null hypothesis that there is no persistence effect, for each of
the 5,000 sets of simulated match results the test statistic τ (=number of matches ÷
number of reversals) is calculated for each of the two types of reversal. By examin-
ing the sampling distributions of the two sets of 5,000 simulated τ, critical values
are established, leading to the acceptance or rejection of the null hypothesis of no
persistence.
The persistence tests are carried out using the data for all forty seasons from
1970 to 2009, and using the same data subdivided into eight subperiods of five
seasons each: seasons 1970–1974, 1975–1979 and so on through to 2005–2009.
Table 3.14 reports the results of these tests. The upper panel shows the results
for WD|L reversals, and the lower panel shows the results for W|DL reversals.
Table 3.14 Tests for persistence in sequences of consecutive match results
Monte Carlo simulations Actual

p0.5 p2.5 p5.0 p95.0 p97.5 p99.5 τ p-value
Sequences without a loss or sequences of losses: ratio of matches played to WD|L reversals
1970–2009 2.194 2.197 2.198 2.213 2.215 2.217 2.197 .0556
1970–1974 2.191 2.198 2.202 2.245 2.250 2.258 2.228 .7284
1975–1979 2.147 2.155 2.159 2.201 2.205 2.213 2.184 .7480
1980–1984 2.135 2.142 2.146 2.187 2.191 2.198 2.153 .3152
1985–1989 2.160 2.168 2.171 2.213 2.217 2.225 2.203 .3988
1990–1994 2.163 2.171 2.175 2.216 2.220 2.228 2.173 .0764
1995–1999 2.187 2.196 2.200 2.242 2.246 2.255 2.205 .2216
2000–2004 2.198 2.205 2.208 2.252 2.257 2.265 2.216 .2964
2005–2009 2.207 2.215 2.219 2.262 2.266 2.275 2.215 .0528
Sequences of wins or sequences without a win: ratio of matches played to W|DL reversals
1970–2009 2.199 2.203 2.204 2.219 2.220 2.223 2.182 .0000
1970–1974 2.193 2.201 2.205 2.248 2.251 2.258 2.200 .0428
1975–1979 2.151 2.158 2.162 2.202 2.206 2.215 2.151 .0104
1980–1984 2.138 2.147 2.151 2.192 2.196 2.205 2.149 .0836
1985–1989 2.162 2.170 2.174 2.216 2.221 2.228 2.175 .1112
1990–1994 2.166 2.174 2.178 2.220 2.225 2.232 2.178 .0948
1995–1999 2.192 2.201 2.206 2.249 2.253 2.262 2.165 .0000
2000–2004 2.204 2.213 2.217 2.261 2.266 2.274 2.211 .0368
2005–2009 2.217 2.227 2.230 2.276 2.281 2.289 2.229 .0792
The columns headed p0.5, p2.5, p5.0, p95.0, p97.5, p99.5 show the 0.5, 2.5, 5,
95, 97.5 and 99.5 percentiles of the sampling distributions of the test statistic
τ under the null hypothesis of no persistence, obtained from the Monte Carlo
simulations.5
Accordingly, a 95 per cent confidence interval for τ under the null hypothesis of
no persistence, based on the results for WD|L reversals, is given by (2.197, 2.215).
The null hypothesis is rejected at a significance level of 5 per cent if τ falls out-
side this range. Similarly, a 99 per cent confidence interval for τ is given by (2.194,
2.217). The null hypothesis is rejected at a significance level of 1 per cent if τ falls
outside this range.
The final two columns of Table 3.14 report the actual values of τ, and the cor-
responding p-values. The p-value is the minimum significance level at which the
null hypothesis of no persistence can be rejected. In the results based on computa-
tions over the entire forty-season period, for WD|L reversals, τ = 2.197 falls just
inside the lower bound of the 95 per cent confidence interval, but outside the lower
bound of the 90 per cent confidence interval (p-value = .0556). The null hypoth-
esis of no persistence cannot be rejected at the 5 per cent significance level; but the
null is rejected at the 10 per cent level. For W|DL reversals, however, τ = 2.182 falls
outside the lower bounds of both the 95 per cent and the 99 per cent confidence
intervals (p-value = .0000). In this case the null hypothesis of zero persistence is
rejected, in favour of an alternative of negative persistence, at any significance
level. W|DL reversals occur more frequently than is expected if the null hypothesis
is true.
In the results based on computations for five-season subperiods, the pat-
tern is similar. For WD|L reversals, the null hypothesis of no persistence is not
rejected at the 5 per cent level for any of the eight five-year subperiods. The null
hypothesis is rejected at the 10 per cent level for two of the eight periods. For
W|DL reversals, in contrast, the null hypothesis of no persistence is not rejected
at the 5 per cent level for four of the eight five-year subperiods. The null hypoth-
esis is rejected at the 10 per cent level for seven of the eight subperiods. In all
cases where the null is rejected, τ is lower than is expected if the null hypothesis
is true. The number of reversals is therefore higher than is expected if the null
hypothesis is true.
Overall the results indicate that sequences of match results are subject to statis-
tically significant, negative persistence effects. On average, sequences of consecutive
wins and sequences of consecutive matches without a win tend to end sooner than
they would if there were no statistical association between the results of consecu-
tive matches after controlling for heterogeneous team strengths. There is, however,
an element of asymmetry in the pattern. The average duration of sequences of
matches unbeaten is higher than the average duration of sequences without a win;
and the average duration of sequences of losses is higher than the average duration
of sequences of wins. Accordingly, the evidence of a negative persistence effect in
the data on W|DL reversals is stronger than it is in the data on WD|L reversals.
Finally, Dobson and Goddard (2003) point out that the procedure described
above is valid if the assumption of no variation in the team strength parameters
within each season is correct. A difficulty arises, however, if this assumption is
incorrect, because the actual and expected numbers of reversals are sensitive to
these parameters. If there is within-season variation in the team strength param-
eters αi in [3.15], the expected numbers of reversals are somewhat lower than
in the case where there is no variation, and the persistence test described above
is biased towards detection of a positive persistence effect. This suggests that
rejection of the null hypothesis of no persistence in favour of an alternative of
negative persistence is a particularly strong result. If there is within-season vari-
ation in the team strength parameters, this test tends to be biased in the opposite
direction.
Conclusion
Chapter 3 has investigated a number of empirical regularities in football match
results data. The problem of measuring competitive balance or competitive
inequality within a sports league has attracted considerable attention in the aca-
demic sports economics literature in recent years. Researchers have applied sev-
eral measures of concentration or inequality, some of which are borrowed from
industrial economics, to sports teams’ win ratio or league points data. Popular
measures include unadjusted and adjusted standard deviations, the Lorenz curve
and the Gini coefficient, the Herfindahl index, and various concentration ratios.
Some studies measure competitive inequality directly, by means of an analysis of
the probabilities for individual match results.
The sources of home-field advantage in any sport are difficult to identify or
quantify empirically, but can be classified broadly as follows: familiarity with the
home team’s stadium and facilities; disruption to the away team’s preparation due
to the need to travel; crowd effects on either the home team (encouragement) or
the away team (intimidation) or the match officials (pressure for decisions favour-
ing the home team); and (in some sports) rules favouring the home team. English
football has witnessed a reduction in the importance of home-field advantage
between the 1970s and the 2000s, which appears to be partially but not entirely
explained by the switch from the award of two to three league points for a win in
the early 1980s.
Several variants of the Poisson distribution and the negative binomial distribu-
tion provide a good description of the distributional properties of football match
results data presented in scores format. The greater flexibility of the negative bino-
mial distribution in allowing for overdispersion is useful in describing the uncon-
ditional distributions of goals scored by the home and away teams (unconditional
in the sense that controls for team quality are not included). Positive correlation
between the goals scored by the home and away teams may be represented by
means of either bivariate distributions, which include a correlation parameter; or
diagonal-inflated distributions, which make adjustments to the probabilities of

low-scoring draws.
Finally, the phenomenon of persistence in sequences of consecutive football
match results is investigated by means of a Monte Carlo analysis, which involves
comparing the actual numbers of ‘reversals’ of sequences of consecutive results,
with the numbers expected if there were no persistence. A comparison between the
simulation results and forty years of English match results data provides evidence
of a negative persistence effect.
Notes
1 For example, regulations in baseball concerning the order in which the home and away teams’
innings take place.
2 Beyond team sports, Balmer, Nevill and Williams (2001) report a highly significant home-
field advantage in the Olympic Games, affecting event groups that were either subjectively
judged or reliant on subjective decisions (boxing and gymnastics). In contrast, little or no
home-field advantage was observed in two objectively judged event groups (athletics and
weightlifting). The nature of the officiating system was found to be key to both the existence
and extent of home-field advantage.
3 In Chapter 4, however, a forecasting model for match results and goals is developed, based
on the explicit formula for the bivariate Poisson distribution.
4 The conditional probabilities in the other columns of the upper panel of Table 3.11 are cal-
culated in the same way. The probabilities are not reported in cases where there were fewer
than fifty sequences of the required duration on which to base the calculation. This limit is
breached for sequences of consecutive wins and consecutive losses (columns (3), (4), (7) and
(8)) of duration greater than (about) eight matches.
5 In the simulations for WD|L reversals for the period 1970–2009, for example, the first row of
Table 3.14 shows that 0.5 per cent of the simulated τ were below 2.194 (and 99.5 per cent of
the simulated τ were above 2.194); 2.5 per cent of the simulated τ were below 2.197 (and 97.5
per cent were above); and so on. At the opposite end of the range of values for τ, 97.5 per
cent of the simulated τ were below 2.215 (and 2.5 per cent were above); 99.5 per cent of the
simulated τ were below 2.217 (and 0.5 per cent were above); and so on.
4 Forecasting models for football
match results
Introduction
Chapter 4 describes the estimation and application of goals-based and results-
based forecasting models for match outcomes in football, recorded in the form
of either goals scored and conceded by the two teams, or in the form of ‘win-
draw-lose’ match results. Both types of forecasting model are estimated by fit-
ting regression models to past match results data. A diagonal-inflated bivariate
Poisson regression is used for the goals-based model, and an ordered probit
regression is used for the results-based model. Both models draw on an exten-
sive set of covariates, reflecting past goal-scoring performance and match
results over the preceding 24 calendar months, the significance of the match
for end-of-season championship, promotion or relegation outcomes, the current
involvement of the teams in the FA Cup or European tournaments, the average
attendances attracted by both teams, and the geographical distance between the
home stadia of the two teams.
Section 4.1 reviews the previous academic literature on modelling the outcomes
of football matches. Section 4.2 describes the specification and estimation of a
goals-based match results forecasting model. Section 4.3 describes the use of the
goals-based model to generate out-of-sample forecasts, in probabilistic form, for
either goals or ‘win-draw-lose’ match results. Sections 4.4 and 4.5 describe the
estimation and application of a results-based forecasting model. Finally Section
4.6 draws some comparisons between the forecasting performance of the goals-
based and results-based models, and between the probabilities generated by these
models and a set of implied probabilities derived from the quoted odds (prices)
of a selection of high-street and internet bookmakers for fixed-odds betting on
match results.
4.1 Previous literature on modelling and forecasting match results in football

There are two distinct strands of empirical literature on modelling the outcomes
of matches in football. The first approach, favoured by many applied statisticians,
79
80 Forecasting models for football match results
involves modelling the numbers of goals scored and conceded in each match dir-
ectly. Once such a model has been estimated, forecasts of win-draw-lose match
results may be derived indirectly, by aggregating the estimated probabilities
assigned to appropriate permutations of goals scored and conceded by the two
teams. A second approach, favoured by some applied econometricians, involves
modelling win-draw-lose results directly, using discrete choice regression models
such as ordered probit or ordered logit.
A win-draw-lose match results data set is effectively ‘nested’ within a goals data
set: the result of a match is established from the goals scored by the two teams, but
the match result by itself does not indicate the numbers of goals scored. Therefore
any direct comparison between the forecasting capabilities of the two types of model
must be based on forecasts of match results: while goals-based models can fore-
cast goals and results, results-based models forecast results only. One possible prior
hypothesis is that a goals-based model should outperform a results-based model,
because the former draws on a more extensive data set than the latter. On the other
hand, given that league points are awarded for results and not for goals (with goals
relevant only for separating teams with equal points totals), insofar as they deter-
mine league points, wins of 1–0 or 6–3 are of equal value. Therefore goals data
might contain more noise than results data. In a goals-based model, the choice of
distributional assumptions and the treatment of the problem of interdependence
between the goals scored by the two teams are complex issues, which are avoided
when discrete choice regression is used to model match results directly. Accordingly,
a results-based model might be expected to outperform a goals-based model, on the
grounds that the model selection and specification issues are more straightforward.
In the academic literature, a number of studies model match results data for
football. Early contributions by Moroney (1956) and Reep, Pollard and Benjamin
(1971) use the Poisson and negative binomial distributions to model the distribu-
tions of the numbers of goals scored per game, along similar lines to the analysis
that is reported in Chapter 3, Section 3.3. The aggregated approach adopted, how-
ever, precludes the generation of specific forecasts for individual matches based on
information about the respective strengths of the two teams concerned. By com-
paring final league placings with experts’ pre-season forecasts, Hill (1974) demon-
strates that individual match results do nevertheless have a predictable element,
and are not determined solely by chance.
Maher (1982) develops a model in which the home- and away-team scores
follow independent Poisson distributions, with means which are the product of
parameters reflecting the attacking and defensive capabilities of the two teams. If
H denotes the goals scored by home team i and A the goals scored by away team
j, the respective probability functions are:

P(H = h) = exp( − α iβ j ) (α iβ j )h /h! and P(A = a) = exp( − γ i δ j ) (γ i δ j )a /a !
where αi and βj reflect the attacking capability of team i at home and the defensive
capability of team j away, and γi and δj reflect the defensive capability of team i at
Literature on modelling and forecasting results 81
home and the attacking capability of team j away. Tests show that γi and δi can be
regarded as proportional to βi and αi respectively, so it is only necessary to estimate
one set of attacking and one set of defensive parameters for each team. This can
be done ex post, after the full set of match results has been observed, using max-
imum likelihood methods. The model does not predict scores or results ex ante.
Although goodness-of-fit tests show that the model provides a reasonably accur-
ate approximation to the data, separate examination of the observed and expected
distributions of the difference between the scores of the two teams reveals a ten-
dency to underestimate the proportion of drawn matches. This is attributed to
interdependence between the scores of the home and away teams, and is corrected
by modelling scores using a bivariate Poisson distribution. The marginal distribu-
tions are the same as before, but allowance is also made for positive correlation
between the home- and away-team scores in each match.
Dixon and Coles (1997) employ Maher’s (1982) modelling approach for dif-
ferent ends: they seek to develop a forecasting model capable of generating ex
ante match outcome probabilities. Instead of using the bivariate Poisson distri-
bution, the marginal Poisson probabilities for the scores of both teams in low-
scoring games (H≤1 and A≤1) are adjusted directly to allow for interdependence,
which, as already seen, appears to be greatest towards the top-left-hand corners
of Tables 3.4 and 3.5. For forecasting purposes, estimation of the vectors of αi
and βj must be based on historical data only. This is achieved using a pseudo-
likelihood function, in which the attacking and defensive parameters are esti-
mated from past scores, weighted by a factor that declines exponentially over
time. Each team’s parameters are updated from match to match, as the scores
from the most recently completed matches enter the estimation. Graham and
Stott (2008) apply a similar estimation strategy to an ordered probit regression
for match results. Dynamic team-strength parameters for each team are estimated
using a likelihood function in which the contribution of past results decays expo-
nentially, with the rate of decay determined by maximising the model’s predictive
capability in a holdout sample.
Dixon and Pope (2004) compare probabilistic forecasts obtained from the
Dixon-Coles model with probabilities inferred from UK bookmakers’ prices for
fixed-odds betting. Using a forecasting approach similar to that of Dixon-Coles,
Rue and Salvesen (2000) allow the attacking and defensive parameters for all
teams to vary randomly over time. The estimates of these parameters are updated
as new data on match outcomes are obtained. Markov chain Monte Carlo itera-
tive simulation techniques are used for inference. Crowder et al. (2002) propose a
procedure for updating the team strength parameters that is computationally less
demanding than the method proposed by Rue and Salvesen.
Koning’s (2000) approach to the modelling of match results describes a collec-
tion of match results retrospectively, instead of generating forecasts or estimated
probabilities prospectively. Koning advocates modelling results directly, rather
than indirectly through scores, partly on grounds of simplicity: fewer parameters
are required, the estimation procedures are more straightforward, and the speci-
fied ordered probit model lends itself quite easily to the inclusion of dynamics or
other explanatory variables.
Karlis and Ntzoufras (2003) apply a variety of bivariate Poisson models to
the task of modelling goal scoring in football matches. In addition to the double
and bivariate Poisson distributions, and diagonal-inflated versions of the latter, a
Poisson difference distribution is also considered, in which the dependent variable
is the difference between the scores of the home and away teams. Using data from
Italy’s Serie A from a single season, Karlis and Ntzoufras estimate an uncondi-
tional model, and a model in which the expectations of the goals scored by each
team are conditional on a limited number of team characteristics.
Bittner et al. (2007) compare the use of the negative binomial distribution and
several generalised extreme value distributions (Weibull, Gumbel and Frechet) to
describe the distribution of goals scored per match. The outcome for each match
is interpreted as the product of a self-affirmation process, in which the probabil-
ity that either team scores within each microscopic interval within the match (for
example, within each minute) depends on the number of goals it has scored pre-
viously. The form of the adaptation relation, which describes the evolution of the
scoring probabilities over time, determines the shape of the distribution for the
total number of goals per match. The adaptation relation accommodates various
types of feedback effect: for example, a team that has scored previously draws
encouragement and is more likely to score again; or a team that is leading plays
defensively in order to protect its lead, and is less likely to score again. Assumptions
concerning the adaptation relation are identified that generate a negative binomial
distribution or a generalised extreme value distribution for the total goals scored
per match.1
The next four sections describe the specification and estimation of forecasting
models for match results expressed in either ‘win-draw-lose’ format, or ‘scores’ for-
mat, and the application of the estimated models to obtain probabilistic forecasts
for scores or results in out-of-sample matches. The two types of forecasting model
are referred to as the goals-based model, which is described in Sections 4.2 and 4.3;
and the results-based model, described in Sections 4.4 and 4.5. An early prototype
of the results-based model is reported in Dobson and Goddard (2001) and is devel-
oped subsequently in Goddard and Asimakopoulos (2004). Goddard (2005) draws
comparisons between the forecasting performance of goals-based and results-based
models with specifications similar to those reported in this chapter.
4.2 A goals-based forecasting model

The bivariate Poisson distribution provides a convenient representation of a data
set in which each observation (match) generates two discrete goals-scored vari-
ables, for the home and away teams, respectively. Let λ1,i,j denote the mathemat-
ical expectation of the number of goals scored by the home team in the fixture
A goals-based forecasting model 83
between home team i and away team j. Similarly, let λ2,i,j denote the mathematical
expectation of the number of goals scored by the away team in the same fixture.
In the model, it is assumed that λ1,i,j and λ2,i,j are both linear functions of a set
of covariates, which are defined using data that are available before the match is
played. Let, SHi,0(=CAj,0) and CHi,0(=SAj,0) denote the goals scored and the goals con-
ceded by home team i, respectively (identical to the goals conceded and the goals
scored by away team j). Following Holgate (1964), the bivariate Poisson joint
probability function for SHi,0 and CHi,0 takes the form:
P(SHi,0 = s,C Hi,0 = c) = exp( − λ1,i,j − λ 2,i,j + λ 3,i,j )

min(s,c)
× ∑ (λ1,i,j − λ 3,i,j )s − k (λ 2,i,j − λ 3,i,j )c − k λ 3,i,j
k
/{(s − k)!(c − k)!k!}
k=0
for s = 0,1,2,3,4 … and c = 0,1,2,3,4 … [4.1]

where exp( ) denotes the exponential function: exp(x) = ex = 2.71828x; and k!
denotes ‘k-factorial’: 0! = 1, 1! = 1, 2! = 2×1, 3! = 3×2×1, 4! = 4×3×2×1, and
so on.
This joint probability function can be interpreted as the product of three univari-
ate Poisson probability functions with means λ1,i,j–λ3,i,j, λ2,i,j–λ3,i,j and λ3,i,j, respect-
ively. λ1,i,j, the expected number of goals scored by the home team, depends on
covariates reflecting the propensities of home team i to score and away team j to
concede goals. Similarly λ2,i,j, the expected number of goals scored by the away
team, depends on covariates reflecting the propensities of away team j to score and
home team i to concede goals. Finally, λ3,i,j = η λ1,i,j λ 2,i,j is the covariance between
SHi,0 and CHi,0. η is an additional parameter to be estimated, for which a positive value
(η>0) is expected.
Preliminary experimentation with the estimation of [4.1] indicated a tendency
for the bivariate Poisson regression to underestimate slightly the probabilities for
low-scoring draws. Accordingly, [4.1] can be modified by introducing additional
parameters, which inflate the (unadjusted) probabilities of low-scoring draws, and
deflate all of the other (unadjusted) probabilities, so that the adjusted probabilities
sum to one. Experimentation suggested that inflation of the probabilities of 0–0,
1–1 and 2–2 draws is justified. The diagonal-inflated formulation of the bivariate
Poisson probability function, containing three additional parameters π, θ1 and θ2,
is as follows:

 H = s,C H = c)
P(S i,0 i,0
= (1 − π 0 )P(SHi,0 = s,C Hi,0 = c) + π 0 (1 − π1 − π 2 ) for{s,c} = {0,0}

P(S = s,C = c) = (1 − π )P(S = s,C = c) + π π for{s,c} = {1,1}
H H H H
i,0 i,0 0 i,0 i,0 0 1
 H = s,C H = c) = (1 − π )P(SH = s,C H = c) + π π
P(S for{s,c} = {2,2}
i,0 i,0 0 i,0 i,0 0 2

P(Si,0 = s,C i,0 = c) = (1 − π 0 )P(Si,0 = s,C i,0 = c)
H H H H
for{s,c} ≠ {0,0},{1,1},{2,2}
[4.2]
where P�(SHi,0 = s, CHi,0 = c) denotes the adjusted probability (and P(SHi,0 = s, CHi,0 = c)
denotes the unadjusted probability) that the home team scores s goals and concedes
c goals.
The full list of covariate definitions used in the calculation of λ1,i,j and λ2,i,j is as
follows.
Fi,y,s
d
= fi,y,s
d
/ni,y, where fi,y,s
d
= team i’s total goals scored in matches played 0–12
months (y = 0) or 12–24 months (y = 1) before current match; within the cur-
rent season (s = 0) or previous season (s = 1) or two seasons ago (s = 2); in the
team’s current tier (d = 0) or one (d = ±1) or two (d = ±2) tiers above or below
the current tier; and ni,y = team i’s total matches played 0–12 months (y = 0) or
12–24 months (y = 1) before current match.
Ai,y,s
d
= ai,y,s
d
/ni,y, where ai,y,s
d
= team i’s total goals conceded, defined for the same y,s,d
as above; ni,y defined as above.
SHi,k = goals scored in k’th most recent home match by team i, for k = 1, …,9.
CHi,k = goals conceded in k’th most recent home match by team i, for k = 1, …,9.
SAi,k, CAi,k = goals scored and conceded in k’th most recent away match by team i,
for k = 1, … 4.
S j,k, CHj,k = goals scored and conceded in k’th most recent home match by team j,
H
for k = 1, … 4.
SAj,k, CAj,k = goals scored and conceded in k’th most recent away match by team j,
for k = 1, … 9.
SIGHi,j = 1 if match is significant for championship, promotion or relegation
issues for home team i but not for away team j; 0 otherwise.
SIGAi,j = 1 if match is significant for away team j but not for home team i; 0
otherwise.
CUPi = 1 if team i is eliminated from the FA Cup; 0 otherwise.
EUINi = 1 if team i is involved in European competition during the current sea-
son, and has not yet been eliminated; 0 otherwise.
EUOUTi = 1 if team i has been involved in European competition during the cur-
rent season, and has been eliminated; 0 otherwise.
DISTi,j = natural logarithm of the geographical distance between the grounds of
home team i and away team j.
APi,s = residual for team i from a cross-sectional regression of the natural loga-
rithm of average home attendance on final league position (defined on a scale
of 92 for the top team in T1 to 1 for the bottom team in T4) s seasons before the
present season, for s = 1,2.
d9700, d0104 and d0508 are 0–1 dummy variables identifying matches played in
seasons 1997–2000, 2001–2004 and 2005–2008, respectively.
CUPj, EUINj, EUOUTj, APj,s are defined as above, for team j.
Table 4.1 reports the numerical values of the estimated coefficients in the linear
equations for λ1,i,j and λ2,i,j, and the four additional parameters η, π0, π1 and π2 ,
Table 4.1 Goals-based forecasting model: estimated coefficients and p-values
Equation for λ1,i,j (expectation of home goals) Equation for λ2,i,j (expectation of away goals)
Covar. Coeff. p-val. Covar. Coeff. p-val. Covar. Coeff. p-val. Covar. Coeff. p-val. Covar. Coeff. p-val. Covar. Coeff. p-val.
0 0
Fi,0,0 .3576 .000 SHi,1 .0141 .016 SIGHi,j .0927 .035 Fj,0,0 .3426 .000 CHi,1 .0119 .042 SIGHi,j −.0928 .004
F+1
i,0,1 .5175 .000 SHi,2 .0064 .270 SIGAi,j −.1369 .000 F+1
j,0,1 .2807 .000 CHi,2 .0036 .533 SIGAi,j .1077 .004
0 0
Fi,0,1 .3629 .000 SHi,3 .0117 .044 CUPi −.0996 .000 Fj,0,1 .4005 .000 CHi,3 .0175 .003 CUPi .0202 .343
F−1
i,0,1 .2326 .000 SHi,4 .0107 .066 CUPj .0468 .066 F−1
j,0,1 .6204 .000 CHi,4 .0030 .609 CUPj −.0141 .514
F+1
i,1,1 .2125 .003 SHi,5 .0017 .767 EUINi .2102 .000 F+1
j,1,1 .1208 .001 CHi,5 .0002 .969 EUINi −.0729 .013
0 0
Fi,1,1 .1318 .005 SHi,6 .0053 .353 EUINj .1650 .000 Fj,1,1 .1456 .000 CHi,6 −.0123 .032 EUINj −.0322 .356
F−1
i,1,1 .0995 .019 SHi,7 .0020 .735 EUOUTi −.1905 .000 F−1
j,1,1 .2354 .000 CHi,7 .0101 .080 EUOUTi .1453 .000
F+2
i,1,2 .1286 .617 SHi,8 .0022 .707 EUOUTj −.0568 .172 F+2
j,1,2 −.0566 .546 CHi,8 −.0004 .949 EUOUTj .0769 .050
F+1
i,1,2 .1958 .004 SHi,9 .0163 .005 ΔAPi,1 .0649 .055 F+1
j,1,2 .1694 .000 CHi,9 −.0010 .865 ΔAPi,1 −.1690 .000
0 0
Fi,1,2 .1532 .001 SAi,1 .0149 .029 APi,2 .1183 .000 Fj,1,2 .1641 .000 CAi,1 .0064 .197 APi,2 −.1396 .000
F−1
i,1,2 .1544 .000 SAi,2 .0020 .763 ΔAPj,1 −.1111 .001 F−1
j,1,2 .1853 .002 CAi,2 .0081 .103 ΔAPj,1 .0840 .003
F−2
i,1,2 −.0324 .693 SAi,3 .0214 .002 APj,2 −.1446 .000 F−2
j,1,2 .1076 .498 CAi,3 .0080 .106 APj,2 .0842 .000
0 0
Aj,0,0 .4277 .000 SAi,4 −.0011 .866 DISTi,j .0327 .000 Ai,0,0 .3121 .000 CAi,4 .0050 .321 DISTi,j −.0266 .000
A+1j,0,1 .2797 .000 CHj,1 .0111 .107 d9700 .0055 .776 A+1i,0,1 .3455 .000 SHj,1 .0061 .214 d9700 .0101 .541
0 0
Aj,0,1 .3622 .000 CHj,2 .0086 .206 d0104 .0036 .853 Ai,0,1 .1922 .000 SHj,2 .0121 .014 d0104 .0212 .201
A−1j,0,1 .4805 .000 CHj,3 .0143 .038 d0508 −.0406 .038 A−1i,0,1 .1073 .005 SHj,3 .0091 .065 d0508 −.0358 .032
A+1j,1,1 .0634 .153 CHj,4 .0123 .069 const. −.2220 .008 A+1i,1,1 .0513 .393 SHj,4 .0018 .721 const. −.1105 .123
0 0
Aj,1,1 .1108 .020 CAj,1 .0147 .013 Ai,1,1 .0317 .423 SAj,1 .0213 .000
A−1j,1,1 .3049 .000 CAj,2 .0083 .156 A−1i,1,1 .0406 .251 SAj,2 .0032 .580
A+2j,1,2 −.0852 .494 CAj,3 .0101 .084 A+2i,1,2 .1710 .458 SAj,3 .0084 .146
A+1j,1,2 .0781 .062 CAj,4 .0036 .533 A+1i,1,2 .1356 .017 SAj,4 −.0012 .828
0 0
Aj,1,2 .1096 .020 CAj,5 .0029 .623 Ai,1,2 .1249 .002 SAj,5 −.0026 .653
A−1j,1,2 .1997 .004 CAj,6 .0027 .645 A−1i,1,2 .0979 .004 SAj,6 .0120 .038
A−2j,1,2 .2965 .103 CAj,7 .0103 .074 A−2i,1,2 .0188 .787 SAj,7 .0085 .135
CAj,8 .0084 .148 SAj,8 .0138 .016
CAj,9 .0057 .322 SAj,9 .0025 .664
Ancillary parameters: η̂ = .0403 (p-value = .000) πˆ 0 = .0224 (p-value = .000) πˆ 1 = .5816 (p-value = .000) πˆ 2 = .1623 (p-value = .000)
estimated using data for seasons 1993 to 2008, inclusive. The coefficient values
are estimated using 32,552 match result observations for all matches played in
T1–T4 of the Premier League and the Football League. In addition to the match
results from these sixteen seasons, data from the two preceding seasons (1991
and 1992), and data from the non-league division immediately below T4 (known
as the Conference, to and from which either zero, one or two T4 teams were rel-
egated and promoted per season), are used in the construction of the covariates
used in the estimation.
In the goals-based model, the average goals scored and goals conceded vari-
ables Fi,y,s
d
and Ai,y,s
d
(for team i and their counterparts for team j), calculated over
the 24 months prior to the current match, are the main indicators of attacking
and defensive capability, or team quality. Positive coefficients on Fi,y,s d
and Ai,y,s
d
are expected. It is assumed that team i’s propensity to score is captured by its
+1
average scoring rate over the previous 12 months, Fi,0,0
0
+ ∑ Fi,0,1
d
, and its average
d = -1
+1 +2
scoring rate between 12 and 24 months ago, ∑ Fi,1,1
d
+ ∑ Fi,1,2
d
. The individual
d = -1 d = -2
components of these sums make separate contributions to the attack-

ing and defensive capability measures. For example, if the current-season
scoring rate is a better indicator of the team’s current attacking capability
than the previous-season scoring rate in the same division within the same
12-month period, the coefficient on Fi,0,0 0
should (and does) exceed the coeffi-
cient on Fi,0,1. Each of the sets of covariates, Ai,y,s
0 d
, Fj,y,s
d
and Aj,y,sd
plays a simi-
lar role in identifying the attaching and defensive capabilities of team i and
team j, using data from preceding matches. The estimated coefficients on
all of these variables are predominantly correctly signed and well defined.
The recent goals scored and conceded variables SHi,k, SAi,k, CHi,k and CAi,k (and their
counterparts for team j) allow for the inclusion of goals data from the few most
recent matches of both teams. The possibility of short-term persistence in team
performance suggests these variables may have particular relevance in helping
predict the outcome of the current match, over and above their contribution to
the average goals scored and conceded covariates Fi,y,sd
, Ai,y,s
d
, Fj,y,s
d
and Aj,y,s
d
.
In general, however, the estimated coefficients on the recent goals scored
and conceded covariates tend to be rather erratic. Data on the home team’s
recent home performance are found to be more useful as predictors than data
on its recent away performance; and similarly the away team’s recent away per-
formance is more useful than its recent home performance. Accordingly, data
are included from the home team’s last nine home matches and last four away
matches, and from the away team’s last four home matches and last nine away
matches.
The identification of matches with significance for end-of-season champion-
ship, promotion and relegation issues is relevant if match outcomes are affected by
incentives. If a match is significant for one team and insignificant for the other, the
A goals-based forecasting model 87
teams may contribute different levels of effort. Of several alternative definitions

of significance that were considered, the chosen algorithm produces the match
significance dummies with the most explanatory power. A match is significant
if it is possible (before the match is played) for the team in question to qualify
automatically for entry into European club competition the following season (for
teams near the top of T1) or be promoted or relegated (for teams near the bot-
tom of T1, and for teams in T2, T3 and T4), assuming all other teams currently
in contention for the same end-of-season outcome take one point on average from
each of their remaining fixtures. The signs and significance of the estimated coef-
ficients on SIGHi,j and SIGAi,j are predominantly consistent with incentive effects
as described above.
The FA Cup is a sudden-death knockout tournament involving both league
and non-league teams. Teams from T3 and T4 enter the cup in the first round,
and teams from T1 and T2 enter in the third round. The final is played at the end
of the league season. Early elimination from the cup might have implications
for a team’s performance in subsequent league matches, though the direction of
the effect is uncertain. A team eliminated from the cup may be able to concen-
trate its efforts on the league, suggesting an improvement in league performance.
Alternatively, elimination may reduce confidence (while progress fosters team
spirit), suggesting a decline in league performance. The signs and significance of
the estimated coefficients on CUPi and CUPj suggest that the second of these two
effects dominates.
Teams that finish in the top few positions in T1, and teams that are successful
in the two major domestic cup tournaments (the FA Cup and the League Cup),
are eligible to compete in European club competition during the following season.
Until the end of the 1999 season, there were three European club tournaments;
but subsequently this number was reduced to two, known (until the end of the
2009 season) as the Champions League and the UEFA Cup. Between the 1993 and
2009 seasons, the number of English participants in the two or three European
club tournaments was never less than five and never more than nine. The propor-
tion of matches that involve teams that participated in European club competition
in the same season is therefore small; though the matches concerned, involving the
most successful teams, are naturally the subject of disproportionate media scru-
tiny. The signs and significance of the estimated coefficients on EUINi, EUOUTi,
EUINj and EUOUTj suggest that current involvement in European competition
has a positive effect on league performance; while involvement earlier in the cur-
rent season (curtailed by elimination prior to the date of the current league fix-
ture) has an effect that is less pronounced, but still positive.
The coefficients on DISTi,j indicate that geographical distance is a significant
influence on match outcomes. The signs and significance of the estimated coef-
ficients on DISTi,j suggest home advantage is stronger in matches between teams
whose home stadia are located long distances apart, and weaker in matches
between teams whose stadia are located near one another. The greater intensity of
competition in local derbies may have some effect in offsetting home-field advan-
tage, while the psychological or practical difficulties of long-distance travel for
teams and supporters may increase home-field advantage in matches between
teams from distant cities.
Finally, the covariates APi,s and APj,s are positive for teams that tend to attract
higher-than-average home attendances after controlling for league position (and
negative in the opposite case). These covariates allow for a ‘big team’ effect on
match outcomes: regardless of the values of other controls, ‘big’ teams are more
likely (and ‘small’ teams less likely) to win, either through the direct influence
of the crowd on the outcome, or because teams with a larger revenue base have
more resources to spend on players. To reduce the effect of temporary variation
in the attendance-performance relationship, the values of these variables for the
two preceding seasons are included in the model. Since the values over successive
seasons tend to be highly correlated, ΔAPi,1 = APi,1–APi,2 (and its counterpart for
team j) is used in place of APi,1 (and APj,1).
4.3 Probabilistic forecasts for match results in ‘scores’ format

For each fixture for which match result probabilities are required, the model evalu-
ates the mathematical expectations of the goals scored by the home and away
teams. These are obtained using a linear combination of a set of covariate values
based on past match results and other data for the two teams, and a set of coef-
ficients that are estimated using all observations in the match results database and
reported in Table 4.1. Once the mathematical expectations of the goals scored by
the home and away teams have been calculated, these expectations are substituted
into a modified version of the bivariate Poisson joint probability function in order
to obtain probabilities for each specific match result.
For expositional purposes, it is convenient to break down the task of cal-
culating the probabilities for a forthcoming fixture into a number of steps, as
follows:
1. Calculation of the numerical values of the model’s covariates for the fixture for
which probabilities are required.
2. Calculation of the expectations of the numbers of goals scored by the home
and away teams.
3. Conversion of the expectations of goals scored into a set of probabilities for
each specific match result.
The structure and layout of Section 4.3 follows these three steps.
Calculation of the numerical values of the model’s covariates

The calculation of the numerical values for each of the covariates will be explained
using (as an illustration) the T1 (Premier League) Hull City vs Tottenham Hotspur
Forecasts in ‘scores’ format 89
Table 4.2 Data for calculation of average goals scored and conceded covariates,
Hull City vs Tottenham Hotspur fixture
Matches Goals Goals Match

Tier played for against result
‘points’
Hull City
2009 season, before 23/02/09 T1 25 31 46 11.0
2008 season, after 23/02/08 T2 13 24 10 8.5
2008 season, before or on 23/02/08 T2 33 41 37 18.5
2007 season, after 23/02/07 T2 13 18 19 6.0
Tottenham Hotspur
2009 season, before 23/02/09 T1 25 26 31 9.5
2008 season, after 23/02/08 T1 12 18 20 5.5
2008 season, before or on 23/02/08 T1 26 48 41 12.0
2007 season, after 23/02/07 T1 11 25 15 8.5
Note: Match result ‘points’ are awarded 1 for a win, 0.5 for a draw, 0 for a loss.
Source: Sky Sports Football Yearbook
fixture from the 2009 season, played on 23 February 2009. It is helpful to note
that all i-subscripts on variables refer to Hull City, and all j-subscripts refer to
Tottenham Hotspur.
(i) Average goals scored and conceded over past 12 months
The covariates Fi,y,s
d
and Ai,y,s
d
(and their counterparts for team j) are the model’s
main team quality controls. For the Hull–Tottenham fixture, Fi,y,s d
(averages of
the goals scored by Hull over the past 24 months, partitioned by time period,
season and division) and Aj,y,s d
(averages of the goals conceded by Tottenham
over the past 24 months, partitioned in the same way) are used in the equa-
tion for λ1,i,j. Similarly, Fj,y,s
d
(averages of the goals scored by Tottenham)
and Ai,y,s (averages of the goals conceded by Hull) are used in the equation
d
for λ2,i,j.
Table 4.2 reports the data for the calculation of the average goals scored and
conceded covariates for the Hull–Tottenham fixture. Using these data, Hull’s cov-
ariate values are as follows:
F0 = 31 /( 25 + 13) = 0.8158 A 0 = 46 /( 25 + 13) = 1.2105
i,0,0 i,0,0
Fi,0,1
-1
= 24 /( 25 + 13) = 0.6316 A -1i,0,1 = 10 /( 25 + 13) = 0.2632
Fi,1,1
-1
= 41 /(33 + 13) = 0.8913 A -1i,1,1 = 37 /(33 + 13) = 0.8043
Fi,1,2
-1
= 18 /(33 + 13) = 0.3913 A -1i,1,2 = 19 /(33 + 13) = 0.4130
Tottenham’s covariate values are as follows:

Fj,0,0
0
= 26 /( 25 + 12 ) = 0.7027 A 0j,0,0 = 31 /( 25 + 12 ) = 0.8378
Fj,0,1
0
= 18 /( 25 + 12 ) = 0.4865 A 0j,0,1 = 20 /( 25 + 12 ) = 0.5405
Fi,1,1 = 48 /( 26 + 11) = 1.2973
0
A 0j,1,1 = 41 /( 26 + 11) = 1.1081
Fj,1,2
0
= 25 /( 26 + 11) = 0.6757 A 0j,1,2 = 15 /( 26 + 11) = 0.4054
Zero values are assigned to Fi,y,s d

, Ai,y,s
d
, Fj,y,s
d
and Aj,y,s
d
for all permutations of values
for d, y and s that are not applicable for the current fixture.
Hull’s goals scored ratio (average goals scored per match) over the 12 months
prior to the Hull–Tottenham fixture is Fi,0,0 0
+ Fi,0,1
-1
= 1.4474. Similarly, Hull’s goals
scored ratio over the period 12–24 months prior to the Hull–Tottenham fixture is
Fi,1,1
-1
+ Fi,1,2
-1
= 1.2826 . Effectively, the definitions of Fi,y,s d
(and of Ai,y,s
d
, Fj,y,s
d
and Aj,y,s
d
)
partition these ratios into components relating to the current season (s = 0), the
previous season (s = 1) and two seasons ago (s = 2), and into components relating
to the team’s current tier (d = 0), one tier above or below the team’s current tier
(d = ±1) and two tiers above or below the team’s current tier (d = ±2).
The partitioning enables each component of the team’s goals ratios to make
different contributions to the evaluation of the expectations of the goals scored
and conceded for the current fixture. For example, the coefficients that are
reported in Table 4.1 attach more weight to goals scored in the current season
than to goals scored in the previous season, or to goals scored two seasons ago.
Goals scored in a higher tier carry more weight than goals scored in the team’s
current tier, which in turn carry more weight than goals scored in a lower tier.
(ii) Goals scored and conceded in the two teams’ most recent matches
The covariates SHi,m and CHi,m (goals scored and conceded in home team i’s m’th
most recent home match), SAi,n and CAi,n (goals scored and conceded in home team
i’s n’th most recent away match), and their counterparts for away team j, allow for
the inclusion of each team’s few most recent home and away results in the calcu-
lation of the expectations of the goals scored and conceded by both teams in the
current fixture. These covariates also contribute towards the values of Fi,y,sd
, Ai,y,s
d
(and their counterparts for team j), and are to some extent correlated with these
covariates. By including the goals scored and conceded in each of the teams’ most
recent matches as separate covariates, however, greater flexibility is achieved in the
use of past match results data to predict future results.
Experimentation indicated that data from the home team’s recent home matches
are more useful as predictors than data from its recent away matches; and similarly
data from the away team’s recent away matches are more useful than data from its
recent home matches. Accordingly, data from the home team’s nine most recent
home matches, and from the home team’s last four most recent away matches are
used. Similarly, data from the away team’s four most recent home matches, and
from the away team’s nine most recent away matches are used.
Table 4.3 Data for calculation of goals scored and conceded in recent matches
covariates, Hull City vs Tottenham Hotspur fixture
Hull City Tottenham Hotspur

Last nine home results Last four away Last four home Last nine away
(most recent first) results results results
D 2–2 D 0–0 D 0–0 L 2–3

L 1–3 L 0–2 W 3–1 L 0–1
L 0–1 L 0–2 D 1–1 L 0–2
L 1–4 L 1–5 D 0–0 L 1–2
W 2–1 — — W 2–0
D 2–2 — — L 1–2
L 0–1 — — W 2–1
L 0–3 — — D 4–4
W 1–0 — — L 1–2
For the Hull–Tottenham fixture, SHi,k and SAi,k (goals scored by Hull in recent
home and away matches) and CHj,k and CAj,k (goals conceded by Tottenham) are
used in the equation for λ1,i,j. Similarly, CHi,k and CAi,k (goals conceded by Hull) and
SHj,k and SAj,k (goals scored by Tottenham) are used in the equation for λ2,i,j.
Table 4.3 reports the data for the calculation of the goals scored and conceded
in recent matches covariates for the Hull–Tottenham fixture. Using these data, the
covariate values are as follows:
Hull City: SHi,1 = 2, CHi,1 = 2; SHi,2 = 1, CHi,2 = 3; … ; SHi,9 = 1, CHi,9 = 0

SAi,1 = 0, CAi,1 = 0; SAi,2 = 0, CAi,2 = 2; … ; SAi,4 = 1, CAi,4 = 5
Tottenham Hotspur: SHj,1 = 0, CHj,1 = 0; SHj,2 = 3, CHj,2 = 1; … ; SHj,4 = 0, CHj,4 = 0
SAj,1 = 2, CAj,1 = 3; SAj,2 = 0, CAj,2 = 1; … ; SAj,9 = 1, CAj,9 = 2
(iii) Significance of the match for end-of-season outcomes
The covariates SIGHi,j and SIGAi,j control for the likelihood that match outcomes
are affected by incentives: if a fixture is significant for one team and insignificant
for the other, the incentive differential is likely to influence the match result. The
algorithm used to assess whether or not a fixture is significant is as follows: the
fixture is significant if it is still possible (before the match is played) for the team
in question to win the championship or be promoted or relegated, assuming that
all other teams currently in contention for the same outcome take one point on
average from each of their remaining fixtures.
In T1, the teams that finish fifth and higher qualify automatically to play in
European club competition in the following season, and the teams that finish
eighteenth and lower are relegated. Therefore for the Hull–Tottenham fixture,
the algorithm assesses whether it is still possible for either team to finish fifth (or
higher) or eighteenth (or lower), assuming the teams currently in those positions
take one point from each of their remaining fixtures.
On the morning of 23 February 2009, Hull were thirteenth in T1 with 29 points
from 25 matches; Arsenal were fifth with 45 points from 26 matches; Tottenham
were sixteenth with 25 points from 25 matches; and Blackburn were eighteenth
with 23 points from 25 matches (with 38 matches to be played in total by each
team). The algorithm establishes that the Hull–Tottenham fixture is significant for
both teams. Therefore SIGHi,j = 0 and SIGAi,j = 0.
(iv) Involvement of teams in FA Cup and European competition
Early elimination from the FA Cup or from Europe may have implications for
a team’s results in subsequent league matches in either direction. On the one
hand, a team eliminated from the cup or from Europe may be able to concen-
trate its efforts on the league, suggesting an improvement in league results. On the
other hand, elimination may cause loss of confidence, suggesting a deterioration
in league results. On 23 February 2009, Hull were currently involved in the fifth
round of the FA Cup; but Tottenham had already been eliminated. Hull did not
participate in European competition during the 2009 season; but on 23 February
2009 Tottenham were currently involved in the third round of the UEFA Cup.2
Accordingly, for the Hull–Tottenham fixture:
CUPi = 0; CUPj = 1; EUINi = 0; EUOUTi = 0; EUINj = 1; EUOUTj = 0
(v) Geographical distance
The greater intensity of competition in local derbies may have some effect in
offsetting home-field advantage in such matches, while the difficulties of long-
distance travel for the away team and its supporters may have the effect of
increasing home-field advantage in matches between teams from opposite ends
of the country.
The geographical distance (as the crow flies, and based on map references)
between the grounds of Hull City and Tottenham Hotspur is 152 miles. The
covariate DISTi,j is the natural logarithm of this number: DISTi,j = ln(152) =
5.0239.
(vi) Average attendance relative to league position over previous two seasons
The covariates APi,s and APj,s are positive for teams that tend to attract higher-
than-average home attendances after controlling for league position (and negative
in the opposite case). These covariates allow for a ‘big team’ effect on match out-
comes: regardless of the values of other controls, ‘big’ teams may be more likely
(and ‘small’ teams less likely) to win, either through the direct influence of the
crowd on the match result, or because teams with a larger revenue base have more
resources to spend on players. To reduce the effect of temporary variation in the
attendance-performance relationship, the values of these variables for the two preced-
ing seasons are included in the model.
To calculate these covariates for the Hull–Tottenham fixture, cross-sectional

regressions of the natural logarithm of each team’s average home attendance
(denoted LNATT) on each team’s final league position (denoted POS) are esti-
mated, using data for all 116 teams that played in the Premier League, the Football
League and the top non-league division during the 2007 season, and again during
the 2008 season. The estimation results are as follows:
2007 season: LNATT = 7.0624 + 0.0311 POS + RESID

2008 season: LNATT = 6.9694 + 0.0323 POS + RESID
APi,1 and APi,2 are the values of the residuals (denoted RESID) from these
regressions for Hull for the 2007 and 2008 seasons, respectively. Similarly,
APj,1 and APj,2 are the values of the residuals for Tottenham for the same two
seasons.
The data for the calculation of these covariates for the Hull–Tottenham fixture
are as follows:
Hull City: 2007 season average attendance = 18,758; position = 76

2008 season average attendance = 18,025; position = 94
Tottenham Hotspur: 2007 season average attendance = 35,739; position = 112
2008 season average attendance = 35,967; position = 106
The covariate values for the Hull–Tottenham fixture are as follows:
APi,1 = ln(18025) – 6.9694 – 0.0323 × 94 = –0.2060

APi,2 = ln(18758) – 7.0624 – 0.0311 × 76 = 0.4134
ΔAPi,1 = –0.2060 – 0.4134 = –0.6194
APj,1 = ln(35967) – 6.9694 – 0.0323 × 106 = 0.0972
APj,2 = ln(35739) – 7.0624 – 0.0311 × 112 = –0.0616
ΔAPj,1 = 0.0978 – –0.0622 = 0.1600
Calculation of the mathematical expectations of the goals scored by the home

and away teams
Once the numerical values of all covariates have been calculated, the expectations
of the goals scored by the home and away teams (λ1,i,j and λ2,i,j) are obtained by
substituting the covariate values into the two linear equations whose coefficients
are reported in Table 4.1.
The mathematical expectations of the goals scored by the home and away teams
are obtained by combining these coefficients with the numerical values of the cov-
ariates for the fixture in question. In total, there are sixty-three covariates (and a
constant term and dummy variables for three four-season periods) on the right-
hand sides of each of these equations. Algebraic expressions for the equations for
λ1,i,j and λ2,i,j are as follows:
Expectation of home team’s goals:

λ1,i,j = (α 0 + α 66 d0508) + α1Fi,0,0
0
+ α 2 Fi,0,1
+1
+ α 3Fi,0,1
0
+ α 4 Fi,0
-1
,1 + … + α 63 DISTi,j
Expectation of away team’s goals:

λ 2,i,j = (β0 + β66 d0508) + β1Fj,0,0
0
+ β2 Fj,0,1
+1
+ β3Fj,0,1
0
+ β 4 Fj,0
-1
,1 + … + β63 DISTi,j
For the Hull–Tottenham fixture:

λ1,i,j = [ −0.2220 + ( −0.0406 × 1)] + ( 0.3576 × 0.8158) + ( 0.5175 × 0 ) + ( 0..3629 × 0 )
+ ( 0.2326 × 0.6316 ) +  + ( 0.0327 × 5.0239) = 1.3347
λ 2,i,j = [ −0.1105 + ( −0.0358 × 1)] + ( 0.3426 × 0.7027 ) + ( 0.2807 × 0 )
+ ( 0.3982 × 0.4005) + ( 0.6204 × 0 ) +  + ( −0.0266 × 5.0239) = 1.5198
Conversion of expectations of goals scored into match result probabilities

The estimated coefficients reported in Table 4.1 are used in conjunction with
equations [4.1] and [4.2] to convert the numerical values of λ1,i,j and λ2,i,j into a grid
containing probabilities for individual scores.
For the Hull–Tottenham fixture, λ1,i,j = 1.3347 and λ2,i,j = 1.5198. The ancillary
parameters are η = 0.0403, π0 = 0.0224, π1 = 0.5816, π2 = 0.1623. This implies:

λ 3,i,j = η λ1,i,j λ 2,i,j = 0.0403 × 1.3347 × 1.5198 = 0.0572
− λ1,i,j − λ 2,i,j + λ 3,i,j = −2.7974
exp( − λ1,i,j − λ 2,i,j + λ 3,i,j ) = exp( −2.7974 ) = 0.0610
λ1,i,j − λ 3,i,j = 1.2776
λ 2,i,j − λ 3,i,j = 1.4627
To illustrate the application of equations [4.1] and [4.2], the calculation of the
adjusted bivariate Poisson probability for a score of Hull 2 Tottenham 3 is shown
in full below.
Using [4.1], the unadjusted bivariate Poisson probability is:

P(SHi,0 = 2, C Hi,0 = 3) = exp( − λ1,i,j − λ 2,i,j + λ 3,i,j )
min( 2 ,3 )
× ∑ ( λ1,i,j − λ 3,i,j )2 − k (λ 2,i,j − λ 3,i,j )3 − k λ 3,i,j
k
/{( 2 − k )!(3 − k )! k !}
k=0
2
= 0.0610 × ∑ 1.27762 − k1.46273 − k 0.0572k /{( 2 − k )!(3 − k )! k !}
k=0
= 0.0610 × [1.27762 × 1.46273 × 0.05720 /{2 !×3 !× 0 !}
+ 1.27761 × 1.46272 × 0.05721 /{1!× 2 !×1!}
+ 1.27760 × 1.46271 × 0.05722 /{0 !×1!× 2 !}]] = 0.0309
Using [4.2], the adjusted bivariate Poisson probability is:

 (SH = 2, C H = 3) = (1 − π )P(SH = s, C H = c ) = (1 − 0.0224 ) × 0.0309 = 0.0301
P i,0 i,0 0 i,0 i,0
A results-based forecasting model 95
Table 4.4 Estimated match result probabilities, Hull City vs Tottenham Hotspur
fixture
Away goals → 0 1 2 3 4 5 6 7
Home goals ↓
0 .0663 .0894 .0670 .0325 .0120 .0035 .0009 .0002

1 .0756 .1281 .0875 .0443 .0168 .0051 .0013 .0003
2 .0472 .0740 .0615 .0301 .0117 .0037 .0009 .0002
3 .0196 .0317 .0255 .0136 .0055 .0017 .0005 .0001
4 .0061 .0102 .0084 .0046 .0019 .0006 .0002 —
5 .0015 .0026 .0022 .0012 .0005 .0002 — —
6 .0003 .0006 .0005 .0003 .0001 — — —
7 .0001 .0001 .0001 .0001 — — — —
Table 4.4 shows the full set of adjusted bivariate Poisson probabilities for the
Hull–Tottenham fixture.
4.4 A results-based forecasting model

The description of the results-based forecasting model starts by introducing
the notation that is used for the dependent variable. Each match represents one
observation in the data set, and the dependent variable can be defined, without
any loss of generality, for the match between home team i and away team j. Let
RHi,k denote the result of team i’s k’th most recent home match from team i’s per-
spective, coded 1 for a win, 0.5 for a draw and 0 for a defeat. k = 0 refers to the
current match (the match between home team i and away team j); and integer
values of k>0 refer to team i’s previous home matches (k = 1 is team i’s last home
match, before the match with team j; k = 2 is team i’s last-but-one home match;
and so on).
In an ordered probit regression model, match results in win-draw-lose format
are used as the dependent variable. Accordingly, the dependent variable in the
ordered probit regression model is RHi,0. Let RAj,0 denote the result of team j’s k’th
most recent away match, defined in the same way but from team j’s perspective.
Accordingly, RAj,0 = 1 − RHi,0. It is assumed that the result of the match between
team i and team j depends upon an unobserved or latent variable denoted yi,j* ; and
an independent and identically distributed disturbance term, εi,j, which follows
the standard Normal distribution.
Home win: RHi,0 = 1 if µ 2 < y*i,j + ε i,j
Draw: RHi,0 = 0.5 if µ1 < y*i,j + ε i,j < µ 2

Away win: RHi,0 = 0 if y*i,j + ε i,j < µ1 [4.3]
The latent variable yi,j* is assumed to be a linear function of a set of covariates,

which are defined using data that are available before the match is played. The full
list of covariate definitions used in the calculation of yi,j* is as follows.
Pi,y,s
d
= pi,y,s
d
/ni,y, where pi,y,s
d
= team i’s total ‘points’ score (calculated by awarding
1 ‘point’ for a win, 0.5 ‘points’ for a draw and 0 ‘points’ for a loss) in matches
played 0–12 months (y = 0) or 12–24 months (y = 1) before the current match;
within the current season (s = 0) or previous season (s = 1) or two seasons ago
(s = 2); in the team’s current tier (d = 0) or one (d = ±1) or two (d = ±2) tiers
above or below the current tier; and ni,y = team i’s total matches played 0–12
months (y = 0) or 12–24 months (y = 1) before the current match.
Pj,y,s
d
= pj,y,s
d
/nj,y, where pj,y,s
d
is team j’s total ‘points’ score, and nj,y = team j’s total
matches played (both defined as above).
RHi,k = result of team i’s k’th most recent home match (coded 1 for a win by team
i, 0.5 for a draw and 0 for a loss) for k = 1, …,9.
RAi,k = result of team i’s k’th most recent away match (coded 1 for a win by team i,
0.5 for a draw and 0 for a loss) for k = 1, …,4.
R j,k defined as above for team j, for k = 1, …,4.
H
RAj,k defined as above for team j, for k = 1, …,9.

SIGHi,j, SIGAi,j, CUPi, CUPj, EUINi, EUOUTi, EUINj, EUOUTj, DISTi,j, APi,s,
APj,s, d9700, d0104, d0508 are all as defined in Section 4.2.
Table 4.5 reports the numerical values of the estimated coefficients in the linear
equation for yi,j* , estimated using data for the seasons 1993–2008, inclusive.
In the results-based model, the average match results ‘points’ variable Pi,y,s d
for
team i, and its counterpart Pj,y,s for team j, calculated over the 24 months prior to
d
the current match, are the principal indicators of team quality. These covariates are
similarly defined, and play a similar role, to the goals scored and conceded covari-
ates, SHi,m, CHi,m, SAi,n and CAi,n, in the goals-based model. The indexing of these vari-
ables allows for separate contributions to the team quality measures from ‘points’
gained in matches played: 0–12 months (y = 0) or 12–24 months (y = 1) before the
current match; within the current season (s = 0) or previous season (s = 1) or two
seasons ago (s = 2); and in the team’s current tier (d = 0) or one (d = ±1) or two (d =
±2) tiers above or below the current tier.
Positive coefficients on Pi,y,s d
and negative coefficients on Pj,y,s
d
are expected. It is
assumed that the quality of team i is captured by its average ‘points’ score over the
+1
previous 12 months, equivalent to its win ratio over this period, Pi,0,0
0
+ ∑ Pi,0,1
d
,
d = -1
and its average ‘points’ score or win ratio between 12 and 24 months ago,
+1 +2
∑ Pi,1,1
d
+ ∑ Pi,1,2
d
. The individual components of these sums make separate contri-
d = -1 d = -2
butions to the team quality measures. For example, if the current-season win ratio is
a better indicator of the team’s current quality than the previous-season’s win
ratio in the same division within the same 12-month period, the coefficient on Pi,0,0
0
Table 4.5 Results-based forecasting model: estimated coefficients and p-values
Covariate Coefficient p-value Covariate Coefficient p-value Covariate Coefficient p-value

0
Pi,0,0 1.6727 .000 RHi,1 .0150 .349 SIGHi,j .1222 .002
P+1
i,0,1 1.9362 .000 RHi,2 .0197 .218 SIGAi,j −.1416 .000
0
Pi,0,1 1.1905 .000 RHi,3 .0407 .011 CUPi −.0643 .007
P−1
i,0,1 .7283 .000 RHi,4 .0236 .141 CUPj .0318 .184
P+1
i,1,1 .8169 .000 RHi,5 −.0050 .756 EUINi .1607 .000
0
Pi,1,1 .6193 .000 RHi,6 −.0341 .033 EUOUTj .0714 .095
P−1
i,1,1 .4572 .000 RHi,7 .0186 .245 EUINi −.1564 .000
P+2
i,1,2 .4563 .561 RHi,8 −.0123 .440 EUOUTj −.0034 .935
P+1
i,1,2 .3555 .072 RHi,9 .0148 .355 ΔAPi,1 .1935 .000
0
Pi,1,2 .3723 .003 RAi,1 .0036 .825 APi,2 .1565 .000
P−1
i,1,2 .3975 .000 RAi,2 .0202 .208 ΔAPj,1 −.1374 .000
P−2
i,1,2 .0757 .707 RAi,3 .0286 .074 APj,2 −.1558 .000
0
Pj,0,0 −1.2529 .000 RAi,4 −.0065 .687 DISTi,j .0437 .000
P+1
j,0,1 −1.1888 .000 RHj,1 −.0384 .017 d9700 .0110 .544
0
Pj,0,1 −.7697 .000 RHj,2 −.0204 .205 d0104 −.0121 .507
P−1
j,0,1 −.4493 .000 RHj,3 −.0271 .093 d0508 −.0472 .010
P+1
j,1,1 −.9749 .000 RHj,4 −.0210 .192
0
Pj,1,1 −.6081 .000 RHj,1 −.0498 .002
P−1
j,1,1 −.3275 .003 RAj,2 −.0008 .961
P+2
j,1,2 −1.4198 .076 RAj,3 −.0315 .049
P+1
j,1,2 −.6729 .001 RAj,4 −.0248 .121
0
Pj,1,2 −.5134 .000 RAj,5 −.0088 .580
P−1
j,1,2 −.3185 .002 RAj,6 −.0048 .763
P−2
j,1,2 −.2332 .261 RAj,7 −.0235 .141
RAj,8 −.0166 .299
RAj,9 −.0478 .003
Cut-off parameters: ˆμ1 = –.3650 μˆ 2 = .3901

should (and does) exceed the coefficient on Pi,0,1 0

. Predominantly, the estimated
coefficients on the covariates Pi,y,s and Pj,y,s are correctly signed and well defined.
d d
Preliminary experimentation indicated that the coefficients on {Pi,y,s d

, Pj,y,s
d
} were
highly significant for y = 0,1 (data from matches played 0–12 months and 12–24
months before the current match); but not for y = 2 (24–36 months before the cur-
rent match).
The recent match results variables RHi,k and RAi,k, and their counterparts RHj,k and
R j,k for team j, allow for the inclusion of match results data from the most recent
A
matches of both teams. The possibility of short-term persistence in team perform-

ance suggests these variables may have particular relevance in helping predict the
outcome of the current match, over and above their contribution to the covariates
{Pi,y,s
d
, Pj,y,s
d
}.
In general, the estimated coefficients on the recent match results covariates
tend to be rather erratic. Experimentation indicated that data on the home team’s
recent home performance are more useful as predictors than data on its recent
away performance; and similarly the away team’s recent away performance is more
useful than its recent home performance. Statistically significant estimated coef-
ficients are obtained for some (but not all) values of k≤9 for RHi,k and RAj,k, and for
some n≤4 for RAi,k and RHj,k. Accordingly, data are included from the home team’s
last nine home matches and last four away matches, and from the away team’s last
four home matches and last nine away matches.
The remaining covariates SIGHi,j, SIGAi,j, CUPi, CUPj, EUINi, EUOUTi,
EUINj, EUOUTj, DISTi,j, APi,s, APj,s play the same role in the results-based fore-
casting model as in the goals-based model; and the patterns in the signs and sig-
nificance of the coefficients on these covariates in Table 4.5 are similar to those in
Table 4.1. See Section 4.2 for commentary on the contribution of these covariates
to the model.
4.5 Probabilistic forecasts for match results in ‘win-draw-lose’ format

Section 4.5 provides a summary description of the use of the results-based fore-
casting model introduced in Section 4.4 to obtain probabilistic forecasts for match
results in ‘win-draw-lose’ format. The method is similar in principle to the method
for the goals-based model that is described in detail in Section 4.3 (and many of
the covariate definitions are the same). Accordingly, the following description of
the method for the results-based model is relatively concise. As before, the method
is illustrated using the T1 Hull City vs Tottenham Hotspur fixture played on 23
February 2009.
For each fixture for which ‘win-draw-lose’ probabilities are required, the model
evaluates the latent variable yi,j* in [4.3]. yi,j* is a linear combination of a set of cov-
ariate values based on past match results and other data for the two teams, and a
set of estimated coefficients.
Forecasts in ‘win-draw-lose’ format 99
In the results-based model, the past match results ‘points’ covariates Pi,y,s
d
and
P are the main team quality controls. Using the data reported in Table 4.2, the
d
j,y,s
covariate values are as follows:

HullCity : Pi,0,0
0
= 11 /( 25 + 13) = 0.2895
Pi,0,1
-1
= 8.5 /( 25 + 13) = 0.22237
Pi,1,1 = 18.5 /(33 + 13) = 0.4022
-1
Pi,1,2
-1
= 6 /(33 + 13) = 0.1304
Tottenham Hotspur: Pj,0,0

0
= 9.5 /( 25 + 12 ) = 0.2568
Pj,0,1
0
= 5.5 /( 25 + 12 ) = 0.1486
Pj,1,1
0
= 12 /( 26 + 11) = 0.3243
Pj,1,2
0
= 8.5 /( 26 + 11) = 0.2297
Zero values are assigned to Pi,y,s

d
and Pj,y,s
d
for all permutations of values for d, y and
s that are not applicable for the current fixture.
The covariates RHi,m and RAi,n (results of home team i’s m’th most recent home
match for m = 1 … 9 and n’th most recent away match for n = 1 … 4), and their
counterparts for away team j, allow each team’s few most recent home and away
results to enter the calculation of yi,j* individually. Using the data reported in Table
4.3, the covariate values are as follows:
HullCity: RH = 0.5;RH = 0;RH = 0;…;RH = 1
i,1 i,2 i,3 i,9
R = 0.5;R
A
i,1
A
i,2 = 0;…;R A
i,4 =0
Tottenham Hotspur: RHj,1 = 0.5;RHj,2 = 1;…;RHj,4 = 0.5

RAj,1 = 0;RAj,2 = 0;RAj,3 = 0;…;RAj,9 = 0
The values for all of the remaining covariates SIGHi,j, SIGAi,j, CUPi, CUPj, EUINi,
EUOUTi, EUINj, EUOUTj, DISTi,j, APi,s, APj,s are the same as those reported in
Section 4.3, and the calculations are not repeated here.
Once the numerical values of all covariates have been calculated, the latent vari-
able yi,j* is evaluated by substituting the covariate values into the linear equation
whose coefficients are reported in Table 4.5:

y*i,j = γ 1Pi,0,0
0
+ γ 2 Pi,0,1
+1
+ γ 3 Pi,0,1
0
+ γ 4 Pi,0,1
-1
+ … + γ 39 DISTi,j + γ 42 d0508
For the Hull–Tottenham fixture:

y*i,j = (1.6727 × 0.2895) + (1.9362 × 0 ) + (1.1905 × 0 ) + ( 0.7283 × 0.2237))
+  + ( 0.0437 × 5.0239) + ( −0.0472 × 1) = 0.0100
To obtain the ‘win-draw-lose’ probabilities, the evaluated yi,j* and the estimated
cut-off parameters μ̂1 and μ̂2 are substituted into a rearranged formulation of [4.3],
as follows:
Home win probability = pi,jH = P(εi,j > µ

ˆ 2 − yî,j* ) = 1 − Φ (µ ˆ 2 − yî,j* )
Draw probability ˆ 1 − yî,j* < εi,j < µ
= pi,jD = P(µ ˆ 2 − yî,j* ) = Φ (µˆ 2 − yî,j* ) − Φ (µ
ˆ 1 − yî,j* )
Away win probability = pi,jA = P(εi,j < µ ˆ 1 − yî,j* )
ˆ 1 − yî,j* ) = Φ (µ [4.4]

where Φ( ) is the distribution function for the standard Normal distribution. For
the Hull–Tottenham fixture, the match result probabilities in ‘win-draw-lose’ for-
mat are as follows:
pH = 1 − Φ( 0.3901 − 0.0100 ) = 0.3519
i,j
pDi,j = Φ( 0.3901 − 0.0100 ) − Φ( −0.3650 − 0.0100 ) = 0.2942
pAi,j = Φ( −0.3650 − 0.0100 ) = 0.3538
4.6 Evaluation of the goals-based and results-based forecasting models

Section 4.6 provides an evaluation of the forecasting performance of the goals-based
and results-based models. Table 4.6 reports the match result probabilities produced
by both models for all ten T1 matches played over the same weekend as the 2009 sea-
son Hull City vs Tottenham Hotspur fixture (21–23 February 2009), together with
probabilities for selected scores produced by the goals-based forecasting model.
Table 4.6 also reports the averages of the implied match result probabilities cal-
culated from the prices for fixed-odds betting quoted by six high-street and inter-
net bookmakers. Let OHi,j denote a bookmaker’s odds for a home win in the match
between team i and team j, quoted in decimal form. For example, OHi,j = 4 implies
a bet of £1 placed on a home win pays £4 (£3 winnings plus return of £1 stake) if
the bet wins. A fair bet would have expected winnings of zero, which is the case in
this example if the probability of a home win is (1/OHi,j) = 1/4. Invariably, however,
bookmakers’ odds contain a margin for costs and profit. This implies a deflation
adjustment is required to calculate a set of bookmakers’ implied probabilities that
sum to one. Let ODi,j and OAi,j denote the decimal odds for a draw and an away win.
The implied probability for a home win is:
oHi,j = (1/OHi,j )/{(1/OHi,j ) + (1/ODi,j ) + (1/O Ai,j )} [4.5]
The implied probabilities for a draw and an away win, denoted oDi,j and oAi,j, are
similarly defined. Table 4.6 provides an indication of the typical degree of vari-
ation in the probabilities that are generated by the goals-based and result-based
forecasting models, and the bookmakers’ implied probabilities. For several fix-
tures, such as Stoke vs Portsmouth, Middlesbrough vs Wigan, Fulham vs West
Bromwich Albion and Newcastle vs Everton, the two forecasting models and the
Evaluation of forecasting models 101
bookmakers were in close agreement. For other fixtures, however, there are some
significant divergences. As shown above, the two forecasting models diverge in
their assessment of the probabilities for the Hull vs Tottenham fixture. In this
case the bookmakers’ implied probabilities split the difference between those of
the models. For the evenly balanced Aston Villa vs Chelsea fixture, the book-
makers took a more pessimistic view than the models of the home team’s pros-
pects. The bookmakers’ view was justified by the match result. For the highly
unbalanced Manchester United vs Blackburn fixture, the home team were even
stronger favourites according to the models than they were according to the
bookmakers. The home team duly won the match, though only by a narrow
margin.
A convenient summary measure of forecasting performance over any number
of matches, which can be applied to a set of probabilities generated by a forecast-
ing model or to a set of bookmakers’ implied probabilities, is pseudo-R2, defined
as the geometric mean of the probabilities assigned to the actual result of each
matches played during the forecast period. Let πi denote the probability that was
assigned to the actual result of match i, and let n denote the number of matches
for which forecasts were generated. Equivalent formulae for pseudo-R2 are as
follows:
n
pseudo-R2 = (π1π 2 … π n )1/n or pseudo-R2 = exp[(1/n)∑ ln(π i )] [4.6]
i=1
Table 4.7 reports ten sets of pseudo-R2 values, each calculated over all matches
played during the 2009 season. This measure is evaluated for the match results
probabilities generated by the goals-based and results-based models, and for a
‘model consensus’ set defined as the averages of the probabilities generated by
these two models. The pseudo-R2 measure is also evaluated for each of the six
sets of bookmakers’ implied probabilities, and for a ‘bookmaker consensus’ set
defined as the averages of the six sets of bookmaker probabilities. The results are
reported separately by tier, and for all four tiers combined.
The overall forecasting performance of the forecasting models and the book-
makers is rather similar, but there is some variation in the pattern by tier. For T1
and T3, the results-based model and the model consensus record higher pseudo-
R2 values than any of the six bookmakers. The performance of the goals-based
model falls somewhere in the middle of the range for the six bookmakers. For
T2 and T4, all six bookmakers record higher pseudo-R2 values than the forecast-
ing models. Overall the bookmakers hold a narrow advantage, but the differences
in forecasting performance appear to be very small. Chapter 12 presents a more
detailed evaluation of the informational efficiency of the six bookmakers’ betting
odds, and examines whether forecasting models of the kind developed in Chapter
4 can be employed to ‘beat the bookmaker’ and generate a positive return on
fixed-odds betting.
Table 4.6 Fitted match result and selected score probabilities, Premier League, weekend of 21–23 February 2009
Date 21/02 21/02 21/02 21/02 21/02 21/02 22/02 22/02 22/02 23/02
Home team Arsenal Aston Villa Bolton Man Utd Stoke Middlesbro’ Fulham Liverpool Newcastle Hull
Away team Sunderland Chelsea West Ham Blackburn Portsmouth Wigan WBA Man City Everton Tottenham
Home pos. 5th 3rd 14th 1st 17th 20th 11th 2nd 15th 12th
Away pos. 10th 4th 8th 18th 13th 7th 19th 9th 6th 16th
Match result probabilities: goals-based model
Home .6819 .3436 .3385 .7729 .3747 .3956 .5674 .5647 .3066 .3131
Draw .2175 .2901 .2982 .1727 .2903 .3225 .2490 .2563 .2915 .2717
Away .1006 .3663 .3633 .0544 .3350 .2819 .1836 .1790 .4018 .4152
Match result probabilities: results-based model
Home .7462 .3771 .3260 .8100 .4099 .3927 .5507 .6527 .3047 .3519
Draw .1757 .2936 .2934 .1388 .2911 .2927 .2606 .2217 .2917 .2942
Away .0782 .3292 .3805 .0512 .2990 .3147 .1888 .1255 .4036 .3538
Implied match result probabilities: averages of six bookmakers
Home .6555 .2958 .3959 .7452 .3794 .3892 .5223 .5989 .3281 .3325
Draw .2266 .2882 .2880 .1770 .2896 .2990 .2714 .2528 .2888 .2830
Away .1179 .4160 .3161 .0778 .3310 .3218 .2063 .1483 .3831 .3845
Selected match scores probabilities: goals-based model
1–0 .1439 .0968 .1038 .1471 .1026 .1420 .1138 .1254 .0930 .0756
2–0 .1402 .0579 .0589 .1644 .0646 .0784 .1009 .1063 .0510 .0472
2–1 .0891 .0770 .0749 .0754 .0811 .0751 .0966 .0951 .0709 .0740
3–0 .0910 .0231 .0223 .1225 .0271 .0289 .0596 .0601 .0187 .0196
3–1 .0600 .0317 .0293 .0585 .0351 .0288 .0589 .0555 .0268 .0317
0–0 .0796 .0867 .0972 .0716 .0872 .1342 .0699 .0798 .0905 .0663
1–1 .1011 .1376 .1406 .0778 .1377 .1437 .1186 .1216 .1380 .1281
2–2 .0319 .0547 .0511 .0209 .0545 .0395 .0498 .0461 .0528 .0615
0–1 .0435 .1007 .1085 .0278 .0957 .1134 .0576 .0619 .1101 .0894
0–2 .0128 .0626 .0643 .0059 .0562 .0501 .0258 .0258 .0714 .0660
1–2 .0270 .0801 .0782 .0142 .0757 .0600 .0489 .0469 .0838 .0875
0–3 .0025 .0260 .0254 .0008 .0220 .0147 .0077 .0072 .0309 .0325
1–3 .0055 .0343 .0320 .0021 .0306 .0184 .0151 .0135 .0375 .0443
Match result 0–0 0–1 2–1 2–1 2–2 0–0 2–0 1–1 0–0 1–2
Source: Sky Sports Football Yearbook; www.football-data.co.uk

Table 4.7 Pseudo-R-square values for forecasting performance, 2009 season: goals-

based forecasting model, and probabilities derived from six betting firms’ prices
T1 T2 T3 T4 All
Forecasting models
Goals-based 0.3827 0.3444 0.3533 0.3429 0.3533
Results-based 0.3839 0.3440 0.3538 0.3429 0.3535
Model consensus 0.3836 0.3445 0.3539 0.3434 0.3538
(average)
Bookmakers
Bet365 0.3834 0.3473 0.3535 0.3450 0.3548
Bet&Win 0.3822 0.3467 0.3533 0.3450 0.3544
Gamebookers 0.3826 0.3476 0.3535 0.3455 0.3549
Ladbrokes 0.3809 0.3473 0.3531 0.3445 0.3541
Sporting Bet 0.3823 0.3467 0.3530 0.3453 0.3544
William Hill 0.3835 0.3484 0.3535 0.3453 0.3553
Bookmaker consensus 0.3826 0.3475 0.3536 0.3452 0.3549
(average)
Conclusion
Chapter 4 has developed goals-based and results-based forecasting models for
match outcomes recorded in the form of either goals scored and conceded by the
two teams, or in the form of ‘win-draw-lose’ match results. Both types of fore-
casting model are estimated by fitting regression models to past match results
data. A diagonal-inflated bivariate Poisson regression is used for the goals-based
model, and an ordered probit regression is used for the results-based model.
Both models draw on an extensive set of covariates, reflecting past goal-scoring
performance and match results over the preceding 24 calendar months, the sig-
nificance of the match for end-of-season championship, promotion or relegation
outcomes, the current involvement of the teams in the FA Cup or European
tournaments, the average attendances attracted by both teams, and the geo-
graphical distance between the home stadia of the two teams. The use of the
fitted models to generate out-of-sample probabilistic forecasts for future match
results in goals or results format has been described in some detail, and illus-
trative match result forecasts for a full round of Premier League (T1) matches
played during the 2009 season have been presented. Comparisons between the
forecasting performance of the models on the one hand, and a selection of high-
street and internet bookmakers on the other hand, indicate that the differences
in forecasting capability are small. The contribution of the forecasting models
developed in this chapter to the design of a betting strategy is examined in more
detail in Chapter 12.
Evaluation of forecasting models 105
Notes
1 Bittner et al. (2009) report further evidence on the tails of the probability distributions of
goals scored by the home and away teams, which are modelled most effectively using extreme
value distributions. See also, Hever, Miller and Rubner (2010) and Ribeiro et al. (2010).
2 Having drawn their FA Cup fifth round match against Sheffield United, Hull went on to
win the fifth round replay, but were eliminated by Arsenal in the sixth round. Tottenham
were eliminated by Manchester United in the fourth round. Having lost the away leg of their
UEFA Cup third round tie against Shakhtar Donetsk, Tottenham were eliminated four days
after the Hull fixture after being held to a draw by Shakhtar in the home leg.
5 Game theory and football games
Introduction
Game theory is the study, by mathematicians, economists and decision scientists,
of decision-making in situations of conflict and interdependence. Most games
played in real life are complex, with multiple strategies, incomplete information
and pay-offs that might not be explicitly specified. By contrast, sports sometimes
give rise to situations in which the structure of a ‘game’ (using the term in the
technical sense) is simple and clearly defined. Accordingly, some economists have
argued that sports such as football, and others, offer a highly promising arena for
the empirical investigation of the propositions of game theory.
In football, the ‘game’ between the kicker and the goalkeeper that is played
out each time a penalty kick is awarded and taken approximates rather closely to
the simple and highly stylised examples typically used to develop the principles of
game theory in economics textbooks. The kicker must decide in which direction
to shoot and the goalkeeper must decide in which direction to dive. Each decides
simultaneously, before knowing the other’s selection. The reward structure is
zero-sum: either the kicker scores; or a goal is prevented because the goalkeeper
saves or the kicker shoots high or wide. Section 5.1 of this chapter examines the-
oretical and empirical research on the strategic choices of kickers and goalkeepers
during those few intense and highly charged moments that elapse between the
referee’s decision to award a penalty, and its execution.
Viewed more generally, the football match in its entirety has many of the char-
acteristics of a strategic and dynamic ‘game’ (again, using the term in its technical
sense). In every match, the two teams are pitched into direct opposition for a
period of play of 90 minutes’ notional duration.1 Each team starts the match with
one goalkeeper and ten outfield players. Team managers or coaches are at liberty
to deploy their outfield players in any formation of their choosing, and to adjust
their team’s formation and style of play at any stage of the match. The immedi-
ate objectives of the two teams throughout the match are symmetric: each team
attempts to score goals and prevent its opponent from scoring. Accordingly there
is a high level of interdependence: both teams’ strategies have implications for
both teams’ chances of scoring and conceding goals.
106
The penalty kick 107
The payoff structure is other than zero-sum, however, for two reasons. First, if
the scores at the end of the match are unequal, three league points are awarded to
the winning team and none to the losing team. If the scores are level, one point
is awarded to each team. Second, a player who commits serious foul play may be
dismissed while the match is in progress. No replacement is permitted, and play
resumes with the dismissed player’s team at a numerical disadvantage in players.
A dismissal also results in the player’s suspension from up to three future matches;
in this case, replacements are permitted. However, a suspension still imposes a cost
upon the team concerned, which is obliged to select from a smaller or weaker pool
of players. The corresponding gain accrues to the team’s future opponents and
not its current opponent.
The remaining sections of this chapter examine the extent to which the in-play
strategic behaviour of football teams (during the course of each match) can be
rationalised in accordance with game-theoretic principles of optimising strategic
behaviour by independent agents when payoffs are interdependent and non-zero-
sum. A theoretical model of strategic choice is described and subjected to empir-
ical scrutiny, using data on the timings of player dismissals and goals scored
in more than 16,000 English league matches played between the 2002 and 2009
seasons.
Section 5.2 describes a game-theoretic model of strategic behaviour for foot-
ball teams, in which the teams choose between an offensive or defensive forma-
tion, and between a violent and non-violent style of play, and can vary these
choices continuously over the duration of the match. Section 5.3 summarises a
data set containing minute-by-minute details of the timings of player dismissals
and goals scored. Section 5.4 describes the specification of an empirical model
for the incidence and timings of dismissals and goals. Section 5.5 reports and
interprets the estimation results. Finally, Section 5.6 uses stochastic simulations
to obtain in-play probabilities for match results conditional upon the state of
the match (goals already scored and dismissals already having occurred) at any
stage.
5.1 The penalty kick

Academic interest in the penalty kick in football, viewed primarily as a vehicle for
empirical tests of the propositions of game theory, has grown rapidly since the
publication of a seminal paper by Chiappori, Levitt and Groseclose (2002). This
study examines the selection of strategies by kickers and goalkeepers during the
execution of penalty kicks. In the theoretical model devised by Chiappori, Levitt
and Groseclose, the kicker selects one of three strategies denoted L, C and R: kick
left, kick centre or kick right. Similarly the goalkeeper selects one of three strat-
egies also denoted L, C, R: dive to the kicker’s left (the goalkeeper’s right), remain
in the centre, or dive to the kicker’s right (the goalkeeper’s left). The following
assumptions are made:
• Kicks taken with the kicker’s natural foot are more effective. Accordingly, kicks
taken by right-footed kickers (the majority) are more likely to score when L is
selected.
• If the kicker selects C, a goal is never scored if the goalkeeper also selects C.
• If the kicker selects L and the goalkeeper selects L, the probability that a
goal is scored is positive but smaller than the probability when the goalkeeper
selects R.
• If the goalkeeper selects L, the probability that a goal is scored is smaller if the
kicker selects C than it is if the kicker selects R.
• Relations similar to those stated in the previous two bullets hold between the
probabilities when either the kicker or the goalkeeper selects R.
The structure of this game is such that there is no pure-strategy equilibrium. If a
right-footed kicker always selects L so as to kick on his stronger side, for example,
it would pay the goalkeeper to always select L. In that case, however, it would
be better for the kicker to select R rather than L. The game does have a mixed-
strategy equilibrium, in which both players make their selections randomly, but
with specific probabilities that depend upon the probabilities that the penalty is
converted for all of the possible permutations of strategies. In equilibrium, the
kicker’s scoring probability is the same whether he kicks L, C or R, and the goal-
keeper’s probability of averting a goal is the same whether he dives L, C or R.
This indifference property is a standard feature of mixed-strategy equilibria. If the
players were not indifferent, then it would pay them to adjust their probabilities
towards more frequent selection of the strategy with the higher scoring probabil-
ity (in the case of the kicker) or the strategy with the higher probability of averting
a goal (in the case of the goalkeeper).
The indifference property leads to several testable propositions:
(i) The right-footed kicker selects L (his natural side) more often than R;
(ii) The goalkeeper selects L more often than R;
(iii) The goalkeeper selects L more often than the right-footed kicker;
(iv) The kicker selects C more often than the goalkeeper.
It is straightforward to show that departures from these propositions lead to
violations of the indifference property. In the case of (i) for example, if the right-
footed kicker selects L and R with equal probability, the goalkeeper would not
be indifferent between L and R, because he would avert a goal more often by
selecting R (diving to the kicker’s weaker side). Likewise in the case of (ii), if
the goalkeeper selects L and R with equal probability, the right-footed kicker
would not be indifferent between L and R, because he would score more often
by selecting L (kicking on his stronger side). Selecting C is highly damaging
for the kicker if the goalkeeper also selects C. For the kicker to be indifferent
between C and either L or R, in accordance with (iv), the goalkeeper must only
select C very rarely.
The penalty kick 109
Using data compiled from match videotapes from the top tiers of the French
and Italian leagues containing 459 penalty kicks, Chiappori, Levitt and Groseclose
subject these propositions to empirical scrutiny. The empirical analysis is subject
to an aggregation problem. While the relationships identified above should hold
for the game played between each individual kicker and goalkeeper, equivalent
relationships might not hold at the aggregate level in data that are compiled over a
heterogeneous collection of players. For example, if Kevin always selects L when
taking a penalty against Gareth, and Kevin always selects R when taking a penalty
against Graham (in contravention of the randomisation strategy that is predicted
by the model), the aggregated data on Kevin’s kicks might appear random, but
Kevin’s selections against either goalkeeper are in fact completely deterministic.
Despite the difficulties created by this aggregation issue, several tabulations
and statistical tests are presented that appear consistent with the main propos-
itions of the model. Table 5.1 reports a summary tabulation of the proportions of
occasions kickers and goalkeepers selected each strategy, and the outcomes (pro-
portions of occasions the kicks were successful) in each case. In accordance with
the indifference property, the proportions of successful kicks and goals averted
are similar for each of the three strategies available to both kickers and goalkeep-
ers. Propositions (i) to (iv) are all supported. Both kickers and goalkeepers select
L more often than R; and goalkeepers select L more often than kickers. Kickers
select C more often than goalkeepers; the latter do so only rarely. In accordance
with the randomisation hypothesis, Chiappori, Levitt and Groseclose find that
the strategy selected for the previous kick taken by the kicker or faced by the
goalkeeper does not predict the strategy chosen for the next kick, after control-
ling for the overall proportions of selections by each player in the data set. While
most footballers might very well be oblivious to the intricacies and technicalities
of game theory, the results of this study suggest that they nevertheless succeed,
through either good intuition or good coaching, in devising optimising strategies
that correspond closely to those that are recommended by the theoreticians.
Using the same data set as Chiappori, Levitt and Groseclose, Coloma (2007)
tests for the validity of the mixed-strategy equilibrium by estimating a simul-
taneous equations regression model in which the dependent variables are linear
probability model-type regressions, with the kicker’s and goalkeeper’s selections
and the outcome (whether the penalty is converted or not) as the dependent
variables, and covariates including a dummy variable for the kicker’s right- or
left-footedness and various other indicators of the state of the match at the time
the penalty is taken. The game-theoretic model imposes various restrictions on
the coefficients of these equations, which are supported by hypothesis tests on the
estimated model.
Palacios-Huerta (2003) presents further empirical evidence to support the
hypothesis that the behaviour of kickers and goalkeepers is consistent with a mixed-
strategy equilibrium and random selection of strategies, based on a larger data set
obtained from TV footage of 1,417 penalty kicks from several European countries
Table 5.1 Observed proportions of penalty kicks, goalkeeper dives and goals
Kicker
Goalkeeper ↓ Left Centre Right Total
Percentages of all penalty kicks classified by kicker and goalkeeper strategies

Left 25.5 10.5 20.7 56.6
Centre 0.9 0.7 0.9 2.4
Right 18.5 6.1 16.3 41.0
Total 44.9 17.2 37.9 100.0
Percentages of successful kicks for each pairing of kicker and goalkeeper strategies
Left 63.2 81.2 89.5 76.2
Centre 100.0 0.0 100.0 72.7
Right 94.1 89.3 44.0 73.4
Total 76.7 81.0 70.1 74.9
Source: Adapted from Chiappori, Levitt and Groseclose (2002)
that were screened on US TV. The larger data set permits more repeat observa-
tions on individual kickers and goalkeepers than were available in the Chiappori,
Levitt and Groseclose study, and more powerful tests of the null hypothesis of no
serial correlation or persistence in the selection of strategies by individual players
over successive kicks.
Palacios-Huerta reports runs tests (Mood, 1940) for the randomness of observed
sequences of strategic choices by individual players. In some other cases where the-
ory suggests that players should randomise their selections, empirical evidence of
negative serial correlation has been reported. The observed choice of first serves by
tennis players, for example, tends to switch too frequently (alternating selections on
successive serves) for consistency with randomisation (Walker and Wooders, 2001).
In the case of penalty kicks, by contrast, Palacios-Huerta is usually unable to reject
a null hypothesis of no serial correlation. There is also evidence consistent with the
indifference property of the mixed-strategy equilibrium: the success rates of both
kickers and goalkeepers are similar for each strategic choice.
In other empirical research on penalty-taking and goalkeeping, Bar-Eli et al.
(2007) analyse data on 286 penalties drawn from TV footage of various European
leagues, for 18 of which the goalkeeper selected C (remain in the centre), 6 of which
in turn were saved. On the basis of this 33 per cent success rate, it is suggested that
goalkeepers dive more frequently than is optimal, and should remain in the centre
more often. This is attributed to an action bias on the part of goalkeepers: the
desire to be seen to be taking positive action (diving) results in suboptimal behav-
iour in the form of a reluctance to select the neutral action of non-movement.
Azar and Bar-Eli (2010) report further analysis of the same data set, in the form
of tests for the validity of the mixed-strategy equilibrium model.
A game-theoretic model of in-play strategic choice 111
Dohmen (2008b) uses data on all 3,619 penalties awarded in the top tier of
the German Bundesliga from its inception until the 2004 season, in order to
examine whether the proportions of penalties not converted due to the kicker
shooting high or wide are sensitive to either the importance of the kick for the
match result, or to home-team status. Kickers are found to shoot high or wide
more frequently when playing at home than when playing away. Using a smaller
German Bundesliga data set comprising 835 penalties, Kuss, Kluttig and Stoll
(2007) investigate whether there are differences in conversion rates of penalties
when the kicker was the victim of the foul that led to the award of the penalty,
and penalties taken by kickers who were not the victims. No evidence is found of
any significant difference in conversion rates. Jordet et al. (2007) report that the
number of penalty misses increases with the importance of the kick, and Jordet,
Hartman and Sigmundstad (2009) find that kickers who were delayed in taking
a penalty, often by the goalkeeper, were more likely to miss. Bauman, Friehe and
Wedow (2010) find that a kicker’s general ability is a reliable indicator of his
success rate.
McGarry and Franks (2000) examine the optimal selection of kickers in pen-
alty shoot-outs, held to determine winners at the end of drawn matches in some
tournaments (usually after 30 minutes’ extra time has been played in addition
to the regulation 90 minutes). In a shoot-out, each team has a minimum of five
penalties, each of which must be taken by a different kicker. If the scores are level
after five kicks each, the shoot-out continues on a sudden-death basis until a win-
ner emerges. In a probability analysis, the later kicks in the initial sequence of five
bear greater weight in determining the match result than the earlier kicks. On this
basis, a case can be made for the later kicks to be assigned to the team’s most pro-
ficient kickers. Carillo (2007) proposes that penalty shoot-outs should be staged at
the end of 90 minutes’ play in matches that are level at that stage, before the match
enters a 30-minute period of extra time. The result of the penalty shoot-out would
only count if the match were still level at the end of extra time. It is suggested that
this proposal would improve the attractiveness of play during extra time, by giving
the team that loses the shoot-out an incentive to play offensively.
5.2 A game-theoretic model of in-play strategic choice for football teams

Section 5.2 describes a game-theoretic model of strategic choice for football teams,
in which the teams (or their managers or coaches on their behalf) choose between
an offensive (attacking) or a defensive formation, and between a violent and non-
violent style of play. These choices are permitted to vary continuously over the
duration of each match. This model of optimising strategic behaviour has three
important antecedents in the literature. First, Palomino, Rigotti and Rustichini
(2000) develop a dynamic game-theoretic model, in which the two teams choose
between defensive and attacking formations, and can alter these choices continu-
ously. Working backwards from the end of the match, the teams’ optimal strategies
conditional on the current state of the match are determined by solving a series of
two-player non-cooperative subgames.
In regressions estimated using data on 2,855 matches played in the top tiers
of the English, Italian and Spanish leagues, covariates measuring team quality
effects, the current score and home-field advantage are all significant determinants
of the probability of scoring. Teams tend to play more defensively when they are
leading than they play when they are either level or trailing. The analysis relies
upon an unproven assumption that if one team adopts an attacking strategy and
the other adopts a defensive strategy, the goal-scoring rate of the attacking team
increases by more than that of the defending team. The attacking and defensive
strengths of the two teams are assumed to be the same; and the payoff structure
at the end of the match is assumed, for simplicity, to be zero-sum. However, the
analysis does not account for what appears to be a strong empirical regularity, that
after controlling for team quality and duration effects, the scoring rates of both
teams when the scores are level tend to be lower than those of teams that are trail-
ing, and not significantly different from those of teams that are leading.
Second, Brocas and Carrillo (2004) develop a dynamic game-theoretic model to
examine the effects of the introduction of three-points-for-a-win, and the so-called
‘golden goal’ rule,2 on the choice of attacking and defensive strategies by football
teams. The model consists of three periods: the first and second halves of regular
time, and (for matches that were level after 90 minutes’ play) extra time. As in the
Palomino, Rigotti and Rustichini model, the teams’ optimal strategies conditional
on the state of the match at the start of each period are established by working
backwards from the end of the match. In the Brocas and Carrillo model, three-
points-for-a-win encourages attacking play during the later stages of matches that
are level, but it may have a perverse effect of encouraging teams to play more
defensively during the early stages.
Third, Banerjee, Swinnen and Weersink (2007) examine the impact on team
strategies in the NHL of a change in the league points scoring system affecting
matches that were tied at the end of regular time.3 Both the theoretical model
and the empirical analysis suggest that the rule change had the desired effect of
encouraging attacking play during overtime. However, it also had a perverse effect
of encouraging defensive play during regular time.
Dobson and Goddard (2010) explore the properties of a theoretical model that
allows football teams to choose between defensive and attacking formations, and
between a non-violent and a violent style of play. This model is reviewed in the
remainder of this section. It is assumed that each match consists of a number
of discrete unit time intervals, such that durations within the match can be rep-
resented by t = 0, …, T where T is the complete match duration. It is both con-
venient and natural to think of each time interval as representing one minute, so
that T = 90.
The payoff for each team at the end of the match is dependent on the number
of league points gained, and the cost of future player suspensions arising from
any dismissals incurred during the current match. To allow for the possibility of
risk-averse behaviour on the part of the teams, the payoffs from the league points
gained are determined through a utility function, which may be either linear in
points gained (risk-neutrality) or concave (risk-aversion). The available league
points are 3 for a win, 1 for a draw (tie) and 0 for a loss. Without any loss of gen-
erality, the utility function may be specified as follows: U(0) = 0, U(1) = 1, U(3) =
1+2λ, where λ = 1 represents risk neutrality and 0 < λ < 1 represents risk aversion.
For simplicity the utility cost of a dismissal (arising from future player suspen-
sions) is a fixed utility deduction per player dismissed, denoted ω.
Let s denote the difference in scores at any time during the match, and let dh
and da denote the numbers of dismissals incurred by the home and away teams
prior to that time, respectively. For notational simplicity, time-subscripts denoting
the point in time within the match are suppressed from s, dh and da. Let the value
functions h890(s,dh,da) and a890(s,dh,da) denote the final payoffs to the home and away
teams at the end of the match. Using the notation introduced above, h890(s,dh,da) =
1+2λ–ωdh for s>0 and for any da; 1–ωdh for s = 0; and –ωdh for s < 0. Similarly, a98 0
(s,dh,da) = –ωda for s>0 (and for any dh); 1–ωda for s = 0; and 1+2λ–ωda for s < 0.
For every minute within the match, it is assumed that each team manager
or coach can select either a non-violent or a violent style of play, and either a
defensive or an attacking team formation. The home team’s strategic choices are
denoted by i, defined as follows: i = 1 denotes (non-violent, defend); i = 2 denotes
(violent, defend); i = 3 denotes (non-violent, attack); and i = 4 denotes (violent,
attack). The away team’s choices are denoted by j, defined in the same manner.
Let p denote the probability that the home team scores a goal during any minute
of the match, denoted t. It is assumed, and it is important to note, that p is condi-
tional on x = dh–da, on i and j, and on t. x-, i-, j- and t-subscripts could be appended
to p to represent the dependence of p on these values; but these subscripts are
omitted in order to keep the notation as simple as possible. Similarly, let q denote
the probability that the away team scores a goal during any minute of the match.
Again, it is assumed that q is conditional on x, i, j and t. Let u denote the prob-
ability that the home team has a player dismissed during any minute of the match.
It is assumed that u is dependent on i and on t (but not on x or on j). Finally, let
v denote the probability that the away team has a player dismissed during any
minute of the match. It is assumed that v is dependent on j and on t (but not on
x or on i). For simplicity, it is assumed that only one goal may be scored and one
player dismissed within each minute.
Stylised numerical examples will illustrate how the teams’ strategic choices are
determined. The home team’s expected payoff at the start of the 90th minute is
dependent on the strategic choices, i and j, of both teams for the 90th minute.
These strategic choices determine the values of p, q, u and v, the probabilities of
a goal being scored or a player being dismissed during the 90th minute. From the
home team’s perspective, let h89 i,j
(s,dh,da) denote the expected payoff at the end of
the 89th minute, conditional on the choices of i and j for the 90th minute and the
Table 5.2 Hypothetical goal-scoring and player-dismissal probabilities for

numerical examples
Hypothetical goal-scoring probabilities

Home team = p Away team = q
j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4

i=1 0.03 0.02 0.05 0.04 0.02 0.03 0.03 0.04
2 0.04 0.03 0.06 0.05 0.01 0.02 0.02 0.03
3 0.04 0.03 0.06 0.05 0.04 0.05 0.05 0.06
4 0.05 0.04 0.07 0.06 0.03 0.04 0.04 0.05
Hypothetical player-dismissal probabilities
Home team = u Away team = v
j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
i=1 0.005 0.005 0.005 0.005 0.015 0.020 0.015 0.020
2 0.010 0.010 0.010 0.010 0.015 0.020 0.015 0.020
3 0.005 0.005 0.005 0.005 0.015 0.020 0.015 0.020
4 0.010 0.010 0.010 0.010 0.015 0.020 0.015 0.020
values of s, dh and da at the end of the 89th minute. Suppose initially that the home
team is leading by one goal after 89 minutes, and no players have been dismissed.
Accordingly, s = 1, dh = 0 and da = 0. h89 i,j
(s,dh,da) can be defined as a probability-
weighted average of the payoffs to the home team after 90 minutes contingent on
the values taken by s, dh and da after 90 minutes, which in turn depend only upon
whether a goal is scored or a player is dismissed during the 90th minute:
i,j
h89 (1, 0, 0 ) = q[vh90 ( 0, 0,1) + (1 − u − v ) h90 ( 0, 0,1) + uh90 ( 0,1, 0 )]
+ (1 − p − q )[vh90 (1, 0,1) + (1 − u − v )h90 (1, 0, 0 ) + uh90 (1,1, 0 )]
+ p[vh900 ( 2, 0,1) + (1 − u − v )h90 ( 2, 0, 0 ) + uh90 ( 2,1, 0 )] [5.1]
A similar expression can be defined for a89

i,j
(s,dh,da), the away team’s expected pay-
off at the end of the 89th minute, conditional on the choices of i and j for the 90th
minute and the values of s, dh and da at the end of the 89th minute:
i,j
a89 (1, 0, 0 ) = q[v a90 ( 0, 0,1) + (1 − u − v )a90 ( 0, 0, 0 ) + u a90 ( 0,1, 0 )]
+ (1 − p − q )[v a90 (1, 0,1) + (1 − u − v ) a90 (1, 0, 0 ) + u a90 (1, 0, 0 )]
+ p[v a90 ( 2, 0,1) + (1 − u − v ) a90 ( 2, 0, 0 ) + u a90 ( 2,1, 0 )]
[5.2]

Table 5.2 specifies some hypothetical values for p, q, u and v in the 90th minute,
conditional on i and j. These values of p, q, u and v are used to evaluate h89
i,j
(1,0,0)
and a89(1,0,0) over all possible permutations of i and j. The values of h890(s,dh,da)
i,j
and a890(s,dh,da) in [5.1] and [5.2] are calculated assuming λ = 0.5 and ω = 0.5.
Table 5.3 identifies the two-person non-cooperative subgame solution for the
two teams’ optimal choices of i and j for the 90th minute. For the home team,
Table 5.3 Determination of the two teams’ optimal strategies for the 90th minute,
home team leading by one goal after 89 minutes
Expected pay-offs after 90 minutes at the start of the 90th minute

Home team = h89
i,j
(1,0,0) Away team = a89
i,j
(1,0,0)
j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
i=1 1.9675 1.9575 1.9575 1.9475 0.0275 0.0350 0.0375 0.0450

2 1.9750 1.9650 1.9650 1.9550 0.0175 0.0250 0.0275 0.0350
3 1.9475 1.9375 1.9375 1.9275 0.0475 0.0550 0.0575 0.0650
4 1.9550 1.9450 1.9450 1.9350 0.0375 0.0450 0.0475 0.0550
Note: Payoffs for dominant strategies are highlighted in bold
i = 2 is a dominant strategy. Reading down the columns of the left-hand panel

of Table 5.3, i = 2 always produces the highest payoff to the home team, no
matter whether j = 1,2,3 or 4 is chosen by the away team. For the away team,
j = 4 is a dominant strategy. Reading across the rows of the right-hand panel
of Table 5.3, j = 4 always produces the highest payoff to the away team, no
matter whether i = 1,2,3 or 4 is chosen by the home team. Therefore the non-
cooperative solution is {i = 2,j = 4}: the home team, which leads and wishes to
avoid conceding a goal, plays defensively and aggressively; while the away team,
which trails and needs to score a goal, plays offensively and aggressively. Since
{i = 2,j = 4} are the teams’ chosen strategies for the 90th minute, the uncondi-
tional expectations of the final payoffs to the home and away teams at the end
of the 89th minute are h889(1,0,0) = 1.9550 and a889(1,0,0) = 0.0350.
The solution {i = 2,j = 4} is a Nash equilibrium, because neither team would
wish to alter its choice in view of the choice that is being made by the other
team. However, the solution {i = 2,j = 4} also has the characteristics of a pris-
oner’s dilemma. There exists an alternative cooperative solution, {i = 1,j = 3},
at which both teams would have a higher expected return than they have at the
non-cooperative solution {i = 2,j = 4}. If both teams were to cooperate by play-
ing non-aggressively, both would become better off, because both would avoid
incurring the increased risk of a dismissal. However, the cooperative solution is
unstable, because both teams would have an incentive to alter their choices in view
of the choice that is being made by the other team. Starting from {i = 1,j = 3} for
example, if the home team defects from i = 1 to i = 2 while the away team selects
j = 3, the home team increases its expected payoff at the away team’s expense. This
gives the away team an incentive also to defect to j = 4, restoring the stable, non-
cooperative solution {i = 2,j = 4}.
Table 5.4 identifies the non-cooperative equilibrium solution in the case where
the scores are level at the end of the 89th minute. The expected payoffs are calcu-
lated by adapting [5.1] and [5.2]:
Table 5.4 Determination of the two teams’ optimal strategies for the 90th minute,
scores level after 89 minutes
Expected payoffs after 90 minutes at the start of the 90th minute

Home team = h89
i,j
(0,0,0) Away team = a89
i,j
(0,0,0)
j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
i=1 1.0075 0.9875 1.0175 0.9975 0.9825 1.0000 0.9725 0.9900

2 1.0250 1.0050 1.0350 1.0150 0.9625 0.9800 0.9525 0.9700
3 0.9975 0.9775 1.0075 0.9875 0.9925 1.0100 0.9825 1.0000
4 1.0150 0.9950 1.0250 1.0050 0.9725 0.9900 0.9625 0.9800
i,j
h89 ( 0, 0, 0 ) = q[vh90 ( −1, 0,1) + (1 − u − v )h90 ( −1, 0, 0 ) + uh90 ( −1,1, 0 )]
+ (1 − p − q )[vh90 ( 0, 0,1) + (1 − u − v )h90 ( 0, 0, 0 ) + uh90 ( 0,1, 0 )]
+ p[vh90 (1, 0,1) + (1 − u − v )h90 (1, 0, 0 ) + uh90 (1,1, 0 )] [5.3]
i,j
a89 ( 0, 0, 0 ) = q[v a90 ( −1, 0,1) + (1 − u − v )a90 ( −1, 0, 0 ) + u a90 ( −1,1, 0 )]
+ (1 − p − q )[v a90 ( 0, 0,1) + (1 − u − v ) a90 ( 0, 0, 0 ) + u a90 ( 0,1, 0 )]
+ p[v a90 (1, 0,1) + (1 − u − v ) a90 (1, 0, 0 ) + u a90 (1,1, 0 )] [5.4]
In Table 5.4, {i = 2,j = 2} is the non-cooperative solution, in which both teams play
defensively and aggressively. As before, the non-cooperative solution is a Nash
equilibrium, because neither team would wish to alter its choice in view of the
choice that is being made by the other team. However, the non-cooperative solu-
tion also has the characteristics of a prisoner’s dilemma. In this case there are two
alternative cooperative solutions, {i = 1,j = 1} and {i = 3,j = 3}, at which both
teams would have a higher expected return than they have at the non-cooperative
solution {i = 2,j = 2}. If both teams were to cooperate by playing non-aggressively,
both would become better off. Once again, however, the cooperative solutions are
unstable, because both teams would have an incentive to defect in view of the
choice that is being made by the other team. Since {i = 2,j = 2} are the teams’
chosen strategies for the 90th minute, the unconditional expectations of the final
payoffs to the home and away teams at the end of the 89th minute are h889(0,0,0) =
1.0050 and a88 9(0,0,0) = 0.9800.
By following similar procedures, a complete set of values for h889(s,dh,da) and
a889(s,dh,da) can be established, for all possible values of s, dh and da. Then, the
two teams’ optimal choices of strategy for the 89th minute can be examined, by
identifying the expected payoffs at the end of the 88th minute conditional on
both team’s choices of i and j for the 89th minute. This procedure establishes
a complete set of values for h888(s,dh,da) and a888(s,dh,da), for all possible values
of s, dh and da. Then, the two teams’ optimal choices of strategy for the 88th
minute can be examined, by identifying the expected payoffs at the end of the
87th minute conditional on both team’s choices of i and j for the 88th minute.
By working backwards in a similar manner, the optimal choices of i and j at any
stage of the match can be determined. The general expressions for the relations
that are used to obtain these solutions (of which [5.1] to [5.4] are particular
cases) are:
hti,j (s, d h , da ) = q[vht +1 (s − 1, d h , da + 1)

+ (1 − u − v )ht+1 (s − 1, d h , da ) + uht +1 (s − 1, d h + 1, da )]
+ (1 − p − q )[vht +1 (s,d h ,da + 1)
+ (1 − u − v )ht+1 (s,d h ,da ) + uht+1 (s,d h + 1, da )]
+ p[vht+1 (s + 1,d h ,da + 1)
+ (1 − u − v )ht+1 (s + 1, d h ,da ) + uht+1 (s + 1, d h + 1, da )] [5.5]
a ti,j (s, d h , da ) = q[v at +1 (s − 1, d h , da + 1)

+ (1 − u − v ) at+1 (s − 1, d h , da ) + u at +1 (s − 1, d h + 1, da )]
+ (1 − p − q )[v at +1 (s,d h ,da + 1)
+ (1 − u − v ) at+1 (s,d h ,da ) + u at+1 (s,d h + 1, da )]
+ p[v at+1 (s + 1,d h ,da + 1)
+ (1 − u − v )at+1 (s + 1, d h ,da ) + u at+1 (s + 1, d h + 1, da )] [5.6]
The two previous examples illustrating the determination of the non-cooperative

solution for the teams’ strategic choices in the 90th minute result in both teams
choosing a violent style of play. Is the resort to violence a pervasive feature of this
model, or are there circumstances in which the teams would be able to adhere to
a cooperative solution involving non-violence? Dobson and Goddard (2010) show
that the Nash equilibrium is more likely to coincide with the cooperative solution
in the earlier stages of matches, because the overall cost to the team of having a
player dismissed is higher if the dismissal occurs near the start of the match than it
is if the dismissal occurs near the end. For example, if a player is dismissed in the
first minute, the team plays virtually the entire match at a numerical disadvantage,
and is significantly less likely to take league points from the match. The team also
loses the right to field the dismissed player in one or more future matches through
suspension. If a player is dismissed in the 90th minute, the cost in terms of future
suspension is the same, but the team suffers virtually no disadvantage in the current
match.
Table 5.5 provides the data for a further stylised example, in the form of a set of
(hypothetical) match result probabilities applicable at the end of the first minute
Table 5.5 Hypothetical match result probabilities at the end of the first minute,
conditional on score and player dismissals, for numerical examples
Home-team player Away-team player

dismissed No player dismissed dismissed
Score after 1 Home Draw Away Home Draw Away Home Draw Away
minute win win win win win win
0–1 0.12 0.30 0.58 0.30 0.30 0.40 0.48 0.30 0.22
0–0 0.22 0.30 0.48 0.40 0.30 0.30 0.58 0.30 0.12
1–0 0.32 0.30 0.38 0.50 0.30 0.20 0.68 0.30 0.02
Table 5.6 Determination of the two teams’ optimal strategies for the first minute
Expected payoffs after one minute at the start of the match

Home team = hi,j0 (0,0,0) Away team = ai,j0 (0,0,0)
j=1 j=2 j=3 j=4 j=1 j=2 j=3 j=4
i=1 0.8975 0.8953 0.8995 0.8973 0.4975 0.4972 0.4955 0.4952

2 0.8972 0.8950 0.8992 0.8970 0.4953 0.4950 0.4933 0.4930
3 0.8955 0.8933 0.8975 0.8953 0.4995 0.4992 0.4975 0.4972
4 0.8952 0.8930 0.8972 0.8950 0.4973 0.4970 0.4953 0.4950
of a match, conditional upon whether a goal is scored or a player is dismissed

during the first minute.4
For the purposes of the example, the goal-scoring and player-dismissal probabil-
ities specified in Table 5.2 are assumed to apply to the first minute. It is assumed
that if no players were dismissed during the first minute, the expectations at the
end of the first minute of the numbers of players dismissed by the end of the
match are 0.4 for the home team, and 0.8 for the away team; as before, the prob-
abilities for goals and dismissals have been exaggerated for the purposes of con-
structing a stylised example.
On the basis of these assumptions, Table 5.6 reports the payoffs that determine
the two teams’ optimal choices of i and j for the first minute. In Table 5.6, {i = 1,j = 1}
is the cooperative solution, in which both teams play defensively and non-vio-
lently. In this case the cooperative solution is a Nash equilibrium, because neither
team would wish to alter its choice in view of the choice that is being made by
the other team. Data on the incidence of goal scoring and disciplinary sanction
within matches, reported in Section 5.3 below, indicate that the award of red cards
increases sharply during the latter stages of play. This empirical pattern is consist-
ent with the central prediction of the theoretical model described in this section,
which suggests a tendency for a shift from a cooperative and non-violent pattern
The timings of player dismissals and goals 119
Table 5.7 Rates of player dismissal and goal scoring conditional on current
duration, English League, T1–T4, 2002–2009 seasons
Players dismissed per minute Goals scored per minute

Match duration, t
(minutes) Home team Away team Home team Away team
0 < t ≤ 10 .00009 .00029 .0119 .0090

10 < t ≤ 20 .00034 .00045 .0137 .0101
20 < t ≤ 30 .00041 .00056 .0142 .0104
30 < t ≤ 40 .00052 .00086 .0142 .0109
40 < t ≤ 44 .00043 .00103 .0148 .0104
44 < t ≤ 45 .00356 .00387 .0539 .0378
45 < t ≤ 55 .00060 .00090 .0156 .0122
55 < t ≤ 65 .00099 .00142 .0161 .0127
65 < t ≤ 75 .00107 .00184 .0164 .0125
75 < t ≤ 85 .00128 .00205 .0169 .0128
85 < t ≤ 89 .00118 .00232 .0174 .0128
89 < t ≤ 90 .00952 .01375 .0861 .0662
of play in the early stages of matches, to a non-cooperative and violent pattern in

the latter stages. This is attributed to a shift in the structure of rewards and penal-
ties over the course of the match, and in particular, a decline over the course of the
match in the cost borne by a team that has a player dismissed, measured in terms
of expected league points foregone.
Dobson and Goddard (2010) report the results of numerical simulations of the
model, in which the strategic choices at all points during the match are determined
more rigorously than was the case in the previous stylised example, by solving the
model backwards from the end of the match to the start. Speaking generally, it is
found that teams that are trailing tend to attack, and teams that are trailing tend to
play violently for as long as there is a realistic chance of salvaging the match. Teams
that are trailing by three goals or more tend not to play violently. The stage at which
a team that is trailing by two goals or more ceases to play violently, implicitly con-
ceding that the match is beyond salvation, depends upon relative team quality and
upon home-field advantage. Stronger teams when trailing play violently for longer
than weaker teams, and home teams when trailing play violently for longer than
away teams. The propensity for teams that are either trailing or level to play vio-
lently and/or to attack is increasing in λ, the risk-aversion parameter.
5.3 The timings of player dismissals and goals

The data sample for the empirical analysis that is reported in this section com-
prises all 16,288 matches played in the four tiers of the English Premier League
and Football League during the eight seasons from 2002 to 2009 (inclusive).5
Tables 5.7, 5.8 and 5.9 report descriptive statistics, which highlight several of the
Table 5.8 Rates of player dismissal and goal scoring conditional on current
difference in scores, English League, T1–T4, 2002–2009 seasons
Difference in scores, s Players dismissed per minute Goals scored per minute
(home goals – away goals)
Home team Away team Home team Away team
s ≤ –3 .00123 .00069 .0182 .0169

s = –2 .00202 .00145 .0183 .0140
s = –1 .00130 .00126 .0170 .0124
s=0 .00057 .00090 .0150 .0113
s=1 .00079 .00187 .0167 .0134
s=2 .00086 .00206 .0182 .0134
s≥3 .00056 .00201 .0213 .0132
Table 5.9 Rates of player dismissal and goal scoring conditional on numerical
disparity in players, English League, T1–T4, 2002–2009 seasons
Numerical disparity in players, x Players dismissed per minute Goals scored per minute
(home players dismissed –
Home team Away team Home team Away team
away players dismissed)
x ≤ –1 .00182 .00238 .0287 .0093

x=0 .00075 .00121 .0159 .0122
x≥1 .00173 .00306 .0123 .0228
key strategic issues that are outlined above. Table 5.7 reports the average numbers
of dismissals per minute and the average rates at which goals were scored by the
home and away teams per minute, at various durations within matches. Table 5.8
reports the average dismissal rates and scoring rates conditional on the current
difference between the scores of the two teams; and Table 5.9 reports the rates
conditional on the current numerical imbalance between the two teams (if any)
due to any players having already been dismissed.
In Table 5.7 the data for the 45th and the 90th minutes are displayed separ-
ately from those for other durations. These cells include all player dismissals
and goals scored during the 45th and 90th minutes, and during stoppage time
(played immediately after the 45th and 90th minutes have elapsed). The data
source does not record the amount of stoppage time played, which is variable
but typically of between two and five minutes’ duration at the end of each
45-minute period. Accordingly the rates of player dismissal and goal scoring
recorded in these cells are several times the magnitudes of those in adjacent
cells.
The timings of player dismissals and goals 121
According to Table 5.7, there is a pronounced upward trend in the numbers

of dismissals over the duration of matches. The rates for 85 < t ≤ 89 (durations
between 85 and 89 minutes) are around 18 times and 9 times those for 0 < t ≤ 10
for home teams and away teams, respectively. There is also a clear, but less heavily
pronounced, upward trend in the numbers of goals scored. The goal-scoring rates
for 85 < t ≤ 89 are around 1.5 times those for 0 < t ≤ 10 for both home teams and
away teams.
Except in the case where either team is leading by more than two goals (for
which the data are relatively sparse), the probability of having a player dismissed
is lowest when the scores are level; is of intermediate value for teams leading by
one or two goals (s = 1,2 for the home team, s = –1,–2 for the away team); and
is highest for teams trailing by one or two goals (s = –1,–2 for the home team,
s = 1,2 for the away team). The probability of either team scoring is lowest when
the scores are level, and is somewhat increased when the scores are unequal. In the
latter case, and provided the current scores difference is not more than two goals,
the scoring probabilities of both teams increase in similar proportions (no mat-
ter which team is leading and which is trailing). However, a team that is leading
by three goals or more appears to experience a large increase in its probability of
scoring further goals.
The tendency for scoring rates to increase over the duration of the match (see
Table 5.7) is well known to commentators, pundits and spectators, and is com-
monly attributed to a tendency for players to become fatigued and commit defen-
sive errors, which lead to goals being scored. Errors due to fatigue could also
produce more fouls in last-ditch attempts to prevent goals, which could in turn
account for the upwardly trended rates of player dismissal.
Since goals in football are relatively infrequent events (in comparison with
scores in other team sports such as rugby or basketball), observations in Table
5.8 for level scores tend to be weighted towards the earlier match durations, while
observations for a non-zero difference in scores tend to be weighted towards later
durations. Therefore player fatigue could also account for some of the variation
shown in Table 5.8. However, the data in Table 5.8 could also reflect a form of
team quality selection effect. Observations of scoring rates when the away team
is already leading (for example) are drawn predominantly from matches in which
the away team is of higher quality than the home team. This means the probability
that the away team scores further goals should be higher than the average. This
team quality selection effect might explain why teams score more frequently when
leading than when scores are level, but it does not explain why teams that are trail-
ing also tend to score more frequently.
Distinct from the player fatigue and team quality selection effects, an alterna-
tive explanation for the variations in scoring rates shown in Tables 5.7 and 5.8 is
suggested by the analysis in Section 5.2. A team that switches from a defensive to
an attacking formation increases both its own and its opponent’s probabilities of
scoring. By leaving itself open to a counter-attack and the possibility of having
to commit a last-ditch foul in order to prevent an opponent from scoring, it also

increases its own probability of having a player dismissed. A team that switches
from a non-violent to a violent style of play disrupts the pattern of play, in a man-
ner that may increase both teams’ scoring probabilities.
As noted in Section 5.2, the dismissal of a player imposes two separate costs
upon his own team. First, the team must complete the match at a numerical disad-
vantage in players (unless an opposing player is also dismissed). Table 5.9 indicates
that there is a large reduction in the scoring probability for a team that is playing
at a numerical disadvantage, and a still larger increase in the probability of con-
ceding a goal. This portion of the cost of a player dismissal is highest for dismiss-
als that occur at the early stages of matches, because the team that loses a player
is exposed to these adverse probabilities for longer. Second, a dismissal results
in the player’s suspension from one or more future matches. This portion of the
cost is independent of the stage of the (current) match at which the dismissal
occurs. Therefore the overall cost of a dismissal in the current match, expressed in
terms of expected league points foregone, is higher for dismissals during the early
stages of matches. A tendency for dismissal rates to increase over the duration of
matches (see Table 5.7) seems consistent with rational strategic behaviour in the
light of this cost structure.
5.4 An empirical model for the in-play arrival rates of player dismissals and goals
In this section we develop an empirical model for the in-play arrival rates of player
dismissals and goals. Following a modelling approach similar to that of Dixon and
Robinson (1998), it is assumed that the arrival rates can be represented as Poisson
processes, such that the probability that a new arrival occurs is independent of the
time that has elapsed since the previous arrival. It is also assumed, for simplicity,
that arrivals of player dismissals and goals are independent of each other.6
Let μk denote the arrival rate for event k, where k = 1 denotes a home-team
player dismissal, k = 2 denotes an away-team player dismissal, k = 3 denotes a goal
scored by the home team, and k = 4 denotes a goal scored by the away team. Let
mk denote the number of minutes that elapse before the next occurrence of event
k. If player dismissals and goals are Poisson processes, mk follows an exponential
distribution, with distribution function Fk(t) = prob(mk ≤ t) = 1–exp(–μkt) and
density function fk(t) = Fk′(t) = μkexp(–μkt).
Section 5.4 reports estimation results for a competing risks model, comprising
hazard functions for the four events (k = 1, … ,4), in which μk are assumed to be
linear in a set of covariates that are time-varying over the duration of the current
match. Spells of continuous play that end in the occurrence of event k are treated
as right-censored in the likelihood functions for the events other than k.
The covariates include team quality covariates obtained from the goals-based
match results forecasting model that is described in Chapter 4, Section 4.2. In order
to generate the team quality covariates for all of the matches played in a particular
Empirical model for arrival rates: dismissals, goals 123
season, the match results forecasting model is estimated using data from the ten
seasons immediately preceding the season in question. Probabilities in win-draw-
lose format for the result of each match within the season are obtained using the
procedure described in Chapter 4, Section 4.3, by substituting into the fitted model
covariate values that relate to the match, all of which are calculated using data
that are available prior to the start of the match. HWINPR and DRAWPR are
probabilities for the current match to end in a home win and a draw. EXPHG and
EXPAG are the expected numbers of goals scored by the home and away teams in
the current match, defined as the fitted values of λ1,i,j and λ2,i,j in equation [4.1].
The full list of covariate definitions for the conditional hazard functions for
home- and away-team player dismissals (μ1 and μ2) are as follows:
RELQUAL = HWINPR+ 0.5×DRAWPR (see above).
UNCERT = RELQUAL×(1–RELQUAL).
DUR = duration (number of minutes elapsed prior to the start of the current
minute) within the current match, measured from 0 to 89.
SECOND = 0 during the first halves of matches, 1 during the second halves.
M45, M90 = 0–1 dummy variables that allow for step changes in the arrival rates
recorded for the 45th and 90th minutes of the match, due to stoppage time.
HOFF, AOFF = 0–1 dummy variables indicating whether either team is currently
experiencing a numerical disadvantage due to one or more players having previ-
ously been dismissed during the current match. HOFF = 1 if the home team is
currently at a disadvantage in player numbers, and 0 otherwise. AOFF is defined
similarly for the away team.
DIFFs = 0–1 dummy variables indicating the goal difference between the home
and away teams at the start of the current minute: DIFF–3 = 1 if the away team
is leading by three goals or more at the start of the current minute, and 0 other-
wise. DIFF–2 = 1 and DIFF–1 = 1 if the away team is leading by two goals or
one goal, respectively. DIFF+1, DIFF+2 and DIFF+3 indicate the home team
leading by one goal, by two goals, or by three goals or more, respectively.
The additional covariate definitions for the conditional hazard functions for goals
scored by the home and away teams (μ3 and μ4) are as follows:
EXPHG and EXPAG = expected numbers of goals scored by the home and away
teams in the current match (see above). These covariates are invariant across all
of the observations pertaining to the current match.
KICKOFF = 1 for the 1st and 46th minutes, and for any minute when a goal has
been scored by either team in the preceding minute, and 0 otherwise.
In the hazard functions for home- and away-team player dismissals, the covariate
RELQUAL measures home-team quality relative to away-team quality, also tak-
ing account of home-field advantage. RELQUAL appears in the player dismissal
hazard functions because previous empirical evidence suggests that players from
lower-quality teams and players from away teams are at greater risk of disciplinary
sanction (Dawson et al., 2007; see also Chapter 10). UNCERT = RELQUAL×(1–
RELQUAL), a standard measure of uncertainty of match outcome, also appears
in the player dismissal hazard functions. In accordance with the theoretical model
and simulation results, and with empirical evidence reported by Dawson et al.
(2007), violent play is more likely when the two teams are closely balanced (in
terms of underlying quality) than when they are unbalanced. Therefore positive
coefficients on UNCERT are expected in both the home-team and away-team haz-
ard functions. RELQUAL and UNCERT are non-varying with duration across
all of the observations pertaining to the current match.
In hazard functions for goals scored by the home and away teams (μ3 and μ4),
EXPHG and EXPAG, the expected numbers of goals scored by the home and
away teams in the current match (see above), control for the relationship between
relative team quality and the scoring rates of the home and away teams. These
covariates are also non-varying with duration across all observations pertaining
to the current match. The covariates RELQUAL and UNCERT do not appear in
the hazard functions for goals scored, because any relative team quality effects on
goal-scoring rates are captured directly by the EXPHG and EXPAG covariates.
The dummy variables M45 and M90 control for the fact that any dismissal that
takes place during stoppage time at the end of each 45-minute period, and any
goal scored during stoppage time, is recorded as having occurred in either the 45th
or the 90th minute. As noted above, the amount of stoppage time added on at the
end of each half is commonly between two and five minutes; although shorter or
longer amounts are possible. The recording system for dismissals or goals scored
during stoppage time ensures that the hazards of these events being recorded as
having occurred in either the 45th or the 90th minute are several times the size of
the hazards for the 44th and 89th minutes (for example), and for other periods of
one minute’s duration. Accordingly, positive coefficients on M45 and M90 are
expected in all estimations.
It is expected that goal-scoring rates should be relatively low in the first minute
(the start of the match), the 46th minute (the resumption of the match following the
half-time interval), and in the minute immediately after a goal is scored by either
team. In each case, play starts or resumes from a kick-off at the halfway line, and
the teams regroup into balanced defensive formations when play kicks off. Goals are
unlikely to be scored during the first few seconds following kick-off, and the scoring
rates for those minutes are reduced accordingly. The dummy variable KICKOFF
takes a value of one for the first and 46th minutes, and for any minute when a goal
has been scored by either team in the preceding minute, and zero otherwise.
5.5 Estimation results and interpretation

Empirical arrival rates for player dismissals
Table 5.10 reports the estimated equations for the arrival rates of home-team player
dismissals, as equations [5.7], [5.8] and [5.9], and away-team player dismissals as
Estimation results and interpretation 125
Table 5.10 Estimation results: player-dismissal hazard functions
Home-team player dismissal Away-team player dismissal

hazard hazard
[5.7] [5.8] [5.9] [5.10] [5.11] [5.12]
RELQUAL −1.576 −1.123 −1.120 1.366 1.004 1.000

−3.82 −2.69 −2.69 3.25 2.37 2.36
UNCERT 7.875 7.623 7.620 4.711 4.539 4.519
3.44 3.33 3.33 2.55 2.45 2.44
DUR .023 .022 .031 .023 .022 .034
16.1 14.9 5.72 20.5 19.0 8.15
SECOND — — .592 — — .521
1.85 2.05
DUR×SECOND — — −.014 — — −.016
−2.06 −3.19
M45 1.844 1.828 1.668 1.444 1.429 1.185
13.5 13.4 9.69 11.2 11.0 7.80
M90 1.766 1.785 1.847 1.628 1.644 1.650
18.4 18.5 17.4 20.7 20.9 19.3
HOFF .435 .360 .364 .950 1.008 1.016
3.09 2.55 2.57 10.3 10.9 11.0
AOFF .386 .465 .474 .267 .223 .222
3.29 3.95 4.02 2.70 2.25 2.24
DIFF–3 — −.072 −.075 — −1.092 −1.047
−0.30 −0.31 −3.41 −3.26
DIFF–2 — .589 .577 — −.167 −.140
5.33 5.17 −1.38 −1.15
DIFF–1 — .410 .393 — −.054 .055
5.26 2.64 −0.75 −0.42
DIFF+1 — −.018 .031 — .352 .071
−0.22 −0.20 5.97 0.60
DIFF+2 — −.127 −.140 — .253 .277
−1.04 −1.14 3.07 3.31
DIFF+3 — −.722 −.723 — .052 .099
−3.18 −3.18 0.42 0.79
DIFF–1×SECOND — — −.592 — — −.539
−1.22 −1.16
DIFF–1×DUR×SECOND — — .008 — — .005
1.31 0.91
DIFF+1×SECOND — — .243 — — −.493
0.48 −1.35
DIFF+1× DUR×SECOND — — −.005 — — .011
−0.70 2.50
Constant −9.451 −9.708 −9.963 −9.951 −9.753 −9.994
−13.3 −13.6 −13.7 −15.39 −15.0 −15.2
Note: z-statistics for the significance of the estimated coefficients are reported in italics.
[5.10], [5.11] and [5.12]. In all of the estimations, the dependent variable is ln(μk)
for k = 1 and 2, the natural logarithm of the arrival rate for home-team dismiss-
als and away-team dismissals, respectively.7 Three alternative specifications are
reported. Equations [5.7] and [5.10] exclude any effects related to the current dif-
ference in scores. Equations [5.8] and [5.11] include dummy variables for the cur-
rent difference in scores. Finally, [5.9] and [5.12] also include several interaction
terms, which permit some of the coefficients to vary between the first and second
halves of matches. The additional terms included in [5.9] and [5.12] are jointly sig-
nificant, and improve the model’s explanatory power. Accordingly, [5.9] and [5.12]
are used in the evaluation of in-play home win, draw and away win probabilities
at various stages of the match, conditional on the state of the match at that stage,
that are reported in Section 5.6. Since the coefficients of [5.7], [5.8], [5.10] and
[5.11] are easier to interpret, however, the following commentary focuses mainly
on these specifications.
In all of the player dismissal arrival rate estimations, the coefficients on DUR,
the linear trend in duration, are positive and significant, reflecting the tendency
for the likelihood of a dismissal occurring to increase over the duration of the
match. The coefficients on M45 and M90 are positively signed and significant, as
expected.
The coefficients on RELQUAL are negative and significant in the home-team
player dismissal equation, and positive and significant in the away-team equation.
These coefficients indicate that weaker teams are more likely to be penalised than
stronger teams. The coefficients on UNCERT are positively signed and signifi-
cant, indicating that players are more likely to be dismissed in matches that are
equally balanced, and less likely in matches where there is a large difference in
quality between the home and away teams. The coefficients on HOFF and AOFF
are positively signed and significant. The coefficients on HOFF are particularly
large in the away-team player dismissal equation. This might perhaps reflect a
tendency for referees who have already dismissed a home-team player in the cur-
rent match to succumb to pressures from the home crowd to ‘even things up’ by
dismissing an away-team player.
When the dummy variables DIFFs (for s = –3,–2,–1,1,2,3) are added to the
specification (comparing [5.8] with [5.7], and [5.11] with [5.10]), the coefficients
and z-statistics on DUR are reduced, but only fractionally. Accordingly, strategic
choices on the part of the two teams dependent on the current difference in scores
do not explain the tendency for the rates of player dismissal to increase over the
duration of the match. Nevertheless, the dummy variables DIFFs do make a
significant contribution to the explanatory power of [5.8] and [5.11], suggesting
that strategic effects are important. The coefficients on DIFF–2 and DIFF–1 are
positive and significant in the home-team player dismissal equation, [5.8]. The
coefficient on DIFF+3 is negative and significant. The pattern for the away-team
player dismissal equation [5.11] is the mirror image of the home-team equation.
The coefficients on DIFF+1 and DIFF+2 in [5.11] are positive and significant,
Estimation results and interpretation 127
but smaller in absolute magnitude than the coefficients on DIFF–2 and DIFF–1
in [5.8]. The coefficient on DIFF–3 in [5.11] is negative and significant.
These findings are consistent with the properties of the theoretical model that
is developed in Section 5.2. The probability of incurring a player dismissal tends
to be higher for a team that is trailing than for the same team either when it is
leading or when the scores are level. Teams when leading tend to play cautiously
(non-violent, defence) in order to minimise both the probability of conceding a
goal and the probability of losing a player through dismissal. The probability of
incurring a dismissal is particularly low when either team is leading by three goals
or more, suggesting that when the match result is beyond reasonable doubt, vio-
lent play tends to fall to a particularly low level. Teams when trailing tend to take
risks (violent, attack), because it is worthwhile bearing an increased probability
of a dismissal in order to increase the probability of scoring. This willingness to
bear additional risk is conditional on the prospects of salvaging the match, how-
ever. Teams trailing by either one or two goals are more inclined to take risks than
teams trailing by larger margins; and the tendency for home teams when trailing
to take risks is stronger than the same tendency for away teams.
Empirical arrival rates for goals scored

Table 5.11 reports the estimated arrival rates for home-team goals ([5.13] to [5.15])
and away-team goals ([5.16] to [5.18]). The dependent variable is ln(μk) for k = 3
and 4, the natural logarithm of the arrival rate for home-team goals and away-
team goals, respectively. As before, three alternative specifications are reported.
Equations [5.13] and [5.16] exclude any effects related to the current difference in
scores. Equations [5.14] and [5.17] include dummy variables for the current diffe-
rence in scores. Finally, [5.15] and [5.18] also include several interaction terms,
which permit some of the coefficients to vary between the first and second halves
of matches and with variation in the team quality control variables. The additional
terms included in [5.15] and [5.18] are jointly significant, and these specifications
are used in the evaluation of in-play home win, draw and away win probabilities
reported in Section 5.6.
The coefficients on KICKOFF, M45, M90, EXPHG, EXPAG and DUR in the
goals equations are all signed as expected and significant, and require no further
comment. The coefficients on HOFF and AOFF are signed in accordance with
the ‘common sense’ view that when a team is playing at a numerical disadvantage
because a player has been dismissed, the likelihood of scoring a goal is reduced
and the likelihood of conceding is increased. The popular theory that a team play-
ing at a numerical disadvantage are more likely to score is firmly rejected by the
empirical evidence.
There is little change in the magnitudes of the coefficients on DUR when
the dummy variables DIFFs (for s = –3,–2,–1,1,2,3) are added to the specifica-
tion (comparing [5.14] with [5.13], and [5.17] with [5.16]). Accordingly and as
Table 5.11 Estimation results: goal-scoring hazard functions
Home-team goal hazard Away-team goal hazard

[5.13] [5.14] [5.15] [5.16] [5.17] [5.18]
EXPHG .599 .615 .841 — — −.252
25.2 25.5 4.25 −1.21
EXPHG^2 — — −.097 — — .008
−1.61 0.11
EXPAG — — −.453 .851 .883 1.186
−1.90 24.2 24.7 4.07
EXPAG^2 — — .088 — — −.184
0.84 −1.53
DUR .003 .003 .004 .004 .004 .004
12.7 10.8 4.53 13.7 11.6 4.16
SECOND — — .174 — — .307
2.55 3.96
DUR×SECOND — — −.003 — — −.004
−2.12 −3.09
M45 1.277 1.276 1.269 1.202 1.199 1.213
37.1 37.0 32.3 29.3 29.2 26.0
M90 1.566 1.568 1.613 1.570 1.574 1.635
52.6 52.6 49.5 46.2 46.3 43.9
KICKOFF −.618 −.623 −.632 −.332 −.338 −.354
−14.85 −15.0 −15.1 −7.90 −8.05 −8.35
HOFF −.324 −.339 −.336 .506 .521 .520
−5.96 −6.20 −6.16 12.6 12.9 12.9
AOFF .486 .494 .498 −.412 −.428 −.425
17.0 17.2 17.4 −8.28 −8.59 −8.53
DIFF–3 — .112 .126 — .037 .042
1.78 1.99 0.56 0.64
DIFF–2 — .131 .135 — −.041 −.045
3.88 3.95 −1.05 −1.14
DIFF–1 — .066 −.168 — −.048 −.077
3.43 −1.48 −2.14 −0.61
DIFF+1 — −.016 .251 — .107 .139
−0.90 2.45 5.35 1.27
DIFF+2 — −.017 −.033 — .098 .111
−0.65 −1.21 3.22 3.58
DIFF+3 — .040 .026 — .074 .108
1.08 0.68 1.59 2.26
DIFF–1×SECOND — — −.098 — — −.068
−0.78 −0.47
DIFF–1×DUR×SECOND — — .002 — — .001
1.08 0.45
DIFF+1×SECOND — — .145 — — −.145
1.27 −1.14
DIFF+1× DUR×SECOND — — −.003 — — .003
−1.91 1.73
DIFF–1×EXPHG — — .139 — — —
1.97
DIFF+1×EXPHG — — −.146 — — —
−2.38
DIFF–1×EXPAG — — — — — .023
0.23
DIFF+1×EXPAG — — — — — −.061
−0.67
Constant −5.289 −5.316 −5.058 −5.642 −5.686 −5.428
−131.0 −129.7 −28.7 −127.0 −126.1 −27.7
commented previously, strategic choices on the part of the two teams dependent
on the current difference in scores do not explain the tendency for the goal-scoring
rates to increase over the duration of the match. As before, however, the dummy
variables DIFFs do make a significant contribution to the explanatory power of
[5.14] and [5.17], suggesting that strategic effects are important. The coefficients
on DIFF–2 and DIFF–1 are positive and significant in [5.14]; and the coefficients
on DIFF+1 and DIFF+2 are positive and significant in [5.17]. The coefficient on
DIFF–1 in [5.17] is negative and significant.
The positive and significant coefficients on DIFF–1 and DIFF–2 in [5.14], and
on DIFF+1 and DIFF+2 in [5.17], are consistent with a tendency for teams that
are trailing to play offensively, and consequently for the likelihood of scoring to be
increased. The negative and significant coefficient on DIFF–1 in [5.17] suggests a
tendency for away teams when protecting a narrow lead to play defensively, redu-
cing the probability that the lead will be extended. For home teams, in contrast,
although the coefficient on DIFF+1 in [5.14] is also negative, this coefficient is
small in absolute magnitude, and not significant.
Looking at the raw data summarised in Table 5.8, the scoring rates of teams
that are leading are higher than those of teams that are level. The scoring rates
reported in Table 5.8 are unconditional on team quality. An implication of Table
5.11 is that after controlling for team quality, the scoring rates of teams that are
leading by one or two goals are slightly lower than those of teams that are level
(although in most cases the coefficients are not significant). Accordingly, the pat-
terns in Table 5.8 should be attributed to a team quality selection effect (see also
Section 5.3). After including controls for team quality in [5.13] to [5.18], the dif-
ferences that are apparent in the raw data, between the scoring rates of teams that
are leading and of teams that are level, are eliminated.
5.6 Stochastic simulations for in-play match result probabilities conditional

on the current state of the match
This section presents a practical application of equations [5.9], [5.12], [5.15] and
[5.18], for the use of team managers, in-play bettors and others who may wish to
obtain in-play probabilities for the final outcome of a match conditional upon the
current state of the match at any stage.
For the purposes of evaluating the in-play match result probabilities, the cur-
rent state of the match comprises two components: (i) the RELQUAL measure
of the relative underlying quality of the two teams, which is based on previous
match results and other data that are observable prior to the start of the match in
question; and (ii) the difference in scores defined by s (using the same notation as
before), and any numerical disparity in players defined by x = dh–da (where dh and
da are the numbers of home-team and away-team players dismissed) at the stage in
the match at which the in-play probabilities are required.
Simulations for in-play match result probabilities 131
Given any set of values for RELQUAL, s and x and starting from any match
duration denoted t, the in-play home win, draw and away win probabilities can
be calculated from stochastic simulations of player dismissals and goals over the
remaining duration of the match (t+1, …,T). The home win, draw and away win
probabilities at the start of the match are computed by running the simulations from
t = 0, over the full 90-minute match duration. In accordance with the underlying
probability model, occurrences of each of the relevant events (k = 1,2 for player
dismissals, k = 3,4 for goals scored) are simulated by means of random drawings
from an exponential distribution with conditional mean μk, which varies over the
remaining duration of the simulated match in accordance with [5.9], [5.12], [5.15]
and [5.18]. The in-play probabilities are the proportions of simulated home wins,
draws and away wins obtained from 10,000 replications of this procedure.
Tables 5.12–5.14 report selected in-play probabilities, calculated for five differ-
ent values of RELQUAL at five different durations within the match, for three
possible values of x and five possible values of s. Table 5.12 reports in-play home
win probabilities; Table 5.13 reports draw probabilities; and Table 5.14 reports
away win probabilities. The values of RELQUAL are the averages within each of
the five quintile ranges formed by ranking the 16,288 matches used in the estima-
tions reported in Tables 5.10 and 5.11 in ascending order of RELQUAL. The
assumed values of UNCERT are calculated by substituting the five RELQUAL
values into the formula UNCERT = RELQUAL×(1–RELQUAL). The assumed
values of EXPHG and EXPAG are the average values within each of the five
RELQUAL quintile ranges.
This procedure produces five sets of values for the relative team quality vari-
ables and the home win, draw and away win probabilities at the start of the match
taking home-field advantage into account: (i) large difference in team quality
favouring the away team (home win/draw/away win probabilities = 0.2974/0.2752/
0.4274); (ii) small difference favouring away team (0.3940/0.2840/0.3220); (iii)
equally balanced teams (0.4488/0.2845/0.2667); (iv) small difference favouring
home team (0.4878/0.2810/0.2312); and (v) large difference favouring home team
(0.5948/0.2452/0.1600). The five match durations for which probabilities are com-
puted are t = 15,30,45,60 and 75 minutes; the three values for the numerical dis-
parity in players are x = 1,0,–1; and the five values for the difference in scores are
s = –2,–1,0,1,2.8
For example, referring to the central panels of Tables 5.12, 5.13 and 5.14 (the
case of equally balanced teams), an opening goal scored by the home team in the
first 15 minutes (with a one-goal lead sustained until the 15th minute) tilts the
in-play home win/draw/away win probabilities for the 15th minute in the home
team’s favour, from 0.4199/0.3070/0.2731 to 0.6995/0.1934/0.1071. For an open-
ing goal scored by the away team in the same period (with the lead also sustained
until at least the 15th minute), the probabilities at the 15th minute tilt in the away
team’s favour, to 0.2010/0.2648/0.5342.
Table 5.12 Home-win probabilities, conditional on relative team strengths and the state of the match at various durations
Home/draw/away x→ +1 0 –1
probabilities, t = 0 ↓
t↓s→ –2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2
15 .0065 .0345 .1248 .3124 .5495 .0312 .1090 .2879 .5707 .7894 .1256 .2956 .5485 .7935 .9323
30 .0057 .0327 .1292 .3437 .6053 .0254 .0899 .2771 .5826 .8216 .0974 .2471 .5090 .7911 .9347
.2974/.2752/.4274 45 .0036 .0264 .1280 .3926 .6941 .0128 .0701 .2565 .6208 .8650 .0589 .1729 .4579 .7945 .9461
60 .0022 .0195 .1238 .4831 .7929 .0088 .0437 .2278 .6675 .9114 .0271 .1114 .3857 .8145 .9673
75 .0004 .0087 .1020 .6126 .9090 .0017 .0205 .1654 .7572 .9623 .0060 .0472 .2721 .8449 .9868
15 .0156 .0583 .1839 .4116 .6530 .0586 .1620 .3693 .6568 .8524 .1980 .3860 .6390 .8551 .9554
30 .0125 .0530 .1775 .4355 .7044 .0435 .1372 .3553 .6611 .8749 .1476 .3304 .5963 .8459 .9621
.3940/.2840/.3220 45 .0070 .0427 .1786 .4919 .7644 .0240 .1002 .3110 .6885 .9007 .0904 .2492 .5318 .8421 .9634
60 .0035 .0312 .1584 .5537 .8472 .0096 .0636 .2775 .7251 .9390 .0421 .1621 .4452 .8457 .9776
75 .0009 .0125 .1262 .6784 .9260 .0023 .0249 .2028 .7947 .9752 .0097 .0723 .3219 .8745 .9882
15 .0220 .0805 .2166 .4544 .6960 .0737 .2010 .4199 .6995 .8798 .2441 .4438 .6813 .8812 .9634
30 .0165 .0680 .2114 .4827 .7417 .0538 .1656 .4009 .7024 .8986 .1883 .3746 .6416 .8752 .9682
.4488/.2845/.2667 45 .0093 .0508 .2043 .5282 .8037 .0338 .1245 .3562 .7224 .9188 .1082 .2806 .5669 .8577 .9762
60 .0051 .0337 .1827 .5823 .8726 .0149 .0796 .3078 .7496 .9480 .0503 .1894 .4857 .8752 .9810
75 .0010 .0153 .1411 .7002 .9481 .0027 .0317 .2152 .8132 .9783 .0108 .0782 .3414 .8914 .9899
15 .0330 .1017 .2579 .5047 .7432 .1001 .2358 .4657 .7436 .9009 .2902 .5105 .7371 .9059 .9742
30 .0233 .0850 .2499 .5319 .7787 .0667 .1950 .4483 .7452 .9135 .2212 .4207 .6823 .8909 .9727
.4878/.2810/.2312 45 .0125 .0676 .2292 .5642 .8312 .0369 .1459 .3870 .7538 .9399 .1365 .3241 .6144 .8838 .9784
60 .0062 .0456 .2115 .6225 .8952 .0171 .0941 .3425 .7719 .9575 .0656 .2148 .5328 .8850 .9866
75 .0008 .0187 .1505 .7159 .9521 .0031 .0357 .2347 .8368 .9812 .0155 .0922 .3751 .8980 .9919
15 .0547 .1549 .3416 .6006 .8208 .1443 .3206 .5621 .8070 .9375 .4050 .6220 .8083 .9404 .9859
30 .0389 .1342 .3336 .6105 .8490 .1104 .2743 .5296 .8034 .9421 .3101 .5351 .7644 .9344 .9871
.5948/.2452/.1600 45 .0223 .0966 .2939 .6383 .8792 .0646 .2008 .4786 .8076 .9565 .1952 .4116 .6931 .9154 .9865
60 .0120 .0621 .2575 .6852 .9226 .0303 .1282 .3958 .8214 .9692 .1045 .2906 .6036 .9143 .9913
75 .0015 .0269 .1903 .7605 .9671 .0077 .0523 .2737 .8589 .9876 .0268 .1320 .4244 .9223 .9934
Notes: Table 5.12 reports home win probabilities enumerated at various values of t, s, x, conditional on relative team quality represented by the home win, draw and
away win probabilities at the start of the match, shown in the first column.
t is the number of minutes’ play completed.
x = dh–da is the number of home-team players dismissed minus the number of away-team players dismissed prior to the t’th minute.
s is the difference between the home-team score and away-team score prior to the t’th minute.
Table 5.13 Draw probabilities, conditional on relative team strengths and the state of the match at various durations
t↓s→ –2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2
15 .0392 .1050 .2288 .2673 .2275 .0881 .2089 .3044 .2509 .1413 .1970 .2771 .2591 .1446 .0516
30 .0300 .1122 .2501 .2972 .2254 .0838 .2014 .3306 .2572 .1275 .1757 .2841 .3055 .1529 .0522
.2974/.2752/.4274 45 .0264 .1171 .3103 .3303 .1992 .0656 .2015 .3972 .2597 .1082 .1552 .3054 .3566 .1619 .0460
60 .0219 .1165 .3943 .3274 .1550 .0426 .1948 .4734 .2569 .0751 .1105 .2887 .4504 .1553 .0297
75 .0080 .1016 .5604 .3063 .0776 .0207 .1557 .6203 .2106 .0346 .0523 .2444 .6045 .1418 .0123
15 .0623 .1498 .2561 .2746 .2005 .1326 .2506 .3169 .2170 .1088 .2254 .2837 .2303 .1102 .0362
30 .0548 .1523 .2909 .2946 .1829 .1234 .2549 .3438 .2309 .0931 .2170 .3015 .2699 .1179 .0326
.3940/.2840/.3220 45 .0432 .1530 .3564 .3042 .1668 .0911 .2525 .4082 .2322 .0785 .1865 .3184 .3372 .1278 .0321
60 .0323 .1549 .4349 .3143 .1191 .0657 .2400 .4822 .2199 .0531 .1430 .3312 .4224 .1363 .0207
75 .0154 .1298 .5898 .2619 .0641 .0294 .1948 .6236 .1818 .0225 .0677 .2773 .5771 .1144 .0115
15 .0703 .1742 .2773 .2735 .1835 .1534 .2648 .3070 .1934 .0905 .2444 .2802 .2104 .0919 .0301
30 .0696 .1855 .3077 .2895 .1667 .1378 .2654 .3360 .2079 .0799 .2366 .3077 .2511 .0974 .0271
.4488/.2845/.2667 45 .0510 .1787 .3610 .3019 .1381 .1133 .2713 .3964 .2105 .0663 .2072 .3326 .3161 .1172 .0215
60 .0384 .1707 .4519 .2992 .1024 .0746 .2562 .4854 .2060 .0467 .1547 .3387 .4074 .1080 .0174
75 .0163 .1416 .6011 .2495 .0459 .0339 .2134 .6273 .1664 .0207 .0744 .3006 .5764 .1012 .0098
15 .0896 .1999 .2998 .2754 .1641 .1763 .2777 .3013 .1804 .0731 .2524 .2550 .1830 .0770 .0224
30 .0788 .1913 .3248 .2732 .1533 .1597 .2922 .3225 .1817 .0698 .2504 .2987 .2293 .0915 .0242
.4878/.2810/.2312 45 .0671 .2016 .3786 .2829 .1267 .1294 .2924 .3968 .1931 .0501 .2264 .3316 .2911 .0978 .0191
60 .0445 .2007 .4552 .2770 .0861 .0894 .2836 .4688 .1899 .0378 .1772 .3495 .3823 .1029 .0125
75 .0198 .1594 .6109 .2380 .0428 .0384 .2239 .6272 .1483 .0179 .0942 .3197 .5450 .0962 .0078
15 .1323 .2356 .3146 .2484 .1286 .2216 .2980 .2783 .1415 .0511 .2593 .2335 .1434 .0499 .0129
30 .1166 .2459 .3327 .2543 .1079 .1953 .3125 .3101 .1535 .0497 .2693 .2732 .1835 .0556 .0114
.5948/.2452/.1600 45 .0899 .2449 .3908 .2554 .0941 .1691 .3173 .3615 .1588 .0383 .2566 .3272 .2474 .0745 .0124
60 .0672 .2353 .4701 .2459 .0662 .1174 .3140 .4592 .1537 .0280 .2132 .3699 .3320 .0767 .0084
75 .0306 .1945 .6121 .2068 .0303 .0561 .2725 .6118 .1310 .0114 .1160 .3545 .5138 .0729 .0064
Notes: Table 5.13 reports draw probabilities enumerated at various values of t, s, x, conditional on relative team quality represented by the home win, draw and
Table 5.14 Away-win probabilities, conditional on relative team strengths and the state of the match at various durations
t↓s→ –2 –1 0 1 2 –2 –1 0 1 2 –2 –1 0 1 2
15 .9543 .8605 .6464 .4203 .2230 .8807 .6821 .4077 .1784 .0693 .6774 .4273 .1924 .0619 .0161
30 .9643 .8551 .6207 .3591 .1693 .8908 .7087 .3923 .1602 .0509 .7269 .4688 .1855 .0560 .0131
.2974/.2752/.4274 45 .9700 .8565 .5617 .2771 .1067 .9216 .7284 .3463 .1195 .0268 .7859 .5217 .1855 .0436 .0079
60 .9759 .8640 .4819 .1895 .0521 .9486 .7615 .2988 .0756 .0135 .8624 .5999 .1639 .0302 .0030
75 .9916 .8897 .3376 .0811 .0134 .9776 .8238 .2143 .0322 .0031 .9417 .7084 .1234 .0133 .0009
15 .9221 .7919 .5600 .3138 .1465 .8088 .5874 .3138 .1262 .0388 .5766 .3303 .1307 .0347 .0084
30 .9327 .7947 .5316 .2699 .1127 .8331 .6079 .3009 .1080 .0320 .6354 .3681 .1338 .0362 .0053
.3940/.2840/.3220 45 .9498 .8043 .4650 .2039 .0688 .8849 .6473 .2808 .0793 .0208 .7231 .4324 .1310 .0301 .0045
60 .9642 .8139 .4067 .1320 .0337 .9247 .6964 .2403 .0550 .0079 .8149 .5067 .1324 .0180 .0017
75 .9837 .8577 .2840 .0597 .0099 .9683 .7803 .1736 .0235 .0023 .9226 .6504 .1010 .0111 .0003
15 .9077 .7453 .5061 .2721 .1205 .7729 .5342 .2731 .1071 .0297 .5115 .2760 .1083 .0269 .0065
30 .9139 .7465 .4809 .2278 .0916 .8084 .5690 .2631 .0897 .0215 .5751 .3177 .1073 .0274 .0047
.4488/.2845/.2667 45 .9397 .7705 .4347 .1699 .0582 .8529 .6042 .2474 .0671 .0149 .6846 .3868 .1170 .0251 .0023
60 .9565 .7956 .3654 .1185 .0250 .9105 .6642 .2068 .0444 .0053 .7950 .4719 .1069 .0168 .0016
75 .9827 .8431 .2578 .0503 .0060 .9634 .7549 .1575 .0204 .0010 .9148 .6212 .0822 .0074 .0003
15 .8774 .6984 .4423 .2199 .0927 .7236 .4865 .2330 .0760 .0260 .4574 .2345 .0799 .0171 .0034
30 .8979 .7237 .4253 .1949 .0680 .7736 .5128 .2292 .0731 .0167 .5284 .2806 .0884 .0176 .0031
.4878/.2810/.2312 45 .9204 .7308 .3922 .1529 .0421 .8337 .5617 .2162 .0531 .0100 .6371 .3443 .0945 .0184 .0025
60 .9493 .7537 .3333 .1005 .0187 .8935 .6223 .1887 .0382 .0047 .7572 .4357 .0849 .0121 .0009
75 .9794 .8219 .2386 .0461 .0051 .9585 .7404 .1381 .0149 .0009 .8903 .5881 .0799 .0058 .0003
15 .8130 .6095 .3438 .1510 .0506 .6341 .3814 .1596 .0515 .0114 .3357 .1445 .0483 .0097 .0012
30 .8445 .6199 .3337 .1352 .0431 .6943 .4132 .1603 .0431 .0082 .4206 .1917 .0521 .0100 .0015
.5948/.2452/.1600 45 .8878 .6585 .3153 .1063 .0267 .7663 .4819 .1599 .0336 .0052 .5482 .2612 .0595 .0101 .0011
60 .9208 .7026 .2724 .0689 .0112 .8523 .5578 .1450 .0249 .0028 .6823 .3395 .0644 .0090 .0003
75 .9679 .7786 .1976 .0327 .0026 .9362 .6752 .1145 .0101 .0010 .8572 .5135 .0618 .0048 .0002
Notes: Table 5.14 reports away win probabilities enumerated at various values of t, s, x, conditional on relative team quality represented by the home win, draw and
Naturally, the later in the match the opening goal (or any goal that establishes a
lead) is scored, the greater is the impact on the match outcome and the greater is
the value of the goal to the scoring team. A goal scored between the 60th and 75th
minutes which gives the home team a one-goal lead (sustained until at least the
75th minute) tilts the in-play probabilities at the 75th minute rather dramatically
in the home team’s favour, from 0.2152/0.6273/0.1575 to 0.8132/0.1664/0.0204.
The away team’s prospects for recovering from a one-goal deficit after 75 minutes
to win the match are considered close to negligible. For a goal scored at the same
stage that gives the away team a one-goal lead (with the lead again sustained until
the 75th minute), the in-play probabilities at the 75th minute tilt in the away team’s
favour, to 0.0317/0.2134/0.7549. The home team’s chances of recovering from a
one-goal deficit after 75 minutes to win the match are somewhat higher than those
for the away team, but still slender.
Similarly, Tables 5.12, 5.13 and 5.14 can be used to determine the cost to either
team (in terms of the outcome of the current match) of incurring a player dismis-
sal. Using the same example, if the away team loses a player after 15 minutes, the
in-play probabilities tilt in the home team’s favour, from 0.4199/0.3070/0.2713 to
0.6813/0.2104/0.1083. If the home team loses a player after 15 minutes, the in-play
probabilities tilt in the away team’s favour, to 0.2166/0.2773/0.5061. Accordingly,
the impact of a player dismissal after 15 minutes on the match result probabilities
is only marginally smaller than the impact of an opening goal being scored at the
same stage.
Naturally, the later in the match a player dismissal creating a numerical disparity
between the teams takes place, the smaller is the impact on the in-play match result
probabilities, because the shorter is the remaining match duration over which one
team plays at a numerical disadvantage. The dismissal of an away-team player in
the 75th minute when the scores are level tilts the probabilities in the home team’s
favour much less dramatically than a goal giving the home team the lead at that
stage, from 0.2152/0.6273/0.1575 to 0.3414/0.5764/ 0.0822. Likewise the dismissal
of a home-team player at the same stage tilts the probabilities by a smaller amount
than a goal giving the away team the lead, to 0.1411/0.6011/0.2578.
As a further illustration, Table 5.15 tracks the model’s in-play probabilities for
two of the 2009 season fixtures included in Table 4.6 (used to illustrate the match
results forecasting models reported in Chapter 4, Sections 4.3 and 4.5), Fulham
vs West Bromwich Albion and Newcastle United vs Everton.
Prior to the Fulham vs West Bromwich match, the goals-based forecasting
model (Sections 4.2 and 4.3) assessed the home win/draw/away probabilities
as 0.5521/0.2521/0.1958. The match remained scoreless until the 61st minute
when Fulham took a 1–0 lead. Fulham scored again in the 72nd minute to
make the final score 2–0. Immediately before Fulham’s first goal, with just over
two-thirds of the match having been completed, the probabilities had shifted
to 0.3639/0.4637/0.1724. The first goal immediately tilted the probabilities in
Fulham’s favour to 0.7999/0.1703/0.0298. Immediately prior to the second
Table 5.15 In-play home-win/draw/away-win probabilities: illustration
Fulham vs West Bromwich Albion Newcastle United vs Everton

Goals or Home Away Goals or Home Away
Minute dismissals win Draw win dismissals win Draw win
5 .5357 .2727 .1916 .3027 .2933 .4040

10 .5319 .2759 .1922 .3040 .3036 .3924
15 .5170 .2884 .1946 .2980 .3093 .3927
20 .5138 .2911 .1951 .2968 .3186 .3846
25 .4980 .3040 .1980 .2923 .3258 .3819
30 .4911 .3199 .1890 .2814 .3433 .3753
35 .4830 .3301 .1869 .2879 .3487 .3634
40 .4644 .3484 .1872 Newcastle .2733 .3696 .3571
45 .4404 .3701 .1895 player sent .1301 .3214 .5485
50 .4160 .4043 .1797 off, 44th min. .1315 .3437 .5248
55 Fulham .4064 .4202 .1734 .1297 .3685 .5018
60 score, .3690 .4639 .1671 .1335 .4097 .4568
65 61st min. .8016 .1720 .0264 .1260 .4438 .4302
70 Fulham .8221 .1586 .0193 .1224 .4917 .3859
75 score, .9823 .0166 .0011 .1071 .5551 .3378
80 72nd min. .9904 .0094 .0002 .0902 .6367 .2731
85 .9961 .0037 .0002 .0691 .7269 .2040
90 .9984 .0015 .0001 .0399 .8509 .1092
Fulham goal, as the match moved into the final 20 minutes, the probabilities had
drifted further to 0.8225/0.1583/0.0192. The second Fulham goal eliminated
most of the remaining uncertainty over the match outcome, with the probabil-
ities tilting even further to 0.9807/0.0184/0.0009 immediately afterwards.
For the Newcastle vs Everton fixture, the forecasting model’s prior probabil-
ities were 0.3073/0.2936/0.3991. The match finished 0–0, with the most note-
worthy event being the dismissal of Newcastle’s Kevin Nolan in the 44th minute.
Immediately before this incident, with nearly half of the allotted time having been
played and the match still scoreless, the probabilities had drifted to 0.2621/0.3852/
0.3527. Immediately after Newcastle were reduced to ten men the probabilities
tilted in Everton’s favour, to 0.1301/0.3214/0.5485. As the second half progressed
with the score remaining goalless, the draw probability steadily increased. By the
80th minute, for example, the probabilities had shifted to 0.0902/0.6367/0.2731.
Conclusion
Chapter 5 has examined the extent to which the strategic behaviour of individ-
ual players or football teams during the course of the matches they play can be
rationalised in accordance with game-theoretic principles of optimising strategic

behaviour by independent agents when payoffs are interdependent.
Whenever a penalty kick is awarded and executed, a two-person non-cooperative
zero-sum game is played out, in the space of a few seconds, between the kicker
and the opposing goalkeeper. Theoretical and empirical research on the strategic
choices of penalty kickers and goalkeepers suggests that, perhaps unknowingly,
the behaviour of the players conforms closely to the predictions of game theory.
Kickers and goalkeepers appear to randomise their choices between kicking and
diving in either direction, and the proportions of penalties converted and penalties
where a goal is averted are insensitive to the decisions of the kicker and the goal-
keeper, as the theory suggests.
Viewed more generally, the football match in its entirety has many of the char-
acteristics of a dynamic ‘game’ in the economist’s sense. At any stage of the
match, teams choose between defensive and attacking formations and between a
non-violent and a violent style of play, so as to maximise their expected payoffs at
the end of the match. Their strategic choices determine the probabilities of scor-
ing and conceding goals at the current stage of the match, and the probabilities of
having a player dismissed.
In the dynamic game-theoretic model that is presented in this chapter, opti-
mal strategic behaviour at any stage of the match is dependent on the current
difference in scores, and on the amount of playing time that has elapsed. Teams
that are trailing tend to attack, and teams that are trailing tend to play aggres-
sively for as long as there is a realistic chance of salvaging the match. Stronger
teams when trailing play aggressively for longer than weaker teams, and home
teams when trailing play aggressively for longer than away teams. Towards
the end of matches in which scores are level, the subgames have a prisoner’s
dilemma structure, and there is a tendency for the teams to defect from non-
violence to violence. This feature of the model is consistent with an observed
tendency for the number of dismissals to increase markedly during the closing
stages of matches.
The model is subjected to empirical scrutiny, using data on the timings of
goals and player dismissals from more than 16,000 English professional league
matches. Several features of the empirical hazard functions for the conditional
arrival rates of player dismissals and goals are found to be consistent with the
theoretical model. In particular, teams that are currently trailing by one goal
or by two goals are willing to bear an increased probability of losing a player
through dismissal, in order to increase the probability of scoring a goal quickly.
Similarly, the goal-scoring rates of teams that are trailing by one goal or by two
goals are significantly higher than they are for teams that are leading or level in
scores. Stochastic simulations are used to obtain in-play match result probabil-
ities, conditional upon the current state of the match at any stage, defined by an
underlying relative team quality measure, the difference in scores and any numer-
ical disparity in players.
Notes
1 The 90-minute duration is notional, because there are frequent discontinuities in play for
various reasons: play is interrupted when the ball travels outside the confines of the pitch,
when a foul is committed or a player is penalised under the offside law, when a goal is scored
and when a player requires treatment on the pitch for injury. Although a few minutes of
stoppage time are added on at the end of each 45-minute period of play, the time added is
invariably less than the time lost through routine stoppages.
2 The ‘golden goal’ rule was used in several elimination tournaments during the 2000s, in
matches that were drawn after 90 minutes’ play and were continued into an additional period
of extra time. Extra time normally lasts for 30 minutes, and extra time continues regardless
of whether any goals are scored. In tournaments where the golden goal rule was applied,
a goal scored during extra time would finish the match immediately, with the scoring team
victorious.
3 Prior to the 2000 season when the new rule was introduced, a team that won a match during
a five-minute sudden-death overtime period (played when the scores were tied at the end of
regular time) was awarded two league points, while its losing opponent received zero points.
The new rule awarded two points to a team that won during overtime, and one point to the
losing team.
4 The solutions for the two teams’ optimal strategies for the first minute should be determined
by solving the model iteratively ‘backwards’, all the way from the 90th minute to the first
minute. In order to construct this example, however, a set of match result probabilities applic-
able at the end of the first minute has been specified arbitrarily. These probabilities determine
the two teams’ payoffs at the end of the first minute conditional on their strategies for the first
minute.
5 The data source is soccernet.esp.go.com. Match information posted on this website records
the match durations (in minutes) at which player dismissals occurred and goals were scored.
The data set was compiled by transcribing these details for each match individually.
6 This is a simplification because a player who commits a foul in his own team’s penalty area,
and by so doing denies the opposing team a goal-scoring opportunity, is liable to be dis-
missed, while the opposing team has an opportunity to score from the resulting penalty kick.
The proportions of all player dismissals and goals that arrive simultaneously in this manner
are relatively low, however. Ignoring this element of simultaneity greatly simplifies the spe-
cification and estimation of the empirical model.
7 Applying the log transformation to the arrival rates ensures that the fitted arrival rates gen-
erated by the estimated model are always positive.
8 Approximations for in-play match result probabilities at other durations, or for matches
where the value of RELQUAL differs from those reported in Tables 5.12–5.14, may be
obtained by means of interpolation or extrapolation based on the reported probabilities.
The accuracy of results obtained by linear interpolation will depend upon how linear or
non-linear are the relationships between RELQUAL or match duration and the reported
probabilities in the nearest vicinity of the interpolated probabilities.
6 English professional
football: historical development
and commercial structure
Introduction
Professional football as a sport has always been inextricably linked to its attributes
as a business, but never more so than during the modern era. Complaints are aired
regularly in the media and elsewhere that players are overpaid; that the transfer
market is out of control; that the whims of wealthy owners are driving many clubs
to penury; that exorbitant ticket prices are driving spectators away from football;
and that the priorities of television are dictating both the strategic and the oper-
ational decisions of football clubs and the sport’s organising bodies.
Chapter 6 presents an overview of the historical development of English club
football as a business, and analyses its current economic, financial and commercial
structure. Section 6.1 describes the competitive structure of the major league and
cup tournaments in which English teams participate, and identifies historical trends
and patterns in the performance of groups of teams distinguished by characteristics
such as geographical location, city size and date of entry into the league. Section 6.2
describes trends in match attendances, and the explanations for changing patterns
of attendance that have been proposed by historians, sociologists and economists.
Section 6.3 provides an overview of the profitability or loss-making propen-
sities of English football clubs. The following three sections describe historical
and current trends in the main revenue and cost items that appear in a football
club’s profit and loss account: namely gate revenues, examined in Section 6.4;
broadcast revenues, examined in Section 6.5; and expenditure on players’ wages
and salaries, and transfer expenditure, examined in Section 6.6. Finally, Section
6.7 describes historical and current trends in the ownership, governance and finan-
cing of English football clubs.
6.1 English professional football: competitive structure and team performance

Origins and competitive structure
The origins of English football can be traced back to the late Middle Ages.
Originally football was a rough and violent game, involving unspecified numbers
139
140 English professional football
of young men pursuing a ball through urban or rural locations, often resulting
in serious damage to property, personal injury and even death. Despite repeated
attempts by both church and state to ban the sport, street or village football
remained popular until the arrival of industrialisation, when newly emerging pat-
terns of urban residence and employment imposed new constraints and discip-
lines. In the first half of the nineteenth century football survived mainly within the
upper-class public schools and universities, where the codes that define the mod-
ern sports of association football and rugby football were developed. By impos-
ing rules and conventions on school sports that previously were perhaps no less
disorderly than their street and village antecedents, reform-minded schoolmasters
such as Sir Thomas Arnold of Rugby sought to inculcate qualities of discipline,
courage and leadership among pupils. Slowly, as successive generations of pupils
graduated from education into adult life, the popularity of sports such as football
and rugby spread beyond the schools and universities. At a landmark meeting
of representatives of a number of London and suburban clubs in October 1863,
English football’s present-day governing body, the Football Association (FA),
was established (Walvin, 1994; Szymanski and Zimbalist, 2006).
English football’s two most durable and important club competitions both
emerged soon afterwards. Fifteen clubs first contested the FA Cup during the win-
ter of 1871–2. The Wanderers, a team comprising players who had attended the
leading public schools and Oxford or Cambridge Universities, defeated the Royal
Engineers in the first final, watched by 2,000 spectators. During the 1880s, foot-
ball’s geographic centre of power shifted away from the predominantly southern
ex-public school and ex-university clubs, towards teams based in manufacturing
towns and cities in the Midlands and the North West. Enlightened factory owners
and employers, many of whom were themselves public school graduates, began
to see the benefits of regular Saturday afternoon holidays for workplace morale
and productivity, and created new opportunities for the development of organ-
ised working-class leisure activities. New football clubs were formed, often at the
instigation of local church leaders, throughout England during the late 1870s and
1880s. A northern team, Blackburn Olympic, won the FA Cup for the first time in
1883, followed by Blackburn Rovers, three-time winners between 1884 and 1886.
The principle of professionalism was accepted by the FA in July 1885. Its rec-
ognition was a key development in the processes leading to the eventual forma-
tion of the Football League in 1888. Twelve teams were members of the league
during its first three seasons, with Preston North End the inaugural champions
in the 1889 and 1890 seasons, followed by Everton in 1891. During the next three
decades, membership of the Football League expanded progressively. In the 1893
season, a second tier of twelve teams was created, while membership of the top
tier was expanded to sixteen. Several of the leading names of the modern era made
their debut appearances in the 1890s, including Manchester United (then Newton
Heath) in the 1893 season and Liverpool in the 1894 season. Woolwich Arsenal
was the first southern club admitted, also in the 1894 season. Total membership
Competitive structure and team performance 141
was increased to thirty-two teams in the 1895 season, then thirty-six in the 1899
season, then forty in the 1906 season.
The league finally achieved a membership of magnitude comparable to the com-
bined strength of the present Premier League and Football League (ninety-two
teams) shortly after the First World War. In the 1921 season, twenty-two teams
from the Southern League were added to the existing two-tier structure (which
had by then expanded to forty-four teams) to form the new Division 3 (South). In
the 1922 season a new Division 3 (North) was created, initially comprising twenty
teams, but with membership increased to twenty-two two seasons later.
The most important structural changes to the league’s format since the 1920s
were as follows:
• A further increase in league membership from eighty-eight to ninety-two teams
in the 1951 season;
• The reorganisation of the lower two tiers into Divisions 3 and 4 from the 1959
season, with membership determined on merit (by promotion and relegation)
rather than by geographical location;
• An increase in mobility within the league with three (rather than two) teams
promoted and relegated between T1 and T2, and between T2 and T3, each sea-
son from the 1974 season onwards, and a play-off system introduced in the 1987
season to determine one of the promotion places for each of T2, T3 and T4;
• Several adjustments to the size of the tiers during the late 1980s and early 1990s,
with automatic promotion and relegation between T4 and the highest tier below
the Football League also implemented in a number of seasons during this period
and routinely during the 2000s; and
• The withdrawal from the Football League of the T1 clubs to form a breakaway
Premier League, starting in the 1993 season. This development has not affected
professional football’s basic competitive structure, but it has had profound
organisational and financial implications.
In addition, there were several changes to the names of the four tiers during the
1990s and 2000s. As was noted in Chapter 3, the official nomenclature at the time
of writing (during the 2010 season) is Premiership, Championship, League One
and League Two. In the rest of this chapter, as throughout this volume, the four
divisions are known as T1, T2, T3 and T4 (tiers one to four).
Since the Second World War, the number of tournaments has proliferated.
Although the FA Cup has continued in a sudden-death format essentially
unchanged since its inception in the 1870s, many more clubs (both amateur and
professional) now enter, with the major clubs competing only in the final stages.
By the 2009 season, the total entry had expanded to 762 teams, with the qualifying
stages starting in August, and the showpiece final played at the end of the domes-
tic season in May.
The most important domestic competition of post-war vintage is the League
Cup. Entry to this sudden-death knockout tournament is open to league clubs
only. The League Cup made an uncertain start with a number of leading clubs
either refusing to enter, or to take the competition seriously, for several years fol-
lowing its introduction in the 1961 season. Aston Villa of T1 defeated Rotherham
of T3 over two legs in the first final, and several other lower-tier teams reached the
final during the 1960s. By the start of the 1970s, the League Cup was established
as a prestigious competition which all clubs considered worth taking seriously.
The prestige of both major domestic cup competitions has suffered in recent sea-
sons, however, with the leading clubs (in particular) often choosing to field reserve
teams in order to keep their star players fresh for league and European fixtures
that they consider to be more important.
Regular competitive football at European level was first introduced in 1955.
Birmingham City and a combined London team both competed in the first Fairs
Cup tournament (the predecessor of the UEFA Cup) staged between 1955 and
1958. Meanwhile, Manchester United was the first English club to enter the
European Cup in the 1957 season. By the 1961 season, both the Fairs Cup and a
new European Cup Winners’ Cup were also operating on an annual basis. From
then until 1985, when English clubs were excluded from European tournaments
following the hooligan-related Heysel stadium disaster at the European Cup Final
in Brussels, the three European competitions attracted a combined quota of up to
eight leading English clubs per season. England’s ban was rescinded in the 1991
season. Since then, European competition has acquired an increasingly prominent
position in the fixture lists of the top clubs. The importance of regular European
participation has been enhanced by rapid growth in the financial rewards available
from increasingly lucrative television contracts. During the 1990s, a league format
was introduced for the early stages of the European Cup, which was renamed the
Champions League. This removed the threat of early elimination, and guaranteed
all entrants a minimum number of fixtures. In the 2009 season, thirty-two clubs,
including up to four from each of the leading European domestic leagues (rather
than just the domestic champions) gained entry to the Champions League. Some
clubs (the champions and runners-up from the major football powers) enter the
league direct, while others (third- and fourth-placed clubs from the major powers,
and champions from lesser powers) participate in a qualifying tournament, which
eliminates many of the weaker aspirants before the main competition gets under-
way. The 1999 season also saw the consolidation of the UEFA and Cup Winners’
Cups into a single UEFA Cup tournament. Following several further changes of
format, the UEFA Cup was retitled as the Europa League at the start of the 2010
season.
Team performance
Although the main focus of this chapter is the financial performance, rather than
the sporting performance, of English football clubs, it is obvious that there are
direct linkages between the two. At the micro level, a club’s capacity to generate
revenue depends on its team’s success on the field of play. On the other hand, a
club’s capacity to strengthen its team by purchasing better players in the transfer
market and by offering remuneration at a level that will attract and retain the best
players depends on the strength of its finances. At the macro level, economists have
argued that the attractiveness to spectators, and therefore the revenue-generating
potential of any sports league, depends on the maintenance of a reasonable level
of competitive balance (see Chapters 1, 2 and 3). Before considering the various
economic and financial aspects of English football’s historical development as a
business, however, it is useful first to identify a few key facts concerning team per-
formance and championship dominance.
Table 6.1 summarises data on the best-performing teams in the league cham-
pionship and the cup (domestic and European) competitions. The first panel
shows the number of championship victories per team in each decade since the
end of the First World War. As an alternative and slightly broader measure of
championship dominance, the second panel shows the top-performing teams in
each decade, scored by awarding three, two and one points respectively to the
teams finishing first, second and third in T1 in each season. The third and fourth
panels show, for each decade, all teams that won the FA Cup and League Cup.
The fifth panel shows the winners in European competition in each decade.
On both sets of measures of championship dominance, Arsenal in the 1930s,
Liverpool in the late 1970s and 1980s and Manchester United in the 1990s and
2000s all achieved a level of dominance that was not matched by any team in
any of the other decades. Many commentators attribute Arsenal’s success in the
1930s to superior training methods and mastery of tactics, initially under the man-
agement of Herbert Chapman until his death in 1934. Continuity of an effective
managerial style, maintained through a series of internal appointments (with Bill
Shankly followed by Bob Paisley, Joe Fagan and Kenny Dalglish) is seen as a key
factor in Liverpool’s success during the 1970s and 1980s. While Alex Ferguson’s
managerial contribution during the 1990s and 2000s is widely acknowledged,
detractors tend to see a degree of inevitability in Manchester United’s success
given the club’s overwhelming financial strength.
Table 6.1 indicates that the 1920s and 1960s were the decades in which the des-
tination of the championship from season to season was most unpredictable.
Although Huddersfield Town was by some distance the most successful team
of the 1920s, this decade was unusual in respect of the total number of teams
that managed to achieve a top three position on at least one occasion. Two teams
shared top honours almost evenly during the 1950s: Wolverhampton Wanderers
and Manchester United. The 1960s and the first half of the 1970s can be seen as
a continuum, with ten different teams winning the championship during sixteen
seasons, and no team winning in consecutive seasons. By winning three of the last
four championships in the 1970s, another six during the 1980s, and one more in
1990, Liverpool established a degree of dominance unprecedented at the time, but
subsequently matched, and perhaps even surpassed, by Manchester United. The
Table 6.1 Historical performance of top English teams in league and cup competition
1920–1929 1930–1939 1947–1959 1960–1969 1970–1979 1980–1989 1990–1999 2000–2009
League championships
Huddersf’ld 3 Arsenal 5 Man Utd 3 Liverpool 2 Liverpool 4 Liverpool 6 Man Utd 5 Man Utd 6
Liverpool 2 Everton 2 Wolves 3 Man Utd 2 Derby 2 Everton 2 Arsenal 2 Arsenal 2
Burnley 1 Man City 1 Portsmouth 2 Burnley 1 Arsenal 1 Arsenal 1 Blackburn 1 Chelsea 2
Everton 1 Sheff Wed 1 Arsenal 2 Everton 1 Everton 1 Aston Villa 1 Leeds 1
Newcastle 1 Sunderland 1 Chelsea 1 Ipswich 1 Leeds 1 Liverpool 1
Sheff Wed 1 Liverpool 1 Leeds 1 Nottm Forest 1
WBA 1 Tottenham 1 Man City 1
Tottenham 1
League championship – points for a top three finish
Huddersf’ld 14 Arsenal 18 Man Utd 19 Man Utd 10 Liverpool 19 Liverpool 24 Man Utd 21 Man Utd 23
Liverpool 6 Sheff Wed 7 Wolves 16 Liverpool 9 Leeds 10 Everton 8 Arsenal 9 Arsenal 15
Burnley 6 Everton 6 Arsenal 8 Tottenham 8 Derby 7 Man Utd 6 Liverpool 7 Chelsea 13
Sunderland 5 Sunderland 5 Tottenham 8 Burnley 7 Arsenal 5 Ipswich 5 Blackburn 5 Liverpool 7
WBA 5 Aston Villa 4 Portsmouth 7 Leeds 7 Nottm Forest 5 Arsenal 4 Newcastle 5 Leeds 1
Everton 3 Derby 4 Preston 5 Everton 5 Everton 4 Aston Villa 3 Aston Villa 4 Newcastle 1
Leicester 3 Man City 4 Blackpool 3 Ipswich 3 Ipswich 2 Nottm Forest 3 Leeds 3
Newcastle 3 Wolves 4 Chelsea 3 Man City 3 Man City 2 Southampton 2 Chelsea 1
Sheff Wed 3 Charlton 3 Liverpool 3 Wolves 3 QPR 2 Tottenham 2 Crystal Pal 1
Arsenal 2 Huddersf’ld 3 WBA 2 Nottm Forest 2 Chelsea 1 Watford 2 Norwich 1
Bolton 2 Preston 1 Burnley 1 Sheff Wed 2 Man Utd 1 West Ham 1 Nottm Forest1
Cardiff 2 Tottenham 1 Derby 1 Chelsea 1 Tottenham 1 Sheff Wed 1
Man City 2 Huddersf’ld 1 WBA 1 Tottenham 1
Tottenham 2 Sunderland 1
Aston Villa 1
Chelsea 1
FA Cup wins
Bolton 3 Arsenal 2 Newcastle 3 Tottenham 3 Arsenal 2 Liverpool 2 Man Utd 4 Arsenal 3
Aston Villa 1 Everton 1 Arsenal 1 Everton 1 Chelsea 1 Man Utd 2 Arsenal 2 Chelsea 3
Blackburn 1 Man City 1 Aston Villa 1 Liverpool 1 Ipswich 1 Tottenham 2 Chelsea 1 Liverpool 2
Cardiff 1 Newcastle 1 Blackpool 1 Man City 1 Leeds 1 Coventry 1 Everton 1 Man Utd 1
Huddersf’ld 1 Portsmouth 1 Bolton 1 Man Utd 1 Liverpool 1 Everton 1 Liverpool 1 Portsmouth 1
Newcastle 1 Preston 1 Charlton1 WBA 1 Man Utd 1 West Ham 1 Tottenham 1
Sheff Utd 1 Sheff Wed 1 Derby 1 West Ham 1 Southampton 1 Wimbledon 1
Tottenham 1 Sunderland 1 Man City 1 Wolves 1 Sunderland 1
WBA 1 Man Utd 1 West Ham 1
Nottm Forest 1
WBA 1
Wolves 1
League Cup wins (from 1961)
Aston Villa 1 Aston Villa 2 Liverpool 4 Aston Villa 2 Chelsea 2
Birmingham 1 Man City 2 Arsenal 1 Arsenal 1 Liverpool 2
Chelsea 1 Nottm Forest 2 Luton 1 Chelsea 1 Man Utd 2
Leeds 1 Tottenham 2 Norwich 1 Leicester 1 Blackburn 1
Leicester 1 Stoke 1 Nottm Forest 1 Liverpool 1 Leicester 1
Norwich 1 Wolves 1 Oxford 1 Man Utd 1 Middlesbro 1
QPR 1 Wolves 1 Nottm Forest 1 Tottenham 1
Swindon 1 Sheff Wed 1
WBA 1 Tottenham 1
European competition wins (from 1956)
Leeds 1 Liverpool 4 Liverpool 2 Man Utd 2 Liverpool 2
Man Utd 1 Chelsea 1 Aston Villa 1 Arsenal 1 Man Utd 1
Newcastle 1 Leeds 1 Everton 1 Chelsea 1
Tottenham 1 Man City 1 Ipswich 1
West Ham 1 Nottm Forest 1 Nottm Forest 1
Tottenham 1 Tottenham 1
Source: Butler (1987), Smailes (1992), Rothmans/Sky Sports Football Yearbook

latter’s championship victory in 1993 ended a barren run stretching back to 1967
and was the first of eleven triumphs in seventeen seasons. By winning the cham-
pionship in 1989 and 1991, Arsenal were instrumental in ending Liverpool’s long
spell of dominance; and further victories in 1998, 2002 and 2004 consolidated
Arsenal’s status as Manchester United’s closest challengers. Chelsea’s emergence
as a major force, with championship victories in 2005 and 2006, saw Manchester
United deprived of the title for three consecutive seasons, before a run of consecu-
tive victories (2007–9) reaffirmed their dominance.
Also notable in Table 6.1 is a shift in demographic base of the most success-
ful clubs. In all decades since the 1930s, clubs from the six largest English cities
(London, Birmingham, Liverpool, Manchester, Sheffield and Leeds) have claimed
the majority of championship wins (at least six out of every ten). As late as the early
1960s, however, Burnley and Ipswich Town repeated the earlier championship suc-
cesses of other smaller-market clubs like Huddersfield Town and Portsmouth. In
the 1970s, Derby County and Nottingham Forest also interrupted the dominance
of the largest-market clubs. Since the 1980s, however, only Blackburn Rovers have
achieved the same feat, in the 1995 season. In any event, as arguably the wealthiest
club in the league at the time thanks to the extraordinary largesse of their benefac-
tor Jack Walker, Blackburn have some cause to be seen as an exceptional case.
Table 6.1 indicates that success at the top of T1 has never previously been as
highly concentrated as it was during the 2000s, when only three different teams
won the league championship and only six achieved a top three finish (two of
which, Leeds United and Newcastle United, did so only once, finishing in third
position in the 2000 and 2003 seasons, respectively). Likewise only five different
teams won the FA Cup. Portsmouth (2008) were the only team to do so from out-
side the small clique of four (Arsenal, Chelsea, Liverpool and Manchester United)
that also dominated the league championship. Perhaps more surprisingly, the
2000s were the least successful decade for English teams in terms of European tro-
phy success since the 1950s. Despite regularly dominating the latter stages of the
Champions League, particularly towards the end of the decade, only Liverpool
(2005) and Manchester United (2008) won the coveted title; and Liverpool (2001)
were the only successful English team in the UEFA Cup.1
As noted in Chapter 3, Section 3.1, FA Cup match results data suggest there was
an increase in competitive inequality during the 1920s and 1930s, and a further
increase that began during the mid-1970s and was still in progress by the end of
the 1990s (Dobson and Goddard, 2004). Table 6.2 reports data on the incidence
of ‘giant-killings’ in FA Cup ties between the 1974 and 2009 seasons, in the form
of the proportions of ties contested by teams from different tiers that were won by
the lower-ranked team. These data are reported in four-season bands, collectively
for all ties contested by teams from different tiers, and separately for each permu-
tation of rankings (for example, T1 versus T2, T1 versus T3, T2 versus T3, and so
on). Ties contested by a league team and a non-league team are included, with
all non-league teams treated as a single category. The final column of Table 6.2
Table 6.2 Incidence of giant-killings in FA Cup ties, 1974–2009
T1 vs T1 vs T1 vs T1 vs T2 vs T2 vs T2 vs T3 vs T3 vs T4 vs
T2 T3 T4 NL T3 T4 NL T4 NL NL All
Proportion of ties won by lower-ranked team

1974–1977 .279 .222 .125 .125 .481 .200 .000 .333 .189 .463 .292
1978–1981 .306 .214 .091 .000 .458 .600 .250 .292 .161 .286 .265
1982–1985 .299 .267 .000 .000 .435 .250 .000 .368 .164 .340 .272
1986–1989 .197 .200 .125 .333 .478 .400 .000 .421 .125 .298 .272
1990–1993 .279 .208 .154 .000 .444 .250 .167 .283 .127 .309 .256
1994–1997 .263 .250 .000 .000 .455 .182 .250 .339 .185 .276 .256
1998–2001 .172 .250 .091 .222 .364 .111 .167 .293 .203 .292 .236
2002–2005 .299 .231 .100 .000 .263 .471 .200 .474 .234 .224 .284
2006–2009 .200 .300 .100 .000 .300 .200 .111 .391 .217 .298 .255
Number of ties
1974–1977 61 27 8 8 27 10 3 60 74 41 319
1978–1981 62 28 11 5 18 10 8 48 62 42 294
1982–1985 67 30 5 4 20 16 5 57 67 53 324
1986–1989 61 20 16 6 20 15 5 57 48 57 305
1990–1993 68 24 13 4 34 8 6 53 55 55 320
1994–1997 80 24 10 7 28 11 8 56 54 58 336
1998–2001 64 28 11 9 22 9 6 58 59 65 331
2002–2005 67 13 20 5 18 17 5 57 64 58 324
2006–2009 70 20 10 7 18 15 9 64 46 47 306
Note: NL denotes non-league. NL is treated as a single category. The column headed ‘All’
refers to all FA Cup ties contested by league teams from different tiers, and ties contested by a
league team and a non-league team.
reports the overall proportion of ties won by the lower-ranked team, and reflects
a declining incidence of giant-killings during the 1970s, 1980s and 1990s. During
the 2000s, however, the downward trend in the incidence of giant-killings has not
continued. This could reflect a cessation of the trend towards rising competitive
inequality; or it could reflect the increasing popularity among the leading teams
(and even among some of the lesser teams) of a policy of resting star players and
selecting reserve-team players for FA Cup ties.
Below we seek to identify some broader patterns in the performance of teams
throughout the entire league. For this purpose, Table 6.3 identifies five groups of
clubs with broadly similar characteristics (Groups 1 to 5, henceforth known as G1
to G5), using three simple criteria to define the clubs that make up each group:
• The club’s home-town population recorded in the 1961 Census of Population;
• The date of the club’s initial entry into the league;
• The club’s geographical location (for which there are two broad categories:
South and Midlands/North).
Table 6.3 Group definitions
Group 1: Clubs from towns with populations larger than 500,000, which entered the
league before its expansion in the early 1920s.
Arsenal, Aston Villa, Birmingham City, Chelsea, Everton, Leeds United,
Liverpool, Manchester City, Manchester United, Sheffield United,
Sheffield Wednesday, Tottenham Hotspur, West Bromwich Albion, West
Ham United (14 clubs).
Group 2: Clubs from towns with populations in the range 250–500,000 in the
Midlands and North (all English regions from the East and West
Midlands northwards).
Bradford City, Bradford Park Avenue, Coventry City, Derby County, Hull
City, Leicester City, Newcastle United, Notts County, Nottingham Forest,
Port Vale, Stoke City, Sunderland, Wolverhampton Wanderers (13 clubs).
Group 3: Other clubs from towns in the south (the South East, South West and East
Anglia regions, plus South Wales) with populations below 500,000, as
well as the smaller London clubs not included in Group 1. Most of these
clubs (except Bristol City, Luton Town, Fulham and Leyton Orient)
joined the League during or after its early 1920s expansion.
Aberdare, Aldershot, Barnet, Bournemouth, Brentford, Brighton and Hove
Albion, Bristol City, Bristol Rovers,Cambridge United, Cardiff City,
Charlton Athletic, Cheltenham Town, Colchester United, Crystal Palace,
Dagenham and Redbridge, Exeter City, Fulham, Gillingham, Ipswich
Town, Leyton Orient, Luton Town, Maidstone United, Merthyr Town,
Millwall, Newport County, Norwich City, Oxford United, Peterborough
United, Plymouth Argyle, Portsmouth, Queens Park Rangers, Reading,
Southampton, Southend United, Swansea City, Swindon Town, Thames,
Torquay United, Watford, Wimbledon/Milton Keynes Dons, Wycombe
Wanderers, Yeovil Town (42 clubs).
Group 4: All clubs from smaller towns in the Midlands/North which entered the
League before 1920.
Barnsley, Blackburn Rovers, Blackpool, Bolton Wanderers, Burnley, Bury,
Chesterfield, Crewe Alexandra, Doncaster Rovers, Gateshead, Grimsby
Town, Huddersfield Town, Lincoln City, Middlesbrough, Oldham Athletic,
Preston North End, Rotherham United, Stockport County, Walsall (19
clubs).
Group 5: Clubs from the Midlands/North which entered the League during or after
the early 1920s expansion.
Accrington Stanley, Ashington, Barrow, Boston United, Carlisle United,
Chester City, Darlington, Durham City, Halifax Town, Hartlepool
United, Hereford United, Kidderminster Harriers, Macclesfield Town,
Mansfield Town, Morecambe, Nelson, New Brighton, Northampton Town,
Rochdale, Rushden and Diamonds, Scarborough, Scunthorpe United,
Shrewsbury Town, Southport, Stalybridge Celtic, Tranmere Rovers,
Wigan Athletic, Workington Town, Wrexham, York City (30 clubs).
Figure 6.1 shows the percentage share of the clubs in each of the five groups in an
aggregate performance score, calculated by awarding 92 points to the club which
finished first in T1, 91 points to the club which finished second, and so on, in
each season. Minor adjustments to the points system are made for seasons when
the number of teams in the league was greater or less than 92. Teams finishing
in equal positions in the old Divisions 3 (South) and (North) are awarded equal
points: first position in either division gains 48 points, second gains 46 points, and
so on.
The relative constancy of G1’s performance score emphasises the fact that
the dominance of the clubs from the largest cities has been a consistent feature
throughout the league’s history, and is by no means a uniquely recent phenom-
enon. In contrast, the G2 clubs from the next city size-band, which include three
championship winners from the 1920s, 1930s and 1950s (Newcastle United,
Sunderland and Wolverhampton Wanderers), experienced a significant decline
in their performance as a group between the late 1930s and late 1950s, and again
during the 2000s.
G3 clubs registered a steady and sustained improvement in average performance
between the 1920s and the end of the 1980s. Their progressive advance represents
the most important long-term shift in the geographical balance of football power.
Since the Second World War, a small majority of the clubs that gained admission
to the league are located in the south.2 More important, however, has been a pro-
gressive rise in the status of many of southern clubs that entered the league during
its early 1920s expansion or later, and were engaged in a catching-up process for
several decades subsequently. Charlton Athletic, Crystal Palace, Ipswich Town,
Norwich City, Portsmouth, Queens Park Rangers, Southampton and Swindon
Town all rose from modest starting positions in Division 3 (South) eventually
to feature prominently in T1 and T2. Collectively, however, the G3 clubs were
not the main beneficiaries of English football’s popular revival during the 1990s.
According to Figure 6.1, G3’s performance peaked in the late 1980s and early
1990s. A sharp decline during the mid to late 1990s preceded a partial recovery
during the 2000s.
G4 consists mainly of northern clubs from smaller towns, all of which were
already established in the two upper tiers at the time of the league’s early 1920s
expansion. These clubs were the main victims of the southern clubs’ upward pro-
gress between the 1920s and 1980s. Many clubs located in traditional industrial
or manufacturing towns especially hard-hit by three major recessions during the
1970s, 1980s and 1990s, have not been able to keep pace with their more upwardly
mobile southern counterparts. Nevertheless, several G4 members (Blackburn
Rovers, Bolton Wanderers, Middlesbrough) have responded adroitly to the oppor-
tunities presented by the recovery in football’s popular appeal. Accordingly, G4’s
performance score improved significantly during the 1990s and this improvement
has been sustained subsequently.
40
35 Group 3
30
Group 1
25
Per cent
Group 4
20
15
Group 2
10
Group 5
0
1922 1925 1928 1931 1934 1937 1947 1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007
Season
Figure 6.1 Percentage shares in aggregate performance of clubs in Groups 1 to 5
The long-term performance of the G5 clubs, many of which were the founder
members of Division 3 (North) in the early 1920s, is in marked contrast to that of
their southern counterparts in G3. From the start, there was an imbalance between
the respective strengths of the southern and northern Division 3 clubs. In the 1920
season, the last before the formation of Division 3 (South), the league comprised
thirty-seven clubs from the North and Midlands, and only seven from the South.
At this point a strong pool of aspiring southern clubs were clamouring for entry;
while most of the larger northern and midland towns were already represented.
This meant that Division 3 (North) was formed from a weaker pool of clubs than
Division 3 (South). Since the 1920s, several G5 clubs lost their league status, and
few of the survivors have enjoyed anything more than the briefest spells in the top
two tiers. During the second half of the 2000s, Wigan Athletic became a notable
exception by achieving and retaining T1 status.
6.2 Match attendances

Trends in attendances
This subsection describes broad trends in English league match attendances at
an aggregate level. A statistical analysis of the long-term determinants of attend-
ance for individual clubs during the post-Second World War period is included in
Chapter 11. Table 6.4 reports data on aggregate English league attendances between
1922 and 2009. During the inter-war period, the trend in attendances mirrored the
Match attendances 151
Table 6.4 English league attendances, aggregate and by tier
Aggregates (millions) Percentage shares

Season Total T1 T2 T3 T4 T1 T2 T3 T4
1922 25.6 12.5 6.1 4.4 2.5 48.8 23.9 17.4 9.9
1923 23.6 10.7 6.2 4.3 2.3 45.5 26.4 18.2 9.9
1924 23.3 10.5 5.8 4.0 3.0 44.9 25.1 17.3 12.7
1925 23.1 10.0 6.7 3.9 2.6 43.2 28.8 16.7 11.2
1926 23.1 10.4 6.1 3.9 2.6 45.3 26.6 16.8 11.4
1927 23.4 10.6 6.5 3.8 2.5 45.2 27.9 16.1 10.8
1928 23.5 10.6 6.9 3.6 2.5 44.9 29.3 15.2 10.6
1929 23.9 10.5 6.7 4.0 2.7 43.9 28.0 16.6 11.5
1930 22.9 10.5 6.3 3.9 2.3 45.6 27.5 17.1 9.8
1931 21.0 9.5 6.1 3.2 2.3 45.0 28.9 15.3 10.8
1932 21.8 9.9 5.6 3.9 2.2 45.7 25.9 18.0 10.3
1933 21.4 9.5 6.1 3.5 2.3 44.6 28.4 16.3 10.7
1934 22.5 10.4 5.8 3.8 2.5 46.5 25.8 16.8 11.0
1935 23.1 10.8 6.1 3.8 2.4 46.8 26.4 16.4 10.5
1936 24.7 11.4 6.9 4.0 2.4 46.1 28.0 16.1 9.8
1937 26.4 11.4 8.0 4.4 2.6 43.0 30.3 16.8 9.9
1938 27.9 11.6 8.6 4.6 3.1 41.7 30.9 16.5 11.0
1939 27.0 11.5 8.6 4.0 2.9 42.4 31.9 14.8 10.8
1947 35.4 14.9 11.0 5.6 3.9 42.1 31.1 15.9 10.9
1948 40.2 16.7 12.3 6.7 4.6 41.5 30.5 16.5 11.4
1949 41.0 17.9 11.1 7.0 5.0 43.7 27.1 17.0 12.2
1950 40.6 17.4 11.7 7.1 4.4 42.8 28.8 17.5 10.9
1951 39.5 16.7 10.8 7.3 4.8 42.2 27.3 18.5 12.0
1952 39.0 16.1 11.1 6.9 4.9 41.3 28.4 17.8 12.5
1953 37.1 16.1 9.7 6.7 4.7 43.2 26.0 18.0 12.7
1954 36.4 16.1 9.7 6.3 4.2 44.4 26.7 17.4 11.5
1955 34.1 15.1 9.0 6.0 4.0 44.2 26.4 17.6 11.9
1956 33.2 14.2 9.1 5.7 4.3 42.7 27.3 17.1 12.9
1957 32.7 13.8 8.7 5.6 4.6 42.2 26.6 17.1 14.0
1958 33.5 14.4 8.6 6.1 4.3 43.1 25.8 18.2 12.9
1959 33.7 14.7 8.6 6.1 4.2 43.6 25.6 18.2 12.6
1960 32.5 14.4 8.4 5.7 4.0 44.4 25.8 17.6 12.3
1961 28.6 12.9 7.0 4.8 3.9 45.3 24.5 16.7 13.5
1962 28.0 12.1 7.5 5.2 3.3 43.1 26.6 18.6 11.7
1963 28.8 12.4 7.4 5.7 3.2 43.3 25.8 19.7 11.3
1964 28.5 12.5 7.6 5.4 3.0 43.8 26.6 19.0 10.6
1965 27.6 12.7 7.0 4.4 3.5 46.0 25.3 16.0 12.7
1966 27.2 12.5 6.9 4.8 3.0 45.8 25.4 17.6 11.1
1967 28.9 14.2 7.3 4.4 3.0 49.2 25.1 15.3 10.3
1968 30.1 15.3 7.4 4.0 3.4 50.8 24.7 13.4 11.1
1969 29.2 14.5 7.4 4.3 3.0 49.8 25.2 14.8 10.2
1970 29.5 14.8 7.6 4.2 2.9 50.2 25.7 14.2 9.9
1971 28.2 14.0 7.1 4.4 2.8 49.6 25.1 15.5 9.8
1972 28.7 14.5 6.8 4.7 2.7 50.5 23.6 16.4 9.5
1973 25.4 14.0 5.6 3.7 2.1 55.0 22.1 14.7 8.3
Table 6.4 (cont.)
Aggregates (millions) Percentage shares

Season Total T1 T2 T3 T4 T1 T2 T3 T4
1974 25.0 13.1 6.3 3.4 2.2 52.3 25.3 13.7 8.7
1975 25.6 12.6 7.0 4.1 1.9 49.2 27.2 16.0 7.5
1976 24.9 13.1 5.8 3.9 2.1 52.6 23.3 15.8 8.3
1977 26.0 13.6 6.2 4.1 2.1 52.2 23.9 15.8 8.1
1978 25.4 13.3 6.5 3.3 2.3 52.2 25.5 13.1 9.2
1979 24.5 12.7 6.1 3.4 2.3 51.7 25.1 13.8 9.4
1980 24.6 12.2 6.1 4.0 2.4 49.4 24.8 16.2 9.5
1981 21.9 11.4 5.2 3.6 1.7 52.0 23.6 16.6 7.7
1982 20.0 10.4 4.8 2.8 2.0 52.1 23.8 14.1 10.0
1983 18.8 9.3 5.0 2.9 1.6 49.6 26.5 15.7 8.3
1984 18.3 8.7 5.4 2.7 1.6 47.4 29.2 14.9 8.5
1985 17.8 9.7 4.0 2.7 1.4 54.7 22.6 15.0 7.7
1986 16.5 9.0 3.6 2.5 1.4 54.8 21.5 15.1 8.5
1987 17.4 9.1 4.2 2.4 1.7 52.6 24.0 13.5 9.9
1988 18.0 8.1 5.3 2.8 1.8 45.1 29.7 15.3 9.9
1989 18.5 7.8 5.8 3.0 1.8 42.3 31.6 16.4 9.7
1990 19.5 7.9 6.9 2.8 1.9 40.5 35.4 14.4 9.7
1991 19.5 8.6 6.3 2.8 1.8 44.2 32.2 14.6 9.0
1992 20.4 10.0 5.8 3.0 1.6 49.0 28.5 14.7 7.8
1993 20.6 9.7 5.9 3.5 1.5 47.2 28.5 16.9 7.5
1994 21.7 10.7 6.5 3.0 1.6 49.1 29.9 13.7 7.3
1995 21.8 11.2 6.0 3.0 1.6 51.4 27.5 13.9 7.1
1996 21.9 10.5 6.6 2.9 2.0 47.7 30.0 13.3 9.0
1997 22.8 10.8 6.9 3.2 1.9 47.4 30.5 14.0 8.1
1998 24.7 11.1 8.3 3.5 1.8 44.9 33.7 14.2 7.2
1999 25.4 11.6 7.5 4.1 2.1 45.8 29.6 16.3 8.3
2000 25.4 11.7 7.8 3.7 2.2 46.0 30.9 14.6 8.5
2001 26.1 12.5 7.9 3.5 2.2 47.9 30.4 13.4 8.4
2002 27.8 13.0 8.4 4.0 2.4 47.0 30.1 14.3 8.6
2003 28.5 12.8 9.3 3.9 2.5 44.9 32.8 13.7 8.6
2004 29.2 13.3 8.8 4.1 3.0 45.6 30.0 14.2 10.2
2005 29.2 12.9 9.6 4.2 2.6 44.0 32.9 14.3 8.8
2006 29.1 12.9 9.7 4.2 2.3 44.2 33.4 14.4 8.0
2007 29.5 13.1 10.1 4.1 2.3 44.2 34.0 14.0 7.8
2008 29.9 13.7 9.4 4.4 2.4 45.8 31.4 14.7 8.0
2009 29.9 13.5 9.9 4.2 2.3 45.3 33.1 14.0 7.7
Note: Until the 1958 season, T3 is Division 3 (South) and T4 is Division 3 (North).

Source: Tabner (1992), Rothmans/Sky Sports Football Yearbook
fortunes of the national economy closely, with a short post-war boom followed by
a period of decline from the 1922 season onwards, and then a gradual recovery
starting in the 1932 season. Following the Second World War, attendances surged,
achieving an all-time high of 41 million in the 1949 season. Again, the post-war
boom was relatively short-lived, and was followed by sustained decline in attend-
ances that continued, almost uninterrupted, until the 1986 season. The down-
ward trend was punctuated only by a few brief interludes, most notably following
England’s World Cup victory in 1966. The decline in aggregate attendance accel-
erated briefly in the early 1980s, but then levelled out. The 1987 season witnessed
an unexpected improvement, which has continued ever since; although the rate of
growth has slowed towards the end of the 2000s. Nevertheless the total attendance
for the 2008 season was the highest recorded for forty years, and the 2009 total
was only fractionally smaller. The growth in aggregate attendance would have been
faster had it not been for capacity constraints that are effective at the refurbished all-
seated stadia of many of the larger-market clubs (see Chapter 11, Section 11.1).
Table 6.4 also shows attendances by tier, and the percentage share of each tier
in the aggregate attendance. The percentage shares by tier were roughly constant
through the 1920s and most of the 1930s. During the last three seasons before the
Second World War, however, the attendances of T1 were 5.5 per cent higher than
during the previous three seasons; those of the clubs outside T1 were 24.1 per cent
higher. Consequently, T1’s percentage share fell by around 5 per cent just before
the Second World War. The story was similar during the post-war boom, but as
aggregate attendances started to decline, the lower-tier clubs shed spectators at a
faster rate than their T1 counterparts. Between the 1949 season when the aggre-
gate attendance peaked, and the 1986 season when the nadir was finally reached,
T1’s aggregate attendance fell by 49.7 per cent, while the combined attendance
of the lower three tiers fell by 67.5 per cent. Consequently, T1’s attendance share
increased by about 10 per cent. Most of the improvement in T1’s relative position
took place between the late 1950s and the mid-1970s.
Between the mid-1970s and the mid-1980s, as aggregate attendances continued
to fall, T1’s share hovered at just over 50 per cent. Since the mid-1980s, T2 has
enjoyed the fastest growth in attendance. T1 attendances did not start to increase
until the 1990 season, fully three seasons after the aggregate figure first started to
rise. While stadium capacity constraints are responsible for restricting the attend-
ance share of T1 to below 50 per cent in all but one season during the 1990s and
2000s, the creation of new capacity through either redevelopment (Manchester
United, Newcastle United) or relocation to new stadia (Arsenal, Manchester City,
Sunderland) has permitted continued growth in T1 attendances.
The percentage shares in aggregate attendance of the clubs in each of the five
groups (defined in Table 6.3) are shown in Figure 6.2. The graphs confirm the
narrowing of differentials between clubs in terms of attendance share immedi-
ately before and after the Second World War, and the widening that has taken
place since, especially between the late 1950s and the mid-1970s. It is noticeable
that the trends in the fortunes of the G1 and G3 clubs, measured in terms of per-
formance on the one hand and attendance on the other, differ strikingly between
Figures 6.1 and 6.2. While the average performance of the G1 clubs hardly
changed between the 1920s and 2000s, their attendance share has fluctuated
45
40
Group 1
35
30
Group 3
25
Per cent
20
Group 2
15
Group 4
10
5
Group 5
0
1922 1925 1928 1931 1934 1937 1947 1950 1953 1956 1959 1962 1965 1968 1971 1974 1977 1980 1983 1986 1989 1992 1995 1998 2001 2004 2007
Season
Figure 6.2 Percentage shares in aggregate attendance of clubs in Groups 1 to 5
markedly, between about 27 per cent (in the early 1950s) and 39 per cent (in the
mid-1980s). G3 clubs in contrast enjoyed significant improvements in perform-
ance until the 1980s, but struggled to achieve a corresponding increase in attend-
ance share. The attendance share of the twenty-seven G3 clubs in the league in
the 1922 season (one of which was in T1, and four in T2) was 22.1 per cent. By
the 1988 season, the number of G3 clubs had increased to 33 (nine of which
were in T1 and six in T2); but G3’s attendance share had only increased rather
modestly, to 29.1 per cent.
Explanations for changes in attendances

Economists, sociologists and social historians have all made strenuous efforts to
explain the trends in attendances that are summarised above. Social and demo-
graphic change, increasing material affluence, the growth of televised football,
football hooliganism, the state of football’s physical infrastructure and the quality
of the entertainment on the field of play, are among the many factors considered
to have contributed to the long post-war decline in football attendances (Walvin,
1994; Russell, 1997). More recently, several of the same factors have begun to
operate in directions that are more favourable from football’s perspective. This
subsection provides a brief survey of the debate in respect of English club football
attendances, starting in the inter-war period.
The geographical bond between English football clubs and their spectators was
undoubtedly stronger in the first half of the twentieth century than it became later.
Specific evidence on the geographical composition of football crowds is rather

sparse, but it seems certain that the proportion of home supporters drawn from
each club’s immediate vicinity was high. For example, twenty-one of the thirty-
three spectators who died in the 1947 Bolton Wanderers stadium disaster were
recorded as living within ten miles of the stadium (Fishwick, 1989). Although
both financial and time constraints militated against regular travel to away
matches in the 1930s and 1940s, Dunning, Murphy and Williams (1988) show that
occasional big games (especially FA Cup ties) were capable of attracting several
thousand travelling spectators. Fishwick (1989) also cites anecdotal evidence that
some Wiltshire spectators preferred to travel regularly to watch Reading, West
Bromwich Albion or Aston Villa rather than see the local team Swindon Town
struggling in the lower reaches of Division 3 (South) during the early 1930s.
In a historical survey focused primarily on the hooliganism phenomenon,
Dunning, Murphy and Williams (1988) characterise the typical football crowd
of the inter-war period as slightly more respectable than its pre-First World
War counterpart. While inter-war crowds remained predominantly working
class and male, there were significant and probably increasing middle-class and
female minorities. Russell (1999) suggests that local rivalries were sometimes
less intense than wider regional hostilities: while spectators often took pleas-
ure in the success of any team in their own region, football provided a con-
venient outlet for resentments arising from the contrasting economic fortunes
of northern and southern England. Although football hooliganism was never
eliminated completely, it seems to have been less pervasive during the inter-war
period than before the First World War. Subsequently it seems to have declined
further during the 1940s and early 1950s, before returning with a vengeance in
the mid-1950s.
Demobilisation and the relaxation of many wartime restrictions after the
Second World War released a widespread and unprecedented pent-up demand for
recreation and entertainment, which manifested itself not only in booming foot-
ball attendances, but also in record-breaking cinema audience figures, and packed
beaches at British seaside holiday resorts during summer. In explaining the subse-
quent sustained decline in football’s popularity as a spectator sport, social histor-
ians have emphasised the impact of broader socioeconomic changes. The growth
of the affluent society saw the emergence of more home-oriented and privatised
patterns of leisure activity, with the range of options extended in particular by
television and the car. Many women became less willing to acquiesce in a home-
based or child-minding function while men’s leisure was centred on the pub, club
or football terrace. The spread of home-ownership led to an increase in the popu-
larity of alternative weekend activities such as DIY and gardening (Dunning,
Murphy and Williams 1988; Walvin, 1994; Russell, 1997).
The increase in incidents of crowd misconduct both inside and outside football
stadia, of which the train-wrecking exploits of Everton and Liverpool supporters
in the late 1950s were an early manifestation, can also be attributed at least in
part to broader social changes. Dunning, Murphy and Williams (1988) interpret
football hooliganism as one unfortunate consequence of a general upsurge in
working-class assertiveness and decline in social deference, stimulated by a new
self-confidence boosted by high and stable levels of employment and increasing
affluence, which also manifested itself positively through the novels, films, plays
and popular music of the period. Whatever the causes, there is no doubt that
over time, hooliganism and the counter-measures taken by clubs and the police
prompted many spectators to switch from the terraces to the living room and TV
set. Football grounds became increasingly the preserve of the more fanatical sup-
porters, as the less committed drew the conclusion that the entertainment on offer
was inadequate to compensate for the inconvenience, discomforts and occasional
physical dangers caused by the hooligan phenomenon.
Rising affluence combined with improvements in both public and private trans-
port, including the spread of car ownership and the construction of the motorway
network, led to important changes in the geographical composition of demand.
With long-distance travel now practical both financially and logistically, the top
clubs began to drain support from their smaller counterparts. In the late 1960s
there emerged a distinctive and vociferous London contingent among the hoo-
ligan fringe of Manchester United’s support. This was symptomatic not only of
the growing hooligan problem, but also of the increasing tendency of the leading
clubs to draw support at a national rather than purely local or regional level.
Increased personal mobility increased the propensity for regular travel to away
fixtures. Demographic changes also eroded the bonds between communities and
their local clubs. Suburbanisation led to population movements away from city
centre districts where most football stadia were located, to be replaced by incom-
ers who did not necessarily have any affinity or interest in the local club, or foot-
ball in general. But paradoxically, spectators who remained loyal to their local
clubs appeared to become increasingly hostile towards the supporters of neigh-
bouring clubs, adding further fuel to the flames of hooliganism.
During the 1970s, a growing atmosphere of financial desperation enveloped
many football clubs. Neither the clubs nor the police seemed able to curb the hoo-
ligan-related incidents, which had become a regular Saturday afternoon occur-
rence, both inside football stadia and outside in town centres and railway stations.
There was a widespread perception of declining standards of conduct on the field
of play. Much of this seemed to reflect unerringly a malaise afflicting English
society on a broader scale, as politicians grappled with problems such as infla-
tion, unemployment, industrial unrest and social breakdown. With the economy
in deep recession by the start of the 1980s, it is unsurprising that the downward
trend in attendances accelerated. Margaret Thatcher’s Conservative government
came increasingly to favour direct intervention to confront football’s problems
head-on, including punitive legislation to tackle hooliganism, and proposals
(never implemented) for compulsory identity cards for all spectators.
English football hooliganism reached its apotheosis with the deaths of thirty-
eight mainly Italian supporters at the Heysel Stadium in Brussels at the Liverpool–
Juventus European Cup Final in May 1985. Heysel’s poor physical condition,
which was similar to that of many major stadia in Britain, was a contributory fac-
tor in this tragedy, as was misbehaviour among the Liverpool section of the crowd.
The declining state of domestic football’s physical infrastructure had already been
demonstrated catastrophically a few days earlier, as fifty-five spectators died when
a wooden stand caught fire during a T3 match between Bradford City and Lincoln
City on the final day of the 1985 league season. But it was only after ninety-six
Liverpool supporters were crushed to death against a collapsed section of perim-
eter fencing at the start of the Liverpool–Nottingham Forest FA Cup semi-final
played at Hillsborough (the home of Sheffield Wednesday) in April 1989, that the
momentum for fundamental change finally became irresistible. One of the main
recommendations of the ensuing report by Lord Justice Taylor (1990), that the
stadia of all clubs in the top two tiers should be converted to all-seated accom-
modation by the start of the 1995 season, was accepted and implemented with
scarcely a murmur of dissent.
During the 1990s, the demand for conversion to all-seated accommodation
provided both the opportunity and the impetus for more general refurbishment,
and in some cases the wholesale reconstruction of England’s major football sta-
dia. The removal of terracing, improvements to stadium facilities, advances in
crowd surveillance technologies, together with a new public mood of revulsion
following the worst excesses of the 1980s, all contributed to the marginalisation
of the hooliganism phenomenon. Russell (1997) is surely correct to draw the obvi-
ous connection between the improvement in standards of crowd behaviour (at
club level at least) and the reversal in the downward trend in attendances. The
level of investment in stadium relocation and redevelopment increased markedly
towards the end of the 1990s. According to data compiled by Deloitte (2009), an
average total investment of just under £200 million per annum has been sustained
throughout the 2000s. Between the 1993 and 2008 seasons, £2.571 billion in total
was invested in stadia and facilities by all Premier League and Football League
clubs. Clubs that relocated to new stadia during this period include Arsenal,
Bolton Wanderers, Coventry City, Derby County, Doncaster Rovers, Hull City,
Leicester City, Manchester City, Middlesbrough, MK Dons, Millwall, Reading,
Southampton, Stoke City, Sunderland, Swansea City and Wigan Athletic. Most
of these relocations involved the abandonment of stadia situated in congested
inner-city or residential locations, which in many cases had been occupied since
the late nineteenth century, in favour of new sites offering better access to road
links and space for car-parking.
Football’s rehabilitation as the most popular and fashionable national sport
has also been aided by skilful exploitation by the industry of selective aspects
of its own ‘heritage’. Whether viewing in person or on television, spectator and
(especially) media interest is now focused to a greater extent than ever before on
a handful of large-market clubs. It is still the case, however, that the majority
of spectators who attend English professional football in person do not watch
Arsenal, Chelsea, Liverpool and Manchester United. As Table 6.4 shows, more
than 54 per cent of the total attendance in the 2009 season went to matches out-
side T1.
English football largely succeeded in ‘rebranding’ itself as a middle-class spec-
tator sport during the 1990s and 2000s. Viewed in a positive light, the 1990s wit-
nessed the re-emergence of
the Victorian tradition of the modern English football stadium as an important source of civic
pride, in terms of its up-to-date amenities and facilities … (and) the popular ‘rediscovery’ of the
much cherished and at least part-mythical traditions of the English sports crowd for its mutual
tolerance and sociability.
Viewed negatively the same period witnessed

the effective end of the local and ‘organic’, the self-administering and creative, English foot-
ball crowd … (and) the emergence of the regulated, individuated, surveilled and high spending
seated bourgeoisie football audience. (Williams, 1997, p243, 244)
It is perhaps surprising to find that empirical evidence on the composition of

football crowds is in limited supply. On the basis of a review of eleven different
surveys conducted at various times over a fourteen-year period between 1983
and 1997 in England and Scotland, Malcolm, Jones and Waddington (2000)
question the common perception that the demographic composition of a typ-
ical 1990s football crowd was markedly different to that of its predecessors in
earlier decades.3 There is no convincing evidence that the age composition of
the typical crowd has changed very much overall. The proportion of female
spectators remained fairly constant, at between 10 and 13 per cent in most of
the surveys. There is some evidence, however, that female spectators tend to be
younger on average than males, and less likely to be married or living with a
partner.
Attempts to trace changes in the social composition of crowds are hampered by
definitional problems and inconsistencies. However, there is little sign of a change
over time in the proportions of unemployed or retired spectators. The proportion
of students appears to have declined, perhaps due to the deteriorating state of stu-
dent finances. In general Malcolm, Jones and Waddington suggest that the com-
position of the typical British football crowd may have changed less than many
commentators assume. By contrast, Tapp and Clowes (2002) and Tapp (2004)
suggest that the representation in the modern-day football crowd of the fanatical
hard-core supporter, who identifies solely with a particular club, may be in a state
of long-term decline. Casual spectators, often with loose allegiances to several
clubs, and interested in the quality of the entertainment on offer as well as prac-
tical details such as ease of access to the stadium and the nature of the arrange-
ments for purchasing tickets, constitute a growing but perhaps unrecognised
element. ‘The commonly held view of football supporters moving more towards a
Financial structure of English football 159
eterogeneous mix of social classes with an increasing number of female support-

h
ers was broadly supported’ (Tapp and Clowes, 2002, p1267).
6.3 Financial structure of English football: overview of profit and loss accounts

Section 6.3 provides an overview of the profit and loss accounts of English
football clubs, and the next three sections describe historical and recent trends
in the main revenue and cost items that appear in a football club’s profit and
loss account: namely gate revenues (Section 6.4), broadcast revenues (Section
6.5), and expenditure on players’ salaries and transfer expenditure (Section
6.6). Table 6.5 summarises the trends in total revenues, wages and salaries
expenditure, and operating profits, of all English league football clubs between
the mid-1990s and the late 2000s. The clubs are subdivided into T1, T2, and
T3 and T4 combined, and the figures are averages per season calculated over
three-season periods.
The data clearly highlight the divergence in financial strength between T1 (the
Premier League) and the other three tiers (the Football League) since the forma-
tion of the Premier League at the start of the 1993 season. For seasons 1994 to
1996, T1 accounted for 66.4 per cent of total Premier League and Football League
revenues. For seasons 2006 to 2008, the figure was 76.3 per cent. Even within the
Football League the story has been similar, with the revenues of T2 consistently
having grown at a faster rate than those of T3 and T4 combined. For seasons
1994 to 1996, T2 accounted for 60.1 per cent of the revenues of T2, T3 and T4
combined. For seasons 2006 to 2008, the figure was 65.5 per cent.
The financial strength of T1 is also reflected in the relativities between expendi-
tures on wages and salaries, and total revenues. During the 1990s the expenditures
of T1 clubs on players’ salaries grew at a faster rate than the clubs’ revenues. Salary
expenditure expressed as a percentage of total revenue increased from below 50
per cent in the mid-1990s to more than 60 per cent by the start of the 2000s. Due
to the Bosman ruling (see Section 6.6, below), the mid-1990s witnessed a major
shift in the balance of power in salary negotiations between the players and the
clubs, favouring the former at the expense of the latter. By the start of the 2000s,
however, the post-Bosman adjustment to players’ salary levels appears to have
been completed. During the 2000s the growth in the T1 clubs’ salary expenditure
proceeded at a similar rate to the growth in revenue.
The ratio of the Football League clubs’ salary expenditure to their revenues was
much higher than the corresponding ratio for the Premier League clubs through-
out the period covered by Table 6.5. The finances of the T2 clubs in particular
seem precarious, with this ratio having risen from just over 70 per cent for sea-
sons 1994 to 1996 to a peak of more than 86 per cent for seasons 2000 to 2002.
The collapse in 2002 of ITV Digital, a broadcaster that held the Football League
broadcast rights at the time (see Section 6.5, below), and the consequent unantici-
pated loss of broadcast revenue, placed many Football League clubs under severe
Table 6.5 Average revenue, wages and salaries and operating profit per season,
three-season periods, English League, 1994–2008 seasons
Revenue Wages and salaries Operating profit

T1 T2 T3/ T1 T2 T3/ T1 T2 T3/
T4 T4 T4
Average values per season, £m

1994–1996 303 92 61 142 66 42 47 −18 −9
1997–1999 572 159 99 305 118 80 84 −22 −29
2000–2002 947 229 138 582 198 113 73 −63 −35
2003–2005 1302 283 155 786 218 113 145 −52 −32
2006–2008 1614 328 173 1006 259 122 139 −77 −26
Percentage growth in revenue and wages and salaries over previous three-season period
1997–1999 88.6 72.6 61.4 115.1 78.5 89.6 — — —
2000–2002 65.6 43.9 39.7 91.0 67.0 42.3 — — —
2003–2005 37.5 23.3 11.7 35.0 10.3 −0.6 — — —
2006–2008 23.9 15.9 11.7 28.1 19.0 8.0 — — —
Wages and salaries and operating profit as percentage of revenue
1994–1996 — — — 46.7 71.8 68.5 15.6 −19.5 −14.1
1997–1999 — — — 53.3 74.3 80.5 14.6 −13.6 −29.6
2000–2002 — — — 61.5 86.2 81.9 7.7 −27.6 −25.3
2003–2005 — — — 60.3 77.1 72.9 11.1 −18.3 −20.5
2006–2008 — — — 62.4 79.1 70.5 8.6 −23.4 −15.3
Source: Deloitte
financial duress. Subsequently the ratio for the T2 clubs has dropped and stabi-
lised to a level just below 80 per cent: a figure that most analysts would still con-
sider dangerously high. For many T2 clubs, there is a temptation to gamble on the
possibility of achieving promotion to T1, and the financial resources that would
flow as a consequence. The gamble involves ratcheting up expenditures on players’
salaries, to levels that would be covered easily if promotion is achieved, but which
are unsustainable otherwise.
The contrast between the Premier League and the Football League is also
reflected, starkly, in the data for operating profits. The Premier League clubs com-
bined turned in positive operating profits in each of the three-season periods cov-
ered by Table 6.5; while the Football League clubs turned in operating losses. The
operating profits data do not include expenditures on player transfers (see Section
6.6, below). The Premier League clubs are heavy net spenders on transfers, while
transfer fees received continued to make an important contribution to the solvency
of many Football League clubs during the 2000s, as was the case in earlier decades.
The revenue and salary data aggregated by tier highlight the divergence in finan-
cial strength between the Premier League and the Football League, but they also
mask divergences within the top two tiers that carry significant implications for
Financial structure of English football 161
Table 6.6 Financial data, leading English clubs and averages by tier, 2008
season, £m
Operating
profit before
Total Salary player Pre-tax
revenue expenditure transfers profit
T1
Manchester United 257.1 121.1 71.8 −44.8
Chelsea 213.6 172.1 −30.9 −84.5
Arsenal 209.3 101.3 48.5 36.7
Liverpool 164.2 90.4 28.4 −40.9
16 other T1 clubs (average) 66.2 43.4 3.7 −7.8
T2
4 ‘parachute’ clubs (average) 27.4 22.1 −4.3 −1.5
20 other T2 clubs (average) 11.1 10.2 −4.4 −3.8
T3
24 clubs (average) 5.2 3.7 −1.0 −0.8
T4
24 clubs (average) 2.7 1.9 −0.3 −0.3
Source: Deloitte
competitive inequality. Based on data compiled by Deloitte (2009), Table 6.6 pro-
vides some further analysis of football clubs’ profit and loss accounts for the 2008
season. For T1, these data are tabulated separately for the four clubs with the lar-
gest total revenues: Arsenal, Chelsea, Liverpool and Manchester United; and for
the other sixteen clubs in the form of cross-sectional averages. The so-called ‘big
four’ are reported separately because the revenue arising from Champions League
participation renders their profit and loss accounts atypical of T1 as a whole.
For T2, separate averages are reported for four clubs that benefited under the
‘parachute’ system, and the other twenty clubs. The parachute system allocates a
share of the Premier League’s broadcast revenues to clubs recently relegated from
T1, during the first two seasons following relegation. Charlton Athletic, Sheffield
United, Watford and West Bromwich Albion were the four T2 recipients of para-
chute payments during the 2008 season, and their profit and loss accounts were
atypical of T2 as a whole. Table 6.7 provides some further detail on the contribu-
tions of UEFA (Champions League) and Premier League TV income to the total
revenues of T1 and T2 clubs.
The data reported in Tables 6.6 and 6.7 emphasise the inequality in financial struc-
ture that exists within the Premier League between the ‘big four’ and the other six-
teen clubs, and between the Premier League as a whole and the other three Football
League tiers. Participation in the Champions League is rewarded by a share of
UEFA’s TV income, worth £34 million to Manchester United, the tournament
Table 6.7 TV and other revenue, leading English clubs and averages by tier,
2008 season, £m
UEFA Premier Other Total

TV League TV revenue revenue
T1
Manchester United 34 50 173 257
Chelsea 29 46 139 214
Arsenal 18 47 144 209
Liverpool 21 45 98 164
16 other T1 clubs (average) — 34 32 66
T2
4 ‘parachute’ clubs (average) — 12 15 27
20 other T2 clubs (average) — — 11 11
Source: Deloitte
winners in 2008, and just over half that amount to Arsenal, the least successful
of the four English participants in 2008 (eliminated at the quarter-final stage).
Champions League participation also yields additional matchday income, and
enhances the value of sponsorship and advertising deals for the clubs involved.
Consequently, the average revenues of the ‘big four’ were in excess of £200 mil-
lion, while the average for the other sixteen Premier League clubs was around
£66 million.
For the average Premier League club, relegation to the Football League
would be expected to eliminate around one-third of this revenue immediately.
The parachute system (yielding a payment of around £12 million per eligible
club) helps cushion what would otherwise be an even larger drop in revenue, of
perhaps 50–60 per cent. Relegated clubs that fail to achieve a rapid return to
the Premier League must downsize to this extent in full over a two-year period.
Parachute payments create a degree of inequality within T2 that is similar to
the situation in T1 due to the Champions League involvement of the elite clubs.
A key difference, however, is that the year-on-year turnover in the T2 clubs
eligible to receive parachute payments is high, and any correspondence that
might exist between the additional revenue and success on the field of play
is variable and uncertain. In contrast, the same four T1 clubs participated in
the Champions League in every season between 2004 and 2009,4 a period dur-
ing which these four clubs also exerted a near-stranglehold upon success in
domestic football.
6.4 Gate revenues and admission prices

Until the 1990s, gate revenues from league matches represented the largest source
of revenue for all football clubs. It was not until the start of the 1990s that clubs’
Gate revenues and admission prices 163
revenue sources became sufficiently diversified that the share of league gate rev-
enues in aggregate revenue fell below 50 per cent for the league as a whole. Over
the longer term, however, gate revenues data provide an accurate representation of
trends in football’s overall revenue-raising capability and performance. An attrac-
tion of a gate revenues data set for researchers is that it can be combined with
match attendance data to calculate an average admission price per club per sea-
son, by dividing total gate revenue by total attendance. Tables 6.8 and 6.9 draw
on a data set comprising annual gate revenues from league matches for all English
league clubs, obtained from Football League archives and (for the later seasons)
from Football Trust, and reported previously by Dobson and Goddard (1998b,
2001). The 1926 season is the first and the 1999 season the last for which these
records were available.
Table 6.8 reports the average admission price per spectator in nominal and real
terms, and average nominal admission prices by tier, for each season between 1926
and 1999. For the period up to the mid-1970s, admission prices for spectators pay-
ing at the gate were subject to a statutory minimum determined by the Football
League, also shown in Table 6.8. At certain times, however, concessions were avail-
able to groups such as juniors, pensioners and the unemployed. Clubs were also
permitted to charge more than the minimum. Many did so for seated accommo-
dation (especially) and, increasingly as time went by, for standing accommodation
as well. The original objective of the League’s minimum admission price was to
prevent clubs from attempting to poach spectators from other clubs in the same
geographical catchment area by cutting prices. Table 6.9 reports aggregate gate
revenue data in nominal and real terms, together with gate revenue shares by tier.
Figure 6.3 shows the percentage shares of the G1 to G5 clubs (see Table 6.3) in
aggregate gate revenue.
Table 6.8 suggests that there was little difference between clubs in admission
prices during the inter-war period, with most clubs charging the league min-
imum of one shilling (£0.05) to all but a relatively small minority of more afflu-
ent, seated spectators. The gentle increase in the real admission price series up
to the mid-1930s and its subsequent reversal reflects the constancy of admission
prices in nominal terms, against a background of retail prices that were falling in
the late 1920s and early 1930s and rising thereafter. When football resumed after
the Second World War, the increase in admission prices was just sufficient to
compensate for wartime price inflation. The post-war attendance boom generated
revenues more than 50 per cent higher than before the war in real terms.
The relative homogeneity of football’s pricing structure was preserved through-
out the 1950s. In the 1960 season, T1 clubs were charging only 25 per cent more on
average than their T4 counterparts, a ratio not much greater than that which had
prevailed during the inter-war period. From the 1960s onwards, however, the story
is one of steadily increasing divergence between the prices of clubs in different
tiers. By the 1970 season, T1 clubs were charging 48 per cent more per spectator
on average than T4 clubs. This gap increased to 64 per cent in 1980, 70 per cent in
Table 6.8 English league average admission prices
Average admission price Average nominal admission price (£)

Minimum
Nominal Real admission
Season (£) (1926 = 100) T1 T2 T3 T4 price (£)
1926 0.06 100 0.06 0.06 0.06 0.05 0.05

1927 0.06 103 0.06 0.06 0.06 0.05 0.05
1928 0.06 106 0.06 0.06 0.06 0.05 0.05
1929 0.06 107 0.06 0.06 0.06 0.05 0.05
1930 0.06 112 0.06 0.06 0.06 0.05 0.05
1931 0.06 118 0.06 0.06 0.06 0.05 0.05
1932 0.06 119 0.06 0.06 0.06 0.05 0.05
1933 0.06 124 0.06 0.06 0.06 0.05 0.05
1934 0.06 123 0.06 0.06 0.06 0.05 0.05
1935 0.06 122 0.06 0.06 0.06 0.05 0.05
1936 0.06 118 0.06 0.06 0.06 0.05 0.05
1937 0.06 114 0.06 0.06 0.06 0.05 0.05
1938 0.06 113 0.06 0.06 0.06 0.05 0.05
1939 0.06 110 0.06 0.06 0.06 0.05 0.05
1947 0.08 120 0.09 0.08 0.08 0.07 0.0625
1948 0.09 118 0.09 0.09 0.08 0.08 0.0625
1949 0.09 120 0.10 0.10 0.09 0.08 0.0625
1950 0.09 120 0.10 0.10 0.09 0.08 0.0625
1951 0.10 112 0.10 0.10 0.09 0.08 0.0625
1952 0.11 112 0.11 0.11 0.10 0.09 0.075
1953 0.12 126 0.13 0.12 0.11 0.10 0.075
1954 0.12 124 0.13 0.12 0.12 0.10 0.075
1955 0.12 120 0.13 0.12 0.12 0.10 0.075
1956 0.13 122 0.14 0.13 0.12 0.11 0.10
1957 0.13 118 0.14 0.13 0.12 0.11 0.10
1958 0.13 115 0.14 0.13 0.12 0.11 0.10
1959 0.14 117 0.15 0.14 0.12 0.11 0.10
1960 0.14 120 0.15 0.14 0.13 0.12 0.10
1961 0.16 135 0.18 0.17 0.14 0.13 0.125
1962 0.18 141 0.20 0.18 0.16 0.13 0.125
1963 0.18 141 0.20 0.19 0.16 0.14 0.125
1964 0.19 141 0.21 0.19 0.16 0.15 0.125
1965 0.21 151 0.24 0.21 0.18 0.16 0.125
1966 0.23 160 0.26 0.22 0.20 0.18 0.20
1967 0.24 162 0.27 0.23 0.21 0.18 0.20
1968 0.26 164 0.29 0.24 0.22 0.20 0.20
1969 0.29 176 0.33 0.26 0.24 0.22 0.25
1970 0.30 172 0.34 0.28 0.24 0.23 0.25
1971 0.35 186 0.41 0.32 0.29 0.27 0.30
1972 0.38 185 0.43 0.35 0.31 0.28 0.30
1973 0.46 209 0.52 0.43 0.38 0.36 0.40
1974 0.53 204 0.59 0.49 0.43 0.39 0.40
1975 0.59 185 0.66 0.60 0.46 0.41 0.40
1976 0.76 202 0.83 0.75 0.61 0.59 0.65
Table 6.8 (cont.)
Average admission price Average nominal admission price (£)

Minimum
Nominal Real admission
Season (£) (1926 = 100) T1 T2 T3 T4 price (£)
1977 0.85 197 0.96 0.83 0.68 0.58 n/a
1978 1.05 224 1.19 0.99 0.82 0.75 n/a
1979 1.18 222 1.29 1.16 0.99 0.91 n/a
1980 1.50 239 1.67 1.53 1.21 1.02 n/a
1981 1.84 261 2.05 1.85 1.43 1.20 n/a
1982 2.03 265 2.32 1.93 1.59 1.35 n/a
1983 2.24 281 2.55 2.17 1.83 1.45 n/a
1984 2.44 291 2.76 2.42 1.91 1.63 n/a
1985 2.77 312 3.21 2.46 2.11 1.79 n/a
1986 2.97 323 3.42 2.63 2.33 2.05 n/a
1987 3.21 336 3.70 2.92 2.48 2.35 n/a
1988 3.56 354 4.26 3.22 2.83 2.51 n/a
1989 3.95 365 4.73 3.71 3.13 2.69 n/a
1990 4.48 378 5.39 4.21 3.42 3.17 n/a
1991 5.32 424 6.46 4.82 3.94 3.70 n/a
1992 6.25 481 7.55 5.47 4.45 4.28 n/a
1993 7.09 537 8.77 6.21 5.11 4.31 n/a
1994 7.55 557 9.17 6.64 5.32 4.49 n/a
1995 8.75 625 11.58 6.46 5.46 5.60 n/a
1996 9.64 672 12.74 7.62 5.97 5.37 n/a
1997 10.74 722 14.59 7.96 6.66 5.72 n/a
1998 11.37 749 14.99 9.30 7.51 5.85 n/a
1999 12.30 799 16.27 9.81 8.64 6.45 n/a
Note: Until the 1958 season, T3 is Division 3 (South) and T4 is Division 3 (North). 1999
season data are based on estimates for some clubs.
Source: Football Trust, Bird (1982), Football League, Premier League.
1990 and 152 per cent in 1999. Together with the eight or nine percentage point
increase in T1’s attendance share between the late 1950s and mid-1970s (see Table
6.4), these relative price increases explain an increase of about 18 per cent in T1’s
gate revenue share over the same period. Similarly, the growth of the revenue share
of the G1 clubs between the late 1950s and mid-1970s (Figure 6.3) was an even
faster rate than the growth in their attendance share (Figure 6.2).
The increases in nominal admission prices in all tiers kept pace with inflation
during the 1950s, but started to outstrip inflation in the 1960s, especially in the
higher tiers. Cost pressures emanating from the removal of the maximum wage
and changes affecting the terms of players’ contracts (see Section 6.6, below) were
partly responsible for the imposition of real increases in admission prices during
the 1960s. When retail price inflation began to soar and attendances continued
Table 6.9 English league gate revenues
Percentage share by tier in aggregate

Aggregate league gate revenues league gate revenues
Nominal Real
Season (£m) (1926 = 100) T1 T2 T3 T4
1926 1.352 100 46.9 26.8 16.3 10.0

1927 1.373 105 47.0 28.2 15.5 9.3
1928 1.410 109 47.2 28.9 14.7 9.2
1929 1.419 111 45.7 28.3 16.1 9.9
1930 1.372 111 47.8 27.5 16.5 8.2
1931 1.246 108 47.4 28.5 15.0 9.1
1932 1.263 112 48.1 25.8 17.6 8.5
1933 1.261 115 46.9 28.8 15.4 9.0
1934 1.321 120 48.9 26.2 16.0 8.9
1935 1.373 123 49.9 26.3 15.3 8.5
1936 1.456 127 48.5 28.2 15.3 7.9
1937 1.575 131 45.2 30.3 16.1 8.4
1938 1.661 136 44.5 30.5 15.6 9.4
1939 1.594 129 45.8 31.1 13.9 9.2
1947 2.933 185 45.0 30.9 14.7 9.4
1948 3.578 205 43.8 30.4 15.7 10.2
1949 3.780 214 45.5 28.2 15.9 10.5
1950 3.801 211 44.3 29.9 16.5 9.3
1951 3.786 192 44.2 28.4 17.3 10.1
1952 4.135 189 44.1 28.5 16.7 10.7
1953 4.541 202 46.4 26.2 16.8 10.6
1954 4.431 196 46.9 27.2 16.4 9.5
1955 4.215 177 47.4 26.5 16.5 9.6
1956 4.355 176 45.4 27.6 16.0 11.0
1957 4.311 167 45.3 26.8 16.1 11.9
1958 4.448 167 46.9 25.6 16.8 10.6
1959 4.557 171 47.2 25.7 16.4 10.7
1960 4.549 168 48.1 26.0 15.8 10.1
1961 4.677 168 49.6 24.8 14.7 10.9
1962 4.981 171 47.6 27.2 16.5 8.8
1963 5.242 176 47.2 26.3 17.7 8.9
1964 5.341 174 48.0 27.1 16.5 8.4
1965 5.829 181 52.4 24.7 13.5 9.5
1966 6.308 189 52.0 24.0 15.2 8.7
1967 6.931 203 55.0 23.8 13.3 7.9
1968 7.682 214 56.7 23.1 11.6 8.5
1969 8.388 222 57.1 22.9 12.2 7.8
1970 8.828 220 56.9 23.9 11.5 7.7
1971 9.960 227 56.9 22.7 12.8 7.6
1972 10.814 229 57.8 21.7 13.4 7.1
1973 11.823 230 61.4 20.4 11.9 6.3
1974 13.174 221 58.8 23.7 11.0 6.4
1975 15.180 205 52.6 29.8 12.4 5.2
1976 18.822 218 57.7 23.0 12.8 6.4
Table 6.9 (cont.)
Percentage share by tier in aggregate

Aggregate league gate revenues league gate revenues
Nominal Real
Season (£m) (1926 = 100) T1 T2 T3 T4
1977 22.220 222 58.6 23.3 12.6 5.5
1978 26.651 246 59.3 23.9 10.3 6.5
1979 28.960 236 56.6 24.6 11.6 7.3
1980 36.911 255 55.0 25.4 13.1 6.5
1981 40.239 248 58.2 23.8 12.9 5.0
1982 40.523 230 59.6 22.6 11.1 6.6
1983 42.096 229 56.3 25.6 12.8 5.3
1984 44.760 232 53.7 29.0 11.6 5.7
1985 49.276 240 63.5 20.1 11.4 5.0
1986 48.901 231 63.2 19.1 11.9 5.9
1987 55.844 253 60.6 21.8 10.4 7.2
1988 63.906 276 54.0 26.9 12.2 7.0
1989 72.885 292 50.7 29.7 13.0 6.6
1990 87.219 319 48.8 33.3 11.0 6.9
1991 103.691 359 53.7 29.2 10.8 6.3
1992 127.329 425 59.2 25.0 10.5 5.3
1993 146.238 480 58.3 24.9 12.2 4.5
1994 163.655 524 59.7 26.3 9.7 4.3
1995 191.610 593 64.4 22.3 9.3 4.0
1996 211.500 639 63.0 23.7 8.3 5.0
1997 244.838 714 64.4 22.6 8.7 4.3
1998 283.313 804 59.4 27.5 9.4 3.7
1999 312.056 879 60.6 23.6 11.5 4.3
Note: Until the 1958 season, T3 is Division 3 (South) and T4 is Division 3 (North). 1999
season data are based on estimates for some clubs.
Source: Football Trust, Football League, Premier League.
to fall, for a few years clubs became more circumspect about raising prices by
more than the rate of inflation. As inflation started to fall in the late 1970s and
early 1980s and attendances continued to plummet, the clubs attempted to protect
their dwindling real revenue base by imposing further price increases. But with
attendances falling even faster than admission prices were rising (in real terms),
football’s capacity to insulate itself from the effects of its declining spectator base
appeared to have reached its limit.
As the recovery in football’s popularity gathered pace during the late 1980s and
1990s, there were further large real increases in admission prices. The clubs argued
that higher prices were justified by improvements in the quality of the product,
with highly talented but expensive foreign players (in T1 at least) viewed from
comfortable seats in glistening new stands, built on top of the ruins of the crum-
bling rain-soaked terraces of yesteryear. Simple economics predicts that prices
50
45
Group 1
40
35
30
Per cent
25 Group 3
20
15 Group 2
10
Group 4
5
Group 5
0
1926 1929 1932 1935 1938 1948 1951 1954 1957 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999
Season
Figure 6.3 Percentage shares in aggregate gate revenue of clubs in Groups 1 to 5
should rise sharply once capacity constraints are reached, and this is precisely
what happened. Pricing behaviour of this kind explains why the G1 clubs’ revenue
share increased sharply during the 1990s (Figure 6.3), while their attendance share
barely changed (Figure 6.2).
The gate revenues series was not published during the 2000s. For T1, however,
Deloitte publish an estimated decomposition of total revenue into its matchday,
broadcast and commercial components. Matchday revenue includes gate revenue
from all matches, and other associated income (from matchday activities such as
catering and programme sales). For the 1999 season, Deloitte’s matchday revenue
figure for T1 is £248 million. Table 6.9 indicates that gate revenues from league
matches contributed £189 million towards the matchday total. Deloitte’s total
revenue estimate for the 1999 season is £670 million; therefore league gate revenue
accounted for 28.2 per cent of total revenue, and all matchday revenue accounted
for 38.0 per cent. For T2, T3 and T4 as a whole, the estimated total revenue for
the 1999 season is £281 million, and the percentage contributions to this total of
league gate revenue and total matchday revenue are 43.7 per cent and 48.4 per
cent, respectively.
Since the 1999 season, the percentage contribution of matchday revenue to total
revenue for T1 has declined further. Deloitte’s estimates for the 2008 season are
matchday revenue of £554 million (123 per cent growth since 1999), and total rev-
enue of £1,932 million (188 per cent growth since 1999). The share of matchday
revenue in total revenue is 28.7 per cent. Corresponding 2008 estimates are not
available for T2, T3 and T4; but with the lower three tiers having experienced far
slower growth in broadcast revenue than the top tier, it seems likely that the per-
centage contributions of matchday revenue in total revenue remained close to the
1999 figure throughout the 2000s.
In a survey carried out during the 1999 season, Clowes and Clements (2003)
find English T1 clubs using a broad range of ticket pricing strategies designed
to maximise the value extracted from spectators with differing degrees of will-
ingness to pay, enabling the clubs to generate more revenue than would be pos-
sible with a broad-brush uniform price structure. Several of these strategies are
described by the economic theory of price discrimination. In this context, sec-
ond-degree price discrimination involves charging each spectator a price that
depends upon the number of matches attended; and third-degree price discrim-
ination involves charging different prices to different groups of spectators.5
Revenue maximisation from the sale of tickets for seats in a stadium of fixed
capacity over a finite number of matches of varying attractiveness requires a
pricing structure that is sophisticated and finely-tuned to local conditions of
spectator demand.
Specific examples of second-degree price discrimination include season tickets
offering a discount on the price the spectator would pay to purchase tickets for
each match individually; and membership schemes requiring the payment of a
fixed membership fee, and allowing the member to purchase individual match
tickets at a discounted price (known in the literature as a two-part tariff). In the
US, some major league teams charge a fee that guarantees the right to purchase
a season ticket for a particular seat over the long term (twenty or thirty years).
Personal seat licences typically lapse if the holder dies or fails to renew the sea-
son ticket. Several attempts to introduce similar schemes by English football
clubs in the early 1990s were unpopular with spectators, and the idea failed to
take off.
Third-degree price discrimination is widely practised, by offering price conces-
sions to specific groups, such as juniors, pensioners, students, the unemployed,
people with disabilities, or families. Other common pricing or ticketing policies
include bundling, whereby tickets for two or more matches can be or must be
bought simultaneously; or proof of purchase of a ticket for one match is required
to purchase a ticket for another match. Typically, a sell-out match is bundled
together with one that is unlikely to sell out. Naturally most clubs charge dif-
ferent prices for seats located in different parts of the stadium; and some clubs
operate a price banding system, with different prices applicable for different
fixtures, dependent on the attractiveness of the opposition. Deloitte (2009) note
that spectator demand is sensitive to economic conditions, and at the height of
the recession of the late 2000s many T1 clubs announced ticket price freezes or
reductions in an effort to sustain attendances in the face of adverse economic
conditions.
Both cause and consequence of the widening financial gulf separating the lead-
ing clubs from the rest has been the erosion of arrangements for revenue sharing
within the league. Until the early 1980s, 20 per cent of the notional receipts from
every league match were paid to the visiting club. Notional receipts were based on
minimum admission charges, with certain deductions permitted (e.g. for costs of
policing and stewarding). Since the 1984 season, however, clubs have retained all
of the proceeds from their home fixtures (Dobson and Goddard, 1998b). Under
another revenue-sharing arrangement, 4 per cent of all notional receipts were paid
into a pool, the proceeds from which were distributed evenly among all ninety-two
league member clubs. The 4 per cent levy was reduced to 3 per cent from the 1987
season onwards, prior to the complete withdrawal from this scheme of the Premier
League clubs at the start of the 1993 season. The 3 per cent levy and arrangements
for equal distribution of the proceeds are still effective within the Football League.
The reduction in the reliance of clubs at all levels on revenues from league
matches is also attributable partly to an increase in the number of cup and other
tournaments. For the top clubs, the revenue contribution of participation in
European competition has grown significantly over time, and especially in recent
seasons. For the rest, the introduction in the 1987 season of the end-of-season
play-offs to determine one promotion place from each tier of the Football League,
culminating (from the 1990 season onwards) in a series of three showpiece finals,
played at the national stadia located at Wembley (London) or Cardiff, has added
a dramatic and lucrative postscript to the domestic season.
Elimination cup competitions provide an implicit cross-subsidy from the richer
clubs to the poorer in several ways:
• A lower-tier club may be able to attract an abnormally large attendance by
being drawn to play a top club;
• The rules for pooling gate receipts allow for a greater element of redistribution
than is the case for league fixtures;
• All clubs share in the proceeds of sponsorship agreements.
As with the explicit arrangements for sharing league gate revenues, however, the
willingness of the leading clubs to provide these implicit cross-subsidies by par-
ticipating wholeheartedly in the two main cup tournaments, the FA Cup and
the League Cup, has been strained during the modern era. The extension of the
European calendar in particular has brought genuine problems of fixture conges-
tion, which have been tackled in a number of ways: by delaying the entry of teams
involved in Europe to the League Cup; by substituting penalty shoot-outs for
replays as a means of settling drawn cup ties; by the clubs themselves choosing to
field teams comprising reserve team players for cup matches; and on one occasion,
by the withdrawal of Manchester United from participation in the 2000 season’s
FA Cup competition. In the eyes of traditionalists, these developments have deval-
ued the two domestic cup competitions, to the detriment of the league’s smaller
member clubs and, in the case of the FA Cup, the non-league clubs as well.
Broadcast revenues 171
6.5 Broadcast revenues

The early history of football on television
For the leading clubs in particular, paying spectators arriving through the turnstiles
on match days represent only one part of the target audience. Pay-TV audiences
form an increasingly important component of present-day football’s customer
base. Section 6.5 describes the evolution of English football’s long-standing and
intricate relationship with the broadcasting media.
The British Broadcasting Company was first established in 1922. According to
its first general manager John Reith, its mission was to bring into every home ‘all
that was best in every department of human knowledge, endeavour and achieve-
ment’ (Taylor, 1965). The company was replaced by the state-owned British
Broadcasting Corporation (BBC) in 1926, funded by means of a flat-rate licence
fee payable by all owners of wireless sets. The BBC’s first television transmis-
sions took place in the late 1930s, though it was not until the 1950s that sufficient
households owned sets for television to begin to play a significant part in national
life. The Independent Television (ITV) network of regional broadcasting com-
panies was created in 1955. Funding was by means of advertising revenues. A
second BBC channel was launched in 1964, to be followed by the introduction
of colour transmissions in the late 1960s. Further free-to-air terrestrial chan-
nels were added in 1982 (Channel 4) and 1996 (Channel 5), both of which are
funded by means of advertising. The two competing satellite broadcasters British
Satellite Broadcasting (BSB) and Sky introduced Britain’s first pay-TV services
in the late 1980s. Sky acquired BSB in 1991 to form British Sky Broadcasting
(BSkyB) under the ownership of Rupert Murdoch’s media conglomerate News
Corporation.
The BBC transmitted English football’s first radio commentary of a match
between Arsenal and Sheffield United in January 1927. Concern on the part of
the Football League, however, about the potentially damaging effects on match
attendances, led to a broadcasting ban being imposed in 1931. Regular radio
coverage of league fixtures did not resume until after the Second World War,
although cup matches and internationals were transmitted regularly during the
1930s. Televised transmission of football began on 9 April 1938 with the broad-
cast of an England–Scotland international match. Three weeks later the FA Cup
final between Sunderland and Preston North End was also televised. Highlights
of league matches were transmitted sporadically from 1955 onwards, but regu-
lar coverage did not start until the launch of the BBC’s recorded highlights pro-
gramme Match of the Day in 1964. This was followed by the launch of regional
highlights programmes on the ITV network in 1968. From then until the late 1970s,
as buyers of broadcasting rights the BBC and ITV appear to have operated an
informal cartel, which limited the fees paid to football by the TV companies. The
proceeds were distributed evenly among all league clubs, each receiving a meagre
£1,300 from the fee of £120,000 paid during the 1968 season. By the 1979 season
the total payment had increased modestly to £5,800 per club, or £534,000 in total
(Goldberg and Wragg, 1991).
The first attempted breach in the BBC–ITV stranglehold occurred in 1978,
when London Weekend Television broke ranks with the other ITV companies,
and attempted to negotiate exclusive rights to cover and distribute televised foot-
ball to the rest of the ITV network. An Office of Fair Trading ruling prevented the
deal from going ahead, but in the subsequently renegotiated contract, the amount
received by each club rose to £23,900 from an aggregate annual fee of £2.2 mil-
lion from the 1980 season onwards. Regular live coverage of league matches,
shared between BBC and ITV, took place for the first time in the 1984 season,
in a two-year deal which provided for coverage of ten matches per season at a
cost of £2.6 million per season. Live coverage was suspended temporarily during
the 1986 season, following the breakdown of negotiations for a new contract.
Football’s parlous financial condition at the time, combined with increasing ten-
sions between the larger and smaller clubs over the distribution of the proceeds,
probably allowed the TV companies to retain the upper hand in the negotiations.
A contract was eventually settled at the start of 1986, reinstating live coverage
during the remainder of the 1986 season and for the following two seasons. The
annual price was £3.1 million for fourteen matches, a smaller amount per match
than under the previous contract. The 1986 contract was also the first to breach
the principle of equal distribution of revenues between all ninety-two league mem-
ber clubs; instead, the T1 clubs received 50 per cent of the proceeds, T2 25 per cent
and T3 and T4 12.5 per cent each.
The arrival of the two satellite broadcasters BSB and Sky in the late 1980s sig-
nalled a significant shift in the relative bargaining power of the football industry
as the seller of broadcasting rights and the television companies as buyers. The
BBC–ITV duopsony was gone for good, and the competition between the TV
companies as purchasers of the rights intensified. In the 1988 round of negoti-
ations, ITV secured exclusive rights to show eighteen matches per season over
the next four seasons, from 1989 to 1992 inclusive. The annual fee was £11 mil-
lion. The balance of bargaining power within football was also changing rapidly.
Threats from the leading clubs to withdraw from the Football League and enter
into separate negotiations secured a settlement in which roughly three-quarters
of the ITV fee went to the T1 clubs. Of this, more than 40 per cent was shared
between the so-called ‘big five’, which were to become the subject of the major-
ity of ITV’s coverage: Arsenal, Everton, Liverpool, Manchester United and
Tottenham Hotspur.
Televised football during the era of pay-TV

As the ITV contract’s 1992 expiry date started to approach, it was clear that the
next round of negotiations would be shaped both by the ambitions of the newly
merged satellite operator BSkyB to secure a significantly greater share of the
coverage, and by the manoeuvrings of the leading clubs anxious to maximise their
own share of the proceeds. Strategically, BSkyB identified the acquisition of regu-
lar live coverage of league football as the key step that would generate sufficient
take-up of satellite services to enable the company to extend subscription charges
beyond films, to sports and other channels. Meanwhile in early 1991, the Football
Association’s Blueprint for the Future of Football announced that at the end of
the 1992 season, the twenty-two T1 clubs would withdraw completely from the
Football League to create a new Premier League, with its own governing body
under the auspices of the Football Association. The Premier League would sell its
own broadcasting rights, separately from any arrangements made by the Football
League on behalf of its seventy remaining members.
BSkyB succeeded narrowly in outbidding ITV for the contract providing
exclusive live broadcast rights during the Premier League’s first five seasons,
from 1993 to 1997 inclusive. The partnership with the BBC inherited by BSkyB
following the BSB–Sky merger was to continue, with the BBC screening regu-
lar Saturday evening highlights of league matches played earlier the same day.
The headline cost of the combined BSkyB–BBC Premier League coverage was
£304 million for the five-year contract; although a smaller sum of around £190
million was eventually paid by BSkyB. Subsequently BSkyB was the dominant
bidder in each of five further auctions of Premier League broadcast rights. The
second auction, covering the next four seasons from 1998 to 2001, yielded pro-
ceeds of £670 million. The third, for which the duration was reduced to three
seasons from 2002 to 2004 in an attempt to allay concerns expressed by the EU
competition authorities, yielded £1.1 billion. In the fourth auction covering sea-
sons 2005 to 2007, the phase of rapid growth in the value of the broadcast rights
was temporarily halted, with BSkyB paying £1.024 billion for the renewal.
For the fifth auction covering seasons 2008 to 2010, the Premier League and the
European Commission agreed a new set of rules requiring the sale of the rights to
be broken down into six packages of twenty-three matches each, with no broad-
caster permitted to purchase all six packages. The packages were of differing lev-
els of attractiveness to broadcasters, based on the timings of the slots (Sunday
afternoon matches attract higher TV audiences than Saturday and Monday eve-
nings), and the priority assigned to each broadcaster in selecting which match(es)
to transmit from each weekend’s schedule. The Irish pay-TV broadcaster Setanta
picked up two of the six packages for £392 million, with BSkyB collecting the
other four (including the prime-time Sunday afternoon slots) for £1.31 billion.
Two years into the three-year contract, however, with the uptake of subscriptions
having lagged persistently behind projections, Setanta defaulted on its payment
to the Premier League, and entered into administration in the UK in June 2009.
The rights to broadcast the forty-six remaining matches for the 2010 season were
acquired by ESPN, the American cable broadcaster owned by Disney, for around
£90 million. With Setanta reportedly having already made an advance payment
of £40 million on its final instalment of £130 million, the Premier League appears
Table 6.10 Revenue from sale of broadcast rights, English Premier League,
average per season, 1993–2010 seasons, £m
1993–1997 1998–2001 2002–2004 2005–2007 2008–2010
Live 38 155 367 341 569

Highlights 5 18 60 35 57
International 9 25 60 97 225
Other domestic — — 80 45 56
Total 52 198 567 518 907
Number of live matches 60 60 106 106 138
Fee per match 0.6 2.8 3.5 3.2 4.1
Source: Deloitte
not to have been embarrassed financially by Setanta’s collapse. Earlier in 2009,

the sixth auction covering seasons 2011 to 2013 resulted in the acquisition of five
of the six packages by BSkyB for £1.623 billion. ESPN assumed Setanta’s com-
mitment for the sixth package for an amount that was close to Setanta’s original
winning bid of £159 million.
Table 6.10 shows estimates of the total value of the Premier League broadcast
rights on a per-season basis over the period from the formation of the Premier
League (start of the 1993 season) to the expiry of the fifth contract (end of the
2010 season). The broadcast rights for live transmission of matches to the UK
accounted for more than 60 per cent of the total proceeds from each of the five
auctions. Other proceeds are from the sale of a highlights package in the UK; sales
of live or highlights packages overseas6; and (in the recent auctions) sales of con-
tent to be transmitted over broadband internet and 3G mobile phone networks, as
well as delayed transmissions by BSkyB of entire matches played earlier the same
day. The highlights package has been held by the BBC throughout the Premier
League era (to the end of the 2010 season) except for seasons 2002 to 2004, when
it switched to ITV. The latter’s £183 million winning bid was more than most
analysts believed to be the true value of the rights at the time, and higher than the
BBC’s subsequent winning bid of £105 million for seasons 2005 to 2007. The BBC
eventually paid a higher price, £172 million, to renew the highlights package for
seasons 2008 to 2010.
The highlights package is prized by BBC because of the importance of Match of
the Day in the corporation’s history, and (perhaps) its national cultural significance.
Amidst the Premier League broadcast rights frenzy, however, the position of the
free-to-air broadcasters BBC and ITV as weak competitors in a cut-throat market
has been cruelly exposed. Since the Premier League breakaway, the chequered his-
tory of the Football League broadcast rights, both live and highlights, follows a simi-
lar trajectory. Between the 1993 and 1996 seasons, ITV screened selected Football
League matches live on a regional basis. Between 1997 and 2000 these rights were
switched to BSkyB. The broadcast revenues of the Football League clubs increased
from around £16 million in 1996 to over £30 million per season from 1997.
Amidst a blaze of publicity, ITV recovered the live Football League rights by
bidding £315 million in the next auction, held in 2000. Live match transmissions
were accorded flagship status in the promotion of a terrestrial pay-TV digital ser-
vice ONDigital, co-owned by the two largest ITV companies Carlton and Granada.
The British public, however, were unenthused, and the company struggled to raise
subscriptions. Following an abortive rebranding as ITV Digital in July 2001, the
company was liquidated in April 2002, with only £135 million having been paid to
the Football League. BSkyB snapped up the rights for the next four seasons, 2003 to
2006, for a reported £95 million. Many Football League clubs endured a period of
financial hardship following the sudden disappearance of anticipated TV revenues.
Between 2007 and 2009 BSkyB and ITV shared the Football League rights, with the
former screening live matches and the latter highlights, for a combined fee of around
£110 million. BSkyB formed a similar partnership with the BBC for seasons 2010 to
2012, with the BBC screening ten live matches per season, and BSkyB showing sixty-
five live matches. The 2010 season thus became the first since 1996 in which English
league football was screened live and free-to-air in the UK. The BSkyB–BBC pack-
age, which also covered the League Cup, generated significantly enhanced proceeds
of £264 million for the Football League.
Implications for competition and welfare

The transactions between the football industry and the broadcast media that are
recounted in the previous subsection, and other similar arrangements operative
in other European countries, have drawn the attention of the regulatory bodies
responsible for competition policy and the courts, at both national and European
level (Cave and Crandall, 2001; Hoehn and Lancefield, 2003; Jeanrenaud and
Késenne, 2006; Parlasca, 2006). The following are the main issues of concern. Is
the migration of top-level sports broadcasting from free-to-air to pay-TV in the
public interest, and are there any grounds for intervention by the public author-
ities to inhibit or prevent this migration from taking place? Is the collective sale
of broadcast rights by sports governing bodies anticompetitive, and should clubs
be required to market and sell their own rights individually? Should broadcast
rights be sold on a basis that grants exclusivity to one broadcaster, or should
the rights be divided between more than one? Is the cross-ownership of football
clubs and media companies anticompetitive, and if so should cross-ownership
be prevented? Many of the arguments in legal cases concerning the application
of competition law to the market for sports broadcast rights are complex and
finely balanced; and as a consequence, a patchwork of arrangements for the
organisation of this market has evolved across Europe. In Italy from the 2000
season onwards, the clubs were required to sell their own rights individually
for pay-TV and overseas TV, while the league continued to sell the rights for
public TV on behalf of the clubs (Baroncelli and Largo, 2006). Individual sale
of broadcast rights also operated in Spain (Ascari and Gagnepain, 2006). In
Germany the courts ruled against collective selling, but were overruled when
parliament granted an exemption. In the UK the principle of collective selling
has been upheld by the courts.
The technical argument that the acquisition of broadcast rights by pay-TV
broadcasters might be contrary to the public interest rests heavily upon the fact
that televised broadcasts of sports events exhibit some, but not all, of the charac-
teristics of public goods. Public goods are non-rivalrous, meaning that one per-
son’s consumption does not diminish the amount of consumption that others are
able to enjoy. Sports broadcasts share this characteristic. Public goods are also
non-excludable, meaning that it is not feasible to exclude anyone from consum-
ing the good if it is supplied. Prior to the development of encryption technol-
ogy, sports broadcasts shared this characteristic; but in the era of pay-TV they
no longer do so. Encryption technology renders it feasible to charge for access to
sports broadcasts; but restricting some people’s access due to inability to pay still
creates a welfare loss, since those people could have been permitted to consume
without inhibiting anyone else’s ability to consume.
It has also been argued, particularly in respect of major sporting events, that
freely available televised sport generates positive externalities, through its cap-
acity to bolster social cohesion and inculcate values of tolerance, respect and team
work (Jeanrenaud and Késenne, 2006). If so, restricting some people’s access on
the basis of inability to pay might create a welfare loss. Accordingly, the UK
government has designated the Olympics, football’s World Cup and European
Championship finals, the English and Scottish FA Cup finals, the Grand National
and Derby (horse-racing), Wimbledon (tennis), the World Cup final (rugby union)
and Challenge Cup final (rugby league) as events that must be broadcast free-
to-air, known colloquially as the Crown Jewels. A further list specifies sports or
events for which highlights must be shown free-to-air. Other European countries
operate similar lists.
Counter-arguments in favour of pay-TV sports broadcasting assert that view-
ers do not avoid paying when watching on free-to-air channels; they simply pay
differently, through their willingness to tolerate exposure to advertising. Free-to-
air broadcasters do not receive the price signals that enable pay-TV broadcasters
to gauge viewers’ preferences accurately, leading to a possible misallocation of
resources. Pay-TV makes it feasible for minority sports, or less popular teams in
popular sports such as football, to receive coverage that would not attract suf-
ficient audiences for free-to-air broadcast. While there is public distaste for the
lavish remuneration of top footballers, fuelled as it is mainly by revenues extracted
directly from the bank accounts of pay-TV subscribers, the sport itself has also
benefited during an era of profligacy, through (for example) investment in sta-
dium redevelopment, and improvements in playing standards due to the ability of
English clubs to attract and reward many of the world’s top footballers.
Many of the issues discussed in the very early economics of sports literature,
reviewed in Chapter 1, resonate through contemporary debates over the issue
of the collective versus individual sale of broadcast rights. Are the clubs or the
league the relevant economic entity that should assume responsibility for the sale?
Proponents of collectivity argue that without the league, or without each other,
no club would have a viable product to sell by itself. Proponents of individuality
express concerns over the inhibition of competition if the rights are sold collect-
ively. The league might act as a cartel in order to raise the price by exploiting its
market power as a monopoly supplier. Proponents of collectivity counter that the
matches of each club individually are not close substitutes for each other: fans
of Manchester United wish to watch their team’s matches, and do not regard
those of Liverpool as an acceptable substitute. Therefore individual selling sim-
ply creates more monopoly sellers. This debate hinges on a question of market
definition: in order to determine whether a monopoly exists, and if so whether it
is being abused, the boundaries of the relevant market must first be defined. The
same issue is also key to the debate surrounding exclusivity (see below).
In favour of collective selling, a saving in transaction costs might be achieved if
the rights are sold by one entity, averting the need for broadcasters to enter into
costly and time-consuming negotiations with each club individually. The league
can develop uniform branding, and thereby present a stronger product, than can
the clubs individually. More contentiously, the collective sale of broadcast rights
favours a more equitable division of the proceeds than would be achievable if each
club sold its own rights individually; and an equitable division may create a posi-
tive externality for the clubs through a reduction in competitive inequality. This
latter argument is rejected by some economists, such as Szymanski (2009), who
argues that the question as to whether or how revenues should be shared is con-
ceptually separate from the question of how the rights that generate the same rev-
enues should be marketed. From a practical perspective, however, it seems likely
that an equitable division of revenues is easier to administer and defend if the sale
is conducted and the proceeds are collected centrally, than would be the case with
individual sales followed by a round of punitive taxation and redistribution.
The European Commission has objected to the practice of selling broadcast
rights on a basis that grants exclusivity to one broadcaster. The Commission’s
stance is reflected in the arrangements that have evolved for the sale of both
the Premier League broadcast rights, as described above, and the rights for the
Champions League. The latter must be divided between at least two broadcast-
ers in each country, one of which must be free-to-air. In England the Champions
League rights have been split between BSkyB and ITV since 2002.
In the absence of regulatory intervention, exclusivity might be attractive not
only to broadcasters, but also to the sports themselves. Exclusive rights might be
worth more to a winning bidder than non-exclusive rights to two or more bidders
combined, if competition between the latter to attract viewers or subscribers pre-
vents them from extracting the full economic surplus. Therefore the sport might
be able to realise higher proceeds if exclusivity is permitted. But exclusivity might

be harmful to consumers if, for example, it reduces the number of matches that
are televised. Under present arrangements only 138 of the 380 Premier League
matches played in each season are screened. However, this argument ignores
the practical difficulties in scheduling matches that would arise from any fur-
ther extension of live coverage. Alternatively, allowing transmissions on Saturday
afternoons while other matches are underway, would reduce attendances at
matches that are not televised. This would be detrimental to the quality of a
‘product’ which relies, at least in part, upon the atmosphere generated by the
crowd. In any event, most viewers have only so many hours’ leisure time available
to devote to televised sport, and many would feel that coverage already far exceeds
the saturation level.
When tested in the courts or subjected to the scrutiny of the competition author-
ities, judgements as to whether exclusivity arrangements are anticompetitive rest
heavily on questions of market definition. Should televised sport in general be
viewed as the relevant market? If so, the award of exclusive rights in one sport is
not anticompetitive, because broadcasters compete within a general market for
sports or other programming content. This interpretation tends to evoke scepti-
cism from football fans who would not regard snooker or volleyball as close or
acceptable substitutes. Should the broadcast rights for individual sports, or indi-
vidual competitions within sports, be seen as separate markets? This approach,
adopted by the European Commission, may be problematic if, as suggested above,
the contents of different packages are not perceived by viewers as close substi-
tutes. Under present arrangements, to receive the maximum possible coverage of
his or her team’s matches, the committed supporter needs to purchase subscrip-
tions with each broadcaster. Such reasoning suggests that markets within sports
subdivide to the level of individual teams, or even individual fixtures (Fikentscher,
2006; Jeanrenaud and Késenne, 2006).
The issue of vertical integration between football clubs and broadcasters came
to a head in England in 1998, when BSkyB launched a £623 million takeover bid
for Manchester United. The bid was referred to the Monopolies and Mergers
Commission (MMC), the predecessor of the Competition Commission. In 1999
the MMC delivered its findings, that the merger would damage competition
between broadcasters, and be detrimental to the wider interests of British football
(Monopolies and Mergers Commission, 1999). Accordingly the UK government
blocked the takeover. Specific grounds for rejection of the bid included concerns
that allowing the same company to operate simultaneously on both sides of the
market for the broadcast rights would damage competition between broadcasters.
Assurances that the company’s broadcasting arm would refrain from influencing
the football arm were considered unenforceable. BSkyB would have an unfair
advantage over rival bidders in future auctions, because a proportion of its bid
would return to the company via the football club. BSkyB’s influence over the
Football’s labour market 179
organisation of English football (for example, the scheduling of fixtures) might

be exploited to the detriment of Manchester United’s competitors. A successful
takeover might encourage retaliatory mergers involving other media groups and
major clubs, widening the financial gap between the latter and the less marketable
members of the Premier League and the Football League (Lee, 1999; Michie and
Oughton, 1999).
BSkyB also played a central role in a landmark case brought to the Restrictive
Practices Court (RPC) by the Office of Fair Trading (OFT) in 1999. The OFT
sought to demonstrate that the Premier League acted unlawfully as a cartel in
negotiating the sale of broadcasting rights on behalf of its member clubs. It was
argued that the Premier League’s decision to award exclusive live broadcasting
rights to BSkyB unfairly restricted the number of live matches screened on televi-
sion and inflated the price paid by subscribers. In a judgement published in July
1999, the RPC took the view that on balance these arrangements did not operate
against the public interest. The outcome of this case depended heavily upon issues
of market definition. Did exclusivity confer monopoly power upon BSkyB as
sole suppliers to a distinct market for a specific programme type, or did it merely
enable BSkyB to compete more effectively with other broadcasters within a wider
market for TV programming in general? The RPC was sympathetic to this latter
argument, suggesting that exclusivity could be helpful in promoting competition
through enhanced programme quality (Cave, 2000).
A more recent challenge to BSkyB’s supremacy emerged from a proposal from
Ofcom, the UK’s media regulator, to require BSkyB to sell content from its pre-
mium sports and movies channels wholesale to other pay-TV broadcasters at
regulated and significantly reduced prices. At the time of writing the cable TV
company Virgin Media is the only wholesale purchaser of BSkyB programming;
but the proposal could enable others, such as BT Vision (transmitted via broad-
band internet and digital terrestrial) to enter this market. Wholesale business is
estimated to account for only a small proportion, around 4 per cent, of BSkyB’s
total revenues. More significantly for consumers, however, the Ofcom proposal
raises the prospect that a price war might break out between pay-TV broadcasters.
If this proposal is implemented, rivals are expected to undercut BSkyB’s prices
for premium content; and BSkyB might be forced to follow suit in order to deter
subscribers from defecting to rival platforms.
6.6 Football’s labour market: players’ salaries and the transfer system

The retain-and-transfer system and the maximum wage were key features of
English football’s labour market from the very earliest days of professionalism.
Both were designed to prevent the clubs with the most resources from acquiring
all of the most talented players simply by outbidding other clubs for their services;
the outcome, it was argued, that might occur if the football players’ labour market
operated without any restrictions. Both survived in their original form until the
maximum wage was abolished in 1961, and the retain-and-transfer system was
substantially overhauled in 1963.
The retain-and-transfer system originated in a Football Association (FA) regu-
lation introduced in 1885, requiring clubs to register their players annually with
the FA. Player registrations immediately became tradeable commodities between
clubs, since unregistered players were not permitted to appear (Morrow, 1999).
All player contracts were renewable annually at the club’s discretion, and clubs
were entitled to retain a player’s registration even if his contract was not being
renewed. In theory and sometimes in practice, this enabled a club to prevent an
out-of-contract player from earning a living. A player could only move if his pre-
sent club was prepared to sell or release his registration. Under this system all
clubs had the discretion to retain individual players for as long as they wished,
as well as the guarantee of financial compensation when players were eventually
permitted to depart.
The incentive for players to seek to move was also restricted by the existence
of the maximum wage, originally fixed at £4 per week when it was introduced in
1901. FA rules introduced simultaneously to outlaw bonus payments were regu-
larly flouted during the next decade, resulting in many players receiving more
than the maximum and a number of clubs being punished by the FA (Russell,
1997). By 1922 the official maximum had risen to £8 per week (£6 during the sum-
mer); meanwhile, the average weekly earnings of male employees in engineering in
1924 were £2.65. By the time the maximum had risen to £20 per week (£17 during
the summer) in 1958, the differential had narrowed further: average weekly earn-
ings for male manual employees in 1958 were £12.83 (Department of Labour and
Productivity, 1971).
By restraining both the growth in players’ salaries, and divergences between the
remuneration of players at the top and bottom of the league, the maximum wage
helped maintain a degree of uniformity in the financial structure of professional
football at all levels. Between the early 1920s and the late 1950s the proportionate
increases in the minimum and average admission price, and the maximum wage,
were similar. Clubs appear to have responded to fluctuations in attendances and
revenues primarily by adjusting the number of players employed on professional
terms. For example, the total number of professional players in England and
Wales increased from 5,000 in 1939 to 7,000 by the end of the 1940s, at the height
of the post-war attendance boom (Fishwick, 1989). Roughly 4,000 of these were
registered with league clubs. By 1961 the number of professionals registered with
league clubs had fallen to just over 3,000 (Sloane, 1969). Circumstantial evidence
also suggests that a uniform financial structure helped restrain growth in competi-
tive inequality (see Section 6.1).
Towards the end of its lifetime, the main economic beneficiaries of the max-
imum wage were the paying spectators. As late as the 1960 season, the average
admission price was only 23 per cent higher in real terms than in 1926, despite the
much larger rise in real disposable incomes that had taken place in the meantime.
It is no coincidence that the pressure for abolition mounted progressively as the
post-war economic recovery gathered pace, and as the magnitude of the specta-
tors’ implicit consumer surplus also increased. As soon as the maximum wage
was lifted and players’ salaries started to escalate, especially at the top end of
the league, admission prices also started to rise. Much of the surplus then began
to shift away from the spectators towards the players (Dobson and Goddard,
1998b).
With both proven and suspected cases of illegal payments to players on the
increase, and with the football authorities under pressure from a campaign organ-
ised by the Professional Footballers Association (the players’ union), the max-
imum wage was eventually abolished in 1961. In 1963, a High Court ruling in
the case of the player George Eastham versus his employer Newcastle United,
adjudged the retain-and-transfer system to be an unreasonable restraint on trade.
From then on, the club holding the player’s registration had to offer a new con-
tract at least as remunerative and of the same duration as the expired contract
(which could be for one or two years) in order to retain his registration. If such
a contract was not forthcoming, the player became a free agent. Transfers, how-
ever, were still at the discretion of the selling club, which retained the ultimate
power to frustrate a player’s desire for a move, so long as it was prepared to con-
tinue to pay at the same level as under the old contract.
As anticipated, the immediate consequence of these changes was salary infla-
tion. Famously, the weekly salary of Fulham and England captain Johnny Haynes
was increased to £100: a rise of more than 400 per cent relative to the old max-
imum. For the large majority of professionals, the immediate effect on salaries
was less dramatic. According to Russell (1997) several leading clubs, including
Liverpool and Manchester United, attempted to maintain their own unofficial
maximum during the 1960s. Many clubs also responded to the increase in costs by
reducing the number of professionals employed: the total fell to 2,400 in the 1967
season (Sloane, 1969).
Nevertheless, over a period of years the cumulative effect on football’s finances
of the decision to lift the lid on players’ salaries was profound. According to data
reported by Szymanski and Kuypers (1999), the inflation-adjusted increase in sal-
aries during the 1950s for clubs with complete records was less than 10 per cent.
Between 1961 and 1974 the corresponding inflation-adjusted increase for the same
clubs was around 90 per cent. Revenues had fallen by about 5 per cent in real terms
during the 1950s. Between 1961 and 1974 the real increase was over 30 per cent.
Between 1961 and 1974 growth in salaries appears to have outstripped growth in
revenues among both the larger and the smaller clubs; but the growth in both was
much faster for the larger clubs than for the smaller. The adjustment of football’s
financial structure to the removal of the maximum wage appears to have been
virtually complete by the mid-1970s. Between 1974 and the late 1980s, the rates
of increase in revenues and salaries were similar. Having crept from below 40 per
cent in the 1950s to above 50 per cent by the mid-1970s, the average ratio of total
salaries to total revenues remained stable thereafter. Naturally, however, the aver-
age data tend to mask considerable intra-club and year-on-year variation.
A further key change to the transfer system was introduced in 1978, when play-
ers were awarded the right to decide themselves whether to move on the expiry of
their contracts. Though the new system was termed freedom-of-contract, in prac-
tice the out-of-contract player was still constrained to some extent. If his existing
club offered a new contract at least as rewarding as the expired contract, it could
still demand a compensation fee, either to be agreed between the two clubs, or (in
cases where agreement could not be reached) to be decided by an FA tribunal.
Arbitration was binding on both parties. This meant that in order to move, an
out-of-contract player had to find a new club that would either agree a fee with his
existing club, or be willing to accept the risks of arbitration.
The 1978 ‘freedom-of-contract’ system survived intact in England and Wales
until the landmark 1995 European Court of Justice ruling in the Jean-Marc
Bosman case, which prompted further movement towards the complete accept-
ance of the principle of freedom of movement for out-of-contract players. The
requirement for the payment of compensation to the former club of an out-of-
contract player signing for a new club was found to be incompatible with provi-
sions in Article 48 of the Treaty of Rome for the freedom of movement of labour.
The existing ‘three-foreigner’ rule, which had the effect of limiting the number of
individuals from countries outside the jurisdiction of each national football fed-
eration allowed onto the field of play at any one time, also contravened Article
48, by restricting the opportunity for European Union (EU) nationals to play for
clubs located in other EU countries.
The Bosman ruling concerning out-of-contract players applied originally only
to player transfers that crossed EU national boundaries. But many European
governing bodies quickly brought into line the regulations governing transfers
between clubs within their national jurisdictions. Since the end of the 1998 season,
all transfers of out-of-contract players over the age of 24 between English clubs
have been uncompensated. The retention of the fee element for out-of-contract
players below the age of 24 recognises that in a completely free player labour mar-
ket, clubs that invest in player development might be unable to realise the proceeds
of their investment due to poaching. The post-1998 English arrangements attempt
to preserve some element of compensation for clubs that discover or develop tal-
ented young players. The European Court of Justice was unconvinced that the
entire pre-Bosman transfer system had been justified by the ‘compensation’ argu-
ment; instead, it took the view that transfer fees often bore little or no relation
to the efforts or costs incurred in developing players (Simmons, 1997; Morrow,
1999).
Following the Bosman ruling, protracted discussions led eventually to an agree-
ment between the EU and the European and international football governing
bodies UEFA and FIFA, which took effect from 2003. The new system involved
compensation for clubs for the training and development of young players moving
before the age of 24; the creation of a ‘transfer window’ before the start of each
season, and a further limited mid-season window, with transfers not permitted
outside these windows; players permitted to move only once per season; a min-
imum contract duration of one year and a maximum of five years; and contracts
only enforced for a period of three years up to the age of 28 and two years after
that age, but with financial compensation payable if a contract is breached outside
these limits by either player or club (Goddard, Sloane and Wilson, 2010).
Fifteen years after the original Bosman ruling, it is clear that the players’ labour
market has changed fundamentally, and in ways that were, for the most part, pre-
dictable at the time. First, the removal of all restrictions on the eligibility of players
from EU countries, as well as non-EU players of international standing, has led to
a huge influx of foreign players now being employed by English clubs, especially
in the Premier League (see Chapter 8, Section 8.1). Many have been attracted by
remuneration perhaps much higher than they would expect to attain elsewhere.
While the post-Bosman influx was spectacular, the trend was already underway
before 1995. The success achieved by Eric Cantona, first at Leeds United and then
at Manchester United, following his transfer from Nimes in 1991, seems to have
been influential in raising awareness within the traditionally insular British game,
of the potential contribution of gifted overseas players.
Second, since the price of acquiring an out-of-contract player is now paid
entirely to the player (rather than shared between the player and his former club),
salary inflation is a predictable consequence of Bosman. Players seeking a new
club are in a stronger position to bargain for high remuneration if their acqui-
sition is not encumbered with the obligation to pay a transfer fee. As Table 6.5
shows, salary inflation in English football was exceptional during the immediate
post-Bosman period of the mid to late 1990s, outstripping even what was also
explosive and historically unprecedented revenue growth. Bosman clearly assisted
many players, especially at the highest level, to capture a much larger share of
the burgeoning financial rewards. For Germany, Feess, Frick and Muehlheusser
(2004) report evidence that the higher the remaining length of a player’s contract,
the higher the transfer fee if the player is sold, and the lower the player’s salary
at his new club. These effects were similar, however, in both the pre- and post-
Bosman periods. Tervio (2006) suggests that restricting the length of enforceable
contracts to three years (or less for older players) would further increase salaries
for all types of player, with the salaries of the most talented increasing the most.
Third, many clubs attempted to protect themselves against the possibility of
losing their best players without compensation by offering longer-term contracts,
especially to their younger stars. Morrow (1999) suggests that this development
may also have contributed to salary inflation in the short term, if a higher ini-
tial level of remuneration is needed to persuade a player to commit for a longer
period. Longer contracts may also be attractive to players, however, since they
insure future earnings against the consequences of serious injury or loss of
form. They may help restrain salary inflation over the longer term, if they are
enforced over their full duration. There are nevertheless incentives for clubs to
offer enhanced terms to their top stars far in advance of the expiry of their pre-
sent contracts, to avoid involvement in an eventual bidding contest for the play-
ers’ signatures. The ‘insurance’ argument also provides incentives for risk-averse
players to accept improved terms. A contracted player who declines an improved
offer received within the lifetime of his existing contract cannot be certain his
value to potential bidders will be as high when the contract eventually expires,
since all athletes are constantly exposed to the hazards of loss of form or serious
injury. Feess and Muehlheusser (2003a; 2003b) distinguish between pre-Bosman,
Bosman and post-Bosman contracts, and find that each type leads to similar levels
of player effort, investment incentive and payoffs, provided the payoffs to players
can be adjusted by varying the contract length. Feess, Frick and Muehlheusser
(2004) report that in Germany, the average contract length increased by about
six months after the Bosman ruling, from around two-and-a-half years to three
years.
A final predicted consequence of Bosman, widely debated at the time, was
its potential to cut off what was an important source of finance for clubs in the
lower tiers: proceeds from the sale of their most talented players to higher-tier
clubs. At best, the erosion of the implicit cross-subsidy generated by the domes-
tic transfer market might widen competitive inequaliy between rich and poor; at
worst, the loss of transfer fee earnings might threaten the survival of many of the
small-market clubs. Ericson (2000) argues that free agency might reduce the qual-
ity of play because the marginal cost of talent increases for small-market clubs,
which no longer receive adequate compensation for producing talent. While the
European Court of Justice accepted that it was legitimate for the football indus-
try to use various means of cross-subsidy to reduce competitive inequality, the
Court felt that the transfer system as it had operated pre-Bosman was neither
a necessary nor an effective means for achieving this objective (Morrow, 1999).
According to Antonioni and Cubbin (2000), the majority of transfers before 1995
involved players who were under contract at the time. Such transfers were not
directly affected by the Bosman ruling, which applied to transfers involving out-
of-contract players.
In fact, there is no clear evidence that Bosman has impacted significantly upon
the flows of transfer expenditure between the Premier League and Football League
clubs. For selected three-season periods between the 1970s and 2000s, Table 6.11
reports data on the net expenditures of T1, T2, T3 and T4 clubs in transfer deals
in which both the selling and buying clubs were English league members. During
both the immediate post-Bosman period (1996–8), and long after the dust from
Bosman had settled (2006–8), there were substantial net expenditure flows from
the Premier League to the Football League. These flows appear to be compar-
able in magnitude (relative to either gate revenue or total revenue) to the net flows
from T1 to T2, T3 and T4 combined that are reported in Table 6.11 during the
Table 6.11 Average net transfer expenditure within the Premier League/Football
League per season, selected three-season periods, £m
T1 T2 T3 T4
1976–1978 1.0 −0.6 −0.3 −0.1

1986–1988 3.6 −1.9 −1.1 −0.6
1996–1998 13.7 −6.9 −3.8 −3.0
2006–2008 47.3 −38.8 −6.1 −2.4
Source: Deloitte
Table 6.12 Average flows of transfer expenditure within the Premier League/
Football League per season, 2006–2008 seasons, £m
To → T1 T2 T3 T4 All tiers
From ↓
T1 115.0 64.1 2.9 0.5 182.5

T2 18.6 20.6 6.3 1.4 46.9
T3 1.6 0.7 1.1 1.0 4.4
T4 0.0 0.3 0.2 0.4 0.9
All tiers 135.2 85.7 10.5 3.3 234.7
Source: Deloitte
pre-Bosman period (1976–8, and 1986–8). For the 2006–8 period, Table 6.12 pro-
vides more detail on the gross flows of transfer expenditure between tiers on a
pairwise basis.
The data reported in Tables 6.11 and 6.12 do not include transfers in which
one of the parties was a Scottish, Irish or overseas club. Table 6.13 reports gross
expenditure on both domestic and foreign transfers for the Premier League and
the Football League for individual seasons from 2001 to 2008. The English trans-
fer market entered a temporary recession following the collapse of ITV Digital in
2002. Although the impact was felt most severely in the Football League, there
was a knock-on effect for the Premier League as well. Revenue growth in the lat-
ter slowed sharply in the mid-2000s, for reasons not unconnected with the ITV
Digital affair. As shown in Table 6.8, the total value of the Premier League broad-
cast rights for seasons 2005 to 2007 was less than in seasons 2002 to 2004. In this
context it appears that transfer expenditure can act as a useful cushion, which can
be reduced faster and more easily than other items of expenditure during periods
of unanticipated austerity.
Table 6.13 Gross transfer expenditure of Premier League and Football League
clubs per season, 2001–2008 seasons, £m
T1 T2, T3, T4 All tiers
2001 364 59 423

2002 323 84 407
2003 187 16 203
2004 386 28 414
2005 340 28 368
2006 435 48 483
2007 492 86 578
2008 664 115 779
Source: Deloitte
Table 6.14 Average gross transfer expenditure and average revenue per season,
three-season periods, 1994–2008 seasons, £m
Gross transfer Total Gross transfer expenditure as

expenditure revenues percentage of revenues
1994–1996 151 456 33.2

1997–1999 282 830 34.0
2000–2002 390 1314 29.7
2003–2005 328 1740 18.9
2006–2008 613 2115 29.0
Source: Deloitte
Table 6.14 identifies the longer-term trend in gross expenditure on both domes-
tic and foreign transfers for the Premier League and football clubs as a whole, in
the form of averages per season over three-season periods. The aftermath of the
ITV Digital collapse is also reflected in these data, with the ratio of gross transfer
expenditure to total revenues for seasons 2003 to 2005 dipping temporarily below
20 per cent. Over the longer term, however, this ratio has been relatively consist-
ent: just above 30 per cent during the mid and late 1990s, and just below 30 per
cent during the early and late 2000s.
6.7 Ownership, governance and finance

When the FA was founded in 1863, football operated on an amateur basis. However,
the competitive nature of the FA Cup, and the potential to generate income from
selling match tickets, created pressures for players to receive payments for playing.
Ownership, governance and finance 187
Professionalism was sanctioned in 1885; and a new competition, the Football

League, was established in 1888. Though the FA was unable to prevent the move
to professionalism, it did succeed in defining club ownership as essentially a not-
for-profit activity. English football’s non-profit ethos was underpinned by FA reg-
ulations limiting the amounts that could be remitted to shareholders and directors
in the form of dividends or salaries; requiring existing chairmen to approve any
major transfer of shares, eliminating the possibility of hostile takeover; and stipu-
lating that a club’s assets would revert to the FA in the event that it ceased playing
football, preventing any club from being taken over and closed in order to realise
a quick profit by selling its stadium (Conn, 1998; Horrie, 2002). Although business
interests would on occasions manifest themselves in the management of English
clubs, the rules were largely successful in ‘protecting’ the clubs from profit-taking
and asset-strippers (Vamplew, 1988).
At the origins of professionalism, football clubs were voluntary organisations,
administered by committees elected by voting members. The development of pro-
fessionalism led to further growth in gate revenues, and in the costs of player
compensation and stadium construction. This in turn created pressures for clubs
to acquire limited liability status. According to Russell (1997), Small Heath (later,
Birmingham City) was probably the first club to become a private company in
1888. By 1921, all but two of the league’s eighty-six member clubs had followed
suit. From then until recently, professional football’s traditional methods of com-
mercial and financial management seemed virtually impervious to pressures for
modernisation emanating from a multitude of social and economic pressures.
For most of the twentieth century, the ownership structure and the financing
and administration of the average football club retained many features originally
acquired during the late Victorian era.
Traditionally, the major shareholders and the majority of directors of football
clubs were individuals drawn from the local business community. Ownership and
control were often handed down through the generations, from father to son. Most
clubs were small, privately owned companies, which tended to be undercapitalised
with little finance raised either from issues of new equity (because owners were
unwilling to dilute their personal control) or from retained profit (because usually
there was no profit). In many cases, however, club owners or directors provided
additional long-term finance through personal loans, which for practical purposes
were similar to equity finance. Bank loans and overdrafts were the other princi-
pal source of (mainly short-term) finance. Although share ownership sometimes
extended beyond the well-heeled middle classes, working-class shareholdings were
usually too small to confer influence over the running of the club, and seem to
have existed primarily as a token of loyalty (Russell, 1997).
The ‘true’ motives of those wealthy individuals who injected large amounts of
personal finance into the local football club, often in return for little more than
a seat on the board of directors, have been debated by academic economists and
sociologists. The economist’s usual profit-maximising assumption seems untenable
given the woeful financial performance of most football clubs, and because direct-
ors were, for many years, prohibited by FA regulations from receiving dividends,
until this rule lapsed in the 1980s (Conn, 1997). No doubt some individuals have
attempted to exploit their influence in football for ends such as political advance-
ment, or to win contracts for their own businesses. In many cases, however, per-
sonal enthusiasm provides sufficient motivation, and success in business provides
the means, to acquire an ownership stake and intervene in the running of the local
football club.
Eventually, historical restrictions on the exploitation of football’s commer-
cial potential were eased or lifted. The ban on directors’ remuneration was
revoked in 1981. At the time, North London property developer Irving Scholar
was representative of a new generation of hard-headed, commercially minded
club chairmen. During the early 1980s Scholar initiated a low-key campaign
to buy out small Tottenham Hotspur shareholders. Legend has it that Scholar
signalled his assumption of control by turning up at a directors’ meeting to
announce himself as the new chairman, much to the astonishment of other
board members. He set about refinancing the debt-ridden club by floating a new
holding company, Tottenham Hotspur PLC on the London Stock Exchange.
The holding company would circumvent FA regulations on dividend payments
by owning the football club as a subsidiary, together with a number of other
leisure, clothing and catering businesses. The flotation in October 1983 raised
£3.3 million: a sum that appeared considerably larger at the time than it may
seem when judged by present-day standards, and sufficient to eliminate the
debts inherited from the previous ownership. Indifferent results in several of
the other diversified businesses over the next few years, however, meant that
the flotation was considered to have been somewhat less than an unqualified
success.
Perhaps because the holding company’s financial performance did not offer
strong encouragement to others, Tottenham remained the only club with a Stock
Exchange listing for several years. Other clubs eventually followed suit, however.
Millwall raised £4.8 million in a flotation in October 1989, and Manchester United
£6.7 million in June 1991. Again, the early experiences of the quoted compan-
ies were discouraging. Within two years, Millwall shares were trading at around
one-tenth of their original value. The team’s relegation from T1 in May 1990,
followed by a further decline in its fortunes within T2, led investors to scale back
drastically their assessment of the company’s value, before trading in the shares
was eventually suspended. Manchester United shares were more than 50 per cent
undersubscribed at the time of flotation in June 1991. Meanwhile, other clubs
were experimenting with alternative methods of raising external finance, such as
bond schemes guaranteeing the right to purchase a season ticket for a specific seat
in perpetuity. Such schemes were not popular with fans, however, and were soon
dropped.
Despite the indifferent results of these early attempts to expand and diversify
English football’s capital base, by the early 1990s events and sentiment were
starting to move in a direction more favourable to the sport’s commercial devel-
opment. Decisive action on the part of the sport’s governing bodies, clubs and
the government over issues such as safety and hooliganism had raised hopes
that the stadium disasters and other problems of the 1980s could finally be lain
to rest. Following a perceptible change of mood, watching football was once
again perceived as a fashionable leisure activity. As the Premier League got
underway, a surge of optimistic sentiment concerning future prospects led to
massive appreciation in the share prices of the two T1 clubs that had already
floated. Between December 1994 and December 1996, Tottenham Hotspur and
Manchester United shares increased in value by around 300 per cent and 400 per
cent, respectively.
Meanwhile, the costs of stadium redevelopment and spiralling player salaries
and transfer fees were creating urgent pressures for clubs to tap new sources of
finance. Circumstances were favourable for the spate of fifteen further flotations
on the London Stock Exchange Official List and Alternative Investment Market
that took place between September 1995 and October 1997. By far the largest
flotation, worth over £50 million, was that of Newcastle United in April 1997.
Deloitte (2009) estimate that Premier League clubs raised approximately £175
million in total through stock market flotations. The combined capitalisation of
the nineteen listed English clubs was £651 million in June 2002. Including clubs
listed on the more loosely regulated Ofex market (subsequently renamed Plus
Market), twenty-three English clubs (and four Scottish clubs) experimented with
some form of stock market listing.
During the 2000s, however, many of the clubs that were listed at the start
of the decade eventually took the decision to de-list. By the end of 2009, only
Preston North End (listed on the London Stock Exchange), Millwall, Tottenham
Hotspur, Watford (Alternative Investment Market) and Arsenal (Plus Market)
retained their listed status. Several clubs de-listed under severe financial dur-
ess: for example, Queens Park Rangers and Leicester City did so upon entering
administration in 2001 and 2002, respectively. Several others de-listed following
the acquisition of their equity by a single purchaser: prominent examples include
Aston Villa, Chelsea, Manchester United and Newcastle United.
With hindsight, it is clear that major inconsistencies existed between the gov-
ernance of the typical English football club, and the model of corporate govern-
ance under public ownership to which listed companies are required to adhere.
Critical comment on the governance of Newcastle United figured prominently
in the financial press throughout the decade of the club’s existence as a listed
company, but many of the issues raised were relevant to listed clubs in general.
Prior to its acquisition by Mike Ashley in June 2007 and subsequent stock mar-
ket de-listing, Newcastle was routinely criticised for the dominance on its board
of representatives of the families of the two major shareholders, Sir John Hall
and Freddy Shepherd. The pre-2007 board was characterised by a lack of inde-
pendence, and a tendency for family interests to ride roughshod over the interests
of smaller shareholders. Directors’ salaries and bonuses were disproportionate to
the financial performance of the company, and were not reviewed by an inde-
pendent remuneration committee. The tendency to signal to supporters an immi-
nent change of team manager through none-too-subtle hints dropped via the local
and national media, apparently standard issue in the media relations toolkit of
many football club chairmen and chief executives, sat uncomfortably with stock
market regulations requiring immediate disclosure of information concerning the
turnover of senior company officials. For financially motivated shareholders, of
course, the bottom line was profitability. Once again, the capability of most listed
football clubs to satisfy stock market expectations was found to be deficient, espe-
cially following a few years’ investor experience of football’s fickle competitive
and financial fortunes.
The issue of vertical integration between media companies and football clubs
first attracted headline attention in Britain in September 1998, when BSkyB’s
takeover bid for Manchester United was announced (see Section 6.5). The MMC
ruling and the UK government’s subsequent decision to block the Manchester
United takeover also torpedoed an agreement for the cable pay-TV company
NTL to take a controlling interest in Newcastle United. Although the direct own-
ership of major English football clubs by media companies has since been ruled
out, several media companies acquired minority stakes in more than one club,
subject to an FA regulation restricting parties with multiple interests to a max-
imum 10 per cent shareholding in each club. Accordingly, BSkyB acquired minor-
ity shareholdings in Manchester United, Leeds United, Chelsea, Sunderland and
Manchester City; Granada Media invested in Liverpool; and NTL in Newcastle
United, Middlesbrough, Aston Villa and Leicester City.
Several clubs entered into arrangements with media companies to broadcast
programmes over broadband internet, which subsequently turned out to be of little
or no commercial value. For example, ITV built up a 9.9 per cent ownership stake
in Arsenal with sucessive investments in 2000 and 2005, and a 50 per cent stake
in Arsenal Broadband. These interests were eventually sold to the US billionaire
investor Stan Kroenke in 2008. By this time, ITV had also divested itself of simi-
lar shareholdings in Liverpoolfc.tv (Liverpool) and MUTV (Manchester United).
Deloitte (2009) estimate that strategic media investment contributed around £300
million to the finances of Premier League clubs during the late 1990s and early
2000s; but by the end of the 2000s the model of direct strategic investment in foot-
ball clubs did not figure prominently on the agendas of media companies.
At the height of football’s stock market flirtation, several clubs experimented
with new mechanisms for raising non-equity finance from institutional inves-
tors (Gerrard, 2006). Securitisation involves the capitalisation of future antici-
pated streams of ticket or other commercial revenue. Bonds are issued, giving
the purchasers a prior claim on the designated revenue stream for interest and
capital repayments. The assets against which the bonds are secured remain in
the ownership of the originating company (the football club). Newcastle United
was the first English club to raise finance using this method, in December 1999.
£55 million was raised by means of securitisation of the club’s gate and hospi-
tality income over the next seventeen years. This capital sum was used primarily
to finance redevelopment of the club’s St James Park stadium, with the seating
capacity raised from 36,000 to 55,000. Other clubs that undertook similar transac-
tions between 2000 and 2003 were Southampton (£25 million), Leicester City (£28
million), Ipswich Town (£25 million), Leeds United (£60 million), Everton (£30
million), Manchester City (£44 million), Tottenham Hotspur (£75 million) and
Norwich City (£15 million). In several of these cases, the securitisation proceeds
were employed to finance stadium redevelopment or the construction of new sta-
dia; but in others the proceeds appear to have been used to finance player acquisi-
tions or general working capital.
Securitisation creates a heightened risk of insolvency, in the event that the
securitised revenue stream drops below the level that is required to service the
debt. While many football clubs’ future income streams are perceived to be rather
stable and predictable, thanks to the loyalties of fans, some of the securitisation
deals seemed analagous to mortgaging the house in order to fund lavish current
consumption. Furthermore, few football clubs are entitled to consider themselves
as realistically exempt from the dangers of relegation. As seen above, relegation
from the Premier League entails a large drop in revenue for even the most keenly
supported club, due to the loss of broadcast revenue. Among the clubs listed
above, Leicester City entered administration in 2002, and Ipswich Town followed
in 2003, both having been recently relegated from the Premier League. Following
Leeds United’s failure to qualify for the Champions League in 2002, the club
entered a rapid and spectacular spiral of decline that included relegation from
the Premier League in 2004, and a default on the capital and interest payments
on the £60 million bond issue in the same year. Bondholders reportedly incurred
substantial losses in the subsequent settlement (Gerrard, 2006).
Against this background, it is unsurprising that only one further football club
securitisation took place between 2004 and 2009. The £260 million capitalisation
of the future revenues of Arsenal’s whole business, undertaken in 2006 to finance
the construction of the club’s Emirates stadium, is something of an outlier in
terms of both timing and magnitude. It was claimed that even in the event that
Arsenal failed to maintain its status as a regular Champions League qualifier, the
debt was structured in a manner that would allow the servicing of the debt to
continue. Over the longer term, however, the need to divert a substantial portion
of the club’s revenue towards interest and capital repayments, rather than expend-
iture on players, seems to increase the likelihood that this theory may eventually
be put to the test.
Player sale-and-leaseback was a further form of financial innovation that was
popular in English football in the late 1990s and early 2000s (Gerrard, 2006).6 For
the duration of his contract with his current club, a player’s registration is an asset
that might be used by the club as security for a bank loan, with the lender enjoying
recourse to other assets in the event of default on the loan or a downgrade in the
value of the player’s registration due to loss of form or injury. Typically, the lender
took out credit insurance against the risk of default. Most arrangements of this
kind were effected in order to spread the upfront cost of a player’s transfer over
the full duration of the contract. Deloitte (2009) estimate that finance amount-
ing to around £150 million was raised between 1999 and 2002. With several such
arrangements also ending in default, however, player sale-and-leaseback appears
to have fallen out of favour with investors after the early 2000s.
In some ways, the acquisition of Chelsea by the Russian billionaire Roman
Abramovich in June 2003 heralded a return to a more traditional model of football
club ownership and finance, albeit on a far more extravagent scale than had ever
been witnessed previously, in English football or elsewhere. According to Deloitte
(2009), Abramovich’s total investment in Chelsea to the end of the 2008 season
totalled £760 million in the form of non-interest bearing ‘soft’ loans. Accordingly,
Chelsea’s balance sheet is heavily laden with debt; but since there are no interest
or capital payments, the club’s expenditure on players’ salaries and transfers is
constrained only by the extent of its benefactor’s extraordinary largesse. Chelsea’s
salary expenditure for the 2008 season was £172 million, 42 per cent higher than
that of the next-highest spender, Manchester United on £121 million. Chelsea’s
pre-tax loss of £85 million was easily covered by a £123 million injection of new
investment during the same season.
At the time of writing, Sheik Mansour bin Zayed Al Nahyan of Abu Dhabi, the
billionaire purchaser of Manchester City in September 2008, appears intent on
emulating and perhaps even usurping Roman Abramovich’s position as English
football’s most prolific benefactor. The Chelsea/Manchester City ownership model
faces a future challenge, however, in the form of the Financial Fair Play proposals
being developed by UEFA at the time of writing, and planned for implementation
from the 2013 season onwards. Aimed at preventing clubs with wealthy backers
from attempting to ‘buy success’ by spending beyond their means, the proposal
seeks to force clubs operating above a certain expenditure threshold to limit their
expenditure on salaries and transfers so as to break even over time. Critics have
argued that by entrenching differentials in market size this proposal would, in
common with the G14 payroll cap discussed in Chapter 2, Section 2.5, have the
effect of increasing (rather than reducing) competitive inequality. Several dec-
ades’ experience of attempts to impose restraints on salary expenditure in North
American major league sports suggest significant monitoring and enforcement
challenges, as well as employment opportunities for lawyers, may lie ahead if this
proposal is implemented.
Manchester United, meanwhile, was the subject of a leveraged buyout in 2005
by the American entrepreneur Malcolm Glazer. The £810 million cost of purchas-
ing the company’s equity, which relieved the previous shareholders of the property
rights to the club’s future anticipated revenue or profit streams, was part-financed
through loans from banks and hedge funds, some of which were secured against
the club’s assets. In contrast to the situation at Chelsea and Manchester City,
the Glazer family’s acquisition of Manchester United appears to have been com-
mercially motivated. During the remainder of the 2000s, increased ticket prices
contributed to sustained growth in revenue, accompanied by (relative) restraint
in salary and transfer expenditure. For the 2008 season the interest payment on
the debt was around £69 million, sufficient to wipe out a large proportion of the
club’s operating profit.
The Glazers claimed they could manage Manchester United’s cash flows in a
manner that would meet their obligations to lenders, and provide sufficient financial
resources to sustain a successful team. The revenue flows upon which this model
depends, however, are sensitive to team performance. If the flow of trophies were to
dry up, it seems that a highly leveraged financial model requiring the generation of
large revenue flows in order to service the club’s heavy debt, could unravel. Liverpool,
reputedly Manchester United’s fiercest rivals on the field of play, have also oper-
ated under American ownership since 2007, in the form of George Gillett and Tom
Hicks. Like the Glazer family, Gillett and Hicks both hold ownership stakes in
North American major league sports franchises. As in United’s case, their £435 mil-
lion acquisition of Liverpool was highly leveraged, leaving the club burdened with
debts that incurred interest charges of £35 million during the 2008 season.
Many of the smaller-market Premier League and Football League clubs are
financed along more traditional lines, through some combination of ‘soft’ loans
from benefactors, and bank loans and overdrafts. As the gradient of the relation-
ship between each team’s competitive position within the ninety-two-club league
hierarchy and the scale of its finances (on both the revenue and cost sides) has
become steeper, it is unsurprising that the number of clubs mismanaging their
finances in a manner culminating in insolvency proceedings, was higher during
the 2000s than ever before. During the 2010 season Portsmouth became the first
Premier League member club to enter administration; although several others had
previously experienced this fate soon after relegation. Deloitte (2009) report that
seventeen Football League clubs underwent insolvency procedings between 1992
and 1999, and a further thirty-five between 2000 and 2009. These figures include a
few clubs that became insolvent more than once. No fewer than sixteen Football
League club insolvencies were recorded in 2002 and 2003, following the demise of
ITV Digital. From the start of the 2005 season, the Football League introduced
an automatic penalty, in the form of a deduction of ten league points, for any club
entering administration.7 The aim was to deter clubs from gaining an unfair com-
petitive advantage by overspending on players, and then escaping the full financial
consequences by declaring insolvency. Points deductions have proven onerous on
the majority of occasions they have been applied: seven of thirteen clubs punished
in this manner between the 2005 and 2009 seasons (inclusive) were relegated at the
end of the season in which league points were deducted.
A radically different ownership model, which has proved successful in some

cases at the lower end of the financial scale, is the supporters’ trust. The UK gov-
ernment’s Supporters Direct initiative, launched in 2000, provides legal and prac-
tical support to supporters seeking to form a trust as a vehicle for involvement in
the ownership and management of a football club. Supporters’ trusts are coopera-
tives, which operate on a one-member one-vote basis. A trust might seek to gain
influence by purchasing shares, or by securing representation on the club’s board.
During the 2000s trusts were formed at more than 150 clubs at all levels in Britian,
with 120,000 members in total, and £20 million raised and invested.
Prominent examples include AFC Wimbledon, FC United of Manchester,
Ebbsfleet United and AFC Telford United. AFC Wimbledon was founded at the
base of football’s pyramid by supporters who refused to accept the relocation in
2004 of the former-Premier League club Wimbledon to its new home in Milton
Keynes. At the end of the 2009 season, the newly formed club achieved promo-
tion to the Conference, the tier immediately below T4 of the Football League.
FC United of Manchester, the club formed by Manchester United supporters
affronted by the Glazers’ takeover, regularly attracted 2,000 supporters to home
matches played at a level three tiers below T4 during the 2010 season. Another
Conference member-club of the late 2000s, Ebbsfleet United (formerly known as
Gravesend and Northfleet United) was acquired for £700,000 in 2007 by the web-
site MyFootballClub. The acqusition was funded through member subscriptions.
Payment of the subscription fee entitled each member to vote on matters such
as team selection and player transfers. Telford United, former members of the
Conference, were one of the few financially stricken clubs to go into liquidation.
In 2004 supporters set up their own club, AFC Telford United, to take its place.
Not all supporters’ trusts have been successful, however. Notts County sup-
porters’ trust relinquished its majority stake in the club in the summer of 2009.
Stockport County, supporter-owned since 2005, overspent on either side of its
2008 promotion to T2, and entered administration in April 2009. Supporters’
trusts at Brentford, Chesterfield and York City lacked either the wherewithal
or the financial resources needed to enable the clubs to flourish, and eventually
relinquished ownership to private investors. At the time of writing, Exeter City is
the only Football League club that is wholly owned by its supporters.
Conclusion
Chapter 6 has presented a historical and current analysis of English club foot-
ball’s commercial, financial and economic structure. From the origins of pro-
fessionalism in the late nineteenth century until the early 1960s, considerable
uniformity in professional football’s financial structure at all levels was preserved
through regulation of both admission prices (which determined the main com-
ponent of revenue) and players’ salaries (the principal item of cost). By the end
of the 1950s, however, as the post-war recovery gathered momentum and general
living standards rose dramatically, the pressure for relaxation of these regula-
tions was mounting. The abolition of the maximum wage in 1961, and the reform
of the retain-and-transfer system in 1963, heralded a period of rising financial
inequality between clubs operating at different levels within the league hierarchy,
as football made the transition towards a new and radically different economic
structure.
During the 1970s and most of the 1980s, English club football’s economic
fortunes were in steep decline on a number of fronts. Pressures emanating from
social and demographic changes, the hooligan phenomenon and the failure of
football clubs themselves to maintain their own physical infrastructure, had taken
their toll on attendances over many years, to the extent that the future survival of
large numbers of clubs (both large- and small-market) seemed to be in jeopardy.
Tragically, however, it was only after the stadium disasters of Heysel, Bradford
and Hillsborough in the mid to late 1980s that the momentum for fundamental
reform became irresistible.
Football’s rehabilitation during the 1990s and 2000s as the most popular and
fashionable national sport in England and elsewhere, has coincided with a fur-
ther sharp rise in financial and competitive inequality between English clubs. This
increase has taken place against a background of rising rather than falling specta-
tor demand at all levels, which has done much to offset the most adverse financial
consequences for the league’s smaller members. The wholesale reconstruction and
conversion of all major stadia to all-seated status, together with the marginalisa-
tion of the hooliganism phenomenon, have strengthened football’s appeal as a
middle-class spectator sport. Mainly as a consequence of technological change in
broadcasting, football’s dual status as both spectator sport and television spec-
tacle has become more realistically reflected in the balance between the revenues it
derives from both sources. By showering unprecedented riches upon the Premier
League, but relatively meagre returns for the Football League, the dispersion of
broadcast income has opened up a financial chasm between the leading clubs and
the rest. Meanwhile the market for TV broadcast rights continues to attract the
critical scrutiny of the competition authorities.
The Bosman ruling heralded the arrival of free agency in the European football-
ers’ labour market. An influx of talented overseas players is generally perceived to
have led to marked improvements in playing standards in England, especially in
the Premier League. However, the liberalisation of the rules governing football’s
labour market has greatly widened disparities between the earnings capability of
the top players and the rest. Many football clubs have become financially unstable,
drawn into competitive bidding wars in order to attract better players they realis-
tically cannot afford. Two decades of experimentation with alternative models of
ownership, finance and governance began in the early 1980s, following the removal
of several longstanding restrictions on the personal remuneration and activities of
English football club owners. At the time of writing it appears, in some quarters
at least, that a reversion to a more traditional model of ownership and control is
underway, albeit with foreign billionaires rather than local merchants, builders
and factory-owners assuming the status of English football’s latest and most pro-
lific financial benefactors.
Notes
1 There were fewer opportunities to win a European trophy after the cancellation of the Cup
Winners’ Cup in 1999. However, the English clubs’ haul of three trophies during the 2000s
looks especially meagre in comparison with the 1970s (nine trophies) and the 1980s (seven
up to and including the 1985 season, after which English teams were banned from European
club competition for the rest of the decade).
2 Post-war league entrants from the south were Colchester United (admitted in the 1951
season), Peterborough United (1961), Oxford United (1963), Cambridge United (1971),
Wimbledon (1978), Maidstone United (1990), Barnet (1992), Wycombe Wanderers (1994),
Cheltenham Town (2000), Yeovil Town (2004) and Dagenham and Redbridge (2008).
Entrants from the North and Midlands were Scunthorpe United and Shrewsbury Town
(both in the 1951 season), Hereford United (1973), Wigan Athletic (1979), Scarborough
(1988), Macclesfield Town (1998), Kidderminster Harriers (2001), Rushden and Diamonds
(2002), Boston United (2003) and Morecambe (2008). These lists do not include several clubs
that exited and re-entered.
3 For comparative data based on surveys carried out in Britain, France, Spain, Austria and
Belgium, see Waddington and Malcolm (1998).
4 In the 2006 season a fifth English club, Everton, also participated in the Champions
League.
5 First-degree price discrimination, which is rarely encountered in practice, would involve
charging a price for each ticket that would depend upon both the identity of the spectator
and the total number of tickets purchased.
6 Solberg and Taylor (2010) examine the lucrative and rapidly-growing market in exported
broadcast rights to foreign countries.
7 Technically these deals were not true sale-and-leaseback arrangements, because under
Football Association regulations the club was required to retain the ownership of the player’s
registration.
8 The Premier League, in which fewer matches are played per club per season, imposes a
nine-point penalty.
7 Determinants of professional
footballers’ salaries
Introduction
Since the early 1960s there has been a series of major institutional reforms to the
organisation of the players’ labour market in English football, starting with the
abolition of the maximum wage in 1961, and culminating in the 1995 European
Court of Justice ruling in the Jean-Marc Bosman case. Some of the broader con-
sequences of these changes are obvious and widely recognised. Spiralling salaries,
especially for superstar players, are a consequence of the progressive shift towards
freedom-of-contract that has been underway throughout this period. Although
the chronology and detail of institutional reform varies between countries, the
same long-term trend has been evident worldwide.
This chapter examines explanations for the exceptionally high salaries earned
by the leading stars in modern-day professional football. Above-inflation increases
in players’ salaries, especially at the highest level, have been a permanent feature
of English football since the abolition of the maximum wage in the early 1960s.
Section 7.1 reports some recent data on footballers’ compensation, and considers
to what extent the standard textbook microeconomic theory of wage or salary
determination in labour markets is capable of explaining the patterns that are
observed. Section 7.2 argues that scarcity in the supply of the highest talent can
only form part of the explanation. Before the introduction of pay-TV, football
reached large TV audiences but could not appropriate the full economic value of
the service provided to individual audience members. Since the late 1980s, how-
ever, this situation has been transformed. The changing composition of effective
demand for football has played a central role in creating the explosion in both
revenue and salaries in football (and several other sports) in recent years.
Salary inflation at the aggregate level has been accompanied by rapidly increas-
ing differentials between the earnings of individual football players, even within
the same club. Section 7.3 reviews a model of intra-team earnings distributions.
To an economist, it seems reasonable to assume that football players, like other
economic agents, respond to incentives. If players’ compensation is related to
their performance, players will exert more effort and invest in developing their
197
198 Determinants of professional footballers’ salaries
skills. The payment of extremely high salaries to certain superstar players may be
a rational strategy for clubs to adopt, if it provides an incentive structure which
maximises effort and investment in skills development by all players, especially in
the early stages of their careers. According to the rank-order tournament model,
the margin of performance between players determines their rank in a hierarchy
of remuneration, but it does not determine the margin of compensation. Players’
salaries at any point in time therefore do not necessarily reflect directly their indi-
vidual contribution to the club’s revenue-earning capability at the same time.
Empirical evidence on levels of football players’ compensation is rather thin on
the ground, due to restrictions on disclosure. In several North American sports,
in contrast, there is sufficient disclosure to permit empirical investigation of
the determinants of salaries at the micro level. Section 7.4 reviews some recent
European evidence on the determinants of footballers’ salaries, and provides a
brief treatment of the North American literature.
7.1 The compensation of professional footballers

Both the levels and the rapid growth of earnings of the top professional football
players are issues of occasional public concern or irritation. According to esti-
mates reported by Deloitte (2009), for the 2008 season the total wages and salaries
bill of the twenty English Premier League clubs was £1.196 billion: a figure that
represents an increase of 290 per cent on the 1998 season figure of £305 million.
In the 2008 season, three clubs (Chelsea, Manchester United and Arsenal) each
reported a total payroll in excess of £100 million.
Table 7.1 reports data from the most authoritative survey of individual Premier
League and Football League footballers’ salaries, compiled by The Independent
newspaper in 2000 and 2006. For the 2006 season, the average reported Premier
League footballer’s salary was £676,000. Of course, this average figure masks
large differences between youth- and reserve-team players on the one hand
and the top star players on the other. The average reported earnings in The
Independent survey do not include performance-related bonuses, which typically
increase a player’s earnings by between 60 and 100 per cent. Neither do they
include the top players’ earnings outside football, from sources such as advertising
and sponsorship.
Various unofficial listings of the world’s highest-earning footballers, taking
account of all sources of income, can readily be found on the Internet. According
to one such list, Lionel Messi (Barcelona and Argentina) was the world’s most
highly remunerated footballer in 2009, with total reported earnings of £29.5 million.
David Beckham (LA Galaxy and England, £27.2 million) and Christiano Ronaldo
(Real Madrid and Portugal, £26.8 million) were placed second and third. In sev-
enth and ninth places respectively, Carlos Tevez (Manchester City and Argentina,
£13.7 million) and Frank Lampard (Chelsea and England, £12.6 million) were the
two highest-placed players employed at the time by English Premier League clubs.
The compensation of professional footballers 199
Table 7.1 Average basic footballer’s salary by tier, English League, 2000 and 2006
seasons, £
T1 T2 T3 T4
2000 season
All players 409,000 128,000 54,600 38,800
2006 season
All players 676,000 195,750 67,850 49,600
Age:
17–18 24,500 22,500 — —
19–20 95,000 43,700 18,950 16,000
21–22 139,000 79,000 52,000 32,350
23–24 582,500 79,200 61,650 43,650
25–26 653,000 136,000 67,600 46,300
27–28 899,500 261,850 71,750 47,300
29–30 806,000 247,000 69,300 50,500
31–32 586,000 247,000 87,000 45,800
33+ 660,500 195,700 72,000 49,600
Position:
Goalkeeper 533,000 179,500 53,500 45,900
Defender 653,000 167,000 61,000 44,400
Midfield 754,000 185,950 79,000 46,800
Forward 806,000 292,900 75,000 67,900
Source: The Independent
Public concern about football players’ earnings tends to have two dimen-
sions: one relating to the implications for the competitive and financial structure
of football itself; and one questioning the appropriateness, or even the morality,
of such disproportionately high rewards accruing to participants in what is seen
as an essentially frivolous occupation. The first of these areas is considered else-
where in this volume (see, especially, Chapters 2 and 6). The second is the prin-
cipal subject of this section. Even allowing for the fact that career durations at
the highest level in professional sports are often brief, is it appropriate that the
leading professional sports stars achieve earnings so obviously disproportionate
to those of other occupational groups, including doctors, nurses, teachers and
police officers? To most observers, of course, the contributions of these latter
groups to society’s well-being appear much more important than those of even
the most popular sports personalities.
Without necessarily attempting to resolve the moral debate to everyone’s satis-
faction, economists can perhaps make a significant contribution to the discussion,
by proposing explanations for observed patterns of professional sports stars’ earn-
ings that are grounded in axioms of rational, optimising behaviour on the part of
players and the clubs that employ them. Two of the most influential theoretical
contributions, written by economists during the early 1980s (Lazear and Rosen,
1981; Rosen, 1981), are reviewed in the following two sections.
One way to approach the subject of sports stars’ salary determination is simply
to consider the market for football players in the same way as any other labour
market, by analysing the interaction between the supply and demand for players.
Market mechanisms allocate players to clubs by matching bids and offers for the
players’ services. If clubs seek to maximise profits, the highest salary a club is pre-
pared to offer a player is the amount he would add to the club’s revenue if he were
signed: his Marginal Revenue Product (MRP). If the player is paid more than his
MRP, the club’s profit is lower than it is if it does not sign the player. If the club
can get away with paying the player less than his MRP, the club’s profit is higher
than it is if it does not sign the player.
The question as to whether a player is paid his MRP depends partly on the insti-
tutional characteristics of the players’ labour market. Under a reserve clause (see
Chapter 1), players do not generally receive salaries equal to their MRP. Because
the player can only negotiate with the club that owns his contract, effectively there is
only one bidder for the player’s services, and the market structure is monopsonistic.
The player’s reservation wage is the next highest salary he can achieve in alternative
employment outside the sport. Provided the salary on offer is equal to or higher than
the reservation wage, the theoretical models predict that the player will accept this
salary offer. Salaries tend to be held down, therefore, to the extent that the players’
compensation may only capture a small proportion of the clubs’ total revenues.
In contrast, under free agency each player is expected to play for the club for
which he has the highest MRP; in other words, the club to which he is most valu-
able. His salary should lie between his MRP with that club and the next highest
MRP that would be attainable if he were to sign for another club. The club to
which the player is most valuable can outbid other clubs for his services, and still
increase its profit by signing him. The theoretical models predict that under free
agency, each player should capture at least his MRP with the club that would place
the second-highest value upon him out of all clubs in the league. Salaries tend to
be bid up to the point where most of the clubs’ revenues are absorbed by players’
compensation (Quirk and Fort, 1992).
One possible objection to this type of analysis when applied to football clubs
arises from its reliance on the profit-maximisation assumption. As seen in Chapter
1, Sloane (1971) argues that it is more appropriate to model the club’s object-
ives using a utility-maximising framework. Football clubs seek to maximise util-
ity, subject to a financial solvency constraint. Arguments in the utility function
include playing success, attendance and profit. In this case, even under a reserve
clause, clubs that attach high value to playing success relative to profit may choose
to sign players for salaries higher than their MRP, in an attempt to increase play-
ing success even though it is not profitable to do so.
Nevertheless, departures on the part of some clubs from profit maximisa-
tion do not seem likely to account for the highly skewed patterns observed in
The economics of superstars 201
the distributions of sports stars’ earnings. If the earnings of star players are to
be explained using marginal productivity theory within a traditional supply-and-
demand framework, it therefore seems necessary to confront head-on the question
of whether the MRPs of leading football players could conceivably be so high as
to justify annual salaries of several million pounds.
7.2 The economics of superstars

Rosen (1981) provides an influential explanation as to how not only sports stars,
but also superstars in other fields such as acting, publishing and music can indeed
generate extremely high MRPs when they reach the pinnacle of their respective
professions. Although Rosen’s analysis is presented in terms of professionals
selling their services directly to the public, many of the insights are relevant at a
general level in the case of professional team sports, in which individual sellers
contract with one another (or with a club) to form a team before their services are
sold collectively to the public. It is important to note at the outset, however, that
when the superstars model is applied to team sports, a further set of complexities
that arise in explaining intra-team earnings distributions are not addressed. This
theme will be pursued in the next section.
Rosen’s objective is to explain not only the highly skewed distribution of earn-
ings within certain professions, but also the unusually high levels of seller concen-
tration that appear to coexist with skewed earnings. In other words it is important
to explain not only the high levels of earnings, but also why employment in the
fields concerned often tends to be restricted to relatively small numbers of indi-
viduals. According to figures reported by Deloitte (2009), for the 2008 season
the annual turnover of English Premier League and Football League clubs was
slightly less than £2.5 billion; but only slightly more than 3,000 professionals were
employed in the clubs’ core activity of playing football. Rosen’s account of these
phenomena rests on two key aspects of consumer tastes and production technol-
ogy, as follows:
• Imperfect substitution between sellers of certain services. The very best talent
is in scarce supply. For consumers lesser talent is not perfectly substitutable for
greater talent, even if there is compensation in terms of quantity. For example,
Manchester United might be three times more talented than Birmingham City,
but watching Birmingham City on three occasions is not equivalent to watching
Manchester United once.
• Scale economies in joint consumption of certain services. The technology of pro-
duction enables very large audiences to be serviced at the same time. The effort
and therefore the production costs incurred by Manchester United’s players are
much the same if they are playing in a half-empty stadium, or in front of a
capacity crowd, or additionally before a large TV audience. The technology,
in other words, has characteristics similar to that employed to supply certain
public goods like street lighting and national defence: production costs depend
only slightly, or not at all, on the number of people consuming the service. But in
contrast to these examples, in football there are no free-rider problems prevent-
ing the seller from appropriating a charge for providing the service. Would-be
audience members who are unwilling to pay (by purchasing a match ticket or a
pay-TV subscription) can easily be excluded from consumption.
The implications of these attributes of tastes and technology for the distribu-
tion of earnings and employment opportunities depend on the precise nature of
the relationship between production costs and audience size. Rosen suggests two
reasons why marginal costs of production may eventually start to increase as the
audience size increases. First, internal diseconomies include all the usual reasons
why marginal costs rise as output increases. In professional football, marginal
costs start to rise when players become tired or injured as the number of matches
played increases, or perhaps as a result of increased stress or psychological pres-
sure in matches played before a large audience. Second, external diseconomies are
factors that reduce the quality of the product as the audience size increases. A
stadium with a capacity of 50,000, for example, might provide a better viewing
experience than one twice that size, due to excessive distance from the pitch in the
latter case. Beyond the stadium capacity, the TV audience consumes a product
that is presumably inferior to that obtained by ‘live’ spectators, though one whose
characteristics do not vary further as the TV audience size increases.
Rosen demonstrates first that imperfect substitution between sellers, combined
with internal diseconomies, induces a skewed distribution of earnings, even if
there are no joint consumption economies. The more talented sellers charge each
consumer a higher price and capture a larger share of the market than their less
talented counterparts. But in the absence of joint consumption economies, talent
differentials do not lead to extremely high seller concentration: all sellers are able
to capture some market share. Under certain simplifying assumptions, both the
price paid by each consumer and each seller’s audience share can be shown to be
linear functions of the seller’s talent. This implies that sellers’ total earnings, the
product of price and audience, are quadratic in talent. The distribution of earn-
ings is therefore more skewed than the distribution of talent.
To account for high seller concentration as well as a skewed earnings distri-
bution, joint consumption economies are also required in the model. Suppose
initially that there are no internal or external diseconomies at all, so joint con-
sumption economies are not limited by the size of the audience. In other words,
the audience size can be expanded indefinitely, without imposing additional
costs upon the seller, and without causing any deterioration in the quality of
the product obtained by the consumer. In this extreme case, the market has the
characteristics of a natural monopoly. The most talented seller is able to drive all
other sellers out of business, and in the long run services the entire market. The
equilibrium price, however, is constrained by the threat of entry. The rent element
The economics of superstars 203
in the equilibrium price, which is proportional to the talent differential between

the most and second-most talented sellers, is modest. Crucially, the total rent
obtained by the most talented seller is substantial, because the latter captures the
entire audience. As Rosen and Sanderson (2001) observe, superstars achieve high
earnings by operating at high volume (by servicing very large audiences) but at
low margins (by charging each audience member only a modest mark-up relative
to less talented suppliers).
The ‘zero diseconomies’ case described above is somewhat unrealistic, but use-
ful because it illustrates the main insight of the Rosen model. The most talented
seller captures the entire market because the production technology permits limit-
less expansion of the audience at zero cost to the seller. This is in stark contrast
to the experience of doctors, nurses, teachers and police officers, who find that the
number of patients they can treat, schoolchildren they can teach and criminals
they can catch on any one day is strictly finite. The key to the high earnings of
superstars, in other words, lies in the vast extent of the audience they can reach at
any one time, afforded by the existence of scale economies in joint consumption
(Rosen and Sanderson, 2001).
In practice, of course, one football team does not dominate the entire market.
The Rosen model suggests that this is not solely due to considerations of jointness
in production or concerns over competitive inequality (see Chapters 1 and 2). It
is also to be expected if joint scale economies in consumption are eventually lim-
ited by internal or external diseconomies of the kind described above. In this case
several sellers obtain market shares in long-run equilibrium. The price each seller
charges is constrained by the (actual not potential) presence in the market of other
sellers with less talent, and the rent element in the price is dependent on the talent
differential. If the rate at which internal or external diseconomies increase varies
inversely with talent, audience share and price both increase with talent, but audi-
ence share increases at a faster rate than does price. As before the high earnings
of the most talented sellers derive principally from the vast size of their audiences,
rather than from the modest price differential paid by each audience member.
An extension of Rosen’s superstars model is discussed by MacDonald (1988),
who considers the career progression of performers who eventually become super-
stars in their fields, despite the fact that in the early stages of their careers they
may have imperfect information about their future performance and future earn-
ings. Aspiring stars enter the profession at a young age and perform cheaply. The
audience feedback they receive provides useful but incomplete information about
their probable future performance. Those who receive negative feedback are per-
suaded to withdraw from the profession. Those who receive positive feedback are
encouraged to continue, and as the favourable notices accumulate their audience
rapidly expands. So too do their earnings, and as in the Rosen model at a faster
rate than the talent gradient would suggest. Although MacDonald seems to have
in mind primarily groups such as actors or musicians, the notion of a football
apprenticeship as an information-gathering process, enabling individuals to take
rational decisions at various stages as to whether to continue in pursuit of a car-

eer as a professional, seems an appropriate description of the path followed by
many youngsters, including the majority that eventually do not make the grade as
professionals.
In summary, the contrast between the configurations of high earnings-low util-
ity achieved by sports stars, and low earnings-high utility achieved by groups such
as nurses and teachers is explained mainly by the technologies under which the
former operate, which enable them to provide services, albeit of modest value, to
very large audiences. The latter groups provide services of much higher value, but
to a strictly finite number of users. The same technological conditions explain the
high levels of seller concentration found in professional sports. Internal and exter-
nal diseconomies of scale, however, impose limits on the audience share of the
most talented sellers, enabling the less talented to capture some share of the mar-
ket. The road from apprenticeship to stardom can be viewed as a type of sequen-
tial information-gathering process, in which cumulative feedback on performance
enables the individual to take rational decisions on whether to continue or with-
draw from the profession.
Rosen’s essentially technological explanation for the high earnings of superstars
is clearly relevant in explaining salary inflation in professional football during the
1990s and 2000s. Before the arrival of pay-TV (see Chapter 6, Section 6.5), English
football reached large TV audiences via free-to-air services provided by the BBC
and ITV. Constrained by this type of ‘public goods’ production technology, foot-
ball was unable to appropriate the full economic price for providing the service to
individual audience members. Since the early 1990s, technological innovation in
broadcasting has transformed this situation. With further moves in the direction
of free agency taking place simultaneously in the footballers’ labour market (see
Chapter 6, Section 6.6), the leading players’ bargaining position in salary nego-
tiations was strengthened immeasurably. The players were able to capture a large
proportion of the rent element in the additional broadcast (and other) revenues,
resulting in rampant salary inflation.
7.3 Rank-order tournaments and intra-team earnings distributions

As noted previously, Rosen’s insights on the economics of superstars do not
address the complexities of intra-team earnings distributions. At first sight, these
can seem as puzzling as the disparities between the earnings of sports stars in
general and those of other groups. In the English Premier League, for example,
the top players can earn many times more than the average first-team player, even
though it seems doubtful whether this differential could be explained by a corre-
sponding differential in MRP.
In a competitive labour market, the intra-team salary structure is equivalent
to a piece-rate system, since each player receives a salary equivalent to his MRP.
Differences in salaries between players in the same club should reflect the fact
Rank-order tournaments, intra-team earnings 205
that those contributing most to club revenues get paid the most. Casual observa-
tion, however, suggests that some players receive large salary increases even when
there is no noticeable improvement in performance. Often, players who might be
judged only a small fraction better than others appear to earn significantly more.
In these cases, the intra-team salary structure seems to be hierarchical. The models
outlined above are unable to explain why top stars can earn ‘n’ times more than
other players, if the talent gradient is shallower than ‘n’, and if their contribution
to revenue is not ‘n’ times as high.
In a highly influential paper on the determination of executive salaries in
corporate business, Lazear and Rosen (1981) argue by analogy that top execu-
tives’ high salaries are equivalent to the outcome of a rank-order tournament.
In situations where it is costly to monitor or measure exactly the productivity
of individual workers, it may be efficient for firms to rank workers according to
their productivity, and reward them according to their rankings. In the corpor-
ate sector, the salaries of vice-presidents are typically much lower than those of
presidents. Yet presidents are often promoted from the ranks of vice-presidents.
On the particular day an individual is promoted, his salary might triple, but it is
difficult to argue that his productivity has also suddenly increased by an equivalent
amount.
According to the Lazear and Rosen model, the individual who achieves promo-
tion to president is like the winner of a contest for which there is a very large first
prize. The president’s salary does not reflect his or her current productivity, but
it does induce all junior employees (from the rank of vice-president downwards)
to perform appropriately, in the hope that one day they too will win the prize and
achieve promotion to president. In other words, the president earns a high salary
not because he is highly productive as president, but because this type of salary
structure provides incentives for all employees to be more productive throughout
their entire working lives.
Scully (1995) adapts this type of analysis in order to examine player salary deter-
mination in several North American sports. It is worth noting that in contrast to
the situation in the corporate sector, the productivity of each player or ‘worker’ is
relatively easy to measure (baseball being perhaps the most obvious case in point).
Both the employer and the customer observe the joint output of the team, and
have ample opportunity to form their own subjective assessments of the individ-
ual contributions of each team member. With reference to English football, this
relationship is aptly described as follows:
The first team player faces over forty public examinations a year before an attendance of thou-
sands of spectators. He will be talked about, shouted at, criticised or praised, given extensive
coverage in the press and radio … he is in the limelight and must be seen to behave as befits his
profession. (PEP, 1966, p134)
Even so, professional team sports players do not appear to be paid on a piece-
rate basis. Scully argues that the rank-order tournament model may provide a
more convincing explanation of compensation scales than the traditional supply-

and-demand model. Clubs that are motivated either by playing success or by profit
need players to develop their skills to their maximum potential, and to exert the
maximum effort on behalf of the team. A compensation structure in which sal-
aries depend upon each player’s ranking relative to others within the same club
provides incentives for all players to compete (effectively against one another) in
pursuit of these aims. Even in sports such as football, where there is high input
complementarity and individual outputs are less easily identified, the rank-order
tournament model may still be applicable.
Historically, there were no restrictions on the number of players an English
League club could register. Under proposals announced during the 2010 season
for implementation at the start of the 2011 season, however, premier league clubs
are subject to a maximum first-team squad size of 25 players, of which no more
than 17 can be over the age of 21 and not home-grown (trained before the age of
21 in England or Wales).
Each club’s first-team squad players compete against one another for a place in
the team. At the start of each game the manager selects one goalkeeper, ten outfield
players and up to seven substitutes, three of whom can be used to replace any of
the eleven players on the field of play at any time during the game, either for tac-
tical reasons or because of injury. The numbers of defenders, midfield players and
forwards among the outfield players depends upon the tactical choices of the man-
ager. Typical formations include 4 –3–3 (four defenders, three midfielders and three
forwards), 4 – 4 –2, 5–3–2, 3–5–2 and 4 –5–1. For each playing position, the squad
will include several players who can provide cover. Especially useful in this respect
are utility players, able to play in more than one outfield position. Performance is
related to athletic endowment such as speed, strength, agility, balance and coordin-
ation; intangible attributes of talent such as determination and concentration; and
player and team investments in enhancing skill through coaching, tactical aware-
ness and practice at passing, shooting, heading and tackling.
Theoretical and empirical models of rank-order tournaments have been applied
extensively to individual sports such as tennis and golf. Competitors face a fixed
prize structure and achieve the next higher reward by moving up by one place in the
ranking of competitors. Scully argues, however, that competition among sports play-
ers for a first-team place is not entirely dissimilar. Those in the first team are always
threatened explicitly by the performance of lower-ranked reserve players, and impli-
citly by the possibility that the club might sign a replacement player for the same
position from outside. If a first-team player experiences a slump in form, a reserve
is substituted into the team or the club acquires a new player. Injury or suspension
for disciplinary transgressions can also lead to the replacement of one player by
another. A replacement who performs well can often capture a first-team place, until
he in turn eventually drops out due to loss of form, injury or suspension.
The rest of this subsection contains an exposition of the rank-order tournament
model using mathematical notation. Apart from minor changes of notation, the
Rank-order tournaments, intra-team earnings 207
model is as presented by Scully (1995). It is assumed that two players (i = 1,2) are
competing for one first-team place. The performance of player i, denoted Qi, is
determined by his athletic endowment and his investment in playing skill, denoted
ti, plus a number of exogenous factors that affect performance randomly, denoted
εi. Random factors include temporary fluctuations in form, fitness and fortune. The
probability that player 1 defeats player 2 as part of the contest for a first-team place,
denoted p, is the probability that player 1’s performance exceeds that of player 2:
p = prob(Q1 > Q2 ) = prob(t1 − t 2 > ε 2 − ε1 ) = 1 − F(t1 − t 2 ) [7.1]
where Qi = ti + εi for i = 1,2; E(ε2 – ε1) = 0; and F( ) is the cumulative distribution
function of ε2 – ε1.
The intra-team salary structure pays SF to the first-choice player and SR to the
reserve player, with SF > SR. The expected salaries of the two players, S1 and S2, are
determined by the probability that player 1 beats player 2 for the first-team place
(p) and the salaries for the two positions (SF, SR):
S1 = pSF + (1 − p)SR ; S2 = (1 − p)SF + pSR [7.2]
Players can improve their skills by means of investment. Because some attributes of
playing talent are natural, the cost of investment, Ci(ti), is assumed to vary between
players. It is assumed that the marginal cost of investment in further skill increases
∂C i
with the level of skill already acquired, so > 0 ; and that marginal cost increases
∂t i
∂2 C i
at a faster rate as the level of skill increases, so > 0 . If player 1 has more natural
∂t i 2
ability than player 2, player 1’s marginal cost of further investment in playing skill
∂C1 ∂C 2
(starting at any given level of skill) is lower than that of player 2, so < .
∂t1 ∂t 2
Players are assumed to invest in the development of playing skills to maximise
expected net income, EYi = Si – Ci(ti). The first-order conditions are:
∂EYi
= (SF − SR )f(t1 − t 2 ) − C i′ (t i ) = 0 (for i = 1,2) [7.3]
∂t i
where f = F′ is the probability density function of ε2 – ε1. Equation [7.3] provides
the Cournot-Nash solution for each player’s level of investment in skill (based on
the assumption that the investment of the competitor is fixed at its current level)
in terms of the salary levels, SF and SR. The equilibrium is illustrated in Figure 7.1.
∂C1 ∂C 2
Player 1 has more natural ability than player 2, so MC1 = < MC 2 = . If the
∂t1 ∂t 2
salary differential is (SF – SR)1, t*1 and t*2 represent the optimal investments in play-
ing talent for each player.
The greater the level of playing skill at the club’s disposal, the more valuable
is each match to the club. By varying the intra-team salary structure (SF, SR), the
MC2
Marginal cost and benefit to players of

MC1
investment in talent (SF − SR )1f ( t1* − t *2 )
(SF − SR )1f ( t1** − t *2* )
0 t *2* t *2 t1** t1* t1, t2

Figure 7.1 Optimum investment in playing talent, rank-order tournament model
club can influence the level of investment by both players in their playing skills.
Increased player investment adds to the value of the club’s matches, but there are
lower and upper bounds to the range within which the club can vary the salaries.
First, SR has a lower bound because player 2 has the option to earn a living outside
the sport. If player 2’s potential earnings outside the sport are C8, this constraint
implies that the lower bound of player 2’s expected earnings is equal to outside
earnings plus player 2’s cost of investment in playing skill:
(1 − p)SF + pSR = C 2 (t 2 ) + C [7.4]
Second, assuming that the club has a financial constraint which prohibits it from
making a loss, there is an upper bound to the total salary bill, equal to the club’s
total revenue.
Scully simplifies the derivation of the equilibrium intra-team salary structure by
introducing a parameter c (0 < c ≤ 1) to reflect the degree of competitiveness in the
players’ labour market. c is the total salary bill expressed as a proportion of total
revenue, as follows:
SF + SR = cTR [7.5]
At the maximum value of c = 1, the players’ salaries capture all of the club’s
revenue. This would apply with free agency in the players’ labour market, so the
players’ bargaining power is maximised.1 The distance by which c falls below one
is a measure of the degree of monopsony power held by the club as a buyer of
players.
Equations [7.4] and [7.5] can be used to derive the following expressions for the
salary levels, SF and SR:
SF = cTR[p/(2p − 1)] − [(C 2 (t 2 ) + C)/(2p − 1)]

SR = [(C 2 (t 2 ) + C)/(2p − 1)] − cTR[(1 − p)/(2p − 1] [7.6]
Determinants of players’ compensation 209
Scully points out two key implications of the model:

• The greater the degree of monopsony power in the players’ labour market, the
narrower is the intra-team salary structure. This follows directly from [7.6],
because SF – SR = cTR/(2p – 1). Conversely, a move towards free agency should
increase the differential between the salaries of the top players and the rest.
• A reduction in the differential between SF and SR reduces the level of invest-
ment in playing skills by both players, and narrows the differential in playing
skills between the first-team and the reserve-team player. This is illustrated by
the move from (t*1, t*2) to (t**
1 , t2 ) in Figure 7.1. Conversely, a move towards free
**
agency should increase the level of investment in playing skills generally, and
increase the differential between the playing skills of the top players and the
rest.
Clearly, the rank-order tournament model of salary determination generates a
number of propositions that are relevant in the case of English football. In con-
trast to the traditional supply-and-demand model, the rank-order tournament
model is capable of decoupling players’ salaries from their MRPs, thereby explain-
ing highly unequal and hierarchical intra-team salary distributions.
The model also predicts that moves towards free agency should lead to a widen-
ing of the gap in compensation between top players and the rest, and should lead
to an increase in investment in playing skills all round. There is no doubt that the
former has occurred in English football since movement towards free agency first
started in the early 1960s. The latter is more difficult to verify conclusively, but
there is overwhelming anecdotal evidence: video footage suggests, and most com-
mentators agree, that modern-day players are much fitter, faster and stronger than
their counterparts from earlier eras.2
7.4 Determinants of players’ compensation: empirical evidence

Football players’ salaries
Empirical evidence on the factors that determine players’ salaries in European
professional football is sparse, because individual players’ compensation details
are considered confidential and are generally not disclosed.
Reilly and Witt (2007) examine salary determination in North American Major
League Soccer (MLS), using data for 361 players obtained from the MLS Players
Union on base salaries prior to the start of the 2007 season, excluding bonuses
and endorsements. A salary regression is estimated, with player characteristics,
performance measures and team effects used as explanatory variables. Player char-
acteristics variables include age, playing position, time at the current club, whether
or not the player is a US citizen, and a player’s racial group. Racial group is deter-
mined by inspecting colour photographs; players are designated as ‘white’, ‘black’,
or ‘mixed race’. Performance variables include the number of prior seasons in
MLS, number of MLS games started in the previous season, number of MLS
games as substitute in the previous season, and dummy variables for whether or
not the player has international caps and whether or not the player has played in
European professional football.
In ordinary least squares (OLS) regressions, around two-thirds of the total vari-
ation in base pay is explained. The results show that playing experience (num-
ber of consecutive MLS seasons) is a more important determinant of salary than
length of time at the current club. Experience of playing in European leagues
produces an average mark-up of 70 per cent compared to a player without such
experience. An international player obtains a mark-up of 67 per cent and strikers
are more highly rewarded than players in other positions. Coefficients on the race
dummies are insignificant. Similar results are obtained from quantile regressions
for median salary.
Using 2008 season data on 193 players, Kuethe and Motamed (2010) also
examine the determinants of salaries in MLS. Superstar status is indicated by
a ‘designated player’ dummy variable. In a concerted effort to recruit star play-
ers, MLS awarded to each team in 2007 a single designated player slot, which
is not constrained by the league’s salary cap (known colloquially as the ‘David
Beckham Rule’). Designated players can be traded, but no team can have more
than two. Data on the salaries of individual players are published annually by
the MLS Players Union. Age and experience are significant in salary regressions
estimated using OLS, but performance measures are insignificant. The effect of
country of origin dummy variables is weak, but the designated player dummy
variable is highly significant, yielding a 928 per cent salary premium. In quantile
regressions for percentiles towards the upper end of the salary distribution, desig-
nated player status yields a similar premium, while several performance measures
are significant.
For European football, several researchers have used sources that offer mar-
ket valuations of players as a proxy for undisclosed salaries. Several studies have
used valuations of German Bundesliga players obtained from the Kicker football
magazine.3 The use of transfer market valuations data as a proxy for players’ sal-
aries has been justified on the basis that in cases where the actual salary is known,
the correlation between the actual salary and the published valuation is high. The
Kicker player valuations are compiled by a stable panel of experts, who have estab-
lished consistent practice over a period of years. There is a close correspondence
between the aggregated valuations divided by a constant factor, and the aggregate
salary expenditure for the top tier of the Bundesliga (Frick, 2003; Torgler, Schmidt
and Frey, 2006). As in North American studies based on actual salary data, regres-
sions are estimated in which the player’s valuation is assumed to depend on experi-
ence, performance and team characteristics.
Lehmann and Schulze (2008) use Kicker player valuations data from the 1999
and 2000 seasons for 651 German Bundesliga players. Recognising the skewness in
the distribution of valuations, the authors report quantile regressions in order to
identify the determinants of the valuations at different points in the distribution.
Performance measures, especially goal scoring and assists, are significant. The
signs and magnitudes of some coefficients differ between different points in the
distribution, validating the use of quantile regression. Media presence has a posi-
tive influence on valuations which diminishes over time, suggesting decreasing
returns to popularity. This finding appears contrary to the hypothesis of a super-
star effect.
Battré, Deutscher and Frick (2009) use a German Bundesliga data set with
Kicker player valuations, comprising 6,147 player-year observations for 1,993 dif-
ferent players over 13 seasons from 1996 to 2008. Performance measures include
matches played, goals scored and international caps, recorded for the player’s
entire career to date and for the most recent season. Other explanatory variables
include age, a team captain dummy, playing position dummies and team-specific
variables. Most of the explanatory variables are significant in OLS, random effects
and quantile regressions, with the most recent performance indicators being far
more important than entire-career indicators. The results support the hypothesis
of a superstar effect.
Bryson, Frick and Simmons (2009) examine the impact of two-footedness on
player valuations. Most players are predominantly right-footed or left-footed, but
some are considered equally effective with both feet. Two-footedness may enhance
a player’s valuation for two reasons: first, the player’s performance might be
enhanced directly; and second, a two-footed player may be more versatile because
he can play in several positions.4 OLS and quantile regressions indicate a premium
for two-footedness, estimated at 15.4 per cent for the European cross-section and
13.2 per cent for the Bundesliga panel.
The focus of the study by Ashworth and Heyndels (2007) is the relationship
between month of birth and Kicker player valuations in the German Bundesliga.
Professional footballers born after the cut-off date used to define each age group
have higher valuations due to a selection effect: survivors are revealed to be of
above-average talent because they have overcome the tendency of the scouting
system to discriminate against them; and they have benefited from a superior foot-
ball education that requires them to compete with older (and therefore physically
stronger) players. Estimation results show clear evidence of a month-of-birth-
related bias: players born late in the year, after the cut-off date of 1 August, are
valued more highly on average. The impact of the month of birth varies with
playing position: the premium for being born late is highest for goalkeepers and
defenders, and is absent for forwards.
Lucifora and Simmons (2003) use 1995 data on Italian footballers to esti-
mate an equation for salaries in which performance variables play a key role.
Pre-season salary data on 533 outfield players, gross of tax, but net of bonuses
and signing-on fees, is obtained from an annex to the newspaper Il Giornale pub-
lished on 15 March 1996. Salaries are explained by experience (including age and
number of appearances over the entire career and for the most recent season),
performance (goal scoring and assists), reputation (international appearances)
and team-specific variables. Most of the explanatory variables have a significant

influence on salaries. A key finding is that salaries are highly convex in goal-
scoring rate and assists, consistent with a superstar effect generated by spectator
enthusiasm for attacking play.
Using 2007 data, Christiansen and Sievertsen (2008) also find evidence of a
superstar effect in Italian football. Salary data on 444 players is obtained from
La Gazzetta dello Sport. Explanatory variables in the salary equation include
age, experience (minutes played in the current and previous seasons, and inter-
national caps), performance (goals scored, shots, assists, tackles and passes, all
for the current and previous seasons), and team-specific variables. The average
relationship between age and players’ salaries is non-linear, with a turning point
at age twenty-four. International experience has a positive impact on salaries,
and experience for national teams other than Italy has a larger effect. The results
also show a positive superstar premium in salaries, of almost 70 per cent. Several
performance criteria are used to identify superstars. In comparison, the estimated
effect on the salary of scoring an extra goal per match over an entire season is
around 67 per cent.
Salaries in North American major league sports

For North American major league sports other than MLS, numerous studies have
examined the determinants of player salaries, as well as closely related topics such
as player exploitation and racial discrimination. One of the first and most influen-
tial studies in this literature is Scully (1974), in which a framework is developed for
drawing comparisons between estimated MRPs for different categories of player
in MLB, and the average salaries of players in the same categories. Scully’s study
dates from the pre-free agency period when the reserve clause was still effective, so
systematic divergence between salaries and MRPs is expected.
Scully obtains estimated marginal productivities for baseball hitters and pitch-
ers using the slope coefficients in a regression of team win ratios on team slugging
averages (for hitters) and team strikeout-to-walk ratios (for pitchers). Dummy
variables are also included in the regression to control for differences between
the NL and the AL, and to differentiate teams according to the extent to which
they were in contention to be divisional winners in the season concerned. Gross
MRPs for hitters and for pitchers are imputed by multiplying the marginal prod-
uctivity estimates by the slope coefficient obtained from a second regression of
team revenues on win ratios. Various additional controls used in the second regres-
sion include an NL–AL dummy, home-town population and a variable reflecting
underlying differences between teams in intensity of support. Net MRPs are cal-
culated by deducting allowances for player training and other factor costs from
the gross MRPs.
Scully then estimates salary equations for hitters and pitchers, using individual
data. Among the explanatory variables are personal career slugging averages and
strikeout-to-walk ratios, career durations, and variables (such as population and

intensity of support) that are also used in the revenue regression to control for
demand factors contributing to monopoly rent. The latter are mainly insignificant
in the salaries regressions, suggesting that the players failed to capture any share
of the monopoly rents under the reserve clause. The comparisons between MRPs
and salaries are drawn by substituting sets of explanatory variables describing
representative hitters and pitchers of mediocre, average and star quality into the
various estimated equations. The MRP-talent gradient is much steeper than the
salary-talent gradient. Consequently, while mediocre players earned more than
their net MRPs, the salaries of average players and star players were only around
20 per cent and 15 per cent of their net MRPs, respectively. This suggests there
was substantial monopsony exploitation of players before free agency. Following
Scully’s pioneering work, numerous authors have used the model as a benchmark
for examining exploitation in North American sports (see, for example, Medoff,
1976; Raimondo, 1983; Zimbalist, 1992; MacDonald and Reynolds, 1994;
Gustafson and Hadley, 1995; Krautmann, 1999; Mullin and Dunn, 2002). Pelnar
(2007, chapter 6) reviews the post-Scully literature on exploitation.
Using an empirical framework similar to that employed by Scully twenty years
earlier, MacDonald and Reynolds (1994) use data on player performance and
earnings in MLB in the 1987 season to show that following the introduction of
free agency, players’ salaries matched their estimated career MRP more closely
than they did during the pre-free agency period. There were still some anomalies,
however. Young players were paid less than their MRP, perhaps due to institu-
tional restrictions on their mobility within the labour market, and senior pitch-
ers (but not senior hitters) were overpaid. MacDonald and Reynolds also present
evidence that differences between the salaries of top- and middle-rank players are
consistent with Rosen’s model of superstars.
Scully (1995) uses data on the individual performance and salaries of baseball
and basketball players before and after the introduction of free agency in both
sports, to examine the impact of this change. It is clear that the salary-talent gra-
dient became steeper during the free agency period, and for both sports the sen-
sitivity of salaries to performance in the current season and to measures of past
experience (included in the regressions to control for talent, investment and effort)
is higher than it was before free agency. For baseball, an average individual per-
formance index exceeds an average individual salary index for up to 547 career
games; thereafter, the salary index exceeds the performance index. This finding
seems consistent with the time-profile of salaries predicted by the rank-order tour-
nament model. For the pre-free agency period, in contrast, the salary index is
higher than the performance index for both inexperienced players (less than 405
games) and players nearing the end of their careers (1,550 games).
Some empirical research has suggested that there are limits to the applicability
of the rank-order tournament model to team sports. An intra-team salary distri-
bution that is too unequal may be divisive, perhaps damaging team cohesion and
spirit to an extent that outweighs the incentive benefit described above. It is cer-
tainly common enough for English football clubs to explain failure to reach agree-
ment with a player they are seeking to sign in terms of a reluctance or refusal to
overturn established internal remuneration structures. Richards and Guell (1998)
argue on similar lines that a high intra-team variance in salaries may have a nega-
tive effect on performance, since it implies a dispersed level of talent and an inabil-
ity to function at a high level as a team. One or two star players do not necessarily
carry an entire team. In their empirical results for MLB over the period 1992–5,
the mean salary has a positive effect on the win ratio and the propensity to win
championships, but an increase in the variance of salaries reduces the win ratio
(with a coefficient that is significant at the 10 per cent level). A greater dispersion
of salaries does not seem to affect the probability of winning the championship,
however. Depken (2000) and Jewell and Molina (2003) report further evidence for
MLB. Frick, Prinz and Wintelmann (2003) find that a higher degree of dispersion
enhances the performance of NBA and NHL teams, but has the reverse effect in
the NFL and MLB. Berri and Jewell (2004) fail to find any evidence of a relation-
ship between salary dispersion and team performance in the NBA.
Conclusion
Professional footballers’ pay is an issue of occasional public concern, irritation
or outrage. By the end of the 2000s, the annual payrolls of the top English clubs
were in excess of £100 million per club; while several of the world’s very highest-
paid footballers registered total annual earnings of more than £20 million each,
taking into account salaries, bonuses, sponsorship and income from other com-
mercial activities. This chapter considers theoretical explanations, put forward by
economists, for the exceptional remuneration that is enjoyed by the leading stars in
modern-day professional football, and other professional team sports.
The theoretical analysis of the economics of superstars addresses the apparent
paradox between the high earnings achieved by sports stars whose contribution to
society’s well-being is perhaps rather modest (despite the adulation they receive),
and the low earnings achieved by groups such as nurses and teachers, whose con-
tribution is far more important. The discrepancy is explained by the fact that
especially since the advent of pay-TV, sports stars are capable of servicing very
large paying audiences simultaneously, incurring little or no incremental cost as
the audience size increases. Nurses and teachers, in contrast, service strictly finite
number of users. The potential for high salary inflation in football is therefore cre-
ated by changes in the level and composition of effective demand, driven in turn
by technological change in broadcasting. Meanwhile successive moves towards
free agency in the players’ labour market have increased players’ bargaining power
in their salary negotiations, enabling them to capture a large proportion of the
rent element in the extra revenues that are being generated.
Another feature of sports stars’ compensation is described by the rank-order

tournament model, which examines intra-team salary distributions. Clubs have
an interest in providing incentives for footballers to maximise investment in their
playing skills, and to maximise their level of effort. High performance arises
from innate ability and from investment in skills and effort. According to this
model, large salary differentials provide all players with the incentive to invest in
the development of their skills, in the hope of securing a first-team place and a
higher salary. In contrast to the traditional microeconomics textbook supply-and-
demand model of the labour market, the rank-order tournament model is cap-
able of decoupling players’ salaries from their MRPs, thereby explaining highly
unequal and hierarchical intra-team salary distributions.
Notes
1 Implicitly for c = 1, there is also an assumption that the club seeks to maximise revenue or
playing success, subject to a zero-profit constraint.
2 It is not at all clear, however, that the best modern-day players are more skilful on average
than players of earlier periods. Natural skill is, of course, mainly an endowed rather than
an acquired attribute, and is therefore not sensitive to changes in the structure of players’
compensation.
3 German-language studies include Lehmann and Weigand (1999), Lehmann (2000), Huebl
and Swieter (2002) and Frick and Deutscher (2009).
4 Data on two-footedness and player valuations at the start of the 2006 season for players in
England, France, Germany, Italy and Spain were obtained from the website www.transfer-
markt.de. The German sample was supplemented by valuation data from Kicker magazine
for several earlier and later seasons.
8 Professional footballers:
employment patterns and racial
discrimination
Introduction
The previous chapter has examined the topic of footballers’ remuneration.
Chapter 8 discusses several other topics concerned with the economics of the pro-
fessional footballers’ labour market. Section 8.1 presents an analysis of patterns
of migration, mobility and career development among English club football’s
regular workforce: the players employed by the ninety-two member-clubs of the
Premier League and Football League. The coverage of the analysis extends from
the mid-1980s to the late 2000s, a period during which the character of English
football has been transformed by the arrival of a large contingent of overseas
players. During this period there appears to have been a shift in the overall bur-
den of responsibility for the development of young players away from the smaller
English clubs towards their larger counterparts; and some sharp regional dispar-
ities have emerged in the prospects for locally born youngsters to become profes-
sional footballers.
Patterns of migration by footballers across national borders reflect a wide range
of influences, which are reviewed in Section 8.2. An analysis of the employment
profile (by country of employment in club football) is presented for the inter-
national footballers who participated in the Finals of the 2000 and 2008 European
Championships. Despite the England national team having failed to qualify for
the Euro 2008 Finals, the English league was the largest single host-country pro-
vider of footballers in these Finals, reflecting England’s recent ascendancy within
Europe as an importer of top-level footballers.
Section 8.3 reviews empirical evidence on racial discrimination in English foot-
ball and elsewhere. A bias in the employment of indigenous black footballers in
favour of teams of higher divisional status may suggest that a form of hiring dis-
crimination affects the opportunities for indigenous black players of average tal-
ent to progress to professional status. It seems likely, however, that any such effect
has diminished over time. Possible sources of discrimination include prejudicial
attitudes on the part of club chairmen or team managers, a lack of opportun-
ities for youngsters from disadvantaged backgrounds to participate in competitive
216
Employment mobility, migration and career structure 217
football, and geographical variations in the ethnic composition of the population

combined with barriers to mobility in the footballers’ labour market.
8.1 Employment mobility, migration and career structure in English football

Section 8.1 presents an analysis of patterns of migration, mobility and career
development among English football’s regular workforce: the players employed
by clubs participating at all levels in the Premier League and Football League. The
principal data source for the empirical analysis is the retained list, published annu-
ally in Sky Sports Football Yearbook (known as Rothmans Football Yearbook prior
to 2003). This source provides an annual snapshot of the employment history and
current status of all players employed with the ninety-two English league mem-
ber clubs at an end-of-season (May) census point, enabling career progression to
be tracked at the micro (individual player) level. In addition, the comprehensive
post-war listing of players’ career records compiled by Hugman (1998) is used to
cross-check the data for the earlier years, and to complete some of the early-career
details for players whose records would otherwise have been left-censored.
Table 8.1 reports the number of players recorded at each end-of-season cen-
sus point from 1986 to 2009, in total and disaggregated between the four tiers
(divisions). Over this period there was a large growth in the numbers of players
employed in all four tiers. Total employment increased by 46.0 per cent, and the
rates of growth by tier were 55.2 per cent for T1, 55.4 per cent for T2, 38.5 per cent
for T3 and 31.9 per cent for T4. Much of the employment growth in the Football
League (T2, T3 and T4) took place during the late 1980s, while employment growth
in the Premier League (T1) took place mainly during the 1990s. Employment of
professional footballers by English clubs peaked between 1999 and 2002. Between
2002 and 2009, total employment fell by around 10 per cent, with most of the
decline concentrated on the Football League (T2, T3 and T4).
Table 8.2 reports an analysis of changes in the age distribution of professional
footballers, based on comparisons drawn at four-year intervals using the 1989,
1993, 1997, 2001, 2005 and 2009 end-of-season censuses. These data suggest that
during the 1990s and 2000s there was a shift in the burden of responsibility for
youth player development away from clubs in T3 and T4, and towards clubs in T1
and T2. During the mid to late 1990s, much of the growth in employment in T1
was concentrated among footballers in the younger age groups. During the 2000s
the trend in the employment share of the under-21 age group in T1 has stabilised
(but with some significant year-on-year variation). Over the period 1989–2009 as
a whole, the employment share of the under-21 age group fell by 2.6 per cent in T3
and 6.7 per cent in T4. This shift may be related to the introduction of free agency
through the Bosman ruling. Despite the existence of safeguards in the form of
financial compensation when players below the age of 24 move between clubs, it
has become more difficult for lower-tier clubs to realise large windfall financial
gains in the transfer market; and it seems likely that the incentive for these clubs
218 Footballers’ employment, and racial discrimination
Table 8.1 Employment totals for professional footballers by tier, English League,
1986–2009
T1 T2 T3 T4 Total Mean age SD age
1986 623 509 501 476 2109 24.0 4.7

1987 649 522 546 536 2253 23.9 4.7
1988 613 574 579 585 2351 24.1 4.8
1989 610 655 600 595 2460 24.1 4.8
1990 627 717 630 616 2590 24.1 4.8
1991 679 721 634 626 2660 24.2 4.8
1992 749 718 670 581 2718 24.3 4.9
1993 779 727 617 546 2669 24.3 4.9
1994 779 792 634 622 2828 24.5 5.0
1995 846 753 654 596 2849 24.5 5.1
1996 830 820 675 686 3011 24.4 5.2
1997 825 812 688 691 3016 24.4 5.2
1998 859 884 732 704 3179 24.3 5.3
1999 980 848 767 728 3323 24.2 5.3
2000 937 910 804 741 3392 24.2 5.4
2001 944 946 785 733 3408 24.3 5.3
2002 952 924 776 772 3424 24.3 5.3
2003 958 848 728 687 3221 24.4 5.3
2004 881 776 676 680 3013 24.4 5.2
2005 803 785 726 714 3028 24.6 5.3
2006 815 771 715 686 2987 24.4 5.3
2007 834 765 659 645 2903 24.5 5.2
2008 939 797 700 651 3087 24.2 5.1
2009 967 791 694 628 3080 24.2 5.0
% growth, 1986–2009 +55.2 +55.4 +38.5 +31.9 +46.0
Note: SD age denotes standard deviation of footballers’ ages.

to invest in the training and development of young footballers has been reduced
(see Chapter 6, Section 6.6).
At the opposite end of the age distribution, between 1989 and 2005 the pro-
portion of professional footballers aged over 32 increased from 5.4 per cent to 9.6
per cent. This increase may reflect the effects of improvements in training regimes
and the treatment of injuries, which have enabled more footballers to extend their
playing careers into their mid or late 30s. By the end of the 2000s, however, there
were signs of a reversal of this trend towards career prolongation. Between 2005
and 2009 the proportion of footballers aged over 32 fell back to 6.6 per cent.
Further analysis of professional footballers’ dates of birth reveals a feature of
the data that is quite striking, and consistent throughout the observation period.
Table 8.3 reports the percentage distribution of all players by month of birth.
This reveals that a large majority (around 60 per cent) of professional footballers
Table 8.2 Percentage distribution of professional footballers in England by age

band and tier, English League, selected years
T1 T2 T3 T4 All
1989
under 21 35.9 28.7 30.0 29.2 30.9
21–24 24.3 28.1 26.0 27.7 26.5
25–28 23.9 23.2 22.0 23.0 23.0
29–32 12.8 15.0 15.2 13.4 14.1
over 32 3.1 5.0 6.8 6.6 5.4
1993
under 21 34.5 25.4 25.0 27.8 28.5
21–24 25.3 28.5 30.0 30.2 28.3
25–28 19.6 24.6 21.6 20.9 21.7
29–32 15.5 14.6 15.6 13.4 14.8
over 32 5.0 6.9 7.9 7.7 6.7
1997
under 21 38.5 29.3 26.5 26.6 30.6
21–24 20.6 22.2 26.3 29.8 24.4
25–28 21.6 22.3 25.4 19.2 22.1
29–32 12.8 17.5 15.4 13.0 14.7
over 32 6.4 8.7 6.4 11.3 8.2
2001
under 21 42.4 34.9 26.8 25.6 33.1
21–24 19.8 20.4 29.8 25.5 23.5
25–28 17.7 18.6 19.4 20.3 18.9
29–32 13.8 18.5 14.9 16.9 16.0
over 32 6.4 7.6 9.2 11.6 8.5
2005
under 21 31.9 30.1 29.1 24.1 28.9
21–24 21.0 24.2 27.8 32.9 26.3
25–28 22.9 18.0 20.7 21.6 20.8
29–32 16.2 16.6 12.4 12.2 14.4
over 32 8.0 11.2 10.1 9.2 9.6
2009
under 21 40.7 28.7 27.4 22.5 30.9
21–24 20.1 25.0 27.5 30.3 25.1
25–28 20.6 24.0 23.6 24.7 23.0
29–32 13.9 15.9 14.7 12.9 14.4
over 32 4.8 6.3 6.8 9.7 6.6
were born during months that fall within the first half of the school year, from
September to February. Children whose birthdays fall in these months are phys-
ically bigger and stronger, and therefore more likely to succeed in school sports,
than children in the same school year with birthdays between March and August,
Table 8.3 Percentage distribution of professional footballers in England by month

of birth, English League, selected years
1989 1993 1997 2001 2005 2009
January 8.9 8.0 8.1 8.5 9.9 11.2

February 7.2 7.4 7.3 7.9 8.0 7.6
March 7.1 7.0 7.5 7.5 7.9 8.5
April 6.3 6.7 6.6 6.7 6.3 6.6
May 6.1 5.6 5.9 6.8 7.8 7.3
June 6.2 6.2 5.6 6.0 5.9 5.8
July 6.0 5.7 4.9 5.4 5.8 5.4
August 6.6 6.7 8.7 8.0 6.9 7.2
September 12.0 12.9 12.1 12.7 12.4 11.2
October 13.0 13.3 12.6 12.1 10.8 10.6
November 11.1 11.2 11.1 10.2 9.5 9.7
December 9.6 9.3 9.6 8.3 8.7 8.8
who nevertheless play in the same school sports teams as their older counterparts.
Although the physical disadvantage of the latter is most pronounced at the young-
est ages, it seems likely that an aversion to sport acquired early in life would effect-
ively terminate an individual’s prospects of becoming a sports professional.
Table 8.4 reports the numbers of players recorded at each end-of-season census
point from 1986 to 2009, disaggregated by country of birth. The categories are
England and Wales, Scotland, Northern Ireland, Republic of Ireland and the rest of
the world. It is worth noting that growth in the employment of players born outside
the British Isles (the rest of the world category) had been underway for a number of
years prior to the Bosman ruling, starting from an exceptionally low base. The dou-
ble signing by Tottenham Hotspur in 1978 of the star Argentinian players Osvaldo
Ardiles and Ricardo Villa was hailed as revolutionary at the time, but would be con-
sidered routine today. By the end of the 1993 season, the first since the formation of
the Premier League, each of the leading clubs had a handful of star foreign players
on its books. The following players all featured prominently in the 1993 season: for
Arsenal, John Jensen (Denmark) and Anders Limpar (Sweden); for Liverpool, Stig
Inge Bjornebye (Norway), Jan Molby (Denmark) and Ronnie Rosenthal (Israel);
for Manchester United, Eric Cantona (France), Andrei Kanchelskis (Ukraine) and
Peter Schmeichel (Denmark); and for Tottenham Hotspur, Nayim (Morocco), Eric
Thorstvedt (Norway) and Pat Van Den Hauwe (Belgium). Even as recently as the
mid-1990s, however, the playing rosters of all English clubs were dominated by
British players. Subsequently the trend in the employment of foreign-born play-
ers has been almost relentlessly upward, with only a temporary lull in the rate of
growth for a few years in the early 2000s. A record total of 744 foreign-born players
(24.2 per cent of total employment) was recorded in 2009.
Table 8.4 Employment totals for professional footballers by birthplace (country),

English League, 1986–2009
England Northern Republic Rest of

& Wales Scotland Ireland of Ireland world All
1986 1820 157 39 33 60 2109

1987 1949 160 41 41 62 2253
1988 2038 160 40 48 65 2351
1989 2126 169 44 50 71 2460
1990 2218 174 49 54 95 2590
1991 2288 167 40 62 103 2660
1992 2328 153 40 69 128 2718
1993 2277 144 46 61 141 2669
1994 2408 156 48 60 156 2828
1995 2415 140 48 64 182 2849
1996 2514 134 51 76 236 3011
1997 2443 140 61 79 293 3016
1998 2486 144 67 98 384 3179
1999 2517 153 69 105 479 3323
2000 2521 152 69 133 517 3392
2001 2427 153 71 155 602 3408
2002 2407 147 81 165 624 3424
2003 2268 133 72 153 595 3221
2004 2129 126 67 135 556 3013
2005 2098 121 65 124 620 3028
2006 2084 103 68 122 610 2987
2007 1993 98 53 124 635 2903
2008 2086 102 58 122 719 3087
2009 2057 104 48 127 744 3080
% growth, 1986–2009 +13.0 −33.8 +23.1 +284.8 +1140.0 +46.0
Among the other categories reported in Table 8.4, employment opportun-

ities for footballers born in England and Wales grew significantly between 1986
and 2000, but most of this gain has subsequently been reversed. The number of
Scottish-born players playing in England has declined, gradually during the 1990s
and more sharply during the 2000s. For Northern Ireland there was employment
growth up to 2002, followed subsequently by a sharp decline. For the Republic
of Ireland the pattern is similar, but with faster growth up to 2002 followed by a
slower decline.
Table 8.5 reports a more detailed analysis of changes in the distribution of play-
ers by country of birth and by tier (division), based on comparisons drawn at
four-year intervals using the 1989, 1993, 1997, 2001, 2005 and 2009 end-of-season
censuses. Most of the employment growth of overseas players is concentrated in
Table 8.5 Total numbers of professional footballers by birthplace (country) and

tier, English League, selected years
T1 T2 T3 T4 All
1989
E+W 484 546 557 539 2126
Scotland 47 64 20 38 169
NI 18 12 8 6 44
Ireland 28 8 7 7 50
RoW 33 25 8 5 71
Total 610 655 600 595 2460
1993
E+W 612 639 535 491 2277
Scotland 55 30 39 20 144
NI 19 11 10 6 46
Ireland 27 12 11 11 61
RoW 66 35 22 18 141
Total 779 727 617 546 2669
1997
E+W 586 639 600 618 2443
Scotland 42 46 23 29 140
NI 22 20 11 8 61
Ireland 40 23 11 5 79
RoW 135 84 43 31 293
Total 825 812 688 691 3016
2001
E+W 527 657 601 642 2427
Scotland 52 46 36 19 153
NI 23 23 18 7 71
Ireland 67 50 28 10 155
RoW 275 170 102 55 602
Total 944 946 785 733 3408
2005
E+W 392 541 579 586 2098
Scotland 33 36 34 18 121
NI 14 23 11 17 65
Ireland 28 58 16 22 124
RoW 336 127 86 71 620
Total 803 785 726 714 3028
2009
E+W 451 505 553 548 2057
Scotland 18 47 23 16 104
NI 22 12 6 8 48
Ireland 33 48 31 15 127
RoW 443 179 81 41 744
Total 967 791 694 628 3080

Table 8.6 Percentage distribution of footballers by birthplace (region) and location

of club (region), English League, 1989
Club location:
Birthplace
(region) Lon SE SW East EM WM YH NW NE Wal Total %
London 43.3 25.8 11.5 23.4 6.4 2.8 4.3 3.2 2.9 5.8 337 13.7
SE 9.2 27.2 5.2 6.3 5.5 2.8 3.4 1.3 5.1 7.2 162 6.6
SW 4.0 7.0 30.5 0.9 1.8 4.0 0.9 2.3 0.7 1.4 114 4.6
East 6.9 2.8 3.4 20.7 3.2 3.2 2.0 1.3 0.0 0.0 110 4.5
EM 2.0 2.3 3.4 5.9 24.1 6.8 6.9 4.6 2.9 4.3 154 6.3
WM 3.2 6.1 10.3 6.3 14.1 33.9 4.3 4.9 3.6 7.2 220 8.9
YH 3.4 5.2 8.6 6.3 15.0 7.2 42.0 8.9 11.6 11.6 316 12.8
NW 4.9 8.0 8.6 6.3 9.1 11.2 10.9 46.4 8.7 11.6 389 15.8
NE 5.2 3.3 5.2 5.4 9.5 8.0 10.9 7.4 38.4 1.4 214 8.7
Wales 3.2 2.8 4.6 4.1 0.9 4.4 2.0 4.4 2.9 44.9 110 4.5
Scotland 5.2 3.3 6.9 5.0 5.5 10.4 7.1 7.4 14.5 4.3 169 6.9
NI 2.9 1.9 0.0 1.8 0.9 1.2 1.4 2.7 2.2 0.0 44 1.8
Ireland 2.9 2.3 0.0 2.7 2.3 0.4 2.3 2.3 2.9 0.0 50 2.0
RoW 4.0 1.9 1.7 5.0 1.8 4.0 1.7 3.0 3.6 0.0 71 2.9
All 2460 100.0
Number of clubs (by tier)
T1 7 1 0 2 2 2 1 3 2 0 20
T2 2 3 3 2 1 5 4 3 1 0 24
T3 2 3 2 1 3 3 2 6 0 2 24
T4 1 0 2 3 1 1 7 6 2 1 24
All 12 7 7 8 7 11 14 18 5 3 92
Note: se = South East, SW = South West, East = East of England, EM = East Midlands,
WM = West Midlands, YH = Yorkshire and Humberside, NW = North West, NE = North
East, NI = Northern Ireland, RoW = Rest of the World.
T1, followed by T2, and then T3 and T4. By 2009, overseas players accounted for
nearly one-half of all employment with Premier League (T1) clubs. Clearly the
leading clubs have greater financial resources to allow them to search internation-
ally for playing talent. While the financial rewards in T1 and (perhaps) T2 may
be sufficient to induce the most talented players to relocate, the financial rewards
in the lower tiers may be insufficient for the less talented. Furthermore, the rules
governing the issue of visas and work permits to non-EU nationals tend to favour
the most talented athletes, but are tilted against the less talented in order to protect
the employment of locally born professionals.
The data reported in Tables 8.4 and 8.5 suggest that there have been significant
changes in employment opportunities for footballers born in England and Wales.
The employment prospects of locally born players might have increased due to the
increase in total employment, but any improvement has been offset by the arrival
of the large overseas contingent. Tables 8.6–8.8 present a cross-tabulation of the

Club location:
Birthplace
London 32.2 14.3 10.8 24.1 9.8 5.8 5.0 3.7 3.5 3.5 392 11.8
SE 12.3 23.7 5.6 5.8 5.1 3.4 2.3 2.0 1.8 5.9 205 6.2
SW 1.5 8.6 26.4 2.9 1.4 4.0 2.7 1.6 0.0 4.7 142 4.3
East 9.3 9.0 2.6 20.3 2.9 3.7 1.1 1.6 0.9 0.0 173 5.2
EM 2.3 3.3 4.3 4.8 19.6 8.3 7.0 3.4 2.6 0.0 186 5.6
WM 3.2 4.9 6.1 6.8 10.1 25.7 5.0 4.7 2.6 7.1 241 7.3
YH 2.1 3.7 3.5 3.9 10.9 6.1 32.5 9.2 11.4 5.9 329 9.9
NW 2.8 4.9 7.4 3.9 7.6 8.9 13.8 39.7 7.9 18.8 479 14.4
NE 1.7 2.9 4.3 2.6 6.2 4.0 7.9 4.3 36.8 5.9 217 6.5
Wales 3.4 3.3 6.9 3.2 2.5 4.0 2.5 4.4 2.6 41.2 153 4.6
Scotland 2.1 5.7 6.5 4.5 3.6 4.6 3.2 6.0 7.0 3.5 153 4.6
NI 2.8 1.2 1.7 3.2 2.2 1.8 1.4 2.7 0.9 0.0 69 2.1
Ireland 2.3 1.2 0.9 1.3 1.8 5.2 4.3 4.0 6.6 1.2 105 3.2
RoW 22.0 13.5 13.0 12.9 16.3 14.7 11.5 12.9 15.4 2.4 479 14.4
All 3323 100.0
T1 6 1 0 0 3 2 2 4 2 0 20
T2 2 2 2 3 0 4 5 5 1 0 24
T3 2 3 2 2 4 2 1 7 0 1 24
T4 3 1 3 3 2 0 5 3 2 2 24
All 13 7 7 8 10 8 13 19 5 3 92
Note: For region definitions, see note to Table 8.6.

distribution of the employment of players by birthplace at the level of standard

English regions (for those born in England), against the distribution by the stand-
ard region in which the player’s current employer is located, for three selected
years: 1989, 1999 and 2009.
A striking feature of Tables 8.6–8.8 is the strength of the association between
birthplace and the geographical location of employment for locally born foot-
ballers. Perhaps this pattern is to be expected among the youngest professional
footballers, who are most likely to be spotted and signed by a local professional
club when playing at school or junior level. At all ages, however, the geograph-
ical influence is strong. The distribution of employment by region of birth looks
very different in every region. In 2009, for example, 32.6 per cent of footballers
employed by London clubs were born in London, and only 2.3 per cent were born
in the North West. By contrast, 3.3 per cent of footballers employed by North
West clubs were born in London, and 39.4 per cent were born in the North West.
Tables 8.6–8.8 indicate some sharp differences by region of birth in the changes
in the employment prospects of English-born players. The numbers of professional

Club location:
Birthplace
London 32.6 19.9 15.8 26.3 9.5 7.9 5.2 3.4 6.4 10.9 418 13.6
SE 7.5 24.9 12.4 4.6 5.7 3.0 2.9 1.6 0.6 1.8 191 6.2
SW 1.0 5.1 23.9 2.9 3.8 2.4 2.0 0.5 1.2 1.8 112 3.6
East 5.4 6.5 3.8 21.7 4.3 5.4 2.6 1.0 1.7 1.8 152 4.9
EM 1.9 1.4 1.3 2.9 26.2 3.3 5.2 2.1 1.7 0.0 125 4.1
WM 1.2 2.2 5.6 5.0 3.3 26.6 4.6 5.9 0.6 7.3 196 6.4
YH 2.7 0.0 3.4 3.8 8.6 2.4 35.3 4.5 5.2 0.0 221 7.2
NW 2.3 6.1 8.1 5.0 9.5 9.7 10.9 39.4 8.7 0.0 452 14.7
NE 1.2 1.1 1.3 2.1 0.5 3.0 5.2 2.5 31.2 1.8 119 3.9
Wales 1.9 1.1 3.8 0.8 1.9 1.8 2.0 2.1 1.2 25.5 71 2.3
Scotland 0.4 2.9 6.8 3.8 5.7 3.6 3.7 3.1 2.3 9.1 104 3.4
NI 0.4 0.4 1.3 2.5 1.0 1.5 0.9 2.5 4.0 1.8 48 1.6
Ireland 2.5 2.5 2.1 5.8 3.3 5.1 4.3 4.7 6.9 7.3 127 4.1
RoW 38.9 26.0 10.3 12.9 16.7 24.2 15.2 26.8 28.3 30.9 744 24.2
All 3080 100.0
T1 5 1 0 0 0 3 1 7 3 0 20
T2 3 2 2 3 2 3 4 3 0 2 24
T3 2 2 4 3 2 2 3 5 1 0 24
T4 3 3 2 1 3 2 3 6 1 0 24
All 13 8 8 7 7 10 11 21 5 2 92
Note: For region definitions, see note to Table 8.6.

footballers born in the East, London, North West and South East regions increased,
by 38.2 per cent, 24.0 per cent, 17.9 per cent and 16.2 per cent, respectively, between
1989 and 2009; but the numbers born in the South West, West Midlands, East
Midlands, Yorks and Humber, Wales and North East regions decreased by 1.8 per
cent, 10.9 per cent, 18.8 per cent, 30.1 per cent, 35.5 per cent and 44.4 per cent,
respectively. Some of these shifts might be explained by shifts in the fortunes of
each region’s football clubs. For example, the number of North West clubs in T1
increased from three in 1989 to seven in 2009. The sharp reduction in the employ-
ment of players born in the North East is noteworthy, in view of this region’s his-
torical tradition as a producer of high-quality playing talent. In this respect there
are some affinities between the recent experience of the North East and Scotland.
Tables 8.1–8.8 provide snapshots of the characteristics of the population of
footballers employed by English clubs at moments in time. It is also possible to
track the career progression of individual players over time. Tables 8.9–8.13 report
empirical transition probabilities based on comparisons between the divisional
status (T1, T2, T3 or T4) of players employed at each of five start-years 1989,
Table 8.9 Four-year employment transition probabilities by tier, English League,

1989–1993
1993 employment status: 1993 employment status:

Not Not
T1 T2 T3 T4 emp. T1 T2 T3 T4 emp.
all ages in 1989 ages 25–28 in 1989

1989: T1 .359 .206 .090 .054 .291 .523 .189 .053 .038 .197
T2 .154 .222 .178 .094 .352 .144 .205 .247 .096 .308
T3 .047 .135 .191 .118 .508 .053 .168 .221 .099 .458
T4 .024 .114 .090 .163 .610 .015 .110 .096 .199 .581
All .145 .170 .138 .107 .440 .182 .169 .156 .108 .385
ages under 21 in 1989 ages 29–32 in 1989
1989: T1 .248 .182 .117 .065 .388 .296 .254 .085 .056 .310
T2 .156 .231 .134 .118 .360 .076 .185 .196 .076 .467
T3 .061 .100 .150 .122 .567 .000 .080 .125 .136 .659
T4 .047 .124 .094 .135 .600 .000 .013 .050 .087 .850
All .135 .161 .124 .108 .472 .085 .130 .118 .091 .577
ages 21–24 in 1989 ages over 32 in 1989
1989: T1 .444 .254 .085 .042 .176 .056 .056 .111 .111 .667
T2 .213 .282 .184 .086 .236 .094 .031 .031 .031 .813
T3 .065 .214 .279 .143 .299 .000 .000 .077 .026 .897
T4 .024 .176 .121 .224 .455 .000 .026 .000 .051 .923
All .180 .231 .169 .126 .294 .031 .023 .047 .047 .852
1993, 1997, 2001 and 2005, and the status of the same players four years later in
1993, 1997, 2001, 2005 and 2009, respectively. Transition probabilities are reported
for all players, and for players in the under 21, 21–24, 25–28, 29–32 and over 32
age bands. Using the first row of Table 8.13 as an example, 29.8 per cent of play-
ers who were employed by a T1 club in 2005 were again employed by a T1 club in
2009; 21.0 per cent of those employed in T1 in 2005 were employed in T2 in 2009;
9.4 per cent and 6.6 per cent of those employed in T1 in 2005 were employed in T3
and T4 in 2009, respectively; and 33.2 per cent were not employed by any English
Premier League or Football League club in 2009.
Among those players who remain in employment, there is a consistent ten-
dency for divisional status to decline with career duration. Players who eventually
become unemployable at the highest level are able to prolong their playing careers
by moving to a lower tier. Throughout Tables 8.9–8.13, the probabilities of playing
in a lower tier after four years (above the main diagonal) exceed the probabilities
of playing in a higher tier (below the main diagonal). The probability of not being
employed at the end of each four-year period is directly related to the player’s
divisional status at the start of the period: T1 players are the most likely and T4
International migration of professional footballers 227

1993–1997

Not Not

1993: T1 .316 .231 .093 .083 .278 .382 .324 .110 .022 .162
T2 .110 .205 .156 .150 .379 .138 .269 .234 .174 .186
T3 .030 .131 .155 .165 .519 .000 .228 .157 .197 .417
T4 .021 .034 .172 .152 .621 .009 .037 .174 .202 .578
All .131 .159 .141 .135 .434 .141 .226 .173 .147 .314
1993: T1 .261 .136 .086 .105 .412 .262 .272 .087 .117 .262
T2 .111 .167 .133 .122 .467 .030 .149 .099 .208 .515
T3 .048 .096 .116 .144 .596 .000 .096 .117 .170 .617
T4 .027 .040 .167 .187 .580 .014 .014 .071 .114 .786
All .134 .116 .120 .134 .497 .084 .144 .095 .155 .522
1993: T1 .385 .297 .110 .077 .132 .257 .114 .000 .086 .543
T2 .148 .250 .173 .138 .291 .021 .063 .021 .104 .792
T3 .061 .133 .233 .183 .389 .000 .042 .042 .063 .854
T4 .031 .043 .248 .124 .553 .000 .000 .053 .053 .895
All .160 .186 .189 .131 .334 .059 .053 .030 .077 .781
players are the least likely to remain employed after four years. Over the entire
period, employment growth has been accompanied by a rise in employment turn-
over. Of 2,460 players employed in 1989, 44.0 per cent were not employed in 1993.
Of 3,028 players employed in 2005, 52.3 per cent were not employed in 2009.
The transition probabilities are non-linear in age. By age band, the five-year transi-
tion probabilities out of employment between 1989 and 1993 were 0.472 (under 21),
0.294 (21–24), 0.385 (25–28), 0.577 (29–32) and 0.852 (over 32). The corresponding
probabilities between 2005 and 2009 were 0.575, 0.383, 0.398, 0.606 and 0.880. The
relatively high rates of transition out of employment for the lowest and highest age
bands reflect the fact that the under-21 age band contains a high proportion of young-
sters who will eventually fail to make the grade as professionals; while the over-32 age
band contains older professionals most of whom are approaching retirement.
8.2 International migration of professional footballers

The migration of professional sports labour at international level, both in foot-
ball and in other sports, is examined by Maguire (1994, 2004), Maguire and Stead

1997–2001

Not Not

1997: T1 .286 .178 .084 .093 .359 .476 .238 .056 .040 .190
T2 .076 .209 .174 .113 .429 .066 .375 .243 .112 .204
T3 .028 .138 .177 .175 .482 .019 .196 .253 .259 .272
T4 .008 .027 .105 .186 .674 .000 .016 .071 .294 .619
All .101 .140 .135 .140 .483 .130 .214 .165 .178 .313
1997: T1 .189 .121 .094 .138 .458 .304 .304 .089 .025 .278
T2 .074 .161 .148 .109 .509 .032 .119 .190 .167 .492
T3 .046 .120 .137 .103 .594 .010 .070 .140 .190 .590
T4 .017 .022 .117 .100 .744 .000 .012 .070 .198 .721
All .095 .111 .121 .116 .557 .074 .120 .130 .151 .524
1997: T1 .338 .200 .110 .103 .248 .186 .093 .000 .023 .698
T2 .148 .265 .181 .103 .303 .015 .031 .062 .046 .846
T3 .035 .159 .194 .188 .424 .000 .071 .071 .071 .786
T4 .010 .057 .161 .238 .534 .000 .000 .027 .067 .907
All .121 .163 .163 .164 .389 .040 .040 .040 .053 .827
(1998), Maguire and Pearton (2000), McGovern (2002), Poli (2006, 2009) and Taylor
(2006). Section 8.2 reviews the main findings of this body of work. For economic
historians and sociologists, patterns of international migration of football players
are ‘socially embedded’. The transactions of football clubs in the transfer market
are heavily influenced by cultural and social ties, and by established historical and
economic relationships. According to McGovern (2002), for example, migration
patterns are mediated through social structures in ways that cannot be explained by
a pure ‘market’ approach to economic action. Taylor (2006, p19) adds:
it is difficult to deny that the ‘historical and cultural roots’ so often alluded to but rarely exam-
ined in detail continue to underpin many of the contemporary systems and networks of football
player migration. Indeed it remains clear that where these players choose to go – and where the
clubs decide to look for players – is not indiscriminate, but often determined by long-established
colonial, cultural, linguistic, social and personal connections.
Maguire (1994, 2004) and Maguire and Stead (1998) also acknowledge the
importance of social, historical and cultural influences, but emphasise economic
factors. Patterns of migration of professional sports players across national bor-
ders reflect the following influences:

2001–2005

Not Not

2001: T1 .227 .184 .118 .090 .381 .425 .310 .069 .034 .161
T2 .091 .180 .117 .125 .486 .125 .289 .141 .164 .281
T3 .013 .104 .167 .141 .575 .008 .147 .171 .163 .512
T4 .001 .058 .074 .152 .715 .000 .069 .130 .145 .656
All .083 .133 .119 .127 .538 .114 .194 .133 .135 .425
2001: T1 .154 .112 .133 .112 .489 .273 .312 .130 .052 .234
T2 .057 .141 .094 .094 .613 .055 .188 .156 .125 .477
T3 .016 .110 .115 .157 .602 .000 .072 .134 .113 .680
T4 .000 .077 .044 .160 .718 .000 .009 .026 .132 .833
All .071 .114 .102 .124 .589 .067 .135 .111 .111 .577
2001: T1 .311 .197 .129 .114 .250 .048 .214 .048 .024 .667
T2 .176 .200 .121 .176 .327 .034 .069 .086 .052 .759
T3 .025 .118 .241 .153 .463 .000 .000 .127 .048 .825
T4 .006 .088 .123 .222 .561 .000 .000 .012 .025 .963
All .113 .146 .159 .168 .413 .016 .053 .066 .037 .828
• The degree of geographical proximity between countries and the ease of travel;
• The residual impact of historical imperial or colonial ties;
• Countries’ attitudes towards their own nationals seeking employment oppor-
tunities abroad;
• Countries’ treatment of foreign nationals seeking employment opportunities
within their own boundaries;
• Salary differentials, offering players the opportunity to increase their earnings
by playing abroad, or (the other side of the same coin) clubs the opportunity to
recruit talented players more easily or cheaply than is possible in the domestic
transfer market;
• The reputation, status and characteristics (including, for example, tactics,
styles of play or the strength of physical competition) of the sport in different
countries;
• The extent to which media exposure raises interest and awareness in the sport
across countries;
• Inter-personal links, which might influence players from the same nation to play
in the same foreign country or join the same club simultaneously.

2005–2009

Not Not

2005: T1 .298 .210 .094 .066 .332 .519 .291 .051 .038 .101
T2 .076 .211 .141 .099 .473 .108 .342 .225 .045 .279
T3 .030 .091 .141 .173 .566 .040 .095 .190 .230 .444
T4 .008 .054 .131 .135 .672 .000 .043 .177 .163 .617
All .088 .137 .129 .122 .523 .127 .173 .171 .131 .398
2005: T1 .195 .170 .130 .075 .430 .182 .236 .091 .091 .400
T2 .065 .159 .107 .079 .589 .042 .146 .115 .177 .521
T3 .030 .086 .137 .168 .579 .028 .014 .056 .169 .732
T4 .025 .063 .107 .069 .736 .000 .000 .053 .200 .747
All .082 .123 .121 .099 .575 .054 .094 .081 .165 .606
2005: T1 .475 .242 .071 .061 .152 .059 .118 .059 .059 .706
T2 .115 .309 .176 .139 .261 .014 .028 .069 .042 .847
T3 .028 .156 .194 .194 .428 .015 .000 .000 .030 .955
T4 .005 .092 .175 .175 .553 .000 .000 .032 .032 .935
All .111 .188 .165 .154 .383 .017 .026 .038 .038 .880
Based upon an analysis of the nationalities of all players participating in the

highest tiers of thirty-two European countries in the 1993 season, Maguire and
Stead (1998) report that 1,295 players, or 11.4 per cent of the total workforce, were
employed by clubs located in countries other than their country of nationality. Of
these, the proportions from non-European countries, non-EU European coun-
tries and EU countries were 27 per cent, 54 per cent and 19 per cent, respectively.
Flows of players into and out of Europe were concentrated most heavily on Latin
American countries. In 1994, there were 297 transfers (at all levels) involving moves
from non-European to European countries. Of these, 46 per cent were from Latin
America, 22 per cent from North and Central America and the Caribbean, 20 per
cent from Asia, 7 per cent from Africa and 5 per cent from Oceania. European
countries located around the Mediterranean rim were the principal destinations
for players imported from Latin America.
Intra-European trade in players was dominated by flows from Eastern Europe
into the EU. According to Maguire and Stead, 47.3 per cent of all non-EU play-
ers playing in the EU were from Eastern Europe. Such flows increased significantly
following the collapse of communism at the end of the 1980s. Previously the majority
of former-communist countries permitted only a handful of players to move to the
west, and in most cases not until the latter stages of their careers. During the 1990s,
apart from the ‘pull’ factor of the higher earnings available in Western Europe, sig-
nificant ‘push’ factors include the brutal ethnic conflicts in the former Yugoslav ter-
ritories, and political and economic turmoil elsewhere, mirrored at the sporting level
by the disappearance or wholesale reconstruction of various leagues and individual
clubs. Players from Croatia and the rump Yugoslav Federation formed the two big-
gest groups among non-EU Europeans playing in the EU in 1993. Their principal
destinations (in descending order) were Greece, Spain, Austria and Portugal.
One of the most striking aspects of Maguire and Stead’s analysis of the intra-
European player flows is the imbalance between the number of players moving
from north to south and the number moving in the opposite direction. In the
1995 season, there were twenty-nine northern Europeans playing in the south, but
only five from southern Europe playing in the north. Speculating on the causes of
this asymmetry, Maguire and Stead suggest that the (actual or perceived) profes-
sionalism, adaptability, language competence and high educational standards of
Dutch and Scandinavian players, together with their physical attributes, tend to
make them attractive targets for leading clubs in any country. In contrast, north-
ern European perceptions of southern Europeans tend to focus on concerns about
temperament or the ability to adapt to the physical demands of the northern style
of play. Whether such attributes are imagined or real is beside the point, since it is
the perception that counts in determining clubs’ recruitment policies.
Poli (2006) examines the geographical origin of migrant footballers to leagues
under the jurisdiction of UEFA using data for the 2003 season. Eastern Europe was
the largest originator of migrant footballers (29.7 per cent), followed by Western
Europe (28.7 per cent), Africa (19.6 per cent) and Latin America (16.9 per cent).
The largest exporter country was Brazil (9.5 per cent of all migrants in UEFA coun-
tries). After Brazil, other prominent exporter countries were Serbia-Montenegro,
France, Argentina, Nigeria, Ukraine, Croatia and Cameroon. The largest importer
of migrant footballers was England: 718 migrants (13.5 per cent of all migrants in
UEFA countries) played for English clubs. After England, other prominent importer
countries were Germany, Italy, Portugal, Belgium, France and Spain.
The recruitment of foreign players is influenced by the standard of football. This
relationship also depends upon the continent of origin. In a five-level hierarchy of
European leagues based upon UEFA rankings, the proportion of migrant foot-
ballers from Latin America decreases from the top level to the bottom. However,
migrant footballers from Africa are more heavily represented in the lower four
levels than in the highest level. For example, 53.3 per cent of migrant footballers
in Romania were from Africa. For Malta, Belgium, Switzerland and Albania, the
corresponding figures were 52.6 per cent, 43.4 per cent, 33.7 per cent and 33.3 per
cent. These data reflect a need for the less affluent clubs to recruit foreign talent
from relatively low-cost sources (Poli, 2006).
Poli (2009) reports that the presence of foreign players in the major European
leagues increased significantly between 1996 and 2006. The percentage of non-EU
players has increased both with respect to the total number of players and rela-
tive to the total number of foreign players. In the 2006 season, 38.4 per cent of
players in the big-five European leagues, England, France, Germany, Italy and
Spain, were migrants (20.2 per cent in the 1996 season). The concentration of for-
eign players tends to be heaviest with the top-ranked teams in each league. In the
2008 season, 52.6 per cent of players employed by teams ranked in the top five in
their domestic leagues were foreign. For middle-ranked teams and bottom-ranked
teams, the corresponding figures were 37.1 per cent and 42.2 per cent, respectively.
Darby and Solberg (2010) and Poli (2010) present further evidence on the experi-
ence of African migrant footballers.
The distribution of foreign-born players by origin in the 2006 season varied
widely among the big-five leagues. For Italy and Spain, the pattern was similar.
The proportions of foreign-born players from Latin America were 50.8 per cent
for Italy and 62.2 per cent for Spain. In both cases, Western Europe was the next
highest source (20.2 per cent for Italy and 25.5 per cent for Spain). For Germany,
the two principal sources were Western Europe (34.9 per cent) and Eastern Europe
(33.1 per cent). For France, the two main sources were Africa (48.3 per cent) and
Latin America (23.3 per cent). For England, the two main sources were Western
Europe (63.6 per cent) and Africa (10.7 per cent).
Tables 8.14 and 8.15 present an analysis of the home countries and countries of
employment of the members of the sixteen national team squads which took part
in the finals of the 2000 and 2008 European Championships (Euro 2000 and Euro
2008), staged jointly by Belgium and the Netherlands in 2000, and by Austria and
Switzerland in 2008. The analysis, along the same lines as that of the 1998 World
Cup reported by Maguire and Stead (2000), provides a snapshot of patterns of
intra-European player migration at the highest level. At Euro 2000 22 players were
permitted in each squad, or 352 players in total. At Euro 2008 the squad sizes were
increased to 23 (368 players in total). In Euro 2000, 177 players were foreign-based
(50.4 per cent of the total). In Euro 2008 this proportion was slightly lower: 176
players were foreign-based (47.8 per cent of the total).
The columns of Tables 8.14 and 8.15 show the sixteen participating countries,
arranged from left to right in ascending order of the numbers of foreign-based
players included in their squads, shown in the final row. In 2008, for example,
Russia selected a squad in which only one player was foreign-based. At the other
extreme, the Czech Republic selected twenty-one foreign-based players. The rows
of Tables 8.14 and 8.15 show the countries of employment of the squad mem-
bers of each country. The first sixteen rows report the data for the participating
countries in each tournament; the remaining rows show players employed in non-
participating countries. Each group of countries is arranged in descending order
of the total numbers of imported players employed, shown in the final column.
The numbers of home-based players in each squad, shown in italics in the relevant
cells of Tables 8.14 and 8.15, are not included in the final row and column totals.
Table 8.14 Cross-tabulation of squad players’ home countries and countries of employment, Euro 2000
Home country
Country of
employment Ita Spa Eng Tur Ger Por Bel Cze Rom Nor Fra Slov Swe Net Yug Den All
England 21 1 2 1 2 3 1 9 6 4 5 5 39
Spain 22 1 1 4 1 6 1 2 3 7 7 33
Italy 22 1 4 1 2 1 5 3 3 5 4 29
Germany 18 4 5 1 2 3 2 1 4 22
Netherlands 3 2 5 2 3 10
Belgium 1 10 1 1 1 3 7
France 1 1 1 7 1 1 5
Portugal 11 1 1 1 1 4
Turkey 19 3 3
Denmark 1 2 3 3
Norway 7 1 1
Yugoslavia 1 4 1
Czech Rep 8 0
Romania 7 0
Sweden 5 0
Slovenia 6 0
Scotland 1 1 2 2 6
Austria 1 5 6
Greece 1 1 1 3
US 1 1 2
Croatia 1 1
Switzerland 1 1
Japan 1 1
All 0 0 1 3 4 11 12 14 15 15 15 16 17 17 18 19 177
Source: World Soccer (June 2000)

Table 8.15 Cross-tabulation of squad players’ home countries and countries of employment, Euro 2008
Home country
Country of
employment Rus Ger Spa Ita Tur Aus Gre Rom Por Fra Pol Net Swit Swe Cro Cze All
Germany 1 19 1 1 1 4 1 2 2 2 3 6 1 6 5 36
Spain 2 18 3 1 2 1 5 3 1 3 1 22
Italy 18 2 3 2 2 1 3 4 17
France 1 1 2 10 4 4 3 1 16
Russia 22 1 1 1 1 2 1 7
Greece 1 1 14 2 1 1 6
Austria 15 1 1 1 3
Portugal 1 1 12 2
Romania 12 1 1 2
Netherlands 9 1 1
Turkey 16 1 1
Switzerland 7 1 1
Czech Rep 2 0
Sweden 6 0
Poland 9 0
Croatia 3 0
England 2 5 2 1 2 4 6 3 7 3 5 2 4 46
Belgium 1 3 4
Scotland 1 1 1 3
Ukraine 1 1 1 3
Denmark 1 2 3
Norway 1 1 2
Bulgaria 1 1
All 1 4 5 5 7 8 9 11 11 13 14 14 16 17 20 21 176
Source: www.thebesteleven.com/2008/06/club-breakdown-for-euro-2008-squads.html
Based on the data for both tournaments, England (a non-participant in 2008)

was by some distance the leading importer of players at this level. This position
reflects the financial rewards the leading English clubs are able to offer the leading
players, as a result of the commercial vibrancy of the Premier League. The number
of foreign-based players based in the English league who participated in the finals
increased from thirty-nine for Euro 2000 to forty-six in Euro 2008. The equivalent
figure for Germany also increased, from twenty-two to thirty-six; but the figures
for Spain and Italy fell, from thirty-three to twenty-two (Spain), and from twenty-
nine to seventeen (Italy). In Euro 2000, all members of the Spanish and Italian
squads were home-based; in the case of the latter, the absence of foreign-based
players appears to have been a conscious policy decision, and may not reflect a
complete lack of suitably skilled foreign-based talent. The tendency for the major-
ity of the leading Spanish and Italian players to remain home-based is reflected
in the Euro 2008 data as well; although by then five Spanish squad members were
employed by teams in the English Premier League, and five Italian squad members
were based in Spain, Germany or France.
England’s apparent ascendancy within Europe as an importer of footballers is
a relatively recent phenomenon, but several longer established features of intra-
European migration patterns are also apparent in Tables 8.14 and 8.15. These
include the popularity of England as a destination for Swedes and the Dutch; the
ability of Swedes to find employment across a wide range of European countries;
and the importance of geographical proximity or historical links in determining
migration flows, especially for players from nations with less economically power-
ful domestic leagues. In Euro 2008 Switzerland, for example, had six squad mem-
bers based in Germany, and the Czech Republic had five; while Portugal had five
squad members based in Spain. Finally, it is interesting to note that the two par-
ticipant countries whose domestic leagues hosted the highest numbers of foreign-
based players from other Euro 2008 squads, Germany and Spain, were also the
two finalists of this tournament.
Using data similar to those presented in Tables 8.14 and 8.15, Frick (2009)
presents an econometric analysis of the impact of the Bosman ruling on player
migration, using data on the percentage of players under contract in a foreign
country at the time of a World Cup or European Championship. To explain the
variation in this percentage he estimates a regression model with a Bosman dummy
(before 1997 = 0; after 1997 = 1) and random country effects. After controlling
for the tournament (World Cup or European Championship) and for a (linear)
time trend, the Bosman ruling is found to have produced a statistically significant
increase in the percentage of squad players under contract in a foreign country.
The time trend is also statistically significant, indicating that the Bosman ruling
had the predicted positive impact on player migration.
Frick (2009) also analyses the impact of the ‘internationalisation’ of national
teams on their performance. The playing strength of the national teams is measured
using their performance in World Cup and European Championship tournaments
staged between 1978 and 2006. A binary team performance measure (based on
qualification for the semi-finals or the final of the tournament) is explained by
the (normalised) percentage of players under contract abroad at the time of the
tournament, with controls for the tournament (as above) and the number of par-
ticipating teams. The estimation method is random effects logit. The probability
of reaching the semi-final or the final of either tournament is unaffected by the
percentage of players employed abroad. According to Frick (2009 p 96)
the increasing competitive pressure exerted on players migrating from Eastern Europe, Africa,
Asia, and South America to the better paying leagues in Western Europe has not yet resulted in
a better performance of their national teams
Similarly, Baur and Lehmann (2008) analyse whether a country’s success is

influenced by the number of national team players that do not play in the home
league and by the number of national team players from other countries that play
in the home league, using data for all thirty-two national teams that qualified for
the 2006 World Cup. Both the number of players that play in a foreign league
(exports) and the number of foreign players that play in the home league (imports)
are positively associated with the country’s FIFA ranking. This suggests that all
countries benefit from trade. For example, Brazil and Italy are both highly ranked,
even though Brazil is a net exporter and Italy is a net importer of footballers.
8.3 Racial discrimination in professional team sports

The investigation of discrimination in professional team sports has motivated many
empirical studies of the determinants of player compensation, especially in the
US. This issue has been of particular interest in North American sports. Before the
Second World War, racial segregation at professional level operated in a number of
sports: the teams in the top leagues employed only white players. The colour bar
was lifted for the first time in both baseball and basketball during the late 1940s, but
there is no doubt that discriminatory practices continued for many years afterwards.
Empirical evidence on the extent of racial discrimination in professional team sports
outside North America is more limited, mainly because data on the remuneration of
individual professional sports players are, for the most part, not publicly available.
In professional team sports, the forms that discrimination can take include
inequality in the level of compensation; inequality in hiring standards; inequal-
ity in the allocation of playing positions; inequality in the prospects for reten-
tion or career progression; and inequality in the availability of opportunities for
endorsements or sponsorship. Possible perpetrators of discrimination are employ-
ers, co-workers or customers. Although in principle it is possible to distinguish
between the different forms and sources of discrimination, in practice the bound-
aries may be blurred. A derisory salary offer, for example, may be tantamount
to a refusal to hire. An employer practising hiring or salary discrimination may
attempt to attribute blame to the attitudes of customers.
Racial discrimination in professional team sports 237
Becker (1971) argues that eventually, economic competition between teams

as employers should eliminate employer discrimination. Teams that discrim-
inate should be forced out of business because they are inefficient. In contrast,
co-worker discrimination should result in equally competitive segregated teams.
Using the ‘joint economies of scale in consumption’ argument (see Chapter 7,
Section 7.2), Rosen and Sanderson (2001) suggest that customer discrimination is
likely to have especially adverse consequences for black athletes. Each customer
may demonstrate only a trivial prejudice, reflected in a small reduction in the price
they are willing to pay to watch a team containing black athletes. But from the
team’s perspective, the total loss of revenue that has to be compensated (by paying
black athletes a salary that reflects customer prejudices) may be huge, because of
the large audience sizes involved. In this case a feasible salary may not exist, and
the black athlete is effectively forced out of the market altogether.
Empirical evidence on racial discrimination: North American professional team sports

Kahn (1991, 2000) reviews a substantial body of empirical literature on discrim-
ination in professional team sports. Empirical tests for salary discrimination
normally use a cross-sectional regression of individual salaries on various prod-
uctivity indicators, and a dummy variable for race or any other characteristic that
is thought to be the subject of discrimination. The availability of accurate prod-
uctivity indicators is vital, since as Kahn points out, any correlation between race
and the measurement error in the productivity component will lead to bias in the
estimated discrimination effect.
Another key issue in detecting discrimination in salary equations concerns the
skewness that is often present in athletes’ salary data. If the salary distribution
for blacks and whites reveals significant disparity in the compensation of each
group above or below the mean, an estimation method that focuses on the condi-
tional mean may misrepresent the relationship between salary and race. Quantile
regression permits the impact of race to be established at different points of the
salary distribution, not just at the mean (Hamilton, 1997; Berri and Simmons,
2009).
For the NBA, there is extensive evidence to suggest that prior to the 1990s,
black players were paid lower salaries than white players at comparable levels of
productivity (Kahn and Sherer, 1988; Koch and Vander Hill, 1988; Brown, Spiro
and Keenan 1991). Using NBA data for the 1995 season, Hamilton (1997) finds
evidence of a 1 per cent across-the-board salary differential between white and
black players. At the lower end of the salary range, whites earned less than blacks;
though when the number of matches is included in the regressions as a control,
the effect reverses. At the upper end of the salary range, whites earned more than
equally qualified blacks. The premium was the same regardless of the race of
the team’s manager, and so is attributed by Hamilton to customer rather than
employer discrimination.
Bodvarsson and Brastow (1998) use salary regressions to test for employer dis-
crimination in the NBA in the 1986 and 1991 seasons. Again the race of the
team’s manager is assumed to convey information about team owners’ atti-
tudes: owners that employ black managers are assumed to be non-discriminators.
The test therefore examines whether black players working for white managers
experienced discrimination. The salary regressions include the percentages of
team members and of the local population that were black, as controls for co-
worker and customer discrimination respectively. Salary discrimination, present
in 1986, had disappeared by 1991. This finding is consistent with the hypoth-
esis that two important structural changes in the interim increased competition
between teams as employers, and reduced the discretion for team owners to dis-
criminate as monopsony purchasers in the labour market. These were the 1988
NBA Collective Bargaining Agreement, which introduced free agency; and the
addition of four new teams, which stimulated competition between teams as
employers of players.
By contrast, Jenkins (1996), Dey (1997), Gius and Johnson (1998) and Eschker,
Perez and Siegler (2004) find little evidence of salary discrimination in the NBA.
Berri and Simmons (2009) find evidence of performance-related salary discrim-
ination against black quarterbacks in the top half of the salary distribution in
the NFL. Several studies of discrimination against Francophone players in the
NHL offer mixed results (Jones, Nadeau and Walsh, 1999; Lavoie, 2000; Curme
and Daugherty, 2004). Speaking generally, it is apparent that evidence of discrim-
ination in North American major league professional team sports has become
weaker over time (Rosen and Sanderson, 2001).
Direct tests for discrimination in hiring in North American professional team
sports are less common than tests for salary discrimination, primarily because of
the difficulties in obtaining data on players who were not hired, as well those who
were. Indirect testing, mostly based on comparisons between the performance of
black and white players, is more common. If blacks tend to outperform whites on
average, this is interpreted as evidence of discrimination in hiring: on average a
black needs to attain a higher standard to be hired than a white. Similar methods
can be used to test for discrimination in the allocation of specific team positions
between whites and blacks.
Empirical studies of hiring discrimination have reported mixed findings.
Bellemore (2001) reports evidence of discrimination against black players in pro-
motion from minor league baseball to MLB between the 1960s and 1990s. The
level of discrimination was lower in years when the number of major league teams
was increased. Discrimination against Hispanics was evident at the start of this
period, but had disappeared by the end. McCormick and Tollison (2001) report
that black players in the NBA were assigned more playing time than comparable
white players. Burdekin, Hossfield and Kilholm-Smith (2005) find evidence of a
correspondence between the racial composition of NBA teams and of the popula-
tions of the cities in which the teams were located.
Jiobu (1988) reports evidence of retention discrimination in MLB during the

1970s and early 1980s, in the form of a tendency for team owners to retain under-
performing white players for longer than black players. Similarly for the NBA,
Hoang and Rascher (1999) report evidence of retention discrimination favouring
white players. Using more recent MLB data, however, Groothuis and Hill (2008)
find no link between race and career duration.
Finally, tests for customer discrimination normally focus on the relationship
between match attendances or revenues and team racial composition. The use of
attendance data can be problematic, because many teams routinely sell all of their
seats, especially in the NFL and NBA. However, accounting for the truncation of
the data, Berri, Schmidt and Brook (2004) find that customer discrimination was
widely prevalent in the NBA in the 1980s, but less so subsequently. Foley and Smith
(2007) suggest that MLB attendances in the cities of Boston, Cleveland, Houston,
San Diego and Saint Louis were reduced when the home team added Hispanic play-
ers to its line-up. Tainsky and Winfree (2010) examine the impact of foreign-born
players on MLB attendances. Using TV audience data for televised NBA matches,
Kanazawa and Funk (2001) find that the number of white players in both the home
and visiting teams influenced ratings. Hanssen and Andersen (1999) and Depken
and Ford (2006) examine the voting records for baseball’s All-Star game. There is
evidence of racial discrimination in the 1970s, but not subsequently. Todd, Brown
and Miles (2002) and Todd (2003) find that race did not influence a player’s pros-
pects for induction into the National Baseball Hall of Fame between the 1960s and
1990s.
Empirical evidence on racial discrimination: English and European football

Szymanski (2000) reports the first empirical test for employer discrimination
against black players in English football. Although football has never operated
a formal colour bar, instances of black players appearing in the Football League
before the 1970s were isolated. There is abundant anecdotal or circumstantial evi-
dence of discrimination against black players throughout the 1970s and 1980s,
on the part of some club officials as well as spectators. As Szymanski reports,
the rate at which first-team opportunities opened up for black players appears to
have varied between clubs, to a degree that seems difficult to attribute to chance.
Since the 1980s the situation appears to have improved, with crowd abuse of black
players becoming a rarer phenomenon than it was in earlier times. Even recently,
however, there is no shortage of anecdotal evidence to suggest that the longstand-
ing tradition of racism in English football is not completely defunct. Sir Norman
Chester Centre for Football Research (2002) notes a catalogue of occasions when
senior officials and football club managers have referred to black players as ‘lack-
ing bottle’, ‘having no stamina’ and ‘exhibiting a lack of discipline and consist-
ency’. In 2004 Ron Atkinson, a high-profile former football manager, left his job
as a TV commentator after a racist comment about Chelsea’s Marcel Desailly was
broadcast on a live feed to several Middle East TV channels while the UK trans-
mission was off-air (Goddard and Wilson, 2009).
In the Szymanski (2000) study, the difficulties created by the non-availability of
individual salary data for football players are circumvented by testing for discrim-
ination using data at team rather than at individual level. Total wages and salaries
per team are available from annual football club company accounts. Using data on
thirty-nine clubs for which complete data were available for the period 1978–93,
Szymanski finds that teams that used a below-average proportion of black players
achieved inferior playing performance (measured by league position) after control-
ling for salary expenditure and for the total number of players used in each season.
Perhaps surprisingly, this finding is stronger for the period 1986–93 than it is for
1978–85. The test, however, is more powerful for the later period because by then
more black players were employed. The evidence of discrimination is also stronger
for the larger clubs (measured by stadium capacity) than for the smaller ones.
How is the finding that teams that under-recruited black players also under-
performed on the pitch to be interpreted? It appears that by refraining from recruit-
ing the most talented players available for a given salary spend, team owners or
managers with a taste for discrimination indulged their preferences by accepting
a lower standard of team performance than they could otherwise have achieved.
The proportion of black players used turns out to be insignificant in attendance
or revenue regressions, suggesting that discrimination originates with the teams as
employers rather than with spectators as customers. A crucial assumption under-
lying the test is that the market for playing talent is efficient, so all teams recruit
from the same pool of players. This may rest uneasily with the evidence that player
recruitment is, in practice, geographically highly segmented (see Section 8.1).
Szymanski suggests that a relatively weak market for corporate control of football
clubs may explain why the more efficient, non-discriminating teams did not drive
the less efficient, discriminating ones out of business through competition.
In a separate study, Preston and Szymanski (2000) find no evidence of a link
between the selection of black players and match attendances, suggesting that
customer discrimination is not responsible for the racial influence on the sal-
ary expenditure-performance relationship. However, the ethnic composition of
the local population is a significant determinant of the team-level proportion of
appearances made by black players. It is not clear whether the racial factor in the
salary-performance relationship truly reflects employer discrimination, or a geo-
graphically segmented labour market, which confers advantages on teams able to
hire from an ethnically diverse pool of talent at local level. Pedace (2008) reports
evidence that English teams with more players of South American origin per-
formed less well than those with fewer such players. However, teams fielding South
American players also attracted higher attendances.
For countries other than England, Wilson and Ying (2003) present a analysis
constructed along similar lines to the Szymanski study, using data for the post-
Bosman period from the highest divisions of several top European leagues. It
is suggested team performance could be improved by hiring more players from

under-represented regions, such as eastern Europe and Latin America. There
is no evidence that spectators react adversely to the selection of non-domestic
playing talent. Frick (2006) estimates that German Bundesliga footballers from
other western European countries, from eastern Europe and from Latin America
received salary premiums of 15 per cent, 30 per cent and 50 per cent respectively,
relative to native players born in Germany. Frick’s (2007) survey article reports
little or no evidence of salary discrimination against foreign footballers in the
European leagues.
Using the Premier League and Football League retained list data published
annually in Rothmans Football Yearbook, Goddard and Wilson (2009) report an
analysis of hiring and retention discrimination in English football. Patterns of
career progression for white and black players are compared, by examining the
career progression of three cohorts of players over the five-year intervals 1986–91,
1991–6 and 1996–2001. The choice of sample period is dictated by the timing of
the publication in the Yearbook of playing-squad photographs of English league
clubs. Publication of these photos started in 1986, and ended in 1998. The photos
allow a binary classification based on skin colour of the large majority of profes-
sional footballers who were employed during this period.1 The same source was
also used in the earlier Szymanski (2000) study. The remainder of this section
describes the empirical model used in the Goddard and Wilson study in some
detail, and presents some of the estimation results.
The econometric model for the conditional probability that a professional
footballer is retained within the data set over a five-year period and, for those
that are retained, the conditional probability distribution for the possible transi-
tions in divisional status, comprises three equations. The dependent variables
are initial divisional status, retention and divisional transition. The dependent
variables in the initial divisional status and divisional transition equations can
each take one of four possible values, depending on the player’s divisional status
(T1, T2, T3 or T4) at the start and end of the five-year period. The specification
of these two equations is therefore analogous to ordered probit. The dependent
variable in the retention equation is binary, so the specification is analogous to
binary probit. Ignoring the interdependence between the three equations, the
specification of the ordered probit regression equation for initial divisional sta-
tus is as follows:

Probability (initial divisionalstatus is T1) = 1 − Φ(µ3 − x i,t )
Probability (initial divisional status is T2) = Φ(µ3 − x i,t ) − Φ(µ 2 − x i,t )
Probability (initial divisional status is T3) = Φ(µ 2 − x i,t ) − Φ(µ1 − x i,t )
Probability (initial divisional status is T4) = Φ(µ1 − x i,t ) [8.1]
The specification of the binary probit regression equation for retention is as

follows:
Probability (player is retained) = Φ(y i,t )

Probability (player is not retained) = 1 − Φ(y i,t ) [8.2]
Finally, the specification of the ordered probit regression equation for divisional
transition, conditional on initial divisional status = j for j = 4, … ,1 (j = 4 for T1;
j = 3 for T2; j = 2 for T3, j = 1 for T4), and the player being retained, is as follows:
Probability (final divisional status is T1) = 1 − Φ(δ j,3 −  j z i,t )

Probability (final divisional status is T2) = Φ(δ j,3 −  j z i,t ) − Φ(δ j,2 −  j z i,t )
Probability (final divisional status is T3) = Φ(δ j,2 −  j z i,t ) − Φ(δ j,1 −  j z i,t )
Probability (final divisional status is T4) = Φ(δ j,1 −  j z i,t ) [8.3]
In [8.1], [8.2] and [8.3] Φ is the cumulative Normal distribution function. xi,t,
yi,t and zi,t are vectors of covariates. α, β and γj are vectors of coefficients to be
estimated. β includes an intercept; α and γj have no intercept. In [8.1], μk (for k =
1,2,3) are cut-off parameters to be estimated. Equation [8.3] is parameterised so
that the divisional transition equation contains a separate set of coefficients and
cut-off parameters for each possible state of the initial divisional status variable.
δj,k (for j = 1,2,3,4 and k = 1,2,3) are cut-off parameters to be estimated.2
The covariates appearing in xi,t, yi,t and/or zi,t are as follows:
CLUBSEA = Total number of different teams the player has been employed by up
to and including his present club at the end-of-season census point, divided by
the number of times the player has appeared in the end-of-season census.
APSEALL = Total number of league appearances in the player’s career to date,
divided by the number of times the player has appeared in the end-of-season
census.
DIV18 = Divisional status (T1 = 4, T2 = 3, T3 = 2, T4 = 1) of the club the player
was employed by at the end-of-season census immediately following his eight-
eenth birthday.
DUM18 = 0–1 dummy variable identifying players not in the data set when aged
18 (for whom DIV18 = 0).
SCOTIRE = 0–1 dummy variable identifying players born in Scotland, Northern
Ireland or the Irish Republic.
O’SEAS = 0–1 dummy variable identifying players born overseas.
RACE = 0–1 dummy variable identifying non-white players.
AGE = Player’s age in years at the time of the end-of-season census.
AGESQ = Square of AGE.
DFR, MFR and FWD = 0–1 dummies identifying outfield playing positions
(defender, midfield and forward).
OUTF’LD = DFR + MFR + FWD.
T2, T3, T4 = 0–1 dummy variables identifying players’ initial divisional status.
The model is estimated separately over the three five-year intervals: 1986–91, 1991–6
and 1996–2001. The data set for each five-year interval includes those players
Table 8.16 Estimation results: initial divisional status
1986–1991 1991–1996 1996–2001
CLUBSEA −1.2254*** −1.3713*** −1.5602***

−7.53 −9.98 −11.48
DIV18 .2696*** .1875*** .1973***
7.72 5.90 6.20
DUM18 .8509*** .6031*** .5342***
6.82 5.44 4.98
SCOTIRE .3683*** .2964*** .5005***
4.11 3.58 5.29
O’SEAS .8647*** 1.2870*** 1.2039***
3.79 8.77 11.40
RACE .1313 .3387*** .2499***
1.06 3.84 3.28
O’SEAS × RACE −.5388 −1.4061*** −.2753
−1.45 −4.70 1.24
Notes: Table 8.16 reports the estimation results for α in [8.1].

*** = significantly different from zero, two-tail test, 1% level;
** = 5% level;
* = 10% level.
z-statistics for significance of estimated coefficients are reported in italics.
Source: Adapted from Goddard and Wilson (2009)
who were employed by a league club and aged over 21 at the start of the five-year
interval, and for whom the RACE variable is observed. Players aged under 21 are
excluded to permit the construction of covariates which proxy for player quality,
using data on the player’s divisional status at age 18 (see below). The estimation
results reported by Goddard and Wilson (2009) are summarised in Tables 8.16 (ini-
tial divisional status), 8.17 (retention) and 8.18 (divisional transition).
Positive and highly significant coefficients on DIV18 in [8.1] reflect the expected
association between natural talent and initial divisional status.3 Several covari-
ates play an important role in [8.2] or [8.3], but are unimportant and are there-
fore excluded from [8.1]. AGE and AGESQ, APPSEA, and the positional dummy
variables DFR, MFR and FWD are virtually irrelevant in [8.1]. Clubs in all four
tiers have similar proportions of old and young players, similar proportions of
experienced and inexperienced players, and similar proportions of players in each
playing position. All of these attributes, however, are highly significant determi-
nants of retention and divisional transition.
In [8.2] the dummy variables for T2, T3 and T4 control for the tendency (also
apparent in Tables 8.9–8.13) for the retention probabilities to vary with divisional
status. The parameterisation of [8.3], which allows separate equations for each of
the four possible values of initial divisional status, effectively interacts the T2, T3
Table 8.17 Estimation results: retention
1986–1991 1991–1996 1996–2001
APPSEA .0307*** .0657*** .0571***

6.67 14.6 13.93
AGE .0038 −.1904* .4300***
0.02 −1.76 4.33
AGESQ −.0035 −.0002 −.0113***
−1.25 −0.11 −6.28
DFR −.4066*** −.7483*** −.4228***
−2.99 −5.82 −3.64
MFR −.3290** −.8297*** −.4832***
−2.35 −6.34 −4.12
FWD −.6384*** −.8221*** −.5486***
−4.54 −6.21 −4.61
SCOTIRE −.1147 −.4560*** −.2319*
−1.01 −4.01 −1.91
O’SEAS −.5788** −.2386 −.6978***
2.15 −1.35 −5.51
RACE .3191* .4271*** −.0382
1.95 3.54 −0.39
O’SEAS × RACE .1378 −.5292 .1703
0.31 −1.35 0.61
T2 −.1513 −.4822*** −.2669***
−1.42 −4.99 −3.04
T3 −.5608*** −.8402*** −.5014***
−5.28 −8.14 −5.43
T4 −1.0975*** −1.5957*** −.9183***
−9.81 −13.78 −9.45
Constant 2.7861 5.3444*** −3.4855***
1.32 3.61 −2.57
Notes: Table 8.16 reports the estimation results for β in [8.2].

** = 5% level;
* = 10% level.
and T4 dummy variables with each of the other covariates, allowing maximum
flexibility in the specification of the divisional transition probabilities. Several cov-
ariates (AGESQ, SCOTIRE, O’SEAS and O’SEAS×RACE) are excluded from
the reported versions of [8.3], and the three positional dummies DFR, MFR and
FWD are consolidated into a single outfield dummy, OUTF’LD.
In [8.2], APPSEA is a significant predictor of a player’s prospects for remaining
in employment. Naturally, players who have made regular first-team appearances
throughout their careers are more likely to be retained than those who have not.
Table 8.18 Estimation results: divisional transition
T1 T2 T3 T4
1986–1991
APPSEA .0343*** .0286*** .0023 .0367***
4.08 3.29 0.21 3.06
AGE −.0783*** −.1396*** −.1332*** −.1904***
−2.71 −5.00 −4.84 −3.02
OUTF’LD .0906 −.4601* .1702 −.0971
0.47 −1.96 0.59 −0.28
RACE .2484 .6796*** .0329 .1277
1.08 2.60 0.10 0.40
1991–1996
APPSEA .0316*** .0259*** .0087 .0364**
3.95 3.55 0.96 1.93
AGE −.0551*** −.0903*** −.1432*** −.1758***
−2.71 −4.68 −4.43 −3.31
OUTF’LD −.2501 −.3757* −.0951 −.6666*
−1.31 −1.92 −0.38 −1.85
RACE .1800 .0200 −.1630 .5797*
1.00 0.11 −0.70 1.82
1996–2001
APPSEA .0381*** .0345*** .0210** .0181
3.93 4.21 2.36 1.45
AGE −.0349 −.1374*** −.1332*** −.1202***
−1.30 −5.73 −4.53 −3.30
OUTF’LD −.0817 −.6604* −.3410 −.4361
−0.39 −1.75 −1.53 −1.15
RACE −.1511 −.0307 −.0996 .3035
−0.91 −0.11 −0.43 1.12
Notes: Table 8.16 reports the estimation results for γj in [8.3].

** = 5% level;
* = 10% level.
APPSEA is also significant in most of the estimations of [8.3]. The inclusion of

AGE and AGESQ in [8.2] allows for a possible quadratic relationship between age
and retention probability. There is some variation in the signs and magnitude of
these coefficients between the three estimations, but in all cases there is a negative
age-retention gradient, which becomes steeper as age increases. In most of the
estimations of [8.3], AGE has negative and significant coefficients, reflecting the
tendency for older players to accept a reduction in divisional status in order to
prolong their professional careers.
The coefficients on the playing position dummies DFR, MFR and FWD are all
negative and significant in [8.2], reflecting the tendency for outfield players’ careers
to be of shorter duration than those of goalkeepers. While playing position has a
large impact on the retention probability, among those players who are retained it
has little further effect on divisional transition.
The coefficients on the birthplace dummies SCOTIRE and O’SEAS in [8.1] sug-
gest a tendency for players born outside England and Wales to gravitate towards
higher-tier clubs (see Section 8.1). The coefficients in [8.2] are negative and in sev-
eral cases significant. Employment with an English club may represent a temporary
phase in the careers of many players born outside England and Wales, who are less
likely to be retained over any five-year interval than their locally born counterparts.
Significant coefficients on O’SEAS in the 1986–91 and 1991–6 estimations of
[8.1] may reflect the fact that pre-Bosman, overseas players employed in the English
league were few in number but typically high-profile. In contrast the insignificant
1996–2001 coefficient reflects much-increased post-Bosman overseas representa-
tion at all levels. The relatively large coefficient in the 1996–2001 estimation of
[8.2] may be attributable (as before) to reduced mobility barriers post-Bosman.
The coefficients on RACE in the estimations of [8.1] for 1991–6 and 1996–2001
are positive and significant. The coefficient for 1986–91 (reflecting the experience
of a smaller number of black players) is insignificant. These findings are similar
to those for the SCOTIRE and O’SEAS dummies, but suggest another, perhaps
less benign, type of selection effect. Black players who appear in the league tend
to gravitate towards teams of higher divisional status and are therefore of above
average talent. This is suggestive of a form of hiring discrimination, affecting
opportunities for black players to become professionals. While the most talented
black players are able to overcome the barriers, the less talented are less likely to
become professionals than their (equally talented) white counterparts (Goddard
and Wilson, 2009). Possible explanations for a selection effect of this kind are
discussed below.
The 1986–91 and 1991–6 coefficients on RACE in [8.2] are positive and signifi-
cant, while the 1996–2001 coefficient is insignificant. The 1986–91 coefficients on
RACE in [8.3] are positive throughout, and significant in the equation for players
starting in T2. For 1991–6 three of the four coefficients are positive, and one is sig-
nificant (at the 10 per cent level). For 1996–2001 three of the four coefficients are
negative and none is significant. Allowing for the smaller numbers of observations
used to estimate the equations in [8.3], these results seem consistent with a pattern
of selection bias associated with race, perhaps diminishing over time.
Negative coefficients on an O’SEAS×RACE interaction in [8.1], one of which
is significant and the other two narrowly short of being so, suggest hiring discrim-
ination appears to affect black UK and Irish footballers more than their overseas
counterparts. Most overseas players who appear in the English league are established
professionals, whose ability to perform at first-team level is already revealed at the
point of hiring. In contrast, young indigenous players undergo a lengthy period of
training and development before their ability to perform at first-team level is revealed,
leaving greater scope for prejudicial attitudes on the part of team owners, managers,
coaches or scouts to be obscured by uncertainty (Conlin and Emerson, 2006). The
O’SEAS×RACE interaction is insignificant, however, in both [8.2] and [8.3].
Uncertainty concerning players’ abilities at the point of hiring is not the only
conceivable source of a selection bias that has a greater impact on hiring deci-
sions for indigenous black players than on those for their overseas counterparts.
For example, Vasili (2000) argues that a decline in the provision of competitive
football in UK state schools since the 1980s has had a disproportionate effect on
the professional career prospects of youngsters from disadvantaged backgrounds,
including a high proportion of black youngsters. According to this interpretation,
the ultimate source of the selection bias lies beyond the confines of the profes-
sional sport altogether.
Alternatively, customer discrimination could underlie a selection effect. The geo-
graphical distribution of employment in football does tend to reflect local popula-
tion characteristics to some extent, presumably due to mobility barriers within the
players’ labour market. Higher-tier clubs are predominantly located in the larger
cities, most of which also have sizeable non-white populations, while many lower-
tier clubs are located in smaller provincial towns where the opposite is true.
The ultimate source of the selection effect affecting the recruitment of indigen-
ous black footballers is unresolved, but it is nevertheless apparent that the results
point in a similar direction to those of a number of other sports discrimination
studies. Football teams that discriminate against black players by failing to recruit
from the complete pool of available talent are likely to bear a cost in the form
of reduced performance (Szymanski, 2000). Several US studies report that black
athletes outperform their white counterparts on average (Kahn, 1991). This find-
ing is commonly interpreted as evidence of hiring barriers. The results reported by
Goddard and Wilson (2009) are consistent with such an effect.
Conclusion
Chapter 8 describes the employment patterns of professional footballers with
English clubs between the mid-1980s and 2009. During this period, the charac-
ter of English football was transformed by the arrival of a large contingent of
overseas players. While the employment of players born in the UK also increased
during the 1990s, it declined during the 2000s. Meanwhile the size of the overseas
contingent continued to increase relentlessly. During this period there appears
to have been a shift in the overall burden of responsibility for the development
of young players away from the smaller English clubs towards their larger coun-
terparts. Within the UK there are some sharp regional disparities in the poten-
tial for locally born youngsters to become professional footballers. Employment
prospects have improved for youngsters born in regions with a preponderance of
top-level professional clubs, such as London and the North West; but prospects
have diminished, in some cases substantially, for youngsters born elsewhere. The
employment prospects of youngsters born in certain regions, such as the North
East and Scotland, with a strong historical reputation for producing high-quality
footballing talent, have been especially hard hit.
Patterns of migration by footballers across national borders reflect a wide range
of influences, including: geographical proximity or distance between countries;
historical links between countries; home-country attitudes towards nationals seek-
ing employment abroad; host-country attitudes towards foreign nationals seeking
employment; salaries and remuneration; the standard and style of play of football
in the home and host countries; the level of media exposure football receives in
each country; and footballers’ personal relationships and social networks. Despite
the England national team having failed to qualify for the Finals of the 2008
European Championships, the English league was the largest single host-country
provider of footballers in the Finals. England’s ascendancy within Europe as an
importer of top-level footballers is a relatively recent phenomenon. The migra-
tion flows that have produced this situation, however, form part of a larger, more
complex and long-standing model of international mobility among professional
footballers throughout (and beyond) Europe.
Empirical evidence on racial discrimination in English football suggests a bias
in the employment of indigenous black footballers in favour of teams of higher
divisional status. This suggests that a form of hiring discrimination may affect
the opportunities for indigenous black players to become professionals. While the
most talented tend to be able to overcome whatever barriers exist, less talented
black footballers find it harder to do so than their (equally talented) white coun-
terparts. Prejudicial attitudes on the part of those making the hiring decision are
a possible source of an effect of this kind. Other sources might include a lack
of opportunities for youngsters from disadvantaged backgrounds to participate
in competitive football, and geographical variation in the ethnic composition of
the population combined with barriers to mobility in the footballers’ labour mar-
ket. Happily, the available evidence suggests that the extent of discrimination in
English football has diminished over time.
Notes
1 The conflation of skin colour with race is common practice in the economics literature on
racial discrimination.
2 The inclusion of the initial divisional status and retention equations controls for sample
selection effects that would otherwise bias the estimation of the divisional transition equa-
tion. If common unobservable factors affect both retention and transition, a single-equation
transition equation would be subject to selection bias (Heckman, 1979). Similarly, divisional
transition probabilities may depend on initial divisional status, but initial status and transi-
tion are also affected by common unobservables. A third set of common unobservables may
affect initial divisional status and retention. See Goddard and Wilson (2009) for full details
of the estimation procedure that corrects for these sample selection effects.
3 DUM18 identifies players not in the data set at age 18, for whom DIV18 is unavailable.
9 The football manager
Introduction
For many years economists have recognised the manager’s role in the production
process. In classical and neoclassical theory, individual firms and consumers are
the fundamental building blocks of the market economy. The theoretical distinc-
tion between the owners, entrepreneurs and managers of firms tends to be rather
blurred, because all are assumed to pursue the same objective of profit maximisa-
tion. In a highly influential contribution, Coase (1937) reinvented the theoretical
role of the firm, by asking why it is that in a free market economy, certain transac-
tions take place outside the domain of the market, within centrally planned and
hierarchical organisations known as firms. Coase’s answer was that for certain
types of transaction, the costs of gathering information and negotiating contracts
prohibit the use of market mechanisms; instead it is more efficient for such trans-
actions to be planned and coordinated consciously. The manager is the individual
within the firm who takes responsibility for this coordinating function.
In the context of team sports, whatever the outcome of the debate as to whether
the league or the individual club is the relevant unit of observation (Neale, 1964;
Sloane, 1971; see also Chapter 1), it is clear that the Coasian story has some merit
in explaining why sports team production is organised outside the market domain.
One can easily imagine that transactions costs would be prohibitive if each foot-
ball player had to enter into a network of bilateral contracts with ten other players
to form a team, and each team (or each set of individuals) had to contract bilat-
erally with the individuals in other teams to formulate a set of rules and produce a
series of fixtures. Section 9.1 discusses the historical development and present-day
characteristics of the football manager’s role as the principal planner and coord-
inator of team affairs.
According to the behavioural theory of the firm, the separation of ownership
from control is crucial to understanding the way modern corporations operate
(Berle and Means, 1932). Here the key insight is that the interests of the owner and
the manager may diverge, with implications for the behaviour and performance of
the firm. While the owner is concerned with profit maximisation, the manager may
249
pursue other objectives. In businesses generally, this conflict of interest is import-

ant because once hired, the manager cannot necessarily be monitored effectively.
If monitoring is imperfect, the manager may devote less than the maximum level
of effort towards attainment of the owner’s objectives: there is a moral hazard
problem (Alchian and Demsetz, 1972; Jensen and Meckling, 1976). Agency prob-
lems (in this case, managerial shirking) seem likely to be less severe in football than
elsewhere. The club’s owners are able to observe the performance of the team each
time a match is played, at least once a week. Monitoring is direct and regular, and
performance is transparent and easy to measure.
In professional football, if the output of the team (measured by its win ratio)
is below the maximum that should be attainable given the playing and financial
resources at the manager’s disposal, this could be because the manager is not exert-
ing maximum effort. Perhaps more likely, this could simply reflect bad ‘matching’,
caused by the owner not having full information about the manager’s character-
istics prior to hiring, or by the manager sending misleading signals during the
hiring process, overstating his true abilities. Production frontier estimation is an
empirical technique for the objective measurement of the manager’s performance
while controlling for team quality. Section 9.2 describes the production frontier
approach to measuring the managerial input, and reviews the literature on pro-
duction frontier estimation for professional sports teams.
While the measurement of managerial efficiency is an interesting exercise, it does
not necessarily capture every aspect of a manager’s performance. Many times, a
manager who was deemed a hero yesterday has been transformed into a villain
today. All football followers know that one of the defining characteristics of the
football manager’s position is its chronic insecurity. The rest of this chapter exam-
ines the relationship between managerial change and team performance. Section
9.3 reports tabulations describing a number of salient features of this relationship.
The factors that are critical in triggering the decision, taken either by the club or
by the manager himself, to terminate a manager’s appointment are investigated
more systematically in the next two sections. Section 9.4 reviews findings from the
previous academic literature on this topic, and Section 9.5 reports estimations of
empirical managerial job departure hazard functions for English football, using
match-level data.
Section 9.6 examines the relationship between managerial change and team
performance from the opposite direction, by considering the extent to which the
removal of a manager influences the team’s performance subsequently. According
to one view, it is simple common sense for a manager to be replaced when his team
is performing badly. Assuming that a more effective replacement can be appointed,
team performance should improve post-succession. An alternative view is that a
change of leadership is usually disruptive, especially if it occurs while the season
is in progress, and tends to make matters worse rather than better. A third view
suggests that on average, the appointment of a new manager has no effect what-
ever on team performance, because performance depends primarily on the playing
The role of the football manager 251
talent available to the team, and not on the managerial input. In this case man-
agerial dismissal is simply a form of scapegoating. Section 9.7 presents an empir-
ical analysis of managerial succession effects in English football.
9.1 The role of the football manager

Most football managers are former professional players. For many football man-
agers, a playing career is the only source of previous work experience; although
some enter management having served an informal apprenticeship as a coach or
assistant manager. Few football managers come into the job without some previ-
ous professional involvement in the sport. The careers of most professional players
start at age 16. Many players therefore have only limited educational attainment,
and few study in further or higher education. This is in marked contrast to the typ-
ical educational attainment of managers in other business sectors. Furthermore,
while the majority of managers in industry or commerce are likely to have had
formal training, only top-level football managers are required to hold a formal
qualification. Since 2003 UEFA has stipulated that anyone employed as a man-
ager in the top tier of any member nation’s league structure for more than twelve
weeks, and anyone who manages a team in European competition, must hold a
UEFA Pro-Licence.1
In the early days of professional football, the manager was the club secretary.
Secretary-managers, as they later became known, ran the club on a day-to-day
basis, while a trainer looked after the players. At this time the club secretary’s
role was limited. In the nineteenth and early twentieth centuries, responsibility for
team affairs rested primarily with the club’s directors and chairman. An ethos of
amateurism still prevailed, even within the burgeoning professional sport. In 1903,
for example, the directors of Newcastle United asked the senior players to select
the team. Most early secretary-managers were from administrative or accounting
backgrounds. They might have played football previously at some level, but they
were employed because they knew how to run things. Unlike football managers in
later periods, most were more highly educated than the players (Carter, 2006).
The modern-day football manager role as the person responsible for team affairs
first began to emerge during the inter-war period. During this period, it was quite
common for a manager to see his players on match-days only. During the rest
of the week, the manager would undertake scouting missions, watch other teams
play, and carry out administrative duties. Walvin (1994) identifies the growth of a
stronger ethos of professionalism within football during the 1920s and 1930s as
instrumental in encouraging directors to begin shifting responsibility for perform-
ance on the pitch towards the professional manager.2
Two individuals from the inter-war period, Herbert Chapman (Huddersfield
Town 1921–5, Arsenal 1925–34) and Major Frank Buckley (Wolverhampton
Wanderers 1927–44) typify the evolving new breed of professional, hands-on
team manager. Before Chapman’s arrival, Huddersfield and Arsenal were both
under-achieving clubs. Having enjoyed only a modest career as a player himself,

Chapman steered Huddersfield to the first two of three consecutive champion-
ships in 1924 and 1925, and then won further titles with Arsenal in 1931 and 1933.
The latter was the first of three consecutive championships; tragically, however,
Chapman died during the course of the 1934 season, aged 55. Chapman’s success
at Arsenal (in particular) was based on a regime of physical fitness, strength and
skill previously unseen in the English game. Chapman’s tactical acumen was also
impressive: Arsenal are widely credited as the first team to adapt fully to the intro-
duction of the present-day offside law in 1925.3
Arsenal came in with the idea of a stopper centre-half. Previously centre-halfs were allowed to
wander, and it now became a pendulum. The stopper centre-halves stayed there and then the
two full-backs were like pendulums on a clock. If play was on one wing, one full-back would
take it and the other one came back to cover behind the centre-half. You had three forwards up
front, two in the ‘V’ point of the ‘W’ with a lot of alternatives. One of those men had two men
in front of him and one at the side. He’d got three alternatives with the ball. (Jack Curtis, quoted
in Taylor and Ward, 1995, p27–8)
Although he won no major honours with any of the seven sides he managed,
Major Frank Buckley is one of the best-known managers of all time. His military
background, disciplined approach and personal demeanour seem to belong to a
bygone era, but other aspects of his style, including his dealings in the transfer
market and his astute handling of the media, mark him clearly as an early proto-
type of the modern-day football manager.4
I soon realised that Major Buckley was one out of the top drawer. He didn’t suffer fools
gladly … his style of management in football was very similar to his attitude in the army … Major
Buckley implanted into my mind the direct method of playing which did away with close inter-
passing and square-ball play. If you didn’t like his style you’d very soon be on your bicycle to
another club. He didn’t like defenders overelaborating in their defensive positions … Major
Buckley also knew how to deal with the press. (Stan Cullis, quoted in Taylor and Ward, 1995,
p31–2)
As football’s popularity grew in the 1920s and 1930s, there was an expansion in
press coverage. Critical press reports on the team’s performance could be trans-
formed into abuse from the terraces. Increasingly, the secretary-manager came to
act not only as a buffer between the directors and the players, but also as a scape-
goat for a run of poor results. During the 1930s, the press regularly reported on
the performance of teams in terms of the manager’s actions. This gave a certain
legitimacy to his role, and fostered a growing sense of self-importance (Carter,
2006).
During the post-war period, Matt Busby personified the arrival of the football
manager centre-stage, as the guiding spirit behind the football club, which itself
came to be seen increasingly as an institution of defining importance to a commu-
nity, a city or even a nation’s self-identity (Ronay, 2009). Busby was appointed to
the post of Manchester United manager in 1945, when the club’s Old Trafford sta-
dium was still badly damaged from wartime bombing. Under Busby’s guidance,
The role of the football manager 253
Manchester United became established as one of the strongest teams in Europe by

the late 1950s, before eight players and three coaching staff were tragically killed
in a plane crash at Munich airport in February 1958. An outpouring of public
grief and sympathy at the time may have been pivotal in Manchester United’s
subsequent emergence as the world’s most popular football club. Within another
decade, Busby had constructed a second great Manchester United team, which
became the first from England to lift the European Cup in 1968. Throughout his
two-and-a-half decades as team manager, prior to his eventual retirement in 1971,
Busby stood at the helm, exuding an air of calm but unassailable authority and
dignity.
Although Busby’s longevity as team manager at a major English club was unpar-
alleled at the time, his career provided a template for a number of others who
emerged during the 1950s and 1960s and went on to secure equivalent legendary
status at their own clubs. In England Bill Shankly (Liverpool, 1959–74), Don Revie
(Leeds United, 1961–74) and Brian Clough (Derby County, 1967–73; Nottingham
Forest, 1975–93) all fall into this category, as does Jock Stein (Celtic, 1965–78) in
Scotland. The charisma and media-awareness of Clough, in particular, formed the
basis for a highly successful and sometimes controversial secondary career in TV
punditry: a path also followed by many lesser managers who shared some of the
same telegenic attributes.
The irresistable growth in the popularity and ubiquity of televised football has
undoubtedly impacted on the football manager’s job specification, with the cap-
ability to communicate effectively via the TV media becoming an essential pre-
requisite. It often seems that the manager serves as little more than a figurehead,
satisfying the media’s insatiable demand for a narrative containing instant heroes
and instant scapegoats. The consequences for the football manager’s job security
in the television era, documented in some detail in Section 9.3 below, are predict-
able enough.
Apart from media relations, the remit of most managers in modern-day pro-
fessional football is the selection, supervision and coaching of playing staff, and
devising the team’s tactics and strategies (King and Kelly, 1997; Lambert, 1997).
During the post-war period, most managers have also taken responsibility for
the buying and selling of players, salary negotiations, and various administra-
tive duties. This multi-functional role of the manager still predominates in many
lower-tier clubs. Among the leading clubs, however, as the scale and complexity of
the financial and administrative aspects of club management has increased, there
has been a shift towards the division of responsibility among teams of specialists
in the various functional areas of management, so that the team manager of a
large-market club typically assumes full responsibility for first-team playing mat-
ters only. The extent of the manager’s influence over player transfers and contract
negotiations varies from club to club. Responsibilities for reserve-team and youth
affairs, as well as coaching for specialist positions such as goalkeeper, are often
delegated among a roster of subordinate coaches and assistants.
Against a background of chronic job insecurity for the typical football manager,
the career trajectories of the two longest-serving contemporary managers at the
time of writing, Alex Ferguson (Manchester United, appointed 1986) and Arsène
Wenger (Arsenal, appointed 1996), can be deemed truly exceptional. Ferguson has
often been characterised as the last representative of a dying breed (Ronay, 2009).
This reputation rests upon Ferguson’s authentic working-class upbringing as an
apprentice toolmaker in the Clyde shipyards; colourful tales of his persona as
dressing-room disciplinarian; a string of prickly and sometimes absurd exchanges
with rival managers conducted through the media; and a propensity for bearing
grudges against individual journalists or media organisations (including the BBC)
that can endure for years. Having accumulated eleven Premier League titles and
two European championships with Manchester United to the end of the 2009
season, Ferguson’s credentials as a successful football manager are, of course,
unequalled in English football, and may never be surpassed.
While Ferguson’s Glaswegian roots, shared with several illustrious predecessors
including Busby, Shankly and Stein, seem to hark back to a bygone age, Wenger
exudes an aura of cosmopolitan sophistication that has proved challenging to
many of English football’s more insular and workmanlike traditions. Despite hav-
ing access to rather slender financial resources, in comparison with Manchester
United at least, Wenger’s Arsenal teams had secured three Premier League titles
to the end of the 2009 season. They will, however, perhaps be most fondly remem-
bered for a fluid and skilful style of attacking play that, at its best, has rarely if
ever been matched in English club football.
9.2 Measuring the managerial contribution: the production frontier approach

As discussed in Chapter 1, ‘production’ in team sports is unlike that in other
business sectors because it cannot take place without the joint cooperation of at
least two teams. The requirement that clubs simultaneously cooperate and com-
pete with one another is almost unique to team sports, and has led to debate
regarding the status of the sports team as a firm in the economist’s sense (Neale,
1964). Much of the North American professional team sports economics litera-
ture assumes an objective of profit or wealth maximisation, and many major
league franchises consistently return large operating profits. In English football,
as seen in Chapter 6, the story is very different. Sloane (1971) suggests it is more
appropriate to think of English football clubs as utility-maximisers. Arguments
of the utility function might include playing success, attendance, media recogni-
tion and sponsorship, as well as profit.
Rottenberg (1956) first proposed the notion of a sporting production function:
(T)he product is the game, weighted by the revenues derived from its play. With game admis-
sion prices given, the product is the game, weighted by the number of paying customers who
attend … the players of one team [are] the factors and all others (management, transportation,
ballparks, and the players of the other team), another. (Rottenberg, 1956, p255)
Measuring the managerial contribution 255
Managers of professional sports teams can influence performance in two ways.

First, taking the collection of playing inputs at his disposal as given in the short
term, the manager seeks to maximise the level of performance achieved using
these inputs. The manager’s direct contribution depends partly on the quality of
his strategic input: the effectiveness of his team selections and tactics. The direct
contribution also depends on the manager’s ability to inspire and motivate play-
ers. Second, the manager can attempt to improve the level of skill of his existing
players over the longer term, or strengthen his squad by dealing in the transfer
market. The manager’s indirect contribution depends partly on his effectiveness as
a coach in training and developing his existing players, and partly on his skill and
astuteness in deciding which players he should buy and sell.
In theoretical terms, it is possible to conceive of each team as facing a produc-
tion set, comprising all technologically feasible means of transforming a given
set of playing inputs into a given level of output (measured by the team’s win
ratio). Within the production set there is a maximum level of output that corres-
ponds to a particular combination of inputs, or there is a minimum combination
of inputs that can be used to produce a particular level of output. Both of these
define the limit of technological possibility. A production function defined using
the maximum efficiency assumption is known as a frontier production function
or a production frontier. The quality of the managerial contribution, or the level
of managerial efficiency, can be measured by examining the team’s actual per-
formance relative to the maximum attainable performance (given the quality of its
inputs) that would be predicted by the production frontier.
If several different teams (or managers) are observed, it is likely that some
will be found not to operate on the technologically feasible boundary (too many
inputs are being used to produce the boundary level of output; or output is below
the maximum that should be attainable). Compared to best practice, these man-
agers are technically inefficient. Neoclassical production functions do not allow
for suboptimal solutions, and so exclude the notion of inefficiency. In contrast,
the production frontier approach acknowledges that most managers are ineffi-
cient, to some extent, relative to best practice. The production frontier approach
identifies the frontier, and locates the position of each manager relative to the
frontier.
Empirical studies of team production begin with Scully (1974), who uses an aver-
age production function, rather than a production frontier, to model team output
(win percent) in baseball, as a function of playing and non-playing (management,
capital and team spirit) inputs. This method of modelling the sporting production
process became the norm in early production function studies (Zech, 1981; Scott,
Long and Somppi, 1985; Atkinson, Stanley and Tschirhart, 1988; Schofield, 1988;
Carmichael and Thomas, 1995). All of these studies assume implicitly that team
production takes place at maximum efficiency. In some cases, one or two manager-
ial variables are included on the right-hand side of the estimated equation. Zech
(1981), for example, includes the number of years spent managing in major league
baseball and the manager’s lifetime win-loss ratio. Carmichael and Thomas (1995)
include the number of years’ coaching experience.
Two approaches to efficiency measurement using production frontiers are preva-
lent in the literature. One approach focuses on measuring managerial efficiency,
while the other concentrates on estimating a team’s efficiency over an entire sea-
son. Production or cost frontiers can be fitted using either econometric (stochastic
frontier analysis) or linear programming (data envelopment analysis, or DEA)
techniques. Frontiers estimated using the former method are stochastic, and those
estimated using the latter are deterministic.
Among early North American studies, Porter and Scully (1982) find that the
contribution of the coach to team performance in baseball is comparable to that
of an individual star player. Scully (1994, 1995) compares coach efficiency in base-
ball, basketball and (American) football. Hofler and Payne (1996, 1997) examine
coach efficiency in the NFL and in the NBA. Ruggerio, Hadley and Gustafson
(1996) calculate efficiency scores for twenty-four coaches with four or more full
seasons’ employment in MLB.
Kahn (1993) and Singell (1993) attempt to separate the direct contribution of
the manager to team performance (through team selection and tactics) and the
indirect contribution (through enhancement of player quality by effective coach-
ing). A conventional production function is used to measure the direct effect.
Kahn proxies for managerial performance using the predicted salary, while Singell
uses managerial experience. The indirect managerial effect is captured by running
separate regressions that compare individual player performance for the season
under review with the historical performance of the player. Kahn focuses on the
effect of a new manager on player performance, while Singell concentrates on the
performance of players once they have moved to a new club.
Dawson, Dobson and Gerrard (2000a, 2000b) employ stochastic frontier
analysis to measure managerial efficiency in English football, using T1 (Premier
League) data. An innovative feature is that managerial spells are the unit of obser-
vation, enabling the effects of within-season managerial changes to be incorpo-
rated. Player talent indices, based on weighted sums of player characteristics and
career statistics (age, career appearances, career goals, number of previous clubs,
and so on), are used to measure the playing input, based on data from before the
season in question. Establishing an ex ante measure of playing quality is import-
ant in order to measure the indirect impact of the manager on team output, meas-
ured by win percent.
Dawson, Dobson and Gerrard (2000a) estimate the temporal variation in man-
agerial efficiency, using a time-varying stochastic production function.5 Both time-
invariant and time-varying managerial efficiency scores are reported. However, the
time-invariant specification is found to provide an inadequate representation of
the data. Using a similar approach, Dawson, Dobson and Gerrard (2000b) inves-
tigate the robustness of estimates of managerial efficiency over a range of estima-
tion methods, model specifications and definitions of input and output variables.
Dawson and Dobson (2002) investigate the sources of variation in managerial

efficiency. First, managers have varying levels of accumulated human capital and
inherent managerial ability. Second, clubs differ in terms of their compensation
packages and incentive structures for players and managers, and according to
the frequency with which they change their managers. Manager human capital
is found to be significant in explaining variations in managerial efficiency, as are
club-specific factors.
Barros and Leach (2006, 2007) use stochastic frontier analysis to measure effi-
ciency scores for twelve English T1 clubs for seasons 1999 to 2003. The cost func-
tion has three factor input prices (for labour, and two capital inputs) and three
outputs (league points, attendance and turnover). An average (time-invariant)
efficiency score of 88 per cent is reported, which suggests that by operating at
maximum efficiency the average club could reduce its costs by around 12 per-
cent without affecting its outputs. Kern and Suessmuth (2005) use stochastic fron-
tier analysis to estimate team production functions for the German Bundesliga.
Ex ante inputs (fixed wages of the players and the coach) are transformed into
ex post sporting success. The output measure takes into account performance in
four competitions (domestic league and cup, UEFA Cup and UEFA Champions
League). There is wide variation in the reported team efficiency scores.
Several recent efficiency studies employ a variant of stochastic frontier analysis,
known as the random frontier model (Greene, 2005). In a standard stochastic fron-
tier analysis, the teams are assumed to have the same technological possibilities. If
these differ between teams, however, measured technical inefficiency may be over-
stated. The random frontier model takes heterogeneity into account, by separating
technical inefficiency from technological differences between teams. Using data on
twelve teams that competed in T1 in England throughout seasons 1999 to 2004,
Barros and Garcia-del-Barrio (2008) compare the efficiency scores obtained from
several cost function specifications, including a homogenous (standard) stochas-
tic frontier analysis and a random frontier model. The random frontier provides
a better representation of the data than the homogeneous frontier. Barros, del
Corral and Garcia-del-Barrio (2008) and Barros, Garcia-del-Barrio and Leach
(2009) estimate efficiency scores for the Spanish La Liga, by estimating a ran-
dom frontier model using data from seasons 1996 to 2005. As before, the random
frontier model provides the best representation of the data. Frick and Lee (2010)
estimate time-varying efficiency scores for German clubs using data from seasons
1982 to 2003.
As noted above, deterministic frontiers have also been used to estimate pro-
duction and cost frontiers based on team sports data, fitted using the DEA lin-
ear programming technique. North American examples include Anderson and
Sharp (1997), Mazur (1994) and Volz (2009) for MLB; and Einolf (2004) for the
NFL and MLB. Haas (2003) uses DEA to measure the productive efficiency of
English T1 teams during the 2001 season. Input measures are team payroll as a
proxy for playing talent, and the manager’s salary as a proxy for the managerial
contribution; however, there is no ex ante measure of team quality or manager cap-

ability. Outputs are league points and revenues. DEA frontiers are fitted based on
both constant returns to scale (CRS) and variable returns to scale (VRS) assump-
tions concerning the production technology.
Barros and Leach (2006) combine sporting and financial variables to estimate a
DEA cost frontier, using English T1 data for seasons 1999 to 2003. Inputs are num-
ber of players, payroll, net assets and expenditure on stadium facilities. Outputs
are league points, attendances and turnover. Efficiency scores are estimated using
both CRS and VRS specifications. Teams that are found to be technically efficient
using the CRS specification are also technically efficient using the VRS specifica-
tion, signifying that the dominant source of efficiency is scale.
Guzmán and Morrow (2007) estimate a DEA frontier and calculate a Malmquist
total factor productivity index to measure changes in productivity over seasons
1998 to 2003 for English T1 teams. Input measures are based on staff payroll
(excluding costs relating to non-football activities), directors’ remuneration and
other operating costs (lease and rental charges, fixed asset depreciation, training
ground costs and professional fees). Outputs are league points and revenues. A
CRS specification is adopted. An average efficiency score of around 85 per cent is
reported, which suggests that by operating at maximum efficiency the average club
could reduce its costs by around 15 per cent. The factor productivity analysis fails
to uncover any evidence of improvements in productivity over time.
Haas, Kocher and Sutter (2004) estimate a DEA frontier using German
Bundesliga data for the 2000 season. Input measures are based on staff payroll
(separately for players and managers); and outputs are league points, total rev-
enues and average stadium utilisation (ratio attendance to stadium capacity).
Both CRS and VRS specifications are adopted. The measured efficiency scores
are not correlated with league position, indicating that athletic performance and
efficiency are not synonymous. Haas (2004) estimates a DEA frontier using North
American Major League Soccer (MLS) data for the 2000 season. Input measures
are again based on players’ and coaches’ payroll; and outputs are league points,
numbers of spectators and revenues. In this case the efficiency scores are highly
correlated with league performance, and scale is the principal determinant of the
level of efficiency.
Espita-Escuer and García-Cebrián (2004) use DEA to measure the efficiency
of teams in converting attacking moves during the match into sporting success,
by applying CRS and VRS specifications to Spanish La Liga data from seasons
1999 to 2001. The input variables are the number of players used, the number of
attacking moves, the number of minutes during which the teams had possession of
the ball, and the number of shots and headers, all of them throughout the length
of the season. Outputs are the number of league points achieved throughout the
season. Efficient teams do not always correspond with those that finished highest
in the league at the end of the season, implying that highly placed but inefficient
teams could have achieved the same results with fewer resources or could have
improved their results with the same resources. Scale effects are found to be signifi-
cant, suggesting that a VRS specification is preferred. Garcia-Sanchez (2007) ana-
lyses the operating efficiency of the attack and defence of Spanish La Liga teams
during the 2005 season. The inputs used are goals scored and goals conceded, and
output is measured by total points obtained at the end of the season.
Boscá et al. (2009) present a comparative analysis of attacking and defensive
efficiency for three seasons (2001 to 2003) in Italian and Spanish football. In a
DEA model, the attacking inputs are shots on goal, attacking plays made by the
team, balls kicked into the opposing team’s centre area and minutes of possession.
The inverse of the attacking inputs are used for the defence. The output for attack-
ing and defensive production is the number of goals scored by a team and the
inverse of the number of goals conceded, respectively. The difference between the
best and worst teams is wider in Italy than in Spain. In the Italian league the best-
ranked teams score more goals, concede fewer goals and obtain more points, on
average. In Spain, by contrast, the correlations between these indicators are gener-
ally smaller. In Italy, league rankings are more highly correlated with measures of
efficiency in defence, rather than attack; but in Spain the opposite applies.
Although DEA has been used widely in football efficiency studies, the linear
programming methodology has two important limitations. First, DEA is reliant
upon an assumption that there is no ‘noise’ (or error) in the data being studied.
Second, although the fact that there is no requirement to specify a particular func-
tional form for the production technology might be considered a strength, the
inability to evaluate the results of a DEA study using the conventional criteria of
statistical inference is a weakness. In an attempt to overcome this problem, Simar
and Wilson (1999, 2007) use a bootstrap methodology to carry out hypothesis
tests on DEA efficiency scores. In a recent application of this approach to sport,
Barros, Assaf and Earp (2010) measure the technical efficiency of twenty foot-
ball clubs in Brazil between 2006–7. Outputs are attendance, total receipts and
points in a league. Input measures are based on operational cost (excluding labour
costs), total assets and team payroll. Efficiency scores are estimated using a VRS
specification. The results reveal that none of the clubs are operating close to full
technical efficiency, a finding that contrasts markedly with the results from stand-
ard DEA estimation. The authors also find that technical efficiency is correlated
positively (negatively) with positive (negative) results on the pitch.
Another variant of production function analysis shifts the focus away from
quantification of the relationship between inputs used and outputs achieved over
an entire football season, towards an analysis of the performance of teams at
match level. In the case of English football, this type of analysis draws on the
Opta Index, first published in 1996, which provides detailed performance statistics
based on records of every touch of the ball in every T1 match, compiled by a team
of analysts using videotapes. The ‘play’ records include shots on goal (on-target,
off-target and blocked); passes (successful and unsuccessful); dribbles/runs (pos-
session retained and lost); clearances, blocks and interceptions; interventions by
the goalkeeper (catches and spillages); tackles; free kicks conceded for foul tack-
les and handballs; and yellow and red cards incurred. Using data from the 1998
season, Carmichael, Thomas and Ward (2000) estimate regressions in which the
dependent variable is the difference in scores, and the explanatory variables are
the differences between the teams in various ‘play’ variables. Most of the esti-
mated coefficients are correctly signed and many are significant, confirming the
importance for match outcomes of accurate shooting, and effective defensive
skills including tackles, clearances and blocks. In a follow-up study, Carmichael,
Thomas and Ward (2001) use the Opta data in a more aggregated format to esti-
mate season-based production functions for each T1 team.
Carmichael and Thomas (2005) examine the empirical relationship between a
number of team play measures, interpreted as inputs to a team production func-
tion, and the numbers of shots and goals scored by each team in each match,
interpreted as outputs. The team play measures include the numbers of passes,
dribbles, runs, tackles, fouls, clearances, blocks and goalkeeper saves. To investi-
gate the origins of home-field advantage in differences in play, the empirical ana-
lysis is carried out separately for home and away teams. The statistical significance
of attack-related team play measures is stronger in regressions for home-team per-
formance; and conversely, the significance of defence-related measures is stronger
in regressions for away-team performance.
9.3 Patterns of managerial change in English football

The analysis of managerial change and team performance draws on data for 1,572
managerial spells with English league clubs recorded between the 1973 and 2009
seasons (inclusive). The database includes eighty-five left-censored spells that
commenced before the end of the 1972 season. The number of spells that started
either at the start of the 1973 season or subsequently, and terminated before the
start of the 2010 season, is 1,411. There were seventy-six right-censored spells that
started before the end of the 2009 season and continued subsequently into the
2010 season.
For the 1,572 managerial spells, Table 9.1 shows the top 50 ranked in descending
order of duration, measured in league matches completed. Win ratios are calculated
(as in previous chapters) by awarding 1 point for a win, 0.5 points for a draw and 0
points for a loss, and dividing the total by the number of league matches played. Of
course, in view of the notorious insecurity of the English football manager’s job
tenure, the longevity of a spell is itself often a meaningful measure of success. In
some cases longevity may reflect attributes or attitudes of the club’s owners, which
influence their willingness either to stick with or dismiss a manager during a lean
spell. A highly atypical example from the modern era, featuring at the very top of
the list, is the extraordinary 1,092-match spell recorded by Dario Gradi of Crewe
Alexandra between 1983 and 2007: a spell that included four promotions and
three relegations as Crewe’s divisional status fluctuated between T2, T3 and T4.
Patterns of managerial change in English football 261
Table 9.1 Managerial spells ranked by duration (matches), English League,

1973–2009 seasons
Rank Manager Team Start year End year Matches Win ratio
1 Dario Gradi Crewe 1983 2007 1092 0.492

2 Alex Ferguson Manchester Utd 1986 — 883 0.708
3 Ted Bates Southampton 1955 1973 792 0.516
4 Brian Clough Nottm Forest 1975 1993 758 0.574
5 Tony Waddington Stoke 1960 1977 701 0.484
6 John Rudge Port Vale 1984 1999 701 0.514
7 Bill Nicholson Tottenham 1958 1974 666 0.582
8 Alan Curbishley Charlton 1991 2006 634 0.513
9 John Lyall West Ham 1974 1989 621 0.507
10 Bill Shankly Liverpool 1959 1974 610 0.650
11 Bobby Robson Ipswich 1969 1982 572 0.547
12 Don Revie Leeds 1961 1974 555 0.660
13 Ron Greenwood West Ham 1961 1974 549 0.468
14 Joe Harvey Newcastle 1962 1975 546 0.508
15 Jimmy Frizell Oldham 1970 1982 545 0.516
16 Alan Dicks Bristol City 1967 1980 542 0.476
17 Bryan Flynn Wrexham 1989 2001 540 0.496
18 Joe Royle Oldham 1982 1994 534 0.504
19 Harry Catterick Everton 1961 1973 499 0.584
20 Arsène Wenger Arsenal 1996 — 494 0.715
21 Lawrie McMenemy Southampton 1973 1985 488 0.550
22 Brian Laws Scunthorpe 1997 2006 447 0.519
23 Graham Taylor Watford 1977 1987 434 0.568
24 Tommy McAnearney Aldershot 1972 1981 427 0.516
25 John King Tranmere 1987 1996 415 0.564
26 Barry Fry Peterborough 1996 2005 414 0.467
27 Bertie Mee Arsenal 1966 1976 413 0.574
28 John Neal Wrexham 1968 1977 409 0.550
29 Harry Redknapp Bournemouth 1983 1992 401 0.510
30 Arthur Cox Derby 1984 1993 397 0.524
31 Jimmy Scoular Cardiff 1964 1973 391 0.474
32 Gordon Milne Coventry 1972 1981 390 0.464
33 Ken Roberts Chester 1986 1976 388 0.509
34 Bob Paisley Liverpool 1974 1983 378 0.692
35 Steve Coppell Crystal Palace 1984 1993 376 0.535
36 Mike Buxton Huddersfield 1978 1986 374 0.537
37 Frank Clark Leyton Orient 1983 1991 368 0.507
38 George Graham Arsenal 1986 1995 363 0.609
39 Lennie Lawrence Charlton 1982 1991 358 0.443
40 Johnny Newman Exeter 1969 1976 358 0.510
41 Ronnie Moore Rotherham 1997 2005 355 0.524
42 Graham Turner Wolverhampton 1986 1994 355 0.569
43 David Pleat Luton 1978 1986 352 0.523
44 Benny Fenton Millwall 1966 1974 347 0.535
45 Neil Warnock Sheffield Utd 1999 2007 341 0.544
Table 9.1 (cont.)
46 George Burley Ipswich 1994 2002 341 0.557
47 Dave Bassett Sheffield Utd 1988 1995 339 0.506
48 Keith Burkinshaw Tottenham 1976 1984 336 0.531
49 Howard Wilkinson Leeds 1988 1996 335 0.584
50 Harry McNally Chester 1985 1992 333 0.482
Source: Rothmans/Sky Sports Football Yearbook, www.soccerbase.com
Others from the modern era featuring prominently in the upper echelons of the
top 50 include Alex Ferguson and Arsène Wenger (see Section 9.1).
Several managers achieved longevity by steering their clubs through one or
more promotion campaigns, before consolidating and achieving further success
at a higher level. Brian Clough (Nottingham Forest), John Rudge (Port Vale), Joe
Royle (Oldham Athletic) and Graham Taylor (Watford) all fall into this category.
Another striking feature of Table 9.1 is the preponderance towards the top end
of the list of a number of long-serving managers of T1 clubs at the start of the
period. No fewer than eight of the twenty-two T1 managers at the start of the
1973 season had been in post for more than ten years. Pride of place among this
group belongs to Ted Bates, whose eighteen years at the helm saw Southampton’s
divisional status raised from the old Third Division (South) in 1955, to T1 status
in the early 1970s, through two promotions achieved in 1960 and 1966. The list of
long-servers at the start of the 1973 season also includes legendary names such as
Don Revie (Leeds), Bill Shankly (Liverpool), Ron Greenwood (West Ham) and Bill
Nicholson (Tottenham). By contrast, some distance behind Ferguson and Wenger
(and failing to make the top 50), the third-, fourth- and fifth-longest serving T1
managers in post at the start of the 2010 season were David Moyes (Everton,
appointed in 2002), Rafa Benitez (Liverpool, 2004) and Tony Pulis (Stoke, 2006).
At the opposite end of the scale, Leroy Rosenior (Torquay) lays claim to the
record for the shortest managerial spell of all time, timed at ten minutes according
to football folklore. In May 2007 it was reported that Rosenior, a former Torquay
manager, was reappointed by club chairman Mike Bateson, who also sold his
majority shareholding in the club on the same day. The purchasers decided imme-
diately to relieve the newly appointed manager of his duties. This spell does not
contain a league match, and therefore does not register in the database that is used
to compile the tabulations that are reported in this section. The shortest manager-
ial spell that does register belongs to Micky Adams, who oversaw three consecu-
tive defeats by Swansea during the 1998 season before resigning, reportedly over a
disagreement about money for signing new players.
Using a measure of success that would perhaps be of greater interest to many spec-
tators than longevity, Table 9.2 shows the rankings of the managerial spells by win
ratio. For the purposes of constructing Table 9.2, spells of less than twenty matches’
duration (to the end of the 2009 season) are excluded.
Table 9.2 Managerial spells ranked by win ratio, English League, 1973–2009 seasons
1 Jose Mourinho Chelsea 2004 2007 120 0.813

2 Avram Grant Chelsea 2007 2008 32 0.813
3 Kevin Keegan Fulham 1998 1999 46 0.761
4 Nigel Pearson Leicester 2008 — 46 0.750
5 Paul Ince Milton Keynes 2007 2008 46 0.739
6 Kenny Dalglish Liverpool 1985 1991 224 0.732
7 Peter Taylor Brighton 2001 2002 34 0.721
8 Simon Grayson Leeds 2008 — 25 0.720
9 Arsène Wenger Arsenal 1996 — 494 0.715
10 Alex Ferguson Manchester Utd 1986 — 883 0.708
11 Phil Scolari Chelsea 2008 2009 25 0.700
12 Bob Paisley Liverpool 1974 1983 378 0.692
13 Rafael Benitez Liverpool 2004 — 190 0.689
14 Joe Fagan Liverpool 1983 1985 84 0.673
15 Colin Appleton Hull 1982 1984 111 0.671
16 Roberto Di Matteo Milton Keynes 2008 2009 46 0.663
17 John Benson Wigan 1999 2000 46 0.663
18 Kenny Dalglish Blackburn 1991 1995 169 0.663
19 Jim Smith Oxford 1982 1985 160 0.663
20 Darren Ferguson Peterborough 2007 — 111 0.662
21 Don Revie Leeds 1961 1974 555 0.660
22 Kevin Keegan Newcastle 1992 1997 205 0.656
23 Peter Taylor Gillingham 1999 2000 46 0.652
24 Osvaldo Ardiles West Bromwich 1992 1993 46 0.652
25 David Webb Southend 1986 1987 33 0.652
26 Bill Shankly Liverpool 1959 1974 610 0.650
27 Roy McFarland Bradford City 1981 1982 70 0.643
28 Claudio Ranieri Chelsea 2000 2004 147 0.643
29 Kevin Blackwell Sheffield Utd 2008 — 60 0.642
30 Steve Coppell Brentford 2001 2002 46 0.641
31 Martin Allen Milton Keynes 2006 2007 46 0.641
32 Mike Newell Hartlepool 2002 2003 29 0.638
33 Billy Davies Preston 2004 2006 87 0.638
34 Howard Kendall Everton 1981 1987 252 0.637
35 David O’Leary Leeds 1998 2002 145 0.634
36 Gianluca Vialli Chelsea 1998 2000 94 0.633
37 Paul Sturrock Plymouth 2000 2004 162 0.633
38 Nigel Spackman Sheffield Utd 1997 1998 34 0.632
39 Bruce Rioch Wigan 2000 2001 34 0.632
40 Barry Fry Barnet 1986 1993 80 0.631
41 Roberto Martinez Swansea 2007 2009 107 0.631
42 Peter Taylor Wycombe 2008 — 46 0.630
43 Ron Atkinson Manchester Utd 1981 1986 225 0.629
44 John Bond Bournemouth 1970 1973 156 0.628
45 Barry Fry Southend 1993 1993 28 0.625
46 Gerard Houllier Liverpool 1998 2004 216 0.625
Table 9.2 (cont.)
47 Bryan Hamilton Wigan 1985 1986 60 0.625
48 Gudjohn Thordarson Stoke 1999 2002 121 0.624
49 Richie Barker Shrewsbury 1978 1978 33 0.621
50 Chris Turner Peterborough 1991 1992 87 0.621
Two Chelsea managers who presided over the club’s exceptional run of success
during the mid to late 2000s, Jose Mourinho and Avram Grant, top the list with
identical win ratios achieved over spells of differing durations. Ironically, Grant’s
brief tenure is viewed in some quarters as unsuccessful, with Chelsea forced to set-
tle for the runners-up position to Manchester United in both the Premier League
and the European Champions League at the end of the 2008 season. A third recent
Chelsea manager, Guus Hiddink (February–May 2009) achieved a win ratio even
higher than those of Mourinho and Grant, but does not feature in Table 9.2 because
his spell contained fewer than twenty matches. Even Phil Scolari, abruptly ejected
from the Chelsea manager’s post in February 2009 a few months after his appoint-
ment at the start of the same season, achieves a creditable eleventh position.
Others featuring prominently in Table 9.2 include Peter Taylor (Brighton,
Gillingham and Wycombe), the only manager to appear three times in the top
50; and Kevin Keegan (Fulham and Newcastle), Kenny Dalglish (Liverpool and
Blackburn) and Barry Fry (Barnet and Southend), each of whom appears twice.
Closely wedged together in ninth and tenth positions respectively are the ubiqui-
tous Wenger and Ferguson; from their respective dates of appointment to the end
of the 2009 season, the former held a hair’s-breadth advantage over the latter on
the win ratio measure. Rafa Benitez (Liverpool) follows a short distance behind.
Table 9.2 also includes several recent appointees who achieved early success during
spells that were incomplete at the end of the 2009 season, such as Nigel Pearson
(Leicester) and Simon Grayson (Leeds).
Audas, Dobson and Goddard (1999) discuss a number of factors contributing
to the chronic insecurity of the English football manager’s position. On the one
hand, match results (the manager’s most important performance measure) are
completely transparent, easily interpreted and instantly in the public domain.6 On
the other hand, the manager’s ability to influence results is constrained by uncer-
tainty concerning the fitness, performance and motivation of players. Whether or
not a group of players will combine to form a successful team has always depended
on a mixture of luck and judgement, in proportions that are ultimately impossible
to fathom. In any event, each match and each season’s league programme are
by their very nature zero-sum affairs, in that are failure for some participants is
inevitable. As characteristics of the manager’s position, the combination of direct
accountability for outcomes that are transparent and public, and imperfect con-
trol over the processes that determine the same outcomes, could not be better
designed to minimise job security.
Whereas inadequate performance in business tends to lead to the loss of custom-
ers followed by events such as internal restructuring, acquisition by new owners
or liquidation, football clubs are highly resilient, even in the face of catastrophic
failure on the field of play or on the balance sheet. This is due mainly to the highly
loyal or fanatical nature of their customer base. Relegation to a lower division,
or even to non-league status, usually requires some restructuring of finances and
personnel, but is invariably accepted by regular spectators in sufficient numbers to
ensure that clubs remain viable as going concerns. In accounting terms, all clubs
benefit from significant (but intangible) goodwill. But such fierce customer loyalty
can generate extreme pressure for the scapegoating of an unsuccessful manager.
Finally, the specialised nature of the football manager’s contribution usually
makes it difficult for an unsuccessful manager to be accommodated elsewhere
within the club by means of an upward, sideways or downward move (as often
happens elsewhere in business). The uniformity of football’s ‘production technol-
ogy’, however, makes the team manager’s function highly transferable between
clubs, since the relevant skills tend to be job-specific and not firm-specific. There
are perhaps a few rigidities within the managerial labour market that should be
considered. At times, many club owners have shown a preference for appoint-
ing individuals with a current or past association with the club to the manager’s
position. This may be because lower transaction costs are associated with the
appointment of an individual who is already known to the owner; or it may be
because an ex-player often brings advantages of credibility or popularity with
supporters. Either way, however, it seems unlikely that the occasional propensity
for clubs to show preference towards insiders (of one kind or another) seriously
undermines the conclusion that the managerial labour market is highly flexible
and competitive.
For the purposes of the empirical analysis that is reported below in Section 9.5,
each of the 1,495 managerial departures recorded in the data set is classified as
either ‘involuntary’ or ‘voluntary’. This distinction is important for identification
of the empirical relationship between team performance and the manager’s job
security. While it seems reasonable to expect that poor team performance would
lead to involuntary departure, voluntary departure is often a consequence of good
performance resulting in the manager receiving a more attractive job offer from
another club. In making the distinction between involuntary and voluntary depar-
tures, it seems unwise to rely too heavily on the club’s original explanation of the
reason for the change as either a ‘dismissal’ or a ‘resignation’. To do so would
probably lead to overcounting of the number of voluntary departures, since many
‘resignations’ actually occur in response to behind-the-scenes pressure from the
chairman or directors. In extremis, this can take the form of an ultimatum to
resign or be sacked (Audas, Dobson and Goddard, 1999).
In classifying managerial departures as either involuntary or voluntary, any

information available on the circumstances at the time of the departure was
assessed. Particularly relevant criteria were the team’s performance prior to ter-
mination; the nature of any new appointment accepted by the manager immedi-
ately after his departure; and the reasons given by the club or the manager for the
latter’s departure. Involuntary departures (the majority) include those that were
at the club’s instigation, as well as resignations for which there was no obvious
explanation other than disappointing results or other pressures emanating from
the manager’s current appointment. Voluntary departures include those for which
there is clear and tangible evidence that the manager instigated the move. Such evi-
dence is usually in the form of an immediate or imminent move to another job at a
club (at home or overseas) of comparable or higher status, or a move to a national
team manager’s position.
A few recent examples will illustrate the method of classification. Despite some
inconsistent results, and press reports of dressing-room discontent, the abrupt
termination of Phil Scolari’s short reign as Chelsea manager (summer 2008–
February 2009) took most observers by surprise. The mode of departure was
clearly involuntary. In contrast, Scolari’s successor Guus Hiddink stated consist-
ently that his appointment was only temporary, to the end of the 2009 season.
Hiddink’s departure, which took place according to schedule, is therefore classed
as voluntary. In summer 2008 Paul Ince left the T3 club Milton Keynes (MK)
Dons voluntarily to take charge of Blackburn Rovers in T1. But with Blackburn
positioned near the foot of the table throughout the first half of the 2009 season,
Ince was ejected from his post, involuntarily, in December 2008. Meanwhile Ince’s
successor at MK Dons, Roberto Di Matteo, was enjoying a successful first season
in charge. When Tony Mobray left the relegated T1 club West Bromwich Albion
voluntarily (to take charge of the Scottish club Celtic) at the end of the 2009 sea-
son, Di Matteo was invited to take charge at West Bromwich. His departure from
MK Dons is therefore classed as voluntary. Ironically, Paul Ince was subsequently
reinstated to his original position as manager of MK Dons.
Table 9.3 reports the numbers of involuntary and voluntary managerial depar-
tures by season; and Table 9.4 reports the same data reported as yearly averages
for bands of four seasons’ duration. Clearly, the level of job insecurity of foot-
ball managers has increased over time. While there is considerable year-on-year
variation in the total numbers of departures, the long-term trend is relentlessly
upward. Significantly, the increase in the total is driven almost entirely by increases
in the numbers of involuntary departures, while there is no clear upward trend in
the numbers of voluntary departures. An increasing tendency for senior clubs to
appoint current or recently retired players directly into managerial posts, or to
recruit managers from overseas, may account for this pattern.
With sixty-one managerial departures, the 1995 season established a new record
for managerial turnover, which has never been surpassed. At the end of the 1995
season, the league structure was changed from 22–24 –24 –22 to 20 –24 –24 –24
Table 9.3 Managerial departures, by season, English League, 1973–2009 seasons
Number of departures
Season Involuntary Voluntary All
1973 26 6 32
1974 18 15 33
1975 23 9 32
1976 22 5 27
1977 29 7 36
1978 35 10 45
1979 23 9 32
1980 20 1 21
1981 32 15 47
1982 32 5 37
1983 31 5 36
1984 32 6 38
1985 29 6 35
1986 32 8 40
1987 30 8 38
1988 26 2 28
1989 23 8 31
1990 33 4 37
1991 34 14 48
1992 29 3 32
1993 32 8 40
1994 29 7 36
1995 48 13 61
1996 30 9 39
1997 35 12 47
1998 41 7 48
1999 31 7 38
2000 37 8 45
2001 39 7 46
2002 49 11 60
2003 31 4 35
2004 43 7 50
2005 37 3 40
2006 42 11 53
2007 39 10 49
2008 39 10 49
2009 43 12 55
Total 1204 292 1496

Table 9.4 Average number of managerial departures per season, four-season bands,
English League, 1974–2009 seasons
Involuntary Voluntary All
1974–1977 23.0 9.0 32.0

1978–1981 27.5 8.8 36.3
1982–1985 31.0 5.5 36.5
1986–1989 27.8 6.5 34.3
1990–1993 32.0 7.3 39.3
1994–1997 35.5 10.3 45.8
1998–2001 37.0 7.3 44.3
2002–2005 40.0 6.3 46.3
2006–2009 40.8 10.8 51.5
teams in the four tiers. The transition was made by creating one fewer promotion
berth and one additional relegation berth in each tier for one season only: a meas-
ure that appears to have created sufficient panic at boardroom level to cost many
managers their jobs. Financial pressures on many T2, T3 and T4 clubs created by
the collapse of ITV Digital (see Chapter 6, Section 6.5) may have been a contribu-
tory factor to a near-repeat of this record number of departures during the 2002
season. It appears that team managers may have faced even greater pressure than
usual to deliver a winning team that would generate gate revenues to compensate
for the disappeared broadcast revenue. Similarly, financial pressures on many foot-
ball clubs, exacerbated by the economic recession of 2008–9, may have contributed
to a further outbreak of boardroom panic, and the third-highest all-time total of
fifty-five departures recorded in the 2009 season.
Table 9.5 reveals a high level of monthly variation in the incidence of managerial
departure. 1,036 of the 1,464 departures recorded between the 1974 and 2009 sea-
sons (70.8 per cent) took place during the course of the season, and 428 (29.2 per
cent) during the close season. Of the 1,036 within-season departures, 152 (14.7 per
cent) were voluntary; and of the 428 close season departures, 134 (31.3 per cent)
were voluntary. Departure within the season is therefore relatively more likely to be
at the club’s behest, while departure during the close season (when most manager-
ial contracts tend to expire) is more likely to be at the manager’s behest. October
and November, when failure to realise overoptimistic pre-season aspirations first
becomes inevitable for many clubs, are the peak months for within-season depar-
tures. December and January, perhaps the latest time at which a new appointment
might reasonably be expected to bring about a change of fortune within the current
season, is another period of high turnover; so too are March, April and May, when
many clubs begin planning for the next season. There appears to be a (slight) lull in
the rate of turnover during the month of February.
Table 9.5 Managerial departures by month, English League, 1974–2009 seasons
August 32 8 40
September 92 13 105
October 123 25 148
November 122 24 146
December 99 25 124
January 105 17 122
February 100 9 109
March 108 16 124
April/May 103 15 118
Close season 294 134 428
Table 9.6 Managerial departures by tier, four-season bands, English League,

1974–2009 seasons
T1 T2 T3 T4 All
1974–1977 31 21 36 40 128
1978–1981 28 31 46 40 145
1982–1985 30 37 39 40 146
1986–1989 28 34 35 40 137
1990–1993 27 43 37 50 157
1994–1997 34 50 47 52 183
1998–2001 31 49 52 45 177
2002–2005 29 44 46 65 185
2006–2009 40 52 59 54 206
Involuntary 203 288 306 356 1155
Voluntary 75 70 79 53 277
Total 278 361 397 426 1464
Table 9.6 suggests that managerial insecurity is greater in the lower tiers than in
the higher tiers. Both the number of involuntary departures and the total number
of departures vary inversely with divisional status. If, as seems likely, recent rele-
gation often triggers managerial departure, the pattern might be explained by the
fact that while no team can be relegated to T1, most teams, with the exception of
those promoted (or in earlier times elected) from non-league, enter T4 as a result
of relegation. In contrast, the incidence of voluntary departure is highest in T3,
which appears to be the most effective launch pad for a successful manager seek-
ing another post at a higher level.
Table 9.7 Average duration (matches completed) of terminating managerial spells,

four-season bands, English League, 1974–2009 seasons
1974–1977 146.5 193.9 159.8

1978–1981 109.1 117.7 111.2
1982–1985 98.7 182.5 111.4
1986–1989 102.1 149.9 111.2
1990–1993 101.4 129.7 106.6
1994–1997 86.3 127.0 95.4
1998–2001 80.3 101.4 83.8
2002–2005 83.1 102.3 85.7
2006–2009 77.8 135.0 89.7
Table 9.7 reports the average durations (in matches completed) of terminating
managerial spells in four-season bands for the period 1974–2009. As expected in
view of the upward trend in the numbers of departures, the trend in the average
duration of managerial spells is relentlessly downward. By the end of the period,
the duration of the average managerial spell was around two seasons. The average
duration figure for seasons 1974–1977 is magnified by the closure of several long-
duration spells with top-level clubs that were noted previously (see Table 9.1). The
era when even a moderately successful manager could expect to remain in post for
a period as long as fifteen or twenty years, a common occurrence before the Second
World War and during the 1950s and 1960s, was drawing rapidly to a close; and sev-
eral star performers from the latter part of this era (Shankly, Nicholson, Catterick,
Revie) were, by the mid-1970s, among a handful of survivors of a dying breed.
Table 9.8 reports the average durations of terminating spells by tier and mode
of departure (involuntary or voluntary), and the average win ratio of terminating
spells. The latter is invariably higher for voluntary departures than it is for invol-
untary departures.
9.4 Determinants of managerial change

Grusky’s (1963) investigation of managerial succession in North American base-
ball appears to have been the first academic study to consider the possibility of a
bi-directional link between managerial change and team performance. For sixteen
professional baseball teams, Grusky obtains a negative correlation between the
number of managerial changes and the average standing of each team, for both of
the periods 1921–41 and 1951–8. In other words, teams that made more manager-
ial changes tended to underperform. Grusky also obtains a negative correlation
between the change in the average duration of managerial tenure between these
two periods, and the change in average standing. The teams that increased their
Determinants of managerial change 271
Table 9.8 Average duration (matches completed) and average win ratio, in
terminating managerial spells by tier, English League, 1974–2009 seasons
Average duration (matches completed) of terminating managerial spells

T1 115.9 183.5 134.1
T2 99.9 135.8 106.9
T3 97.8 133.2 105.4
T4 78.1 82.7 78.7
All tiers 95.2 137.3 103.4
Average win ratio of terminating managerial spells
T1 0.471 0.525 0.485
T2 0.457 0.529 0.471
T3 0.455 0.546 0.475
T4 0.420 0.510 0.432
All tiers 0.447 0.529 0.463
rate of managerial turnover the fastest tended to experience a decline in average

standing. Grusky (1963) and Gamson and Scotch (1964), in a comment on the
Grusky paper, go on to identify three possible theories of managerial turnover:
• The common sense theory. When a team is underperforming, the manager is
held accountable and is likely to be replaced. If a more effective replacement is
hired, or if the incoming manager can learn from the mistakes of his predeces-
sor, team performance should improve post-succession. This would be sufficient
to explain Grusky’s finding of negative correlation between managerial turn-
over and team performance.
• The vicious circle theory. Again, poor performance tends to trigger manager-
ial change. But the disruptive effect of a change of leadership, particularly if it
occurs within-season, tends to make matters worse rather than better. Faltering
teams become trapped in a vicious circle of high managerial turnover and declin-
ing performance. Grusky’s empirical results therefore reflect a bi-directional
relationship between turnover and performance.
• The ritual scapegoating theory. On average, the appointment of a new manager
makes no difference to team performance. The latter depends primarily on the
supply of playing talent and not on the managerial input. A change of manager
is not and cannot be expected to improve matters overnight. The supply of play-
ing talent is determined primarily by the effectiveness of scouting and player
development programmes, for which club owners are ultimately responsible.
Grusky’s inverse relationship between turnover and performance is an effect of
scapegoating: during a poor spell, owners can appease disgruntled spectators
and perhaps deflect attention from any shortcomings in their own contribution,
by offering the manager’s head in ritual sacrifice.
Grusky himself argues in favour of the vicious circle, bi-directional causal-
ity theory, on the grounds that it captures a wider range of possible interactions
between performance and succession than the common sense theory, in which
causality is unidirectional. Gamson and Scotch are more hesitant to discard the
latter, but in the absence of any clear evidence as to whether a change of manager
actually does affect team performance, argue that the ritual scapegoating explan-
ation should be adopted as a working hypothesis. This debate provided sufficient
motivation for a number of other researchers to seek to determine whether or
not a change of manager affected team performance after the change took place.
These studies are reviewed in Section 9.6.
Scully (1992, 1994, 1995) and Fizel and D’Itri (1997) investigate the common sense
theory for several North American professional team sports, using more sophisti-
cated empirical techniques than were available to the first-generation researchers
whose findings are outlined above. In developing a model of managerial departure
in MLB, the NBA and NFL, Scully views the role of the manager as twofold. First,
he seeks to maximise points scored subject to the quality of his team’s offensive
skills relative to his opponents’ defensive skills, and to minimise points conceded
subject to the quality of his team’s defensive skills relative to his opponents’ offen-
sive skills. Second, he seeks to convert the realised points scored and conceded into
the maximum attainable win ratio. Scully assumes that all managers perform the
first of these tasks with equal efficiency. But there are differences between managers
in terms of the efficiency of the conversion of points into win ratios. Each man-
ager achieves an efficiency score based on the residuals, ε, from a regression of the
form:
ln( W ) = β0 + β1ln(S / OS) + ε [9.1]
where W is the win ratio and S and OS are points scored and conceded,
respectively.
Scully (1995) obtains ε from OLS estimation of [9.1]. It is shown that career
efficiency is positively related to career length, demonstrating (indirectly) that
efficiency does influence the decision to terminate the manager’s appointment.
Evidence is also found that in baseball and basketball, player-managers are less
efficient than their non-playing counterparts, and that average efficiency increased
over time in all three sports.
Scully (1995) models the decision to terminate the manager’s appointment dir-
ectly. In principle the decision to terminate or retain should be based on a com-
parison between the actual and the maximum attainable win ratio, given the
quality of playing resources at the manager’s disposal. In practice, however, many
other factors also come into play: owners’ subjective judgements about the quality
of their own and their opponents’ players; the manager’s relationships with both
the owner and the players; the manager’s past experience; the possibility that the
owner may wish to sacrifice the manager primarily in order to send signals to play-
ers, supporters or the media; and last but not least, the influence of luck on team
performance.
In Scully’s empirical estimations, the dependent variable is a 0–1 dummy that
indicates either continuation or termination of the manager’s position within or at
the end of each season. The independent variable is the team’s league standing at
the end of the season (or at the time of departure for within-season departures).
Although this measure is crude because it takes no account of either playing
strengths or managerial efficiency (as defined above), it is justified on the grounds
that ‘[f]or both owner and manager, [the club’s standing] is the bottom line’ (Scully,
1995, p161). No distinction is made between involuntary and voluntary manager-
ial departures. The linear probability model and logit and probit are used for esti-
mation. The main finding, that in both sports the decision to terminate is highly
sensitive to league standing, is robust across all estimation methods.
Scully (1994) suggests several technical improvements and refinements in the
estimation of managerial efficiency, and the modelling of the decision to termin-
ate the manager’s appointment. Equation [9.1] is estimated as a stochastic frontier
in order to obtain the managerial efficiency scores. The latter (rather than league
standings) are the main explanatory variable in regressions that model the decision
to terminate the manager’s appointment, making this decision dependent directly
on the manager’s own performance, rather than on player quality. Regressions are
estimated in the form of survivor functions. The main finding is that efficiency
exerts a strong positive influence on average managerial survival time.
Fizel and D’Itri (1997) investigate the decision to terminate the coach’s
appointment in US college basketball. Efficiency scores are generated by com-
paring win ratios with standardised measures of own-team playing talent and
opponent strength. Playing talent is measured ex ante and independently of
team performance, using a ‘talent index’ compiled from expert assessments
of each player’s talent when he first entered college. Ex ante measurement of
inputs avoids most of the simultaneity problems inherent in the usual ex post
measures based on observed performance. Simultaneity problems normally arise
because player performance depends partly on the coach’s contribution, which
the model seeks to capture in the error term and not in the input measures. In
probit regressions indicating either continuation or termination of the coach’s
position, and estimated separately for involuntary and voluntary departures,
Fizel and D’Itri find that efficiency and playing talent both affect the probabil-
ity of involuntary departure in the expected direction. The number of years’
service is positively related to the probability of departure. The significance of
the efficiency and playing talent variables disappears, however, if the win ratio is
also included among the covariates. This might suggest that a lack of adequate
playing resources is an ineffective mitigating plea for a coach failing to achieve a
sufficient proportion of wins.
In the recent North American literature, attention has focused on the extent to
which managerial survival is influenced by the race of the coach (manager). For
example, Volz (2009) shows that after controlling for team performance and per-
sonal characteristics, baseball managers from ethnic minorities are on average 9.6
percentage points more likely to survive to the following season. Accordingly, the
relatively small number of ethnic minority managers in MLB is not due to a lower
probability of survival. A similar result is obtained by Mixon and Trevino (2004),
in a study of US college football coaches. Other things being equal, black coaches
face a significantly lower dismissal probability than their non-black counterparts.
However, Kahn (2006) and Fort, Lee and Berri (2008) identify no significant dif-
ferences in the firing and retention of NBA coaches by race.
For football, Audas, Dobson and Goddard (1999) examine the causes of man-
ager departure (both voluntary and involuntary) in the English league by esti-
mating competing-risks hazard functions specified in accordance with the Cox
proportional hazards model. The job departure hazard is found to be dependent
on both recent match results, and the win ratio achieved over the entire managerial
spell. The probability of being fired is positively and significantly correlated with
age, but not with previous managerial experience. Section 9.5 presents results from
the estimation of a similar empirical model, based on an updated version of the
data set that was used in the 1999 study.
Bachan, Reilly and Witt (2008) focus on the determinants of (involuntary) man-
agerial departure using English match-level data for ninety-one clubs covering sea-
sons 2003 to 2005. A discrete-time logit regression is used, on the grounds that it
is less restrictive in dealing with time-varying covariates than the competing-risks
specification. The fact that each manager is observed on several occasions permits
the inclusion of controls for unobservable manager heterogeneity (ability), in the
form of random- or fixed-effects. The teams are separated into two groups: those
positioned in or near the relegation zone at the start of each month, and those that
were not. The probability that the manager of a relegation-threatened team will be
dismissed is estimated to be 3.6 percentage points higher than the probability in
the case of a more successful team.
Salomo and Teichmann (2000) estimate logit regressions to model managerial
departure, using German Bundesliga data for the period 1979–98. As in other
studies, poor performance in the most recent few matches increases the probability
of departure. The appointment of a new board president increases the likelihood
that the manager will be dismissed; and managers of teams exposed to high levels
of media coverage (especially when performing badly) are more likely to be dis-
missed. Hautsch et al. (2001) estimate probit regressions, Weibull regressions and
Cox proportional hazard functions, using German Bundesliga data for the period
1963–98. The managers of top teams have significantly longer survival times than
the managers of middle- and lower-ranked teams. The probability of dismissal is
higher for older managers, while the survival probability increases with the num-
ber of previous jobs. More recently, Frick, Barros and Passos (2009) apply four
specifications to German Bundesliga data from seasons 1982 to 2003. In all speci-
fications, higher-paid managers have the same prospects of survival as low-paid
managers; but the managers of large-market teams are at higher risk of dismissal
than managers of small-market teams.
De Dios Tena and Forrest (2007) examine twenty within-season (involuntary)
managerial departures in the top division of the Spanish La Liga during the 2003,
2004 and 2005 seasons, by estimating probit regressions. The prospect of possible
relegation is a major cause of involuntary managerial departure. Other explana-
tory variables include the time of season (represented by match round), and a
measure of ‘managerial efficiency’ (number of places by which a club’s current
league position is superior to its ranking in terms of its (wage) budget). Dummy
variables identify teams having already dismissed a manager in the current season,
and a defeat in the team’s most recent match. The coefficient on the latter is sig-
nificant, suggesting that the timing of the decision to remove a manager is heavily
influenced by the most recent match result.
Most of the studies reviewed above use either logit or probit regression, or para-
metric or semi-parametric hazard function models. In constrast, Frick, Barros
and Prince (2010) estimate a mixed logit model, in which there is variation in the
coefficients of the model between different managerial spells. The idea is to allow
for heterogeneity into the manner in which the characteristics of the manager or
club determine the probability of dismissal. German Bundesliga data are used,
covering 398 team-season observations featuring 39 clubs over seasons 1982 to
2003. The probability of managerial departure (both involuntary and voluntary)
is positively related to the manager’s salary and the team payroll, and negatively
related to the win ratio. Previous managerial experience reduces the probability of
departure.
A novel and interesting perspective on the topic of managerial tenure is
provided by Aidt et al. (2006), who fit power-law distributions to data on the
durations of 7,183 managerial spells recorded in several sports and countries
over the period 1874–2005: football (England, Switzerland, France, Spain
and Germany), baseball (US and Japan), and American football (the NFL).
According to a power-law distribution, the probability than any individual
managerial spell in sport i achieves duration t (in years) and terminates before
t+1 takes the form
p(t) = b(i)t a(i) [9.2]
where a(i) is the key parameter that describes the shape of the power-law dis-
tribution, and b(i) is a normalising constant. It has been shown elsewhere that
power-law distributions characterise a wide variety of both natural and social phe-
nomena. Furthermore, the emergence of distributions of this kind can be repli-
cated by means of stochastic simulation, in which the underlying processes that
generate the data are entirely random. As explained by Aidt et al. (2006, p698), ‘a
surprising implication is that factors, such as talent, effort choices, and selection
and matching processes that normally play a role in tenure, are not essential for
understanding the dynamic evolution of hiring and firing in competitive sports’.
Aidt et al. develop a theoretical model that is capable of generating a power-law
distribution for job tenure, in which the processes that determine success or failure
are completely random. The core model is based upon a round-robin tournament
that links the managers (and teams) in a simple network. For example, twenty
managers (teams) play against each other twice, once at home and once away,
and in each game the win-draw-lose probabilities correspond to those observed in
English football over the long term. The probabilities are independent of the iden-
tities of the two managers. Managerial turnover is governed by a number of rules
that relate job tenure to performance.
The model is initialised with twenty randomly selected managers, each with
two attributes, reputation and tenure. Each manager’s reputation when the sea-
son starts is a random positive integer with a constant probability distribution
between 0 and the poaching threshold (see below), and each manager starts with a
random tenure duration distributed uniformly between one and forty years. Each
manager’s reputation is enhanced by a win and diminished by a loss. Job tenure
depends upon the evolution of the reputation variable. The manager remains in
post while reputation remains within a lower bound (the ‘firing’ threshold) and an
upper bound (the ‘poaching’ threshold). Tenure also terminates due to relegation,
or retirement (when the age variable exceeds an ‘ossification’ threshold).
With appropriate choices for the threshold parameters, the model generates a
power-law distribution for tenure duration, with exponents in the range 2<a(i)<3,
which correspond closely to the empirical estimates. By demonstrating that the
observed distribution of managerial spell durations corresponds closely to simu-
lated distributions that can be generated using random numbers only, Aidt et al.
lend credence to the notion that success or failure in competitive sport is a matter
of chance. The activity of hiring and firing managers may involve little or no more
science than either playing roulette or picking lottery numbers.
9.5 Estimation of hazard functions for managerial departure

Section 9.5 reports the estimation results for involuntary and voluntary manager-
ial job departure hazard functions. The hazard function estimations update those
reported by Audas, Dobson and Goddard (1999), by extending the data set to the
end of the 2009 season. There are a few minor changes to the variable definitions
and model specification.
We begin by describing the essentials of duration analysis, including the specifi-
cation of hazard functions for events such as job departure. Keifer (1988) provides
a comprehensive review of this topic. The following functions can be defined to
describe the proportions of spells that have survived or terminated after various
durations:
Estimation of hazard functions for managerial departure 277
St = the proportion of spells that survive to duration t;

Ft = 1 – St = the proportion of spells that fail to survive to duration t;
ft = St – St+1 is the unconditional probability that any spell terminates between
durations t and t+1, so St+1 = St – ft;
ht = ft/St is the conditional probability that a spell that has survived to duration t
terminates between durations t and t+1, so St+1 = St(1–ht).
St is the survivor function, and Ft is the distribution function of duration. ht is the
hazard function: the probability that a managerial spell that has survived until
duration t will terminate between durations t and t+1. In other words, ht is the
probability of managerial departure between t and t+1, conditional upon survival
to duration t. ft is the probability function of duration: the probability that any
spell will terminate between durations t and t+1. In other words, ft is the uncon-
ditional probability of managerial departure between t and t+1. Crude numerical
estimates of St, Ft, ft and ht can be obtained from tabulations of the distribution
of spells by duration. This type of estimation is purely descriptive, however. To
model the data, it is necessary to explain variations in these functions (especially
the hazard function, ht) using other explanatory variables or covariates.
When extending this simple framework in order to model the football man-
ager’s job departure hazard, a complication arises because it is necessary to dis-
tinguish between the two modes of departure: involuntary and voluntary. This
means that effectively, there are two separate hazards at work depleting the num-
ber of survivors among any initial cohort of managers, and two separate hazard
functions to be estimated. Extending the previous notation, f 1t and f 2t are defined
as the unconditional probabilities that any one of the original spells terminates
between durations t and t+1 involuntarily and voluntarily, respectively. h1t and h2t
are the corresponding hazards, measuring the conditional involuntary and vol-
untary departure probabilities for spells that have already survived to duration t.
Then St+1 = St – f t1 – f t2, and Ft = 1 – St as before.
A further extension necessitates making the hazards specific not only to the dur-
ation, but also to the spell in question. This means replacing h dt with hi,t
d
, the condi-
tional probability that, having survived to duration t, managerial spell i terminates
between durations t and t+1, either involuntarily (d = 1) or voluntarily (d = 2).
hi,td can then be made dependent on a number of covariates, reflecting the perform-
ance of the manager and the team to which spell i belongs, as well as human capital
attributes of the manager (such as his age or previous managerial or playing experi-
ence). One way of expressing the dependence of hi,td on other covariates is by using
Cox’s (1972) proportional hazards specification:

h d = h d exp(x  )
i,t t i,t d [9.3]
The terms in [9.3] are as follows:

~
hdt = the baseline hazard of departure by mode d (involuntarily if d = 1, voluntarily
if d = 2);
xi,t = a vector of covariates reflecting the performance and characteristics of the

team and manager at duration t within the i’th managerial spell;
βd = a vector of coefficients to be estimated. The cross-product xi,t′βd determines
whether the actual hazard of departure by mode d is above or below the base-
line hazard.
Maximum likelihood estimation is used to obtain estimates of the coefficients
in βd. The estimated coefficients can be interpreted in a manner similar to those
obtained from other types of regression: each coefficient reflects the sensitivity of
the hazard to variation in the covariate concerned. t-subscripts refer to matches,
and the job departure indicator variables show, for each match t, whether the
manager’s appointment was terminated (involuntarily or voluntarily) between the
completion of match t and match t+1.
Job departure hazards in the match-level model are calculated over sequences
of match results for each team. This implies that each match generates two obser-
vations, one for each team, and therefore appears in the data set twice. The model
does not allow the probability that a manager departs between matches t and t+1
of his reign to be influenced by the standard of opposition his team met in match
t (or in preceding matches). Involuntary departure is usually triggered by a run
of several poor results, over which the standard of opposition rapidly tends to
converge towards the divisional average. The model also does not allow for inter-
dependence between the departure probabilities for managers who are in post
with different clubs at the same time. In practice, however, such interdependence
might be significant. There are numerous instances when, having just lost the ser-
vices of its manager to Club A, Club B has offered the post to the manager of
Club C, which in turn has attempted to secure Club D’s manager as his successor.
Nevertheless, the incidence of chain-reactions of this kind is probably lower today
than it has been in the past, thanks partly to an increased tendency for clubs to
appoint foreign managers, as well as serious attempts that have been made by the
football authorities (including the League Managers’ Association) to discourage
unauthorised poaching of managers currently under contract with other clubs.
The full set of covariate definitions for the job departure hazard functions is as
follows:
Ri,t-k = result of match t–k within the current managerial spell, defined for k = 0 …
17, Ri,t-k = 1 denotes a win, 0.5 denotes a draw, 0 denotes a defeat.
Pi,t = league position of the team going into match t (92 = 1st in T1, 91 = 2nd in
T1, … , 1 = 24th in T4).
P̃i = league position of the team going into the first match of the current man-
ager’s spell.
Ai,t = age of the manager in years at the time of match t.
Mi = number of months of previous management experience with English league
clubs of the manager, prior to the date of his current appointment.
Si = number of years the manager had previously been employed as a player with
English league clubs, prior to the date of his current appointment.
Oi = 1 if the manager had management experience in Scotland or in any foreign
country, prior to the date of his current appointment; 0 otherwise.
Ii = 1 if the manager had represented his country at international level as a player,
prior to the date of his current appointment; 0 otherwise.
Yt = linear time trend (0 if match t was played in 1973 season, 1 if match t was
played in 1974 season, … , 36 if match t was played in 2009 season).
Augi,t = 1 if match t took place in August, 0 otherwise. Other month-within-season
dummy variables are similarly defined for months September to May. If match t
played during April or May is the final match of the season, Apri,t = Mayi,t = 0.
Close-season departures are the reference category. The coefficient on each
month-within-season dummy variable represents the difference between the
hazard for the relevant month, and the close season hazard.
Table 9.9 shows the crude estimates of the survivor, distribution, hazard and
probability functions, St, Ft, ft and ht respectively, for the 1,578 managerial spells
recorded between the 1973 and 2009 seasons, over duration intervals of thirty
matches. The first four columns indicate that 258 of the 1,578 spells terminated
before 30 matches had been completed, and a further 21 were right-censored at a
duration of fewer than 30 matches. Therefore 1,299 spells achieved a duration of at
least 30 matches. Of these, 364 terminated before 60 matches had been completed,
and a further 17 were right-censored at a duration between 30 and 59 matches,
leaving 918 spells that achieved a duration of at least 60 matches; and so on. These
data are used to calculate the survivor function (the probability that any spell
survives to duration t), the distribution function (the probability that any spell
fails to survive to duration t), the hazard function (the unconditional probabil-
ity of departure between durations t and t+30), and the probability function (the
probability of departure between durations t and t+30 conditional on survival to
duration t).
The final two columns of Table 9.9 report two disaggregated survivor functions,
calculated using the data for seasons 1973–1992 and 1993–2009 separately. These
two columns provide an indication of the implications of the increased rate of
managerial job turnover witnessed in the latter period for the survival rates of
managerial spells to various durations. During the 1973–92 period, for example,
52.8 per cent of all managerial spells lasted for at least 90 matches, and 22.2 per
cent lasted for at least 180 matches. The corresponding percentages for the 1993–
2009 period were 35.7 per cent and 9.3 per cent, respectively.
Table 9.10 reports the estimated hazard functions. All coefficients are reported in
hazard ratio format. This means we report the estimated values of exp(βd,j), where
βd,j is an element of the vector of coefficients βd in [9.3]. The hazard ratios represent
the multiplicative effect on the baseline hazard of a unit increase in the covariate;
for example, a hazard ratio of 1.2 indicates that the hazard increases by 20 per cent
Table 9.9 Distribution of complete and right-censored managerial spells by duration (in matches), crude estimates of the survivor,
distribution, hazard and probability functions, English League, 1973–2009 seasons
Right- Probability
Spells censored Hazard function:
that spells, still function: probability of Survivor Survivor
Duration survived Departures live at t+s unconditional departure in function, function,
in to between for s<30 at Survivor Distribution probability of (t, t+30) 1973–1992 1993–2009
matches, duration durations t end of 2009 function, function, departure conditional on seasons seasons
t t and t+30 season St Ft in (t, t+30), ht survival to t, ft only only
0 1578 258 21 1.000 0.000 0.165 0.165 1.000 1.000

30 1299 364 17 0.835 0.165 0.236 0.282 0.891 0.780
60 918 241 14 0.600 0.400 0.159 0.265 0.691 0.510
90 663 180 11 0.441 0.559 0.121 0.273 0.528 0.357
120 472 130 6 0.320 0.680 0.089 0.278 0.408 0.238
150 336 111 2 0.231 0.769 0.077 0.331 0.314 0.154
180 223 71 2 0.155 0.845 0.050 0.320 0.222 0.093
210 150 30 0 0.105 0.895 0.021 0.200 0.161 0.055
240 120 31 1 0.084 0.916 0.022 0.259 0.129 0.043
270 88 21 1 0.062 0.938 0.015 0.242 0.096 0.032
300 66 17 0 0.047 0.953 0.012 0.258 0.077 0.021
330 49 12 0 0.035 0.965 0.009 0.245 0.056 0.016
360 37 6 0 0.027 0.974 0.004 0.162 0.047 0.011
390 31 7 0 0.022 0.978 0.005 0.225 0.040 0.008
420 24 — — 0.017 0.983 — — 0.033 0.005

if the covariate increases by one. The hazard ratio is greater than one for βd,j>0, and
smaller than one for βd,j<0. The z-statistics are for tests of H0: βd,j = 0.
Columns (1) and (2) of Table 9.10 show estimates of the involuntary departure
hazard function using the data for seasons 1973–2009. Column (1) includes all
of the covariates; column (2) omits those covariates with estimated coefficients
that are close to zero. Columns (3) and (4) show the results for the estimation
of the same model as in column (2), using the data for seasons 1973–1992 and
1993–2009 separately. Finally, columns (5) and (6) report the voluntary departure
hazard functions using the data for seasons 1973–2009. Column (5) includes all of
the covariates; column (6) omits those covariates with estimated coefficients that
are close to zero. Since the number of voluntary departures is smaller than the
number of involuntary departures, it is not possible to obtain reliable estimates of
the voluntary departure hazard functions over shorter periods. Accordingly the
latter are not reported.
In the involuntary job departure hazard functions, the importance of current
and recent match results is reflected in negative and significant coefficients on
Ri,t-k (hazard ratios below one) for up to seventeen matches prior to the current
match. The absolute magnitude of the negative effect diminishes fairly regularly
as the number of lagged matches increases. The most recent matches therefore
have a greater effect on the hazard than the more distant ones, as expected. In
general, the coefficients on Ri,t-k confirm that the decision to terminate a man-
ager’s appointment involuntarily is often heavily influenced by short-term consid-
erations. A striking feature of the comparison between the involuntary departure
hazard functions for 1973–1992 and 1993–2009 is a large increase in the strength
of the relationship between the team’s most recent match result Ri,t and the job
departure hazard. This appears suggestive of an ultimate form of short-termism,
with the timing of the decision to terminate a manager’s appointment, if not the
decision itself, being more heavily influenced by the last match result in the latter
period than it was previously.
Among the other covariates of the involuntary job departure hazard, the move-
ment in the team’s league position since the start of the manager’s spell is a highly
significant determinant of the involuntary hazard: managers who are able to
achieve and sustain an improvement relative to the team’s league position at the
commencement of the spell have significantly greater job security. The coefficients
on the age variables suggest that the probability of involuntary departure increases
from around age 37 upwards. The involuntary departure hazard is inversely related
to the amount of previous management experience prior to the commencement of
the incumbent’s current spell. This might reflect a form of selection effect: man-
agers who have been appointed to more than one post during their careers tend
to be of higher quality on average than those whose current appointment is their
first. The other manager human capital variables are all insignificant in the invol-
untary hazard function. The coefficient on the time trend is highly significant, as
are the coefficients on the month-within-season dummies. The negative signs on
Table 9.10 Involuntary and voluntary managerial job departure hazard functions: estimation results
Seasons 1973– 1973– 1973– 1993– 1973– 1973– 1973– 1973– 1973– 1993– 1973– 1973–
2009 2009 1992 2009 2009 2009 2009 2009 1992 2009 2009 2009
Mode of inv. inv. inv. inv. vol. vol. inv. inv. inv. inv. vol. vol.
departure (1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Ri,t .278 .278 .380 .205 .864 .875 (Pi,t − P̃i) .966 .966 .964 .969 1.017 1.018
−16.2 −16.2 −8.75 −13.7 −0.94 −0.87 −14.6 −14.6 −10.7 −9.10 3.38 3.79
Ri,t–1 .393 .393 .394 .389 .880 .905 Ai,t .928 .928 .848 .956 1.144 1.143
−12.7 −12.7 −8.57 −9.30 −0.83 −0.65 −1.70 −1.69 −2.39 −0.72 1.16 1.14
Ri,t–2 .568 .568 .671 .495 .767 .786 Ai,t2 1.001 1.001 1.002 1.001 .998 .998
−8.02 −8.02 −3.89 −7.77 −1.72 −1.57 2.31 2.32 2.82 1.14 −1.36 −1.35
Ri,t–3 .528 .529 .509 .542 .766 .784 Mi .998 .998 .999 .998 1.003 1.003
−8.92 −8.90 −6.34 −6.22 −1.70 −1.57 −2.49 −2.68 −1.29 −2.25 1.80 1.88
Ri,t–4 .535 .535 .537 −.535 .941 — Si 1.000 — — — .987 .987
−8.82 −8.81 −5.95 −6.34 −0.39 −0.02 −1.04 −1.12
Ri,t–5 .655 .656 .617 .683 1.034 — Oi 1.078 — — — 1.438 1.468
−6.09 −6.08 −4.69 −3.99 0.22 0.56 1.40 1.50
Ri,t–6 .630 .631 .571 .700 1.045 — Ii 1.056 — — — 1.365 1.371
−6.61 −6.60 −5.39 −3.72 0.28 0.82 2.15 2.17
Ri,t–7 .678 .679 .643 .706 1.192 — Yi,t 1.021 1.021 1.017 1.015 1.011 1.011
−5.55 −5.55 −4.26 −3.61 1.12 7.25 7.45 2.19 1.76 1.63 1.62
Ri,t–8 .673 .673 .665 .689 .933 — Augi,t .023 .023 .010 .042 .006 .005
−5.69 −5.68 −3.94 −3.90 −0.44 −21.2 −21.2 −13.8 −14.7 −9.99 −10.0
Ri,t–9 .777 .777 .755 .765 1.148 — Sepi,t .063 .063 .042 .085 .020 .020
−3.63 −3.63 −2.72 −2.80 0.88 −22.7 −22.8 −15.7 −15.5 −12.1 −12.2
Ri,t–10 .856 .857 .812 .924 .900 — Octi,t .085 .085 .080 .090 .047 .047
−2.26 −2.24 −2.03 −0.83 −0.67 −22.4 −22.5 −15.7 −15.6 −13.2 −13.3
Ri,t–11 .823 .824 .728 .901 .848 — Novi,t .108 .108 .096 .117 .045 .044
−2.83 −2.82 −3.10 −1.10 −1.05 −20.4 −20.4 −14.2 −14.2 −12.6 −12.7
Ri,t–12 .791 .792 .817 .747 1.351 — Deci,t .077 .077 .085 .076 .047 .046
−3.36 −3.35 −1.98 −3.00 1.89 −21.9 −21.9 −14.5 −15.6 −12.9 −13.0
Ri,t–13 .855 .856 .911 .843 1.313 — Jani,t .101 .100 .124 .082 .044 .043
−2.24 −2.23 −0.91 −1.76 1.70 −19.8 −19.8 −12.7 −14.7 −11.7 −11.8
Ri,t–14 .798 .798 .763 .832 1.118 — Febi,t .086 .086 .087 .086 .015 .015
−3.21 −3.20 −2.62 −1.88 0.71 −20.9 −20.9 −13.8 −15.2 −10.6 −10.6
Ri,t–15 .898 .899 .851 .923 1.115 — Mari,t .067 .067 .070 .064 .022 .022
−1.53 −1.51 −1.56 −0.81 0.68 −23.8 −23.8 −16.4 −17.0 −13.1 −13.2
Ri,t–16 .881 .882 .806 .932 1.025 — Apri,t .051 .051 .052 .049 .010 .010
−1.81 −1.80 −2.08 −0.72 0.15 −24.2 −24.2 −16.8 −16.9 −11.6 −11.7
Ri,t–17 .868 .869 .836 .904 1.359 — Mayi,t .072 .072 .069 .080 .019 .019
−1.99 −1.97 −1.73 −1.01 1.90 −11.1 −11.1 −8.99 −6.07 −5.49 −5.49
obs. 150,348 150,348 81,172 69,176 150,348 150,348
departures 1,255 1,255 584 671 246 246
the latter reflect the pattern that the probability of departure during the close sea-
son is significantly higher than the probability during any particular month within
the season (even though the total probability of within-season departure is much
higher than the close season probability).
The few most recent match results appear to play a considerably less import-
ant role in determining the voluntary job departure hazard than they do in
determining the involuntary hazard. In column (5) the coefficients for the most
recent and four previous match results are all negative (hazard ratios below one),
and two of these five coefficients are significant at the 10 per cent level. For
more than four matches prior to departure, there is a mix of positive and nega-
tive coefficients, very few of which are significant. Ri,t–k are therefore omitted
from column (6) for k≥4. Any suggestion that the voluntary departure hazard
increases if recent results are poor is perhaps counterintuitive, given that volun-
tary departure occurs when a manager accepts a better job offer from elsewhere.
Audas, Dobson and Goddard (1999) suggest that while a run of poor results
might not be helpful in attracting alternative job offers, it might influence a man-
ager’s willingness to accept rather than turn down such an offer if one is forth-
coming. However, the statistical evidence for a negative relationship between
recent match results and the voluntary departure hazard is weak at best. In the
more parsimonious specification reported in column (6), none of the coefficients
on Ri,t–k is significant.
The change in league position since the start of the manager’s appointment is a
highly significant determinant of the voluntary departure hazard, with a reversal
of sign compared with the corresponding coefficient in the involuntary depart-
ure hazard. Managers who have achieved an improvement relative to the team’s
league position at the start of the spell are more likely to leave voluntarily. The
coefficients on the linear and quadratic terms in the manager’s age also exhibit a
reversal of signs compared with the corresponding coefficients in the involuntary
departure hazard. The probability of voluntary departure declines from around
age 34 upwards. Also subject to a sign-reversal is the coefficient on previous man-
agement experience prior to the commencement of the current spell. As before,
this seems likely to reflect a form of selection effect, with experienced managers
being of higher quality on average, and therefore more likely to receive further job
offers triggering voluntary departure.
The coefficients on the previous playing experience variable are negatively
signed (hazard ratios below one) but insignificant in columns (5) and (6). The
coefficients on the overseas management experience dummy are positively signed
(hazard ratios above one), but just fall short of being significant. However, the
coefficients on the international playing experience dummy are positively signed
(hazard ratios above one) and significant, suggesting that a high-profile former
playing career may be helpful for a manager who is already in post towards secur-
ing further managerial job offers. The coefficients on the time trend are posi-
tive but fall just short of being significant; while the negative coefficients on the
The effect of managerial change on team performance 285
month-within-season dummies (hazard ratios below one) are all highly significant
and larger in absolute terms than their counterparts in the involuntary depart-
ure hazard function. This is because the proportion of voluntary departures that
take place during the close season is higher than the corresponding proportion
of involuntary departures.
9.6 The effect of managerial change on team performance

Section 9.6 investigates whether, on average, a change of manager tends to lead
to an improvement in team performance. In a comment on Grusky’s (1963) art-
icle (see Section 9.4), Gamson and Scotch (1964) identify a key issue that must
be addressed by any empirical investigation of this question, by means of an
analogy.
If we compared average rainfall in the month preceding and the month following the perform-
ance of the Hopi rain dance, we would find more rain in the period after. The dance is not
performed unless there is a drought, so such a comparison would be misleading. Nevertheless,
this ‘slump-ending’ effect may help to account for the tenacity of belief in the effectiveness of
the ritual. (Gamson and Scotch, 1964, p71)
Team performance, like rainfall and many other natural and man-made phe-
nomena, exhibits a tendency for mean-reversion or regression towards the mean
over time. Successful teams do not remain successful forever; and teams that are
unsuccessful eventually find the wherewithal to improve. This means that on aver-
age, the performance of the most successful teams at time t will tend to deteri-
orate at time t+1, and the performance of the least successful teams will tend to
improve. If teams that change their manager are predominantly teams that are
unsuccessful at the time they make the change, some improvement in their average
performance in the season following the change is to be expected as a result of
the mean-reversion effect. This is the case even if the manager is just a figurehead,
who makes no contribution whatever to his team’s performance. Any investigation
of the relationship between managerial change and subsequent team performance
must therefore control for this mean-reversion effect.
Gamson and Scotch control for mean-reversion in a relatively crude but straight-
forward manner. In their comparison of the win ratios before and after twenty-
two mid-season managerial changes in MLB which took place between 1954 and
1961, match results during the two weeks prior to each change are excluded from
the calculations. In thirteen of the twenty-two cases, the team performed better
under the new manager than it did up to two weeks before the removal of the old
manager. This finding could be consistent with either the common sense or the
ritual scapegoating theories of managerial succession (see Section 9.4), but it does
support the vicious circle theory, which would anticipate deterioration in post-
succession performance. The assumption that the omission of two weeks’ results is
sufficient to exclude the mean-reversion effect is rather arbitrary, however.
Using the same data, Grusky (1964) tackles the mean-reversion problem by
comparing the post-succession win ratio for the remainder of the season with the
win ratio from the previous season. Grusky also distinguishes between inside suc-
cession, involving promotion of one of the club’s existing coaches or players to
the manager’s position, and outside succession, involving the appointment of an
outsider. Whereas inside succession was associated with an improvement in per-
formance on average, outside succession tended to be followed by further decline.
Allen, Panian and Lotz (1979) investigate similar issues, using a more extensive
MLB data set, covering the period 1920–73. Analyses of variance and covari-
ance are employed to identify the proportion of the variation in per-season win
ratios that can be explained by managerial succession. The mean-reversion effect
is controlled by including the previous season’s win ratio in the analysis of covari-
ance. Succession effects are relatively small, but statistically significant. Whereas
close-season succession tended to produce an improvement in performance in the
following season, within-season succession had the opposite effect. Teams tended
to perform better following insider succession than they performed following out-
sider succession. Multiple succession (two or more changes of manager within
the same season) had a damaging effect on performance. By attributing the entire
season’s performance of teams experiencing a within-season change of manager
to the successor, however, the methodology seems to be biased in favour of identi-
fying deterioration in performance relative to the previous season.
In a study of managerial succession effects in the NFL for the period 1970–8,
Brown (1982) uses panel techniques to estimate a multiple regression model, in
which current season performance depends on lagged performance and a man-
agerial succession dummy. All (unspecified) cross-sectional variation in organ-
isational structure between teams, which might affect both performance and the
decision whether or not to terminate the manager’s appointment, is controlled in
the error structure. As in other studies, teams that changed their manager within-
season are found to have underperformed during the same season. A match-level
comparison between the win ratios of teams that experienced an early-season
slump and changed their manager within-season, and the win ratios of a control
group that experienced a similar slump but did not change their manager, reveals
a similar recovery pattern for the two groups. This finding seems most consistent
with the ritual sacrifice theory of managerial succession.
Jacobs and Singell (1993) address the more general question as to whether
managers make a difference to the performance of the organisations they work
for. Using MLB data covering the period 1945–65, Jacobs and Singell model
team win ratios using a variety of measures of playing and managerial inputs. In
common with several other studies, managerial change is found to have a dam-
aging impact on the team’s win ratio post-succession. A set of individual man-
ager dummies, as well as measures of managerial experience, is significant in the
regressions that model win ratios. Managerial effects are also found to have influ-
enced changes in the individual performance of players who were traded between
The effect of managerial change on team performance 287
teams. Jacobs and Singell infer that managers do make a significant difference to
team performance.
Fizel and D’Itri’s (1997) study of performance and managerial turnover in US
basketball (see also Section 9.4) reports panel estimates of regressions that inves-
tigate the separate effects of involuntary and voluntary managerial departure on
performance, measured by win ratios. Among the covariates, the previous season’s
win ratio controls for mean-reversion. Interactions between the manager change
dummies and the team’s playing talent index, the incoming manager’s past experi-
ence, and his efficiency score from the preceding season are included. The latter
two variables allow for the possibility that not only the succession event itself, but
also the human capital attributes of the incoming manager, influence team per-
formance. Playing talent and managerial efficiency (but not experience) are highly
significant determinants of post-departure performance. It is unclear, however,
why performance should only be influenced by the new manager’s attributes in the
season immediately after his appointment, and not thereafter. Involuntary and
voluntary managerial departure both have a significant and negative impact on
subsequent performance.
White, Persad and Gee (2007) investigate the effects of within-season coach
turnover on team performance in the NHL from 1989 to 2003. Within-season
turnover leads to improved team performance in the short term, even when inex-
perienced coaches replaced experienced coaches. Hill’s (2009) MLB study finds
that managerial succession has a negative effect on team performance. Frequent
changes of manager are particularly damaging, although the relationship between
frequency of turnover and performance is non-linear.
Audas, Dobson and Goddard (1997) isolate the effect of a change of manager in
English football by drawing direct comparisons between the post-departure match
results of a group of teams that changed their managers, and the corresponding
results of a matched control group of teams that retained their managers, each of
which experienced a similar run of results to a team that did change its manager.
There is some evidence that the teams that changed their managers recovered from
a poor run less quickly than the teams in the control group, suggesting that within-
season managerial change tends to have a disruptive effect.
The literature on succession effects in football does not produce unequivocal
results. Using data for the Dutch top division, Bruinshoofd and ter Weel (2003)
compare the recovery of teams that had sacked their managers during a poor run
of results with that of a control group that had not. In each case, the recovery rate
of the first group was actually worse than that of the control group, casting doubt
on the efficacy of a dismissal strategy. In a study of over 8,000 matches played
in three divisions of the Belgian league, Balduck and Buelens (2007) found that
many of the teams whose performance had declined over a period of around two
months had dismissed their managers. Within four games under the direction of
a new manager, team performance improved. However, further analyses revealed
that this increase was due to mean reversion. A control group comprising teams
that experienced a similar dip in performance without dismissing their managers

showed a similar improvement.
Koning (2003) models individual match outcomes (difference in goals scored
between the two teams) in the Dutch Premier League as a function of quality indi-
cators for the two teams, and a dummy variable to reflect which team was playing
at home. The parameters are permitted to vary following a change of manager. For
the 1994 season, there is evidence of significant post-succession improvements in
both team quality and home-field advantage. The same results were not obtained
for any of the four following seasons, however, suggesting that team owners tend
to dismiss managers too readily. The firing decision may be taken for reasons other
than an expected improvement in performance, such as scapegoating in response
to pressure from supporters or the media. Also for the Dutch League, ter Weel
(2006) finds no evidence of any improvement in team performance following a
managerial change. The quality of the manager does not seem to matter in pre-
dicting managerial turnover.
Audas, Dobson and Goddard (2002) model match outcomes in England using
an ordered probit model. To establish whether match outcomes are influenced by
recent changes in manager they include among the regressors variables that reflect
recent changes of manager. Separate dummies are used for first match since the
change, second match since the change, and so on. The total impact is measured by
summing the coefficient estimates on these dummy variables. Managerial change
tends to have a negative effect on team performance in the remaining weeks of the
season. This provides further evidence for the scapegoating hypothesis of man-
agerial change: during a poor run of results, directors may dismiss the manager to
appease club supporters rather than in any real hope of turning around the club’s
fortunes in the short term.
Using Spanish La Liga data, de Dios Tena and Forrest (2007) (see also Section
9.4) use a similar empirical methodology, but distinguish between the impact of
managerial change on home and away match results. The effect of managerial
change on the results of away matches played over the remainder of the season is
insignificant, but there is a small and statistically significant improvement in home
match results. This finding is interpreted to suggest that while a new manager is
unable to produce any immediate technical improvement in the team, managerial
change may tend to rekindle home-team supporters’ enthusiasm, enhancing the
crowd’s contribution to home-field advantage.
Flores, Forrest and de Dios Tena (2008) present a similar analysis using data
from Argentina. Average away performance following a dismissal was typically
below expectations in view of club status and recent form. The adverse effect on
home results was less pronounced, suggesting a home-away differential in the suc-
cession effect that is similar to the results for Spain. A principal-agent model is
developed to describe a scapegoating explanation for managerial dismissal under
circumstances such that the decision-taker does not anticipate any improvement in
team performance. Wirl and Sagmeister (2008) use Austrian Premier League data
Managerial succession effects in English football 289
on almost 2,000 matches played during seasons 1995 to 2004 (inclusive), and 87
managerial changes, most of which were within-season. Managerial change tends
to increase the probability that the home team wins, irrespective of whether the
home team or the away team changed its manager. Although these effects are con-
sistent with those reported by de Dios Tena and Forrest (2007) and Flores, Forrest
and de Dios Tena (2008), they are small and not statistically significant.
Wagner (2010) considers whether the introduction of the three-points-for-a-win
rule in the German Bundesliga in the 1996 season influenced the managerial suc-
cession effect. The data cover 12,488 matches played by 48 teams under 281 differ-
ent managers during a period spanning seasons 1964 to 2003. The analysis is based
on a comparison between the league points achieved over four-match sequences
before and after each managerial change. As expected, a direct comparison of the
average points gained indicates a sharp post-departure improvement in perform-
ance. The magnitude of this effect was greater under three-points-for-a-win: the
differences in average league points per match between the pre- and post-departure
four-match sequences was 0.29 points for 1964–95 and 0.72 points for 1996–2003.
This effect does not disappear when the comparison is made with reference to a
control group that was as similar as possible to the teams that changed their man-
agers, except for the fact that they did not make a change.
De Paolo and Scoppa (2008) evaluate the effects of within-season manager-
ial change on team performance in Italy’s Serie A from seasons 2004 to 2008.
Match results are modelled using a dummy variable for home matches, two rela-
tive team quality measures (difference between the teams’ final league positions
in the current season, and difference between league points earned prior to the
current match), and a dummy variable indicating a change of manager previously
within the current season. Team and season fixed effects are included. The esti-
mate of the effect of the change of manager is obtained by comparing the average
performance of the team under the old manager and new manager (within the
same season). There is no evidence that managerial change produces any signifi-
cant improvement in team performance within the same season.
Hughes et al. (2010) aim to disentangle short-run and long-run effects of man-
agerial change, using English Premier League data covering seasons 1993 to 2004.
Managerial change may create a brief reprieve in poor performance, before under-
lying weaknesses reassert themselves in the long term.
9.7 Managerial succession effects in English football

Section 9.7 reports estimates of regressions that investigate the effects of manager-
ial change on subsequent team performance in English football, using an adapted
version of the results-based forecasting model that was developed in Chapter 4. In
particular, we investigate whether there is any statistical foundation for the wide-
spread belief in ‘new manager syndrome’, a tendency for football teams that part
company with their managers to start winning immediately after doing so. To this
end, dummy variables indicating the timing of within-season managerial depar-

tures are added to the forecasting model, in order to discover whether managerial
change has a discernible impact on individual match results immediately after the
change takes place.
The model specification is similar to that of the results-based forecasting model,
developed in Chapter 4 for the purpose of generating match result forecasts. With
this kind of specification, the tendency for mean-reversion is controlled, probably
more comprehensively than in most other specifications detailed above, by the
inclusion of a multitude of lagged match result and win ratio measures, which
reflect information on team performance over the past twenty-four months. Using
a match-level performance measure, it seems reasonable to restrict the investiga-
tion to within-season managerial changes only. Close season managerial changes,
which can offer players up to three months’ opportunity to adapt to the methods
of the incoming manager, seem unlikely to be disruptive to the same extent as
within-season changes.
The model developed in Chapter 4, Sections 4.4 and 4.5, is modified by add-
ing a set of suitably defined managerial change dummy variables. The effects of
a change of manager on the results of the next twenty matches (within the same
season) following the change are investigated. The manager change dummy vari-
ables are defined as follows:
Dik = 1 if team i changed its manager between the kth and k-1th matches before the
current match and if both these matches were played within the same season
as the current match for k = 1 … 20, and 0 otherwise.
Preliminary inspection of the data suggested that it was not feasible to identify
a separate effect for involuntary and voluntary departures, as there were too few
of the latter for reliable estimated effects to be obtained. As the managerial change
dummy variables are only operative for matches played within the same season as
the change, the maximum value of k is restricted to 20. The number of managerial
changes beyond which more than twenty matches were still to be played within the
same season is also too small for reliable estimated effects to be obtained for k>20.
In Chapter 4, the results-based forecasting model generates a latent variable, y*i,j ,
for the match between home team i and away team j, whose sign and value indicate
the direction and extent to which the relative strengths of the two teams should
influence the match result (see [4.3]). Since the effect on y*i,j of a recent change of
manager by the away team should be roughly equal but opposite in sign to the
effect of a change by the home team, it is possible to combine Dki and Dkj into a
single dummy, Zi,jk = (Dik − Dkj ). The coefficient on Zi,jk reflects the combined effect
of recent managerial changes (if any) by either team on the match result. This pro-
cedure reduces the number of separate coefficients to be estimated, and increases
the reliability of the estimates.
With reference to the variable definitions given in Sections 4.2 and 4.4, the latent
variable yi,j* is assumed to be a linear function of the following covariates:
• The past match results ‘points’ covariates, Pdi,y,s and Pdj,y,s.

• The recent match results covariates, Ri,k
H
, RAi,k, Rj,k
H
, Rj,k
A
.
• Dummy variables for the significance of the match for end-of-season outcomes,
SIGHi,j and SIGAi,j.
• Dummy variables for elimination from the FA Cup, CUPi, CUPj.
• The geographical distance covariate DISTi,j.
• The average past match attendance covariates APi,s, APj,s.
• A complete set of individual football season dummy variables.
The model is estimated using data for the seasons 1973 to 2009, inclusive. The
individual-season dummy variables control for the long-term shift in the direction
of reduced home-field advantage, identified in Chapter 3, Section 3.2.7 The expres-
sion for y*i,j is as follows:
20
y*i,j = {above covariates} + ∑ φk Z ki,j [9.4]
k =1
Table 9.11 reports the estimated values of the coefficients φk, obtained by esti-
mating [9.4] using all match observations from the 1973 to 2009 seasons (inclusive)
for which complete data are available. The estimated coefficients and standard
errors on all other covariates are similar to those reported in Chapter 4 (see Table
4.5); and to conserve space these coefficients are not reported here. The estimates
of the additional coefficients on the manager change dummy variables, Zki,j , and
their p-values, are reported in columns (1) and (2) of Table 9.11. The estimated
coefficient for the match immediately following a change of manager is negative
and significant at the 5 per cent level. This suggests that a within-season manager-
ial change is disruptive in the very short term: a team that changes its manager
tends to underperform, to an extent that is discernible on conventional statistical
criteria, in the match after the change takes place.
Among the other results in columns (1) and (2), the coefficients on Zi,j6 and Z11 i,j
are also significant at the 5 per cent level. Elsewhere there is a mix of positive and
negative estimated coefficients, but the overall preponderance is negative. It can be
argued, however, that the impact of a managerial change on any individual result is
of less interest than the cumulative impact on results over the remaining weeks or
months of the season. Column (3) therefore reports the cumulative values of the φˆ k’s
in column (1); and column (4) reports the corresponding p-values. For example,
row 1 of column (3) is φˆ 1 = –0.0819; row 2 is φˆ 1 + φˆ 2 = –.0819–.0289 = –0.1109;
and so on. The estimated cumulative effect of a managerial change on subsequent
results is uniformly negative for up to twenty matches after the change takes place,
and most of the reported cumulative effects are statistically significant.
Overall, these results suggest that on average, a change of manager that takes
place within-season tends to have an adverse effect on the results of matches played
during the remaining weeks or months of the same season. We find no statistical
evidence to support the popular belief in the ‘new manager syndrome’, that the
Table 9.11 The managerial succession effect: estimation results
Number of
matches
following Cumulative
managerial Effect on y*i,j p-value effect on y*i,j p-value
departure (1) (2) (3) (4)
1 −.0819 0.025 −.0819 0.025

2 −.0289 0.434 −.1109 0.034
3 −.0213 0.567 −.1322 0.042
4 .0004 0.991 −.1318 0.083
5 −.0084 0.824 −.1402 0.103
6 −.0914 0.017 −.2317 0.015
7 −.0519 0.177 −.2836 0.006
8 −.0075 0.847 −.2911 0.010
9 −.0584 0.139 −.3495 0.004
10 −.0019 0.962 −.3515 0.006
11 −.0913 0.024 −.4428 0.001
12 −.0127 0.754 −.4556 0.002
13 −.0345 0.405 −.4901 0.001
14 .0532 0.210 −.4369 0.006
15 .0134 0.756 −.4236 0.011
16 .0349 0.424 −.3887 0.027
17 .0367 0.412 −.3520 0.055
18 −.0465 0.310 −.3986 0.037
19 .0149 0.748 −.3836 0.054
20 −.0077 0.871 −.3913 0.059
performance of teams that part company with their managers tends to improve
immediately afterwards. The statistical evidence points in the opposite direction,
although quantitatively the average effect is very small: the peak cumulative effect
on y*i,j of around –0.49 translates into a loss of just under one league point over
thirteen matches post-departure.
Why, then, is within-season managerial turnover such a common occurrence
in English football? Audas, Dobson and Goddard (2002) consider several pos-
sible explanations. One possibility is a tendency for team owners to overestimate
their own abilities to achieve an improvement in performance by appointing a
more effective successor. A second explanation is that team owners tend to take
the longer view, and are willing to incur short-term costs in the form of slightly
reduced performance, in the expectation of realising long-term benefits. A final
explanation, which permits team owners to act in some sense rationally, while
still adopting very short time horizons, is that if a quick improvement in perform-
ance is required to stave off the threat of relegation, a change of manager might
represent a final ‘throw of the dice’, in the form of a gamble based on an increase
in the variance of performance post-departure. The increase in variance, due to
the uncertainty over the immmediate impact a new manager might achieve, could
increase the probability that the improvement in performance required to avoid

relegation is forthcoming, even though a negative mean effect on performance is
identified in the results reported in this section.
Conclusion
The job description of the football team manager includes the selection, super-
vision and coaching of playing staff, and the formulation of tactics and strat-
egies. Many football managers, especially in professional football’s lower tiers,
are also responsible for the buying and selling of players, salary negotiations and
various administrative duties. In many upper-tier clubs some or all of these tasks
have been delegated to functional specialists, while the manager is supported in
his responsibilities for playing affairs by a team of subordinate coaches and assist-
ants. At the large-market clubs in particular, but at all levels to some extent, the
manager needs all the communication skills of a polished media performer. In
almost every respect, the early twenty-first-century football manager would have
been unrecognisable to his late nineteenth-century and early twentieth-century
forerunners, the club secretary and the secretary-manager. Several of the giants of
twentieth-century football management in England, including Herbert Chapman
and Frank Buckley during the inter-war period, and Matt Busby, Bill Shankly,
Don Revie and Brian Clough during the post-war years, personify the changes
that have taken place, and have helped define the modern-day football manager’s
remit and key attributes.
In principle, the football manager’s contribution to team performance can be
broken down into a direct and an indirect component. Taking the collection of
players at his disposal as given, the manager’s direct contribution is to maximise
performance through astute team selection, superior tactics and powers of motiv-
ation. Over the longer term, the manager’s indirect contribution is to enhance his
existing squad, by coaching players so as to enhance their skills, and by effective
dealings in the transfer market. Despite the fact that the output of the professional
sports team is easily measured by win ratios or league standings, and many data
on the characteristics of the major inputs (players and manager) are easily avail-
able and in the public domain, the indivisibility of the team effort which ultimately
determines performance poses a major challenge for any researcher seeking to iso-
late and measure precisely the contribution made by the manager to the effective-
ness of the team’s performance. Empirical results can depend heavily on the way
in which the inputs are measured, on the specification of the production function,
and on the distributional assumptions concerning the ‘inefficiency’ and random
components of the error term; and on the estimation method employed.
The relationship between managerial change and team performance has been
investigated by means of a statistical analysis of the causes of managerial depart-
ure, and of managerial succession effects. A statistically significant link between
individual match results and the involuntary departure hazard is found for up
to sixteen matches prior to the current match. The involuntary departure hazard
decreases with an improvement in league position; decreases with the age of the
manager; but does not seem to depend on other human capital attributes of the
manager. The few most recent match results are also significant in the voluntary
departure hazard function, but the effect is rather weak. Managers that have
improved their team’s position are more likely to leave voluntarily, and previous
managerial experience and international recognition as a player both increase the
likelihood of voluntary departure.
In order to measure the succession effect of managerial change on team perform-
ance immediately after the change takes place, it is important to control for a natural
tendency for mean-reversion in team performance. No team carries on losing forever,
so a team that has recently experienced a poor spell should be expected to improve,
whether it changes its manager or not. To measure the average effect of a change of
manager on post-succession performance, this natural tendency to improve should
not be attributed to the decision to terminate the previous manager’s appointment.
Only if teams tend to improve by more than the expected amount is it correct to
infer that removing the manager has a beneficial average effect. The empirical ana-
lysis suggests within-season managerial departure tends to be disruptive in the short
term. On average, and after controlling for the mean-reversion effect, a team that
changes its manager tends to underperform in its next few matches.
Notes
1 The Pro-Licence is the highest coaching qualification in football. The course takes one
year to complete and consists of a minimum of 240 hours, of which 90 hours are prac-
tical. Participants study football-specific modules, as well as employment law, finance, the
media and technology, business management and club structure. The overall purpose of the
Pro-Licence is to improve the skills needed to effectively manage top level players. Not all
established managers in England have the licence. Harry Redknapp and Alex Ferguson, for
example, have been awarded an FA coaching diploma in recognition of their experience.
2 This shift did not happen quickly at all clubs. For example, the directors of Newcastle United
and Liverpool were involved in selecting the team until well into the 1950s (Carter, 2006).
3 A player is offside if, when the ball is played forward to him, fewer than two opposing play-
ers stand between him and the goal line. Before 1925, a player was offside if there were fewer
than three opposing players.
4 Under Buckley’s management, Wolverhampton Wanderers finished runners-up in both the
league and FA Cup in 1939. Buckley characteristically wore ‘plus-fours’ and brogue shoes,
and could easily be mistaken for a farmer. His military record included service in the Boer
War (Turner and White, 1993).
5 Examples of time-varying stochastic production function models include those of Cornwell,
Schmidt and Sickles (1990), Battese and Coelli (1992) and Lee and Schmidt (1993). Lee and
Berri (2008) discuss these models in the context of team sport, namely in an application to
US basketball production.
6 In contrast, the types of performance measure on which business managers are judged, such
as sales, growth or profit, may be opaque and subject to interpretation.
7 The dummy variables for involvement in European competition are not included, due to gaps
in the data required to define these variables for a few of the earlier seasons.
10 The football referee
Introduction
Football referees are much maligned individuals. They are routinely criticised by
managers, players, journalists and spectators for being incompetent, inconsistent
and biased. The decisions referees make (often taken in a split-second) can be
crucial for a team’s prospects of achieving success, while the financial implica-
tions of success or failure for individual clubs can be enormous. This helps to
explain why the actions of referees today are more intensely scrutinised than ever
before. Football authorities are under pressure to take steps to ensure that refer-
eeing decisions are fair, consistent and accurate. The intense criticism of referees
in recent times has been reflected in a number of academic papers investigating
sources of bias and inconsistency in referee decision-making in various sports and
countries.
In this chapter, Section 10.1 describes the historical evolution of the football
referee, and the referee’s role in modern-day football. Repeated calls for the use of
video or other forms of technology to assist or adjudicate in resolving contested
or controversial incidents, and for refereeing duties to be shared between more
officials, have so far been resisted by football’s governing bodies. Accordingly, the
referee remains the ultimate authority on the field of play, and exercises consid-
erable discretion when officiating games. For example, in the case of foul play the
referee has the discretion to decide whether the foul merits a caution, in the form
of a yellow or red card. The discretion given to referees may encourage favourit-
ism in their decision-making.
Economics does not have a lot to say about favouritism and decision-making,
although there is a small theoretical literature that considers favouritism as a
source of inefficiency in principal-agent relationships. One problem in developing
the literature on favouritism has been a lack of empirical analysis due to insuf-
ficient data. However, research by sports economists and others has offered a
way forward, since sports data are suitable for testing hypotheses about bias or
favouritism in agents’ decision-making. A key aspect of the sports literature is the
role played by social pressure as a source of bias. Recently this literature has been
295
extended by examining referee decisions in an international context, allowing the

impact of nationality on decision-making to be examined. Section 10.2 reviews
the empirical literature on referee favouritism and bias.
Finally, Section 10.3 reports an update of previous work by the authors of this
volume on the incidence of disciplinary sanction in English football. Several ques-
tions concerning possible sources of inconsistency and bias in refereeing standards
are examined. A refereeing consistency hypothesis, that the incidence of disciplin-
ary sanction does not vary between referees, is rejected. The tendency for away
teams to be penalised by referees more than home teams cannot be attributed
solely to the home-field advantage effect on match results, and appears to be asso-
ciated with a form of refereeing bias favouring the home team.
10.1 The role of the football referee

The task of the football referee (and his two assistants) is to enforce the Laws of
the Game as laid down by the Federation of International Football Associations
(FIFA). When the Laws are broken the referee applies appropriate sanctions,
and his decisions and actions regarding facts connected with play are final. A
study of the European Football Championship in 2000 found that referees made
137 observable interventions on average during a match, including awarding free
kicks, penalties, corners, throw-ins, and halting play for serious injury (Helsen and
Bultynck, 2004). The referee also acts as timekeeper, and can postpone, stop or
suspend a match for reasons associated with the weather, or the conduct or safety
of the spectators. Referees are licensed and trained by national governing bodies
that are members of FIFA. Each national organisation can recommend its top
officials for appointment as FIFA officials. Each national organisation determines
the manner of training, ranking and advancement of officials from junior through
to professional matches.
The situation today is a world apart from that which prevailed in the early
days of organised football. Prior to the formation of the Football Association
in England in 1863, matches were played without a referee. When the first Laws
of Football were published in December 1863, it became clear that officials were
needed to enforce the new laws. At this time it was common for two umpires (one
nominated by each team) to be appointed to officiate at matches. The umpires
had no right to interfere with play, but they could be ‘appealed to’ by the players.
A similar system still operates in cricket at present. The umpires were empowered
to award a free kick for handball in 1873, and for other offences in 1874. Also in
1874, the umpires were empowered to order a player to leave the pitch for ‘persist-
ent infringement of the rules’. Naturally the two umpires did not always agree on
decisions, creating a role for a neutral observer or ‘referee’. From 1880 the referee
was appointed by mutual agreement between the two clubs. He was required to
‘keep a record of the game’ and act as timekeeper. He was empowered to cau-
tion players adjudged guilty of ungentlemanly conduct, provided umpires were
The role of the football referee 297
present. If a player continued to transgress, or was guilty of violent play, the ref-
eree could order him to leave the pitch.
In 1891 the Football Association decided that the referee would be the sole judge
of fair play. The referee could award free kicks at his own discretion, and for the
first time he was permitted onto the pitch. It was no longer necessary for players
to appeal to the referee for a decision. Each club could still nominate an umpire
to assist the referee, but the umpires would be located on the touchlines (at the
edges of the pitch) and would not enter the pitch. The umpires eventually became
known as linesmen, who used flags to indicate decisions. Most early referees and
linesmen were usually non-playing members of football clubs. Controversial deci-
sions often resulted in claims that the referee or linesmen had been influenced by
club loyalties. Gradually they became independent officials without links to the
clubs.
As professional football spread to most parts of the world during the early
twentieth century, the ‘referee and two linesmen model’ was adopted universally.
As a sport football has developed greatly in terms of rules and tactics; but the
evolution of the football referee has been rather limited in comparison. For 90
minutes while the match is underway and within the confines of the field of play,
the role of today’s football referee is not so very different from that of the ‘man
in black’ a hundred years ago or more. As football eventually embraced the mod-
ern digital television era, however, with regular live transmission of high-profile
matches on dedicated sports channels that use multi-angled slow-motion replays
to illuminate endless post-match studio panel discussions, the public profile of the
early twenty-first-century referee officiating at the highest level would have been
unrecognisable to his predecessor a century ago. Referees, just like managers and
players, are subject to intense media and public scrutiny and criticism, and often
cruelly and ruthlessly scapegoated for minor errors of split-second judgement
concerning incidents witnessed at high speed without the benefits of hindsight
conferred by television replays. Indeed, football managers are often among the
chief perpetrators of scapegoating, although rules exist in many countries placing
limits on the level of criticism of referees that is deemed acceptable.
In one important respect, however, the status of the referee at the highest level
has changed during the modern era, through the development of a cadre of full-
time salaried professional referees. For most of the history of football, referees
were ‘amateurs’, paid a notional match fee plus expenses in return for their ser-
vices. In England many referees were drawn from the ranks of schoolteachers, the
police and comparable professions. As recently as the 1990s it was not unusual for
a teacher (for example) to complete a regular day’s work in the classroom, then
drive to a football stadium located in another city, officiate as the referee in an
evening match played in front of 30,000 or 40,000 spectators, and be paid a fee of
less than £100 for this service.
Professionalism was introduced in England at the start of the 2002 season,
when the English Premier League introduced payment of a full-time salary to
referees placed on the Premier League list. The annual salary is around £33,000
plus match fees. Including the match fee element, the referee’s remuneration may
be as high as £60,000. In principle a poorly performing referee can be demoted
from the list at any time, but in general the list is renewed and predetermined at
the start of each season. The move to professionalism was motivated by the view
that professional referees would adopt a more professional attitude to their foot-
balling commitments. Freed from the time constraints of full-time employment in
another profession, they would have time to work on their fitness by training with
footballers, and discuss aspects of their performance with their peers and with
football managers and players.1
10.2 Favouritism and referee behaviour

Referees are employed to act as impartial agents for a principal, typically the
national football association or governing body. Yet, as any football follower is
aware, referees are frequently accused of favouritism and bias. For an economist,
it is natural to look to agency theory for insights as to why agents might make
biased or partial decisions, and for recommendations as to how incentives might
be structured so as to discourage undesirable behaviour on the part of agents
(Prendergast, 1999; Laffont and Mortimont, 2002). Though favouritism has sig-
nificance for agency theory, it has received only limited attention in the theoretical
literature. In a seminal contribution, Prendergast and Topel (1996) derive condi-
tions under which ‘supervisors’ favour ‘workers’ in a manager-supervisor-worker
hierarchy. The utility function of the supervisor depends on the payoff of their
worker(s), whose non-verifiable performance can be observed by the supervisor.
Managers can also observe (privately) workers’ performance and can monitor
supervisors’ reports. In this set-up, where preferences for favouritism are exogen-
ous, stronger incentive pay for workers reduces the accuracy of supervisors’
reports, so that favouritism depends on the incentives offered to the worker.
Empirical studies of favouritism in economics are few in number, because it is
difficult, if not impossible, to identify favouritism leading to a misallocation of
resources. However, in the past few years football has proved a fruitful setting for
this kind of analysis. Studies using football data tend to focus on two decisions
of the agent: the decision of the referee to add on time at the end of regular time
(after 90 minutes has elapsed); or the decision to caution players (award red and
yellow cards).
Using data from the Spanish La Liga, Garicano, Palacios-Huerta and
Prendergast (2005) report a tendency for referees to add on more time at the end
of close matches (one goal difference between the teams at the end of the match)
when the home team is trailing than when the home team is leading. When the
home team is behind by one goal, the average added-on time that is announced
is 35 per cent above the norm; but when the home team is ahead by one goal, the
average added-on time that is announced is 29 per cent below average. A similar
Favouritism and referee behaviour 299
pattern is obtained from an analysis of the comparison between the added-on

time that was announced ex ante and the time that was actually added on ex
post in matches that were level at the start of added-on time, but one of the
teams then scored during added-on time (meaning that the scoring team would
be advantaged if the match were ended very soon afterwards). When the away
team scored, on average 15 per cent more time was added on than when the home
team scored.
Similar findings have been reported in several other studies. Sutter and Kocher
(2004) analyse time added on in close matches in the top division of the German
Bundesliga during the 2001 season. After controlling for events during regular
time that should influence the amount of time added on (player substitutions and
the award of yellow and red cards), if the home team is ahead by one goal (score
margin of +1) or the scores are level, added-on time is significantly less on average
(by about 30 –50 seconds) than it is if the home team is behind by one goal (score
margin of −1). There is also evidence of a bias in the award of penalty kicks,
favouring home teams. Scoppa (2008) obtains similar results for Italy’s Serie A
during seasons 2004 and 2005. The average added-on time is 3.62 minutes; but
referees tend to allow significantly more added-on time when the home team is
trailing by one goal (4.19 minutes, or about half a minute more than the average).
This effect is smaller, but still statistically significant, after controlling for factors
that may influence the amount of added-on time, such as the number of player
substitutions during regular time.
There is also evidence of favouritism in the incidence of disciplinary sanction
(award of red and yellow cards), which tends to be lower for home teams in the
top tier of the Bundesliga (Buraimo, Forrest and Simmons, 2010), in the English
Premier League (Dawson et al., 2007) and in European cup matches (Dawson and
Dobson, 2010) than it is for away teams. Under current rules, a yellow card, also
known as a booking or caution, is awarded for less serious transgressions. There
is no further punishment within the match, unless the player commits a second
similar offence, in which case a red card is awarded and the player is expelled for
the rest of the match (with no replacement permitted, so the team completes the
match one player short). A red card, also known as a sending-off or dismissal,
is awarded for more serious offences, and results in immediate expulsion (again,
with no replacement permitted). After the match, a red card leads to a suspension,
preventing the player from appearing in either one, two or three of his team’s
next scheduled matches. A player who accumulates five yellow cards in different
matches within the same season also receives a suspension.
Dawson et al. (2007) report an analysis of various possible forms of referee bias
in the English Premier League covering seven seasons. Dawson et al. amalgamate
yellow and red cards into a discrete points measure. Their model specification
allows for tests of several hypotheses concerning patterns in the incidence of dis-
ciplinary sanction. The principal hypotheses of interest are: (i) home-field advan-
tage hypothesis – the tendency for away teams to incur more disciplinary points
than home teams is solely a corollary of home-field advantage; (ii) refereeing con-
sistency hypothesis – the average incidence of disciplinary sanction does not vary
between referees; (iii) consistent home-team bias hypothesis – the degree to which
away teams incur more disciplinary points than home teams on average (after con-
trolling for home-field advantage) does not vary between referees.
The evidence shows that the tendency for away teams to incur more disciplin-
ary points than home teams cannot be explained solely by home-field advantage.
Even after controlling for team quality, a (relatively strong) away team can expect
to collect more disciplinary points than a (relatively weak) home team with the
same win probability. Therefore, the statistical evidence points in the direction of a
home-team bias in the incidence of disciplinary sanction. Following on from this,
there is variation between referees in the degree of home-team bias, and this vari-
ation contributes to the overall pattern of refereeing inconsistency.
Instead of using the complete match as the unit of observation, Buraimo, Forrest
and Simmons (2010) use minutes of play within matches. The dependent variable
is the probability of the home or away team receiving a yellow (red) card within
each minute. The results confirm a bias favouring home teams in the incidence
of disciplinary sanction, after allowing for relative team quality and home-field
advantage. Using FA Cup data, Downward and Jones (2007) show that a higher
number of first yellow cards were awarded against the away team, and there was a
non-linear relationship between crowd size and the number of yellow cards.
In much of this literature, alleged favouritism on the part of referees is attributed
to the social pressure applied by the home crowd. This explanation accords with
the view of some economists that social environment affects individual behav-
iour (for example, Akerlof, 1980; Bernheim, 1994; Becker and Murphy, 2000). In
this literature, social interactions are offered as an explanation for various forms
of socioeconomic behaviour, cultural practices and consumption patterns. For
most football referees, it seems likely that social pressure influences choices sub-
consciously. This is the interpretation of Dohmen (2008a), echoing Sutter and
Kocher (2004), in a study of the German Bundesliga covering seasons 1993 to
2004. Home-team bias is influenced by the size of the crowd (absolute size), the
attendance-to-capacity ratio (relative size of the crowd) and the proximity of sup-
porters to the pitch (the presence of a running track). More time is added in close
matches when the crowd is physically close to the field of play. Home teams are
more likely to be awarded a penalty that is disputed, with the physical distance
between the crowd and the pitch important to this decision. Pettersson-Lidbom
and Priks (2010) examine the behaviour of referees in Italian football at a time
(2006–7) when some teams were forced to play home matches behind closed doors
because their stadiums were substandard. The authors find referees punish away
players more harshly than home players when matches are played in front of spec-
tators compared to when they are not. This result is interpreted as evidence that
referee behaviour is influenced by social pressure.
Favouritism and referee behaviour 301
The impact of social pressure on the incidence of disciplinary sanction has

also been examined. Buraimo, Forrest and Simmons (2010) find the size of the
crowd has no statistically significant effect on the award of yellow and red cards
against either home or away teams in the English Premier League and the German
Bundesliga. In the latter, however, the presence of a running track, which cre-
ates greater distance between the crowd and the pitch, increases the numbers of
yellow and red cards awarded against the home team. For the Premier League,
Dawson et al. (2007) find the incidence of disciplinary sanction against home
teams is greater when the match attendance is larger. Dawson and Dobson (2010)
find that the size of the attendance relative to the stadium capacity matters more
than the absolute attendance: in European club matches, more cards are awarded
against both teams when the attendance is near capacity. The presence of a run-
ning track increases the incidence of disciplinary sanction against the home
team.
Boyko, Boyko and Boyko (2007) focus on the importance of referee bias in
influencing home advantage in the English Premier League during seasons 2003 to
2006. In their primary analysis they consider the influence of team ability, crowd
size and the referee for the match on home-field advantage (measured as the dif-
ference between goals scored by the home and away teams). The larger the match
attendance, the greater the home-field advantage; and there are statistically signifi-
cant differences between referees in the extent of home-field advantage. However,
Johnston (2008) fails to replicate the findings of Bokyo, Boyko and Boyko using
data from the 2007 season. The data set used in the earlier study included all
matches, many of which were played before capacity crowds with no variation
in the observations for each home team. When the analysis is conducted over
matches in which there was significant variation in attendance, there is no signifi-
cant relationship between attendance and home-field advantage. In the Johnston
study, there is no significant variation between referees in the magnitude of the
home-field advantage effect.
Nevill, Balmer and Williams (2002) showed a recording of forty-seven chal-
lenges/tackles in a match between Liverpool and Leicester City to forty referees.
Each referee was asked to classify the tackles as legal or illegal. One group of
referees viewed the recording with the soundtrack (including the crowd’s reac-
tion) switched on, while a second group viewed silently. The results showed that
the officials in the audible noise group were more likely to rule in favour of
the home team (calling, on average, 15.5 per cent fewer fouls). The decisions
of those in the audible group were also more in line with those of the original
match referee.
Rickman and Witt (2008) consider whether financial incentives may help check
the influence of social pressure on behaviour. As noted above, English Premier
League referees were accorded professional status from the start of the 2002 sea-
son. Home-team bias in the amount of added-on time at the end of matches of the
kind identified by Garicano, Palacios-Huerta and Prendergast (2005) and others

essentially disappeared following the introduction of professionalism. The higher
remuneration associated with professional status, together with increased moni-
toring, appears to have acted as a deterrent to showing (implicit) favouritism. A
similar argument is made by Lucey and Power (2004) in a study of added-on time
in Italy (2003 season) and the US (2003 season). Although there is some evidence
of favouritism in close matches, there is no clear evidence that social pressure is the
explanation. In both countries, referees are relatively highly rewarded and closely
monitored, giving the agent (the referee) a pecuniary incentive to be reappointed
by the principal (national organisation).
The role of nationality in influencing referees’ decisions is difficult to identify,
not least because of the difficulties involved in disentangling the interplay between
the referee’s nationality, the nationalities of the teams and the reputation of the
domestic leagues in which they compete. Akerlof (1997) and Akerlof and Kranton
(2000) suggest individual decisions are influenced by one’s own identity, and by
the perceptions of others. This notion seems particularly relevant in the context of
football refereeing decisions that rely upon split-second judgements under condi-
tions of uncertainty.
Dawson and Dobson (2010) analyse the award of yellow and red cards in
European club football (Champions League and UEFA Cup). This study addresses
the impact of social pressure on the incidence of disciplinary sanction, but also
exploits the international dimension of the data by examining whether decision-
making is influenced by nationality and reputation. Consistent with previous
studies, social pressure, in the form of crowd density and stadium architecture
(presence of a running track), is an important influence on behaviour.
The referee’s nationality, the teams’ nationalities and the reputation of their
domestic leagues all appear to influence refereeing decisions. In particular, it
appears there is a greater tendency for referees from footballing countries with
smaller populations to favour home teams: officials from Holland, Norway,
Scotland and Sweden awarded fewer yellow and red cards against home teams.
Greek referees tended to penalise (home and away) teams more than average;
and Romanian, Italian and Spanish teams (playing either at home or away) were
penalised more than average. Teams playing Italian opponents were penalised less
than average; but teams playing Spanish and Portuguese opponents were penal-
ised more than average. Some of these effects tended to offset the effects of league
reputation; but some may be due, at least in part, to interactions between different
playing styles, rather than implicit favouritism on the part of referees.
The finding that (national) identity helps shape the decisions of referees is con-
sistent with some theoretical research on identity and individual decision-making
within a utility maximising framework. This finding is also relevant for agency
theory. If social pressures tilt the agent (referee) towards actions that result in
undesired outcomes (favouritism), it may be optimal for the principal (national
football association) to reduce the agents’ discretion (Dohmen, 2008a).
Disciplinary sanction in the English Premier League 303
Table 10.1 Numbers of yellow cards incurred by the home and away teams, English
Premier League, 1997–2009 seasons
Away team
Home team 0 1 2 3 4 5 6 7 Total
0 355 483 316 165 60 15 1 0 1395

1 226 505 456 270 110 37 7 2 1613
2 124 331 315 223 92 44 8 2 1139
3 39 136 168 112 63 24 6 6 554
4 7 27 59 48 25 12 2 0 180
5 2 9 15 15 4 3 0 1 49
6 1 1 2 4 1 1 0 0 10
Total 754 1492 1331 837 355 136 24 11 4940
Source: The Football Association
Table 10.2 Numbers of red cards incurred by the home and away teams, English
Premier League, 1997–2009 seasons
Away team
Home team 0 1 2 Total
0 4211 415 13 4639

1 225 56 5 286
2 7 5 0 12
3 3 0 0 3
Total 4446 476 18 4940
10.3 The incidence of disciplinary sanction in English Premier League football

This section presents an analysis of the incidence of disciplinary sanction, using
data on the 4,940 Premier League matches played during the 13 English football
seasons from 1997 to 2009, inclusive. The empirical methodology is the same as in
Dawson et al. (2007), and the results that are reported in this section provide an
update of those reported in the earlier study.
Tables 10.1 and 10.2 show the frequency distributions for the numbers of yellow
cards and red cards incurred by the home and away teams in the 4,940 matches.
The dependent variables in the estimations that are reported below are the total
numbers of disciplinary ‘points’ incurred by the home (i = 1) and away (i = 2)
teams in match j for j = 1 … N, denoted {Z1,j, Z2,j} and calculated by awarding
one point for a yellow card and two for a red card. Only two points (not three) are
awarded when a player is dismissed for having committed two cautionable (yel-
low card) offences in the same match. This metric accurately reflects the popular
notion that a red card is in some sense equivalent to two yellow cards. In fact,
this notion was literally true of 47.1 per cent of the red cards awarded during the
observation period (391 out of 831 dismissals in total), which resulted from two
cautionable offences having been committed in the same match. Table 10.3 reports
the sample frequency distribution for {Z1,j, Z2,j}, with the rows and columns for
Zi,j≥5 consolidated into a single row and a single column.
In the applied statistics literature, several methods are available to model pro-
fessional team sports bivariate count data, where each match yields two values
of a discrete dependent variable (one for each team: commonly the number of
goals scored as in Chapter 3 of this volume, but the disciplinary points dependent
variable in the present case has the same structure). A description follows of the
probability models that are considered as candidates for the disciplinary ‘points’
dependent variables {Z1,j, Z2,j}.
Let fi(zi) = P(Zi,j = zi) for zi = 0,1,2, … denote the marginal probability function
for Zi,j for i = 1,2 and j = 1 … N. As in Chapter 3, the two candidate distributions
for fi(zi) are the Poisson distribution and the negative binomial distribution. The
formulae for fi(zi) in each case, together with the corresponding expressions for the
expected value of Zi,j and the variance of Zi,j, are as follows:
Poisson distribution: f i (z i ) = exp( − λ i,j )λ zi,ji /z i!

Expected value and variance: E(Z i,j ) = λ i,j var(Z i,j ) = λ i,j
Negative binomial distribution:
fi (z i ) = [Γ (ρi + z i )/{z i!Γ (ρi )}]{ρi /(λ i,j + ρi )} ρi {λ i,j /{λ i,j + ρi )}zi
Expected value and variance:
E(Z i,j ) = λ i,j var(Z i,j ) = λ i,j (1 + κ i λ i,j ) where κ i = 1/ρ i [10.1]
where exp( ) denotes the exponential function and Γ( ) denotes the gamma
function.
As noted in Chapter 3, the key distinction between the Poisson distribution and
the negative binomial distribution is the ancillary parameter ρi (or κi = 1/ρi) which
allows for overdispersion. The degree of overdispersion is greater for the discipli-
nary ‘points’ data than it is for the goals data. The sample mean values of Z1,j and
Z2,j are 1.4348 and 1.9731, and the sample variances are 1.6724 and 2.1392.
Overdispersion is a key feature of the data that are summarised in Tables 10.1,
10.2 and 10.3. There are, however, two additional features that require atten-
tion when formulating an appropriate statistical model to describe these data.
The first additional feature is the existence of a pattern of positive correlation
between Z1,j and Z2,j. The numbers of matches appearing on or near the main
diagonals of these tables (the diagonal running from the top left to the bottom
right of each table) are larger than would be expected if the data were generated
strictly in accordance with either the Poisson distribution or the negative bino-
mial distribution. If one team incurs no cards or very few cards, the probability
Table 10.3 Sample frequency distribution for the bivariate disciplinary points
dependent variable, {Z1,j,Z2,j}
Z2,j
Z1,j 0 1 2 3 4 5+ Total
0 340 443 277 172 85 26 1343

1 215 460 413 258 119 75 1540
2 124 311 292 215 104 90 1136
3 35 126 160 119 79 54 573
4 10 36 64 65 32 26 233
5+ 4 19 34 30 15 13 115
Total 728 1395 1240 859 434 284 4940
that the other team does so as well is somewhat increased; and similarly, if one
team incurs many cards, the probability that the other team does as well is some-
what increased. The sample correlation between Z1,j and Z2,j is +0.2549. A positive
correlation might reflect a tendency for teams to retaliate in kind if the opposing
team is guilty of a particularly high level of foul play. Alternatively, a common
perception among pundits and supporters is that some referees, having penalised
a player from one team, will often seek an opportunity to penalise an opposing
player soon afterwards, perhaps in an attempt to pre-empt any perception of ref-
ereeing bias.
As before the existence of correlation between Z1,j and Z2,j is addressed by gen-
erating a ‘synthetic’ bivariate model, in which the univariate probability distribu-
tions for the ‘points’ incurred by the home and away teams are linked using the
Frank copula (Lee, 1999; Dawson et al., 2007). Let Fi(zi) denote the univariate dis-
tribution functions for Zi,j corresponding to fi(zi).2 The bivariate joint distribution
function is defined as follows:
G[F1 (z1 ),F2 (z2 )] = P(Z1,j ≤ z1 , Z 2,j ≤ z2 )
1  {exp[φF1 (z1 )] − 1}{exp[φF2 (z2 )] − 1}
= ln 1 +  [10.2]
φ  exp(φ ) − 1
The bivariate joint distribution function G[F1(z1),F2(z2)] expresses the probability

that the home team incurs z1 or fewer disciplinary ‘points’ and that the away team
incurs z2 or fewer disciplinary ‘points’. The ancillary parameter φ determines the
direction of correlation between Z1,j and Z2,j: for φ < 0 the correlation is positive,
and for φ>0 the correlation is negative.3 The bivariate joint probability function
f(z1,z2) = P(Z1,j = z1,Z2,j = z2) expresses the probability that the home team incurs
(exactly) z1 disciplinary ‘points’ and the away team incurs (exactly) z2 disciplinary
‘points’. The bivariate joint probability function is obtained iteratively from the
bivariate joint distribution function, as follows:
f(0,0) = G[F1(0),F2(0)]
f(z1,0) = G[F1(z1),F2(0)] – G[F1(z1–1),F2(0)] for z1=1,2, …
f(0,z2) = G[F1(0),F2(z2)] – G[F1(0),F2(z2–1)] for z2=1,2, …
f(z1,z2) = G[F1(z1),F2(z2)] – G[F1(z1–1),F2(z2)]
– G[F1(z1),F2(z2–1)] + G[F1(z1–1),F2(z2–1)] for z1,z2=1,2, … [10.3]
The second additional feature of the data summarised in Tables 10.1, 10.2 and
10.3 is a consistent tendency for the numbers of matches in which no yellow or red
cards are issued (in the top-left-hand cells) to be larger than would be expected if
the data were generated strictly in accordance with either the Poisson or the nega-
tive binomial distribution. This feature is addressed by applying a zero-inflated
adjustment to the joint probabilities that are obtained from the copula function.
The zero-inflated joint probabilities are:

f(0,0)  1,j = 0,Z 2,j = 0) = (1 − π )P(Z1,j = 0,Z 2,j = 0) + π
= P(Z
 ,0) = P(Z
f(z  1,j = z1 ,Z 2,j = 0) = (1 − π )P(Z1,j = z1 ,Z 2,j = 0) for z1 = 1,2,…
1
 z ) = P(Z
f(0,  1,j = 0,Z 2,j = z2 ) = (1 − π )P(Z1,j = 0,Z 2,j = z2 ) for z2 = 1,2,…
2
f(z1 ,z2 ) = P(Z

 1,j = z1 ,Z 2,j = z2 ) = (1 − π )P(Z1,j = z1 ,Z 2,j = z2 ) for z1 ,z2 = 1,2,… [10.4]
Following Dawson et al. (2007), the zero-inflated bivariate negative binomial dis-
tribution is used as a basis for estimating an unconditional model for the numbers
of disciplinary ‘points’ incurred by each team in each match. The unconditional
model does not take into account any information about the teams, the team man-
agers, the referee or the attendance; instead, the probabilities generated from the
unconditional model merely reflect averages for the incidence of disciplinary sanc-
tion across all matches. Therefore in the unconditional model, it is assumed that λi,j
= λi for i = 1,2 and j = 1 … N: the values of the parameters λ1,j and λ2,j are the same
for every match. The fitted values of all of the parameters are as follows:

ˆλ1 = 1.4578 ˆλ 2 = 1.9985 κˆ1 = 0.1019 κ ˆ 2 = 0.0292 φˆ = −1.5586 πˆ = 0.0135
Table 10.4 reports the fitted bivariate probabilities for each pair of values for z1
and z2, evaluated by substituting the estimated parameters into the zero-inflated
joint probability function specified above. The fitted probabilities are compared
with the observed proportions of matches in each cell, which are calculated using
the data reported in Table 10.3. The fitted probabilities appear to be accurate. This
suggests that the zero-inflated bivariate negative binomial distribution provides a
good representation of the disciplinary ‘points’ data.
Again following Dawson et al. (2007), the zero-inflated bivariate Poisson dis-
tribution is used as a basis for estimating a conditional model for the numbers
of disciplinary ‘points’ incurred by each team in each match. The conditional
Table 10.4 Unconditional model: fitted bivariate probabilities and observed

proportions for the numbers of disciplinary ‘points’ incurred by the home and away
teams
Z2,j
Z1,j 0 1 2 3 4 5+ Total
Fitted bivariate probabilities

0 .0688 .0849 .0611 .0310 .0134 .0078 .2669
1 .0486 .0932 .0867 .0528 .0251 .0153 .3216
2 .0232 .0529 .0621 .0460 .0246 .0160 .2249
3 .0092 .0230 .0308 .0260 .0152 .0104 .1145
4 .0034 .0087 .0125 .0112 .0068 .0048 .0475
5+ .0017 .0043 .0063 .0059 .0037 .0026 .0245
Total .1549 .2671 .2595 .1729 .0888 .0568 1.0000
Observed proportions
0 .0688 .0897 .0561 .0348 .0172 .0053 .2719
1 .0435 .0931 .0836 .0522 .0241 .0152 .3117
2 .0251 .0630 .0591 .0435 .0211 .0182 .2300
3 .0071 .0255 .0324 .0241 .0160 .0109 .1160
4 .0020 .0073 .0130 .0132 .0065 .0053 .0472
5+ .0008 .0038 .0069 .0061 .0030 .0026 .0233
Total .1474 .2824 .2510 .1739 .0879 .0575 1.0000
model takes into account information specific to each match that is relevant for
determining the incidence of disciplinary sanction in the match concerned. In the
conditional model, ln(λi,j) is specified as a linear function of a set of covariates.
For the unconditional model, the use of the negative binomial probability model
(with a zero-inflated adjustment) is required to represent the overdispersion in the
sample data for {Z1,j, Z2,j}. For the conditional model, in contrast, the covariates
are largely successful in identifying the sources of overdispersion, rendering the
use of the more complex negative binomial probability model unnecessary.
The covariate definitions for the conditional model are as follows:
qj = home win probability + 0.5 × draw probability for match j. Probabilities

are generated from the results-based forecasting model reported in Chapter 4,
Sections 4.4 and 4.5.
qj(1–qj) = uncertainty of match outcome measure.
attj = reported attendance at match j.
sigi,j = 0–1 dummy variable, coded 1 if match j is significant for end-of-season
championship, European qualification or relegation outcomes, for the home
(i = 1) or away (i = 2) team. Matches with significance for end-of-season out-
comes are identified using the algorithm described in Chapter 4.
DMi,m,j = 1 if match j falls within managerial spell m for the home (i = 1) or away
(i = 2) team; 0 otherwise. m = 1 … 92 represents managerial spells that con-
tained at least thirty Premier League matches within the observation period; the
matches in fifty other spells that contained fewer than thirty matches in total
form the reference category.
DRr,j = 1 if match j is officiated by referee r; 0 otherwise. r = 1 … 37 represents
referees who officiated at least thirty matches within the observation period;
thirteen other referees who officiated fewer than thirty matches each form the
reference category.
DSs,j = 1 if match j is played in season s; 0 otherwise. s represents seasons 1998 to
2009 inclusive; the 1997 season is the reference category.
The model specification allows for tests of several hypotheses concerning patterns
in the incidence of disciplinary sanction, as follows:
H1: The home-field advantage hypothesis (HFAH). The tendency for away teams
to incur more disciplinary points than home teams is solely a corollary of
home-field advantage: the tendency for home teams to win more frequently
than away teams.
H2: The refereeing consistency hypothesis (RCH). The average incidence of discip-
linary sanction does not vary between referees.
H3: The consistent home-team bias hypothesis (CHTBH). The degree to which
away teams incur more disciplinary points than home teams on average (after
controlling for home-field advantage) does not vary between referees.
H4: The time consistency hypothesis (TCH). The average incidence of disciplinary
sanction is stable over time.
H5: The audience neutrality hypothesis (ANH). The incidence of disciplinary sanc-
tion is invariant to the size of the crowd inside the stadium.
The estimated conditional model is reported below. z-statistics for the estimated
coefficients, based on robust standard errors, are shown in parentheses. The esti-
mated coefficients on the season, referee and managerial spell dummy variables
are not reported.
In( λ̂ 1,j ) = −0.1547 − 1.0576 q j + 4.3470 q j (1 − q j ) − 0.0108 sig1,j + 5.6582 att j

− 0.59 − 7.27 5.81 − 0.16 2.06
12 92 37
+ ∑ βˆ 1,s DSs,j + ∑ δˆ 1,m DM1,m,j + ∑ γˆ 1,r DRr,j
s =1 m =1 r =1
In( λˆ 2,j ) = −1.2221 + 1.0348 q j + 4.9446 q j (1 − q i ) + 0.0210 sig 2,j + 4.2741att j

− 4.86 5.06 7.94 0.37 3.336
12 92 37
+ ∑ βˆ 2,s DSs,j + ∑ δˆ 2,m DM2,m,j + ∑ γˆ 2,r DRr,j
s =1 m =1 r =1
φˆ = −1.5395 π̂ = 0.0065 [10.5]

Relative team quality and home-field advantage

The specification of the conditional model allows for a quadratic functional form
for the relationship between qj and ln(λ ˆ i,j). This relationship is parameterised so
that the coefficients reported are for qj, a weighted sum of the home team’s win
and draw probabilities after allowing for home advantage, and qj(1–qj), a measure
of competitive balance.
According to the home-field advantage hypothesis (HFAH), the propensity for
away teams to collect more disciplinary points on average than home teams is
solely a corollary of the home advantage effect on match results. If the HFAH is
correct, the expected incidence of disciplinary sanction for a stronger-than-aver-
age away team should be the same as that for a weaker-than-average home team,
if the two teams’ win probabilities are the same after taking home-field advantage
into account. In other words, (λ ˆ 1,j − λ
ˆ 2,j) should be zero if qj = 0.5, qj(1–qj) = 0.25,
and all other covariates in the bivariate Poisson regression are set to their sample
means. In fact, substituting this set of covariate values into the fitted model pro-
duces (λ ˆ 1,j − ˆλ2,j) = –0.2813. The standard error of (ˆλ1,j − ˆλ2,j) is 0.0477; therefore
ˆ 1,j − ˆλ2,j) is significantly less than zero. The difference of –0.2813 accounts for
(λ
just over one-half (52.3 per cent) of the total differential in the average dis-
ciplinary ‘points’ awarded against the home and away teams, ( ˆλ1,j − λ ˆ 2,j) =
1.4348 – 1.9732 = –0.5383. Therefore just under one-half (47.7 per cent) of the
total disciplinary ‘points’ differential is explained by home-field advantage. The
remainder is left unexplained by home-field advantage, and accordingly is attrib-
uted to refereeing bias favouring the home team.
Another implication of the HFAH is that all coefficients in the home-team equa-
tion in [10.5] should be identical to their counterparts in the away-team equation,
except for the intercept, and the coefficients on qj which should be equal and oppo-
site in sign (because the weighted sum of the away team’s win probability and the
draw probability is one minus this weighted sum for the home team, or 1 – qj). In
other words, if there is no home-team bias, the equations that predict the numbers
of disciplinary ‘points’ incurred by the home and away teams should be identical;
and away teams incur more ‘points’ than home teams because on average away
teams score worse than home teams on the win probability measure, qj. This impli-
cation of the HFAH is rejected by a Wald test of the appropriate cross-equation
equality restrictions on the coefficients of [10.5], which yields χ2(145) = 264.6
(p-value = .0000).
Importance of the match for end-of-season outcomes

The incidence of disciplinary sanction for either team might be affected by the
importance of the match for end-of-season championship, European qualifica-
tion or relegation outcomes. A team that still has end-of-season issues at stake
might be expected to be more determined or aggressive than a team with noth-
ing at stake. In the definitions of the dummy variables sigi,j, the algorithm that
determines whether a match is significant for either team assesses whether it is

arithmetically possible (before the match is played) for the team to win the cham-
pionship, qualify for European competition or be relegated, if all other teams
currently in contention for the same outcome take one point on average from
each of their remaining fixtures. Alternative algorithms, based on more optimis-
tic or pessimistic assumptions concerning the average performance of competing
teams over their remaining fixtures, alter the classification of a small proportion
of matches at the margin, but the implications of such minor variations for the
estimation results reported here are negligible.
Dawson et al. (2007) report that the coefficient on sig1,j (for the home team) is
insignificant, but the coefficient on sig2,j (away team) is positively signed and signifi-
cant. Dawson et al. suggest that away teams may feel able to ‘ease off’ in unimport-
ant end-of-season matches; but home teams, conscious of their own crowd’s critical
scrutiny, feel obliged to demonstrate maximum commitment at all times, even when
no end-of-season issues are at stake. In the present estimations based on a larger data
set, however, neither of the coefficients on sig1,j and sig2,j is significant. Accordingly,
there seems to be no evidence that the importance of the match for end-of-season
outcomes has any bearing on the incidence of disciplinary sanction.
Individual teams and managerial spells

Differences between football teams in playing personnel, styles of play and tactics
represent a further possible source of variation in the incidence of disciplinary
sanction. For each of the thirty-nine teams that played in T1 (the Premier League)
between the 1997 and 2009 seasons, Table 10.5 reports the average numbers of yel-
low and red cards and disciplinary ‘points’ awarded in home and away matches.
The correlation between the win ratio and the average disciplinary ‘points’ is 0.255.
This makes Norwich City, with the lowest average ‘points’ of all teams included in
Table 10.5, achieved in the 2005 season when finishing bottom of T1, something
of an outlier. Ipswich Town, who competed in T1 in the 1999 and 2000 seasons,
also register an average ‘points’ total significantly lower than any of the other thir-
ty-seven clubs included in Table 10.5. Stoke City, well known for a robust physical
style of play during the 2009 season, their first in T1 since the mid-1980s, recorded
the highest average ‘points’ total of all teams included in Table 10.5.
Of course, the playing styles of individual teams may vary over time; but with
twenty-two players (plus substitutes) participating in every match, it is impossible
to control for every change of playing personnel in a match-level empirical ana-
lysis of the incidence of disciplinary sanction. However, managerial spells can be
used as a proxy for football team-related factors that might produce differences
in disciplinary ‘points’ data. This approach can be justified on the grounds that
managers are primarily responsible for tactics and playing styles. For teams man-
aged by only one individual for the matches included in the data set (Arsenal and
Manchester United, and several teams that played in T1 for one or two seasons
Table 10.5 Average numbers of yellow and red cards awarded per match by team, English Premier League, 1997–2009 seasons
Home matches Away matches All matches

No. of Win Average Average Average Average Average Average Average
seasons ratio yellows reds ‘points’ yellows reds ‘points’ ‘points’
Stoke City 1 .434 2.11 0.11 2.26 1.79 0.16 2.11 2.18
Nottm Forest 2 .336 2.00 0.03 2.03 2.05 0.13 2.24 2.13
Derby County 7 .387 1.80 0.06 1.89 2.17 0.07 2.27 2.08
Leeds United 8 .539 1.66 0.05 1.72 2.26 0.12 2.39 2.06
Hull City 1 .355 1.53 0.16 1.79 2.26 0.00 2.26 2.03
Blackburn Rovers 11 .480 1.57 0.12 1.75 2.07 0.13 2.26 2.01
West Ham Utd 11 .474 1.59 0.07 1.69 1.98 0.12 2.17 1.93
Barnsley 1 .329 1.42 0.16 1.68 2.00 0.11 2.16 1.92
Sunderland 8 .372 1.54 0.04 1.61 2.02 0.13 2.22 1.91
Sheffield Utd 1 .368 1.63 0.00 1.63 1.95 0.11 2.16 1.89
Bolton Wanderers 9 .462 1.40 0.07 1.53 2.12 0.08 2.24 1.88
Coventry City 5 .424 1.34 0.06 1.45 2.03 0.12 2.21 1.83
Middlesbrough 12 .451 1.45 0.07 1.57 1.95 0.10 2.09 1.83
Wigan Athletic 4 .418 1.47 0.09 1.64 1.84 0.11 1.99 1.82
Chelsea 13 .695 1.37 0.06 1.46 1.98 0.11 2.15 1.80
Everton 13 .488 1.47 0.09 1.62 1.80 0.12 1.98 1.80
Birmingham 5 .421 1.39 0.11 1.57 1.80 0.14 2.02 1.79
Wolverhampton 1 .342 1.58 0.05 1.63 1.89 0.05 1.95 1.79
Crystal Palace 2 .336 1.55 0.08 1.66 1.76 0.08 1.89 1.78
Arsenal 13 .715 1.31 0.05 1.38 1.92 0.14 2.12 1.75
Tottenham 13 .486 1.27 0.05 1.35 1.80 0.11 1.95 1.65
Fulham 8 .446 1.29 0.11 1.43 1.69 0.09 1.84 1.64
Watford 2 .270 1.55 0.08 1.68 1.50 0.05 1.58 1.63
Table 10.5 (cont.)
Home matches Away matches All matches

No. of Win Average Average Average Average Average Average Average
seasons ratio yellows reds ‘points’ yellows reds ‘points’ ‘points’
Newcastle Utd 13 .505 1.11 0.07 1.23 1.81 0.12 2.01 1.62
Manchester City 8 .442 1.13 0.07 1.25 1.79 0.13 1.96 1.61
Southampton 9 .436 1.29 0.08 1.40 1.61 0.10 1.77 1.59
Portsmouth 6 .443 1.37 0.08 1.46 1.61 0.05 1.68 1.57
Aston Villa 13 .521 1.12 0.05 1.20 1.78 0.08 1.91 1.55
Leicester City 7 .442 1.19 0.05 1.26 1.69 0.11 1.85 1.55
Sheffield Wednesday 4 .431 1.18 0.05 1.29 1.58 0.12 1.72 1.51
Manchester Utd 13 .752 1.11 0.02 1.15 1.67 0.11 1.84 1.50
Reading 2 .428 1.16 0.11 1.29 1.50 0.11 1.68 1.49
West Brom Albion 4 .313 1.30 0.05 1.39 1.43 0.11 1.58 1.49
Bradford City 2 .316 1.37 0.00 1.37 1.50 0.05 1.55 1.46
Charlton Athletic 8 .441 1.21 0.09 1.36 1.43 0.07 1.55 1.46
Liverpool 13 .648 0.95 0.04 1.01 1.64 0.08 1.75 1.38
Wimbledon 4 .438 1.18 0.07 1.29 1.32 0.04 1.39 1.34
Ipswich Town 2 .480 0.68 0.03 0.74 1.29 0.05 1.37 1.05
Norwich City 1 .342 0.89 0.05 0.95 0.89 0.05 1.00 0.97

only), the individual team effects cannot be distinguished from the managerial
spell effects.
In order to investigate the statistical significance of individual team and man-
agerial spell effects, [10.5] is estimated first with individual team dummy variables
in place of the managerial spell dummy variables; in other words, with the restric-
tions δi,m = δi,n imposed if managerial spells m and n were with the same team,
for i = 1,2. In this estimation, a Wald test of H0:δi,m = 0 for i = 1,2 and m =
1 … 92 in [10.5] yields χ2(76) = 288.3 (p-value = .0000), indicating that the indi-
vidual team effects are significant. A Wald test for the validity of the restrictions
δi,m = δi,n yields χ2(108) = 173.5 (p-value = .0001), indicating that the managerial
spell dummy variables contain information that is relevant for explaining the vari-
ation in the incidence of disciplinary sanction, over and above the individual team
dummy variables. A Wald test for H0:δi,m = 0 for i = 1,2 and m = 1 … 92 in [10.5]
(with no restrictions imposed upon δi,m) yields χ2(186) = 441.9 (p-value = .0000),
indicating that the managerial spell effects are highly significant.
Individual referees
Inconsistency in the standards applied by different referees is among the most fre-
quent causes of complaint from football managers, players, supporters and media
pundits. Table 10.6 summarises the average numbers of disciplinary points per
match awarded against the home and away teams and against both teams com-
bined, by each of the thirty-seven referees who officiated at least thirty Premier
League matches during the observation period. There appears to be consider-
able variation between the propensities for individual referees to take disciplinary
action. For example, the most prolific referee (Mike Reed) averaged 4.541 discip-
linary ‘points’ over 85 matches, and the most lenient (Keith Burge) averaged 2.526
‘points’ per match over 57 matches.
Several of the referees who appear towards the top end of Table 10.6 were
involved in controversial incidents that may have hastened the end of their referee-
ing careers. For example, in February 2000 Mike Reed was temporarily suspended
for apparently masking celebratory gestures when Liverpool scored against Leeds
in a match he was officiating. In March 1998 Gary Willard awarded three red
cards to players from the home team in a match between Barnsley and Liverpool,
and was chased by an interloper on the pitch. Similarly in May 1999 Rob Harris
awarded three red cards to West Ham players in a home match against Leeds, and
narrowly escaped being attacked in his car after the match. Both of these two ref-
erees also committed errors involving the award of red cards outside the Premier
League: Willard failed to send off a player to whom two yellow cards had been
issued in a UEFA Cup match; and Harris permitted a team to continue playing
with eleven men after a player was simultaneously sent off and substituted in an
FA Cup match.
Table 10.6 Average numbers of yellow and red cards and disciplinary ‘points’ awarded
per match by referee, English Premier League, 1997–2009 seasons
Both
Home team Away team teams
No.
of Average Average Average Average Average Average Average
matches yellows reds ‘points’ yellows reds ‘points’ ‘points’
Reed, Mike 85 1.718 0.047 1.788 2.529 0.165 2.753 4.541

Willard, Gary 60 1.783 0.083 1.900 2.200 0.100 2.350 4.250
Harris, Rob 52 1.577 0.115 1.750 2.173 0.115 2.327 4.077
Barber, Graham 168 1.583 0.071 1.690 2.155 0.125 2.327 4.018
Riley, Mike 282 1.525 0.106 1.709 2.110 0.128 2.309 4.018
Dean, Mike 193 1.653 0.093 1.798 1.891 0.140 2.114 3.912
Dowd, Phil 148 1.378 0.081 1.527 2.034 0.155 2.291 3.818
Clattenburg, Mark 87 1.759 0.092 1.897 1.839 0.034 1.874 3.770
Wilkes, Clive 30 1.267 0.067 1.400 2.133 0.133 2.333 3.733
Styles, Rob 211 1.474 0.104 1.645 1.844 0.156 2.076 3.720
Knight, Barry 75 1.373 0.133 1.600 1.867 0.120 2.080 3.680
Bennett, Steve 225 1.422 0.067 1.516 1.920 0.142 2.107 3.622
D’Urso, Andy 119 1.311 0.118 1.479 1.899 0.160 2.109 3.588
Messias, Matt 45 1.444 0.044 1.489 1.956 0.089 2.089 3.578
Mason, Lee 39 1.487 0.077 1.641 1.744 0.103 1.923 3.564
Poll, Graham 276 1.395 0.080 1.514 1.909 0.105 2.040 3.554
Bodenham, Martin 44 1.455 0.000 1.455 1.955 0.068 2.045 3.500
Lodge, Stephen 102 1.333 0.029 1.392 1.892 0.137 2.108 3.500
Webb, Howard 150 1.400 0.047 1.473 1.920 0.080 2.027 3.500
Jones, Peter 112 1.304 0.063 1.411 1.875 0.071 1.991 3.402
Rennie, Uriah 176 1.517 0.057 1.614 1.580 0.114 1.773 3.386
Ashby, Gerald 33 1.152 0.030 1.212 1.939 0.121 2.152 3.364
Wilkie, Alan 81 1.284 0.037 1.358 1.864 0.074 1.975 3.333
Elleray, David 129 1.186 0.054 1.295 1.736 0.147 1.984 3.279
Atkinson, Martin 98 1.306 0.071 1.418 1.765 0.061 1.857 3.276
Barry, Neale 157 1.242 0.045 1.318 1.828 0.089 1.949 3.268
Dunn, Steve 178 1.253 0.045 1.331 1.803 0.062 1.882 3.213
Marriner, Andre 58 1.069 0.069 1.155 1.914 0.103 2.052 3.207
Wiley, Alan 248 1.399 0.040 1.448 1.617 0.073 1.742 3.190
Alcock, Paul 78 0.936 0.038 1.000 1.897 0.090 2.026 3.026
Winter, Jeff 165 1.073 0.036 1.121 1.800 0.073 1.891 3.012
Foy, Chris 127 1.205 0.039 1.260 1.583 0.102 1.717 2.976
Gallagher, Dermot 194 1.103 0.062 1.191 1.562 0.103 1.716 2.907
Walton, Peter 100 1.150 0.030 1.200 1.610 0.050 1.700 2.900
Durkin, Paul 169 1.089 0.053 1.166 1.355 0.059 1.450 2.615
Halsey, Mark 217 0.917 0.065 1.037 1.346 0.088 1.498 2.535
Burge, Keith 57 0.877 0.000 0.877 1.561 0.070 1.649 2.526
Note: Referees who officiated at fewer than thirty matches between the 1997 and 2009 seasons
(inclusive) are not included in Table 10.6.
Does the variation in refereeing standards suggested by Table 10.6 consti-

tute statistical evidence of inconsistency in refereeing standards? The refereeing
consistency hypothesis (RCH) imposes zero restrictions on the coefficients on the
individual referee dummy variables DRr,j, which identify matches officiated by the
thirty-seven referees listed in Table 10.6. In [10.5], a Wald test of H0:γi,r = 0 for i =
1,2 and r = 1 … 37 yields χ2(74) = 307.7 (p-value = .0000). Therefore the RCH is
rejected, suggesting that there was significant variation in standards between ref-
erees.4 Individual referees who contribute prominently to the rejection of the RCH
are Paul Alcock and Keith Burge (significantly lower-than-average propensity to
penalise home-team players); Mark Clattenburg, Mike Dean, Mike Reed and Mike
Riley (high propensity to penalise home-team players); Paul Durkin (low propen-
sity to penalise away-team players); and Phil Dowd, Mike Reed and Mike Riley
(high propensity to penalise away-team players).
The rejection of the HFAH suggests there is a bias favouring the home team in
the incidence of disciplinary sanction, even after controlling for home-field advan-
tage in match results. With the RCH also having been rejected, it is relevant to
examine whether there are significant differences between referees in the degree of
home team bias. In other words, do variations in the degree of home-team bias on
the part of different officials contribute to the observed pattern of refereeing incon-
sistency? The consistent home-team bias hypothesis (CHTBH) imposes the restric-
tion that the corresponding coefficients on the individual referee dummy variables
in the home- and away-team equations are the same. The CHTBH would imply
that the rate at which away teams tend to incur more disciplinary points than home
teams does not vary between referees. In [10.5], a Wald test of H0:γ1,r = γ2,r for
r = 1 … 37 yields χ2(37) = 74.2 (p-value = .0003). Therefore the CHTBH is rejected
at a significance level of 0.01, although the rejection of the CHTBH is somewhat
closer to the borderline than the rejection of the RCH. Individual referees who
contribute prominently to the rejection of the RCH are Mark Clattenburg (average
home-team and away-team ‘points’ of 1.897 and 1.874, respectively) and Uriah
Rennie (1.614 and 1.773, respectively). Mark Clattenberg is the only referee among
those listed in Table 10.6 who awarded more disciplinary ‘points’ against home
teams than against away teams.
Time consistency of players’ behaviour and refereeing

The individual football season dummy variables DSs,j are included in the con-
ditional model primarily as a control for changes over time in the content and
interpretation of the rules relating to the award of yellow and red cards. The key
changes during the observation period are detailed in Table 10.7. Most of the
changes have increased the range of offences that are subject to disciplinary sanc-
tion, although occasionally there has been movement in the opposite direction.
Some of the changes, such as directives concerning punishment for severe tackling
or serious foul play, should be considered as changes to the interpretation of rules
Table 10.7 Rule changes and changes of interpretation, by season
Season Rule changes/changes of interpretation
1997 Referees are reminded to severely punish the tackle from behind.
1998 Failure to retreat the required distance at free kicks and delaying the restart of play are to be punishable with a yellow card.
1999 The tackle from behind which endangers the safety of an opponent is to be punished with a red card.
The red card offence of denying an opponent a goal-scoring opportunity is changed to denying an opposing team a goal-scoring
opportunity (widening the scope of this offence).
2000 Simulation (diving, feigning injury or pretending that an offence has been committed) is to be punishable with a yellow card.
Referees are reminded to punish racist remarks with a red card. Swearing is also an offence warranting a red card.
2001 Offensive gestures are to be punishable with a red card.
2002 There is some relaxation of the rule requiring referees to issue a yellow card if a player celebrates a goal by removing his shirt.
However, celebrations that are provocative, inciting, ridiculing of opponents or spectators or time wasting remain punishable with
a yellow card.
Referees are reminded to punish intentional holding or pulling offences with a yellow card.
2003 Referees are reminded to be strict in punishing simulation and the delaying of restarts, especially if players remove shirts for any
length of time celebrating a goal.
2004 Increased onus is placed on the fourth official to draw attention to incidents of violent conduct that have been missed by the other
three officials.
2005 The removal of shirts to celebrate goal scoring is to be punishable with a yellow card.
2006 Any tackle that endangers an opponent must be punished with a red card.
Cards may only be shown while referee and players are on the field of play. Incidents in the tunnel or elsewhere are to be reported as
misconduct.
2007 Lunging tackles with one or two feet, and the use of elbows, are to be punishable with a red card. Shirt-pulling and holding offences
are to be punishable with a yellow card.
2008 New restrictions on goal celebrations indicate that a player who ‘covers his head or face with a mask or similar item’ are to be
punishable with a yellow card.
2009 It is reiterated that high or reckless tackles, including those where the foot is raised, are to be punishable with a red card. Players
guilty of holding and pushing in the penalty area should receive a warning for a first offence, followed by a yellow card for a repeat
offence.

Table 10.8 Average numbers of yellow cards, red cards and disciplinary ‘points’
awarded per match by season, English Premier League, 1997–2009 seasons
Home team Away team Both teams

Average Average Average Average Average Average
yellow red Average yellow red Average yellow red Average
cards cards ‘points’ cards cards ‘points’ cards cards ‘points’
per per per per per per per per per
Season match match match match match match match match match
1997 1.305 0.026 1.350 1.808 0.084 1.934 3.113 0.111 3.284
1998 1.303 0.058 1.405 2.016 0.124 2.189 3.318 0.181 3.595
1999 1.582 0.074 1.695 2.147 0.116 2.316 3.729 0.189 4.010
2000 1.411 0.055 1.497 1.932 0.129 2.118 3.342 0.184 3.616
2001 1.355 0.084 1.487 1.800 0.084 1.921 3.155 0.168 3.408
2002 1.247 0.084 1.389 1.803 0.103 1.955 3.050 0.187 3.345
2003 1.326 0.071 1.432 1.703 0.124 1.882 3.029 0.195 3.313
2004 1.266 0.053 1.350 1.579 0.100 1.726 2.845 0.153 3.076
2005 1.108 0.071 1.218 1.605 0.084 1.745 2.713 0.155 2.963
2006 1.313 0.068 1.424 1.774 0.129 1.968 3.087 0.197 3.392
2007 1.395 0.045 1.461 1.829 0.095 1.974 3.224 0.139 3.434
2008 1.345 0.082 1.476 1.855 0.079 1.976 3.200 0.160 3.453
2009 1.355 0.068 1.468 1.797 0.097 1.945 3.153 0.166 3.413
that already existed. Others, such as directives concerning punishment for exces-
sive or highly stylised goal celebrations, have clearly been introduced in response
to changes in fashionable behaviour on the part of players: rules did not exist pre-
viously, because players did not behave in such a manner.
Table 10.8 reports the average numbers of yellow and red cards and disciplin-
ary ‘points’ incurred by the home and away teams per match by season. There
appears to be little or no trend in the overall incidence of disciplinary sanction,
despite the increase in the range of sanctionable offences. Witt (2005) suggests that
when there is an addition to the list of sanctionable offences, players may modify
their behaviour so that the numbers of cautions and dismissals remain approxi-
mately constant. Alternatively, referees may tend to modify their interpretation of
the boundaries separating non-sanctionable from sanctionable offences, and those
separating cautionable from dismissable offences, so as to maintain an approxi-
mately constant rate of disciplinary sanction.
The directive issued at the start of the 1999 season making the tackle from
behind punishable by automatic dismissal is the only rule change that appears
to have had a discernible impact on the data that are summarised in Table 10.8.
The mean incidence of disciplinary sanction is higher for 1999 than for any of the
other twelve seasons reported in Table 10.8. Within the 1999 season as well, the
process of adjustment to the new disciplinary regime is visible in the data: during
the first three months of the 1999 season the average disciplinary points incurred
by both teams per match was 4.3, while the average for the rest of the season was
3.9 (see also Witt, 2005). Although this directive has remained in force subse-
quently, the incidence of disciplinary sanction returned to levels similar to those
experienced before the directive came into effect. There appears to have been
a slight dip in the incidence of disciplinary sanction in the 2004 and 2005 sea-
sons: the only two seasons reported in Table 10.8 in which the average number
of disciplinary ‘points’ per match dropped to around 3.0. Subsequently, how-
ever, the average has reverted to a level that is very similar to the average for the
1997–2009 period as a whole.
In order to test the time consistency hypothesis (TCH) that the average incidence
of disciplinary sanction is stable over time, the null hypothesis (expressed in terms
of the coefficients of the conditional model) is H0:βi,s = 0 for i = 1,2 and s = 1998 to
2009 (inclusive). A Wald test yields χ2(24) = 82.3 (p-value = .0000), suggesting there
was significant season-to-season variation in the incidence of disciplinary sanction.
This rejection of the TCH is driven primarily by the coefficients in the home- and
away-team equations for 1999 (when, as noted above, the incidence of disciplinary
sanction was significantly higher than the average), and the coefficient in the home-
team equation for 2005 (when the incidence of disciplinary sanction was signifi-
cantly lower than the average). With these exceptions, the TCH can be accepted
in respect of all of the other coefficients on the individual football season dummy
variables; and the TCH therefore receives qualified support from these results.
Match attendance
Under the audience neutrality hypothesis (ANH), the incidence of disciplinary
sanction is unaffected by the size of the crowd inside the stadium.5 Conversely,
a large attendance might be expected to add to the intensity or excitement of the
occasion, resulting in more determined or aggressive play by either or both teams,
or in stricter interpretation or application of the rules by the referee. Alternatively,
a large attendance, presumably dominated by supporters of the home team, might
put pressure on the referee to treat disciplinary transgressions by the home team
more leniently, and those by the away team more severely. In order to investi-
gate the ANH, the covariate attj, defined as the reported attendance at match j, is
included in the regressions for ln(λi,j).
The estimated coefficients on attj in [10.5] are both positive and significantly
greater than zero at the 0.05 level. The ANH is rejected, but there is no evidence
of any tendency for referees to treat the home team more leniently when the crowd
size is large; in fact, the coefficient in the home-team equation is larger than the
coefficient in the away-team equation; but the difference is not statistically signifi-
cant. The positive coefficients on attj are consistent with the notion that the behav-
iour of either the players or the referees is influenced by the size of the crowd, in a
manner that produces a higher incidence of disciplinary sanction in matches with
larger attendances.
Conclusion
Football referees are routinely criticised by managers, players, journalists and
spectators. Split-second decisions taken by referees can have enormous financial
consequences, due to the fine line that exists between success and failure in foot-
ball. The actions of referees have never been more intensely scrutinised than they
are today. Football referees are often accused of favouritism and bias; and these
accusations have been the subject of several academic studies in recent years.
A number of empirical studies of bias in football refereeing focus on the deci-
sion of the referee to add on time at the end of regular time (after 90 minutes has
elapsed). Several such studies report a tendency for referees to add on more time at
the end of close matches (one goal difference between the teams at the end of the
match) when the home team is trailing than when the home team is leading. Other
empirical studies examine the decision to caution or dismiss players (award yellow
and red cards). In this chapter, we have reported estimations for the incidence of
disciplinary sanction against footballers in English Premier League matches, in
the form of a comprehensive statistical analysis of patterns in the award of yellow
cards and red cards over a thirteen-year period.
In the estimations for the numbers of yellow and red cards awarded against the
home and away teams, it is found that relative team strengths matter: underdogs
tend to incur a higher rate of disciplinary sanction than favourites. The incidence
of disciplinary sanction tends to be higher in matches between evenly balanced
teams, and in matches that attract high attendances. Home teams play more
aggressively in front of larger crowds, but crowd size does not influence the inci-
dence of disciplinary sanction against the away team.
Individual referee effects make a significant contribution to the explanatory
power of the model, indicating that there are inconsistencies between referees
in the interpretation or application of the rules. There is evidence of variation
between referees in the degree of home-team bias; and this variation contributes
to the overall pattern of refereeing inconsistency. The tendency for away teams to
be penalised by referees more than home teams cannot be explained solely by the
home-field advantage effect on match results. Even after controlling for team qual-
ity, a (relatively strong) away team can expect to collect more disciplinary points
than a (relatively weak) home team with the same win probability.
Notes
1 Other countries with professional referees include France, Mexico, Brazil, the Netherlands,
Spain and Italy. In 2007 referees in Italy received a salary of €70,000 per year or €35,000 per
year depending on experience. Referees additionally receive €3,500 per game in Serie A and
€2,000 per game in Serie B (plus expenses). The internationally elite referees who officiated
at the World Cup in 2006 were paid $40,000 each for the tournament, double the amount
in 2002. At Euro 2008, referees were paid €10,000 per match officiated. This represents an
increase of almost 60 per cent compared to Euro 2004.
2 In other words, F1(z1) = P(Z1,j ≤ z1) expresses the probability that the home team incurs z1
or fewer disciplinary ‘points’ in match j, for all possible values of z1 ( = 0,1,2, …). Similarly,
f2(z2) = P(Z2,j ≤ z2) expresses the probability that the away team incurs z2 or fewer disciplinary
‘points’.
3 The construction of the bivariate Poisson or negative binomial distributions using the Frank
copula demands some further comment. The ancillary parameter φ allows for unrestricted
(positive or negative) correlation between Z1,j and Z2,j. In contrast, the standard bivari-
ate Poisson distribution constructed by combining three random variables with univariate
Poisson distributions, and several alternative formulations of the bivariate negative binomial
distribution described by Kocherlakota and Kocherlakota (1992), are capable of accommo-
dating positive correlation only. The positive sample correlation coefficient in the present case
notwithstanding, there appears to be no compelling case for defining the bivariate distribu-
tions in any manner that would exclude the possibility of obtaining a negative correlation.
4 Since the conditional model includes controls for team quality and other potential influences
on the incidence of disciplinary sanction, the rejection of the RCH should not be attributable
to any non-randomness in the assignment of referees to matches: for example, the tendency
for referees with a reputation for toughness to be assigned to matches at which disciplinary
issues are anticipated by the authorities. However, the use of 0–1 dummy variables for the
individual referee effects might represent a simplification. This procedure does not allow for
duration dependence in referees’ performance. Duration dependence might arise if referees
modify their behaviour as they gain in experience, or if the removal of unsatisfactory referees
by the football authorities creates a form of survivorship effect, such that the behaviour of
long-serving referees differs on average from the behaviour of all referees.
5 Dawson et al. (2007) also examine whether the incidence of disciplinary sanction differs
between matches that are screened live on TV, and those that are not. There is no evidence
of any difference, and the live TV dummy variable has been omitted from the estimations
reported in this chapter.
11 Spectator demand for football
Introduction
A number of broad trends in league match attendances in English football at the
aggregate level are identified in Chapter 6. During the post-Second World War
boom, league attendances surged, reaching an all-time high of 41 million in the
1949 season. The boom, however, was relatively short-lived. It was followed by
a period of sustained decline that continued, almost uninterrupted, until the
1986 season, when attendances fell to 16.5 million. Subsequently there has been
a steady and sustained improvement. By the 2009 season, total attendances had
increased to 29.9 million. Undoubtedly, the growth in attendance since the late
1980s understates the growth in demand, because many of the leading clubs are
capacity-constrained and could sell volumes of tickets that in some cases would be
far in excess of existing stadium capacities.
Chapter 6 also reviewed the academic debate about the causes of the long
post-war decline in football attendances, and its recent reversal. Social and demo-
graphic change, increasing material affluence, the option to watch football on tele-
vision rather than in person, crowd misbehaviour, the deteriorating physical state
of many of football’s stadia, and the dubious quality of the some of the fare on
offer on the field of play are among the many factors considered to have contrib-
uted to the decline in the popularity of attending live football. More recently,
improved facilities in all-seated stadia, together with the near-eradication of the
hooligan problem, have helped strengthen football’s appeal as a middle-class
spectator sport. In addition, a sport so heavily steeped in history and tradition
could hardly have failed to benefit from the ‘heritage’ fad during the 1990s and
2000s. Improved standards of entertainment on the field of play, boosted by a
large influx of talented overseas players, have also contributed much to football’s
popular resurgence.
Social history and sociology provide many useful insights into the causes of fluc-
tuating football attendances. Econometric modelling of variations in the attend-
ances of individual clubs, both season-by-season and match-by-match, has been
a subject of interest for sports economists, especially in the UK, since the 1970s.
321
Section 11.1 provides a non-technical review of the empirical literature, focusing

on the issues of variable definition, model specification, estimation and interpret-
ation that are faced by researchers in this area.
Using the match attendance and revenues data set described in earlier parts
of this volume, Sections 11.2 and 11.3 present an analysis of variations in club-
specific average season attendances during the post-Second World War period.
There are two stages to the empirical analysis. At the first stage, a model of the
demand for attendance is estimated, reflecting the influence of four factors upon
demand: short-term loyalty, team performance, admission price and entertain-
ment (proxied by goals scored). The model also produces rankings of all clubs
according to their estimated base levels of attendance. At the second stage, the
cross-sectional variation between clubs’ base levels of attendance, and between
their short-term loyalty, performance, price and entertainment coefficients, is
explained in terms of socioeconomic, demographic and football-related charac-
teristics of each club and its home town.
11.1 Econometric analysis of football attendances

As noted in the introduction to Chapter 11, statistical or econometric modelling
of variations in attendances has attracted considerable attention from academics
since the 1970s. Hart, Hutton and Sharot (1975) published the first economet-
ric analysis of patterns of attendance at English football matches. At around the
same time, Demmert (1973) published an investigation of US baseball attend-
ances, while Noll (1974) compared the determinants of attendances at four US
professional team sports: baseball, basketball, football and hockey. Since the
appearance of these early studies, the topic has continued to receive the attentions
of econometricians on a regular basis. This section reviews this literature, focus-
ing primarily on studies that have analysed football attendances, but also drawing
insights from other sports where applicable. For previous reviews of the literature
on the demand for sports in general or football in particular, see Cairns (1990),
Borland and MacDonald (2003) and Downward, Dawson and Dejonghe (2009).
Measuring the demand for football attendance

Attendance data, usually announced by the home club while the match is in pro-
gress, provides the main source of data for the dependent variable in econometric
models of the demand for football attendance. Academic researchers have not
hesitated to exploit voluminous and easily accessible attendance data sets, while
acknowledging that some difficulties can arise when interpreting attendance data
as a measure of spectator demand. Most sports attendance studies fall into one of
two categories: first, those that model individual match attendances; and second,
those that model entire season or annual attendances (either for the league as a
whole or for individual clubs).
Econometric analysis of football attendances 323
Football attendance data are usually aggregated over two types of specta-
tor: season ticket holders, who pay for an entire season’s admission in advance;
and purchasers of tickets for individual matches, who may have decided to attend
as early as a month or so in advance, or as late as the match day itself. Clearly the
factors that influence the decisions of members of either group to attend will tend
to differ. Once a season ticket has been purchased, the further costs arising from
the decision to attend each match relate to time and transport. The disincentives
for season ticket holders to attend are small, unless the home team’s performance
is so poor that it is less painful to stay at home! Match ticket purchasers, on the
other hand, are likely to be more selective in their choice of fixtures, and more
sensitive to fluctuations in recent home-team performance or the quality of the
visiting team. It is regrettable that most match attendance data do not distinguish
between the two groups. Becker and Suls (1983) estimate separate regressions for
match attendances and season ticket sales for MLB during the 1970s. Their find-
ings for the relationship between team performance and demand are quite simi-
lar in both cases. For football, the annual attendance data set used by Simmons
(1996) distinguishes between season ticket and match ticket sales. The match-level
Scottish Premier League attendance data used by Allan and Roy (2008) distin-
guishes between season ticket attendees, and home- and away-team supporters
who purchased individual tickets.
Published match attendance data also include groups of ticket holders who may
have paid different prices for seats of varying quality. Before the advent of all-
seated stadia, reported match attendance data were aggregated over both seated
and standing spectators. Again, the nature of the aggregation tends to restrict or
preclude the investigation of several important issues, including measurement of
price and other elasticities of demand for tickets in different price categories, or
in different parts of the stadium. Using survey data, which disaggregates attend-
ances into standing and seated components, Dobson and Goddard (1992) find
that the attendance of standing spectators was more variable than that of their
seated counterparts. Standing attendances were particularly sensitive to the recent
performance of the home team, and whether the match was significant for cham-
pionship outcomes.
As an alternative to published match attendance data as a measure of spectator
demand, Forrest, Simmons and Feehan (2002) use data from the 1996 season FA
Premier League National Fan Survey. Questionnaires were sent to supporters of
the twenty T1 clubs, and respondents identified their place of residence, and the
number of home matches attended. Concentric zones around each club’s stadium
are defined for distances of up to 60 miles for clubs with smaller catchment areas,
or 200 miles for clubs with larger catchment areas; and ticket sales per head of
population within each zone are estimated using the survey results and the aggre-
gate attendance data for each club.
Most studies analyse attendance by estimating single-equation regressions,
interpreting the resulting equations as demand functions. This is acceptable if it
can be assumed that supply is perfectly elastic. If not, estimation of the coeffi-
cients of the demand equation may be problematic. Suppose, for example, attend-
ance depends on admission price, but admission price also depends upon expected
attendance, which might be the case if clubs increase their ticket prices for the
most popular fixtures. One of the key assumptions of regression analysis, that
the covariates are uncorrelated with the disturbance term, is violated, resulting
in biased and inconsistent coefficient estimates in the demand equation, unless
an estimation procedure is used which takes account of the endogeneity of the
admission price variable.
Using annual data on gate revenue and team performance, Dobson and Goddard
(1998a) use Granger causality tests to investigate whether most attendance mod-
els are correct in assuming, implicitly, that there is a unidirectional relationship
running from variations in team performance to variations in attendance (and
therefore revenue). Alternatively, causality might also run in the opposite direc-
tion, if teams that enjoy high attendance and revenue are able to ‘buy success’,
either through transfer dealings or by offering salary inducements to retain the
best players. In both this study and an earlier study by Davies, Downward and
Jackson (1995) for English rugby league, causality from revenue to performance
is found to be stronger than causality in the reverse direction. These findings raise
some doubts as to whether the single equation models used in most annual attend-
ance studies capture adequately the dynamics of the attendance-team perform-
ance relationship.
The treatment of stadium capacity constraints seems to be one of the least sat-
isfactory aspects of the early empirical sports attendance literature. In cases where
the stadium is filled to capacity, the reported match attendance figure is not a
true representation of actual spectator demand, which is unobservable. Ordinary
least squares (OLS) estimation of a demand equation including the observations
on matches where the attendance was capacity-constrained produces biased and
inconsistent estimates of the coefficients of the demand equation. If the capaci-
ty-constrained observations are omitted from the sample, then the estimation is
subject to a form of sample selection bias, again resulting in biased and incon-
sistent coefficient estimates. The correct technical solution requires the use of
tobit regression, rather than OLS. Tobit is applicable in cases where the depend-
ent variable of a regression equation is truncated or censored. The specification
of the likelihood function takes account of the distinction between the observed
attendance, which is right-censored in the case of matches where the attendance
was capacity-constrained, and the actual demand for attendance for such matches,
which is unobservable.
Table 11.1 presents data on stadium capacity utilisation for the twenty clubs
that played in T1 during the 2009 season. For each club, the stadium capacity and
average league attendance for the 1979, 1989, 1999 and 2009 seasons are reported.
Capacity utilisation for each club is average attendance expressed as a percentage
of capacity. Several 2009-season T1 clubs were playing in the league’s lower tiers
Table 11.1 Stadium capacity, average attendance, capacity utilisation, 2009 season T1 clubs, 1979, 1989, 1999 and 2009
Stadium capacity Average attendance Capacity utilisation Tier

1979 1989 1999 2009 1979 1989 1999 2009 1979 1989 1999 2009 79 89 99 09
Arsenal 60000 57000 38500 60355 36371 35595 38024 60040 60.6 62.4 98.8 99.5 1 1 1 1
Aston Villa 48000 48100 39217 42573 32838 23310 36937 39812 68.4 48.5 94.2 93.5 1 1 1 1
Blackburn 47500 21956 31367 31367 8640 8891 25773 23479 18.2 40.5 82.2 74.9 2 2 1 1
Bolton 43000 29000 25000 28101 24772 5528 18240 22486 57.6 19.1 73.0 80.0 1 3 2 1
Chelsea 60000 43900 35421 41841 24782 15731 34754 41589 41.3 35.8 98.1 99.4 1 2 1 1
Everton 58000 50271 40200 40158 35456 27765 36202 35667 61.1 55.2 90.1 88.8 1 1 1 1
Fulham 42000 19400 19250 26600 10135 4938 11387 24344 24.1 25.5 59.2 91.5 2 3 3 1
Hull 42000 19797 12439 25404 5238 6666 6051 24816 12.5 33.7 48.6 97.7 3 2 4 1
Liverpool 52318 45628 45362 45522 46406 38574 43321 43611 88.7 84.5 95.5 95.8 1 1 1 1
Manchester City 52500 51993 31458 47726 36203 23500 28261 42899 69.0 45.2 89.8 89.9 1 2 3 1
Manchester Utd 58504 56385 56387 75769 46430 36488 55188 75304 79.4 64.7 97.9 99.4 1 1 1 1
Middlesbrough 42000 30647 35000 35100 18459 19999 34386 28429 44.0 65.3 98.2 81.0 1 1 1 1
Newcastle 40480 37703 36834 52387 20834 22921 36690 48750 51.5 60.8 99.6 93.1 2 1 1 1
Portsmouth 46000 29664 19179 20688 10123 10201 11973 19830 22.0 34.4 62.4 95.9 4 2 2 1
Stoke 40000 35812 24054 28383 19125 9817 12732 26960 47.8 27.4 52.9 95.0 2 2 3 1
Sunderland 53500 37775 42000 49000 25454 14878 38745 40168 47.6 39.4 92.3 82.0 2 2 2 1
Tottenham 52000 34258 36236 36534 34902 24467 34149 35929 67.1 71.4 94.2 98.3 1 1 1 1
West Bromwich 38600 35000 25396 28003 26517 12757 14585 25828 68.7 36.4 57.4 92.2 1 2 2 1
West Ham 39500 35556 26054 35303 25778 20738 25639 33701 65.3 58.3 98.4 95.5 2 1 1 1
Wigan 30000 12500 7290 25138 6701 3151 4250 18350 22.3 25.2 58.3 73.0 4 3 3 1

ten, twenty or thirty seasons earlier, and these details are also shown. In the 1979
season, many clubs played in stadia whose physical layout had hardly changed
since the late nineteenth or early twentieth centuries. Typically, large banks of
standing-only accommodation (terraces) were located behind the goals at both
ends of the stadium; and seated accommodation (sometimes together with further
terracing) was located in grandstands built along the sides of the pitch. The ter-
races allowed many clubs to accommodate larger maximum attendances than they
can today; but for most clubs capacity attendances were the exception and not the
rule. Consequently capacity utilisation in 1979 was far lower than it was in 2009.
By the time of the 1989 season, the physical landscape of English football
had already started to change, with many clubs responding to sharply declining
attendances throughout the 1970s and early 1980s, and persistent crowd-control
problems arising from the football hooliganism phenomenon, by lowering their
stadium capacities. Capacity utilisation remained relatively low, however. Urgent
impetus towards the physical overhaul of most football stadia was provided by the
Hillsborough stadium disaster of April 1989; and within a further ten years a radi-
cal transformation was largely complete, with terracing removed from the stadia
of virtually all T1 and T2 clubs (and from some but not all lower-tier clubs). With
attendances generally up and stadium capacities in some cases sharply down, the
capacity utilisation data for the 1999 season show about half of the clubs reported
in Table 11.1 able to sell virtually every available ticket for every match. In 2009 the
picture was similar, with several clubs having, in the meantime, invested heavily in
the expansion of existing stadium capacity, or relocation to new stadia (see also
Chapter 6, Section 6.2).
Some attendance studies include stadium capacity as an explanatory variable
(Noll, 1974; Schollaert and Smith, 1987; Kahn and Sherer, 1988; Brandes, Franck
and Nuesch, 2008). Clearly, however, the resulting model cannot be interpreted as
a demand equation, as it contains a mix of demand and supply side variables. In
some studies it is assumed that the number of matches attracting capacity attend-
ances is sufficiently small for the problem to be ignored. This approach may have
been reasonable at times in the past, when many stadium capacities were higher
and attendance was lower.1 Recently, as Table 11.1 suggests, many of the leading
clubs sell out regularly for all but the least significant matches. Although the esti-
mation of a tobit regression is valid in cases where the distribution of the depend-
ent variable is constrained or truncated, a data set containing a reasonable mix of
constrained and unconstrained attendances is still required. At present, the infor-
mation content in a recent season’s array of home match attendances for clubs
like Arsenal, Chelsea and Manchester United is zero (unless, of course, one merely
wishes to identify the stadium capacity).
Demographic and geographic determinants of attendance

A link between the size of the market from which each club draws its support
and match attendances seems to be both theoretically and intuitively obvious.
Empirically, however, it is less straightforward either to define each club’s catch-

ment area, or to come up with a market size measure that is not arbitrary to some
extent. Defining the size of the market is made difficult by the progressive erosion
over time of the geographical segmentation of supporters, especially of the lead-
ing clubs. Manchester United, for example, are now famous (or notorious) for
drawing their support from all regions of the UK, and far beyond.
Even if markets could be segmented geographically, measurement is made dif-
ficult by the arbitrary nature of the municipal boundaries used to compile most
local population statistics, and by the need to make adjustments in cases where
two or more clubs are situated in close proximity. Despite these difficulties, in prin-
ciple it should be possible to identify empirically a link between local population
and attendance by comparing a cross-section of clubs. It may be more difficult to
do so for a single club over time, since local populations change perhaps far more
slowly than other factors that influence a club’s attendance. In fact, many (but not
all) match attendance studies have included cross-sectional market size measures
for either or both of the home and away teams among their explanatory variables.
Where such measures are used, almost invariably they produce regression coef-
ficients which are of the expected sign and statistically significant.
By estimating separate match attendance equations for each home team, Hart,
Hutton and Sharot (1975) and Cairns (1987) both avoid the need to include a
home-team market size measure. For the away team, Hart, Hutton and Sharot use
the male populations of the parliamentary constituencies surrounding each club’s
ground. Arbitrary adjustments are made to subdivide the data in cases where two
or more clubs are in the same or neighbouring geographical areas. Jennett (1984)
and Baimbridge, Cameron and Dawson (1996) use population measures for local
authority districts. Walker (1986) focuses specifically on city size effects on attend-
ance, using travel-to-work-area population definitions based explicitly on observed
patterns of personal mobility and travel.
Because of the arbitrary nature of any population indicator, Peel and Thomas
(1988) abstain from using such measures. Dobson and Goddard (1992) side-step
the difficulties of measuring the home team’s catchment area by including home-
team dummy variables. All influences on attendance that are time-invariant (from
match to match) and specific to the home team are captured in the dummy vari-
able coefficients. These may include socioeconomic and demographic influences
(population, age structure, earnings and unemployment), and sporting influences
(the strength of the club’s historical record or traditions, or its success over the
long term in building up a base level of support in the local or wider community).
An advantage of this approach is that all such influences are taken care of, and
the other estimated coefficients are not contaminated by omitted variable bias. A
disadvantage is that after dumping all such factors into the dummy variable coeffi-
cients, their individual effects on attendance cannot subsequently be disentangled.
Following Hart, Hutton and Sharot (1975), the geographical distance between
the stadia of the home and away teams is used in most attendance studies, to allow
for the positive ‘local derby’ effect on attendances, and the negative effect of long
distances on the propensity of away supporters to attend. The distance variable

is typically a highly significant determinant of attendance (with a negative coeffi-
cient). There are slight variations in the approaches of Jennett (1984), who divides
the away-team population by distance to obtain a composite market size and dis-
tance measure, and Cairns (1987) who uses a dummy variable for local derbies.
Price, income and unemployment

In general, match attendance models have difficulties in identifying a relationship
between variables such as admission prices, per capita income, average earnings, or
the local unemployment rate, and match attendances. Cross-sectionally, the rela-
tionship tends to be ambiguous, partly because some of the most keenly supported
clubs (such as Liverpool, Everton, Newcastle United and Sunderland) are located
in cities or regions of relatively low per capita income and high unemployment.
Consequently, such clubs may charge lower admission prices than clubs of similar
status in more affluent locations. This does not imply, however, that a reduction
in per capita income, an increase in unemployment or a rise in admission prices
would be expected to increase attendance. Most match attendance studies lack an
adequate time dimension to be capable of identifying the relationship between
variations in price, income or unemployment over time, and corresponding varia-
tions in attendance.
Among the match attendance studies that have used explanatory variables in this
category, Jennett (1984) finds that the unemployment rates of the home and away
team’s local area are both negatively related to attendances for Scottish Premier
League matches. The model was estimated, however, using data for a period
(1975–81) when attendances were falling and unemployment was rising sharply.
Under these circumstances, a negative and significant regression coefficient does
not necessarily reflect a causal link between unemployment and attendance. The
problems involved in interpreting regression models estimated using time-series
data which is non-stationary or trended are discussed below.
Baimbridge, Dawson and Cameron (1996) include regional unemployment, and
linear and quadratic terms in earnings and admission price in an English Premier
League match attendance equation for the 1994 season. Counter-intuitively (but
possibly for the reasons discussed above) the coefficient on unemployment is posi-
tive and significant. The coefficients on earnings seem plausible, and suggest that
attendances increase with average earnings whenever the latter are above about
£400 per week. The model fails, however, to identify an economically meaningful
price effect.
In an analysis of match-level attendances in Spain’s La Liga, Garcia and
Rodriguez (2002) show that price elasticities are underestimated if an attendance
equation is estimated using single-equation OLS, without taking account of the
endogeneity of the price variable. The problem of simultaneity bias is addressed
by means of an instrumental variables estimation. A separate equation for price
is estimated, dependent on a large number of current and lagged covariates. The

attendance demand equation is then estimated using the fitted values from the
price equation, in place of the original price variable. The fitted values should
be uncorrelated with the disturbance term of the attendance equation, while the
price variable is correlated with the disturbance term if price is partly dependent
on attendance. OLS estimation of the attendance equation produces estimated
price elasticities for all clubs that are negative but smaller than one in absolute
value. These estimates are inconsistent with profit-maximising pricing behaviour.
Instrumental variables estimation produces larger and more plausible price elasti-
cities, that are greater than one in absolute value for some (but for not all) clubs.2
As noted above, Forrest, Simmons and Feehan (2002) use survey data to esti-
mate ticket sales per head of population within concentric geographical zones
around each English T1 club’s stadium. Travel costs as a function of distance are
estimated, taking account of the costs of travel by car and by public transport, and
the time costs of travel. Estimated travel costs from each zone are combined with
average ticket prices to obtain a total price for attendance, and demand is shown
to be price-sensitive: the greater is the distance and the higher is the total price,
the lower are ticket sales per head of population. An attempt is made to control
for spatial variation in the affinity felt for each club by including as a control the
proportion of the population who attended at least three away matches. However,
the affinity variable is also highly correlated with distance from the stadium, and
therefore with the price variable. The coefficients on the price variable are used to
compute price elasticities, which are generally larger and perhaps more plausible
than those reported in several other demand studies. It is not completely clear,
however, that the affinity variable is sufficiently finely calibrated to disentangle the
price effect fully from the effect of heterogeneity in the tastes of the population
within each zone.
Several time-series or panel studies, based on annual attendance data, have
attempted to identify relationships between price, income and unemployment var-
iables, and football attendances. Bird’s (1982) study, which reports equations for
attendance aggregated across the entire league and by division, was the first of its
kind for English football. In an attempt to capture the full cost of attending, Bird’s
price measure combines the league’s minimum admission price with an impli-
cit price index for motoring and other transport, taken from the UK’s national
income accounts. Total consumer expenditure is used as an income measure. The
results suggest a price elasticity of demand for football attendance of –0.22, and
an income elasticity of –0.62.
Bird’s negative income elasticity implies that attending football is an inferior
good. As incomes increase, consumers tend to give up attending football, and
follow other, perhaps more expensive or ‘up-market’ leisure activities instead. As
a description of what actually occurred between the 1950s and 1970s, this seems
plausible (see Chapter 6). However, developments in time-series econometrics since
Bird’s paper was published raise doubts about the statistical validity of some of
the empirical relationships. Specifically, although it is correct to say that incomes

rose and football attendances fell between the 1950s and the 1970s, nothing more
than this can be inferred from a negative and significant coefficient obtained in a
crude regression of attendance on income. It is not possible to infer that the rise
in income ‘caused’ or even ‘influenced’ the decline in attendance in any sense. All
that can really be said is that income and attendance were trended in opposite
directions during the period in question.
Szymanski and Smith (1997) estimate attendance and revenue equations as
part of a larger model, which attempts to capture both demand and supply side
influences on the trade-off between team performance and profit faced by club
owners. In an attendance estimation pooled over forty-eight clubs using data for
the period 1974–89, an estimated price elasticity of –0.76 is obtained. When a set
of time dummy variables is included, this estimate drops to –0.34. No income or
unemployment measure is included among the explanatory variables.
Under what circumstances can reliable inferences be drawn from estimations
involving trended or non-stationary time-series variables?3 Using data on the
annual attendances of individual clubs, Dobson and Goddard (1995) suggest de-
trending the data prior to estimation, by expressing all variables as deviations from
their annual averages across clubs. A model estimated using this approach, which
eliminates the possibility that the estimated coefficients will identify spurious rela-
tionships between trended variables, is presented in Section 11.2. Simmons (1996)
and Dobson and Goddard (1996) tackle the same problem using cointegration
techniques. In this case, the idea is that if there is a genuine relationship between
two or more non-stationary time-series variables, either or both of the following
features should be apparent in the data:
• A statistical relationship should be discernible between the year-on-year changes
(first-differences) in the same variables. If income really does influence attend-
ance, it should be possible to identify a link between (say) an above-average rise
in income in a particular year, and an above-average rise or fall in attendance
in the same year (or perhaps in the following year). If apart from both being
trended or non-stationary, income and attendance are otherwise unrelated, no
short-run relationship should be identifiable between their first differences.
• Relationships should be discernible between the long-run movements in the var-
iables. If, for example, the rate at which incomes are increasing slows down for a
few years, other things being equal there should also be a corresponding reduc-
tion in the rate at which attendances are declining, if the income effect is nega-
tive as Bird claims. The fact that soon after Bird’s study was published, incomes
continued to increase while attendances started to rise, suggests that this kind of
pattern would not have been identified had the estimation included later data. If
a group of variables can be identified, each of whose long-run movements can
be adequately accounted for by co-movements in the others, then a cointegrat-
ing relationship is said to hold between the group of variables.
Dobson and Goddard (1996) report an estimated model for annual attendances
based on data from the mid-1950s to the early 1990s. The data for individual clubs
are pooled across all clubs located in each standard region for purposes of esti-
mation of a set of regional attendance equations. Regional unemployment rates
are used as an explanatory variable, in place of income or consumer expenditure.
There is evidence of a long-run, cointegrating relationship between variations
in regional unemployment and annual attendances, but no evidence of a long-
run price effect. Conversely, price changes are found to have a negative short-
run impact on changes in attendance, while movements in unemployment have no
short-run effect. Overall, there is little evidence to suggest that the coefficients of
the attendance equation vary significantly from region to region.
In a set of club-specific time-series estimations for nineteen clubs using data
from the 1960s to the early 1990s, Simmons (1996) finds evidence of a negative
price effect on attendance in the long term, and limited evidence of a long-run
income effect which, where apparent, tends to be positive rather than negative. To
obtain economically meaningful and statistically significant estimated coefficients
in the long-run part of the model, the sets of variables included in the cointegra-
tion tests are permitted to vary between clubs. The evidence of a negative price
effect is consistent, and the estimated long-run price elasticities are higher in abso-
lute value than in other studies: above –0.5 for ten clubs, and above –1.0 for two.
There is some evidence of a negative price effect in the short run, while the short-
run income coefficients include both positive and negative signs.
Team quality and uncertainty of outcome

In most empirical studies, it is clear that there is a strong relationship between
team quality and match attendance. Hart, Hutton and Sharot (1975) use the cur-
rent league positions of the home and away teams as explanatory variables, and
find that while highly placed away teams attract higher attendances, the home pos-
ition coefficients are generally insignificant. By estimating club-specific equations
over a longer period, thereby obtaining greater variation into the home position
variable, Cairns (1987) obtains significant coefficients on both the home and away
position variables; and by pooling their data across clubs Walker (1986), Peel and
Thomas (1988) and Dobson and Goddard (1992) obtain similar results. In add-
ition to league position variables, Walker (1986), Cairns (1987) and Dobson and
Goddard (1992) include the number of points gained from the three, four or five
most recent matches played by either or both teams. A recent run of good results
is usually found to have a significant and positive effect on attendance.
On the whole, the issues involved in incorporating team quality measures into
a season attendance model are straightforward. The estimations of Dobson and
Goddard (1995, 1996), Simmons (1996) and Syzmanski and Smith (1997) using
club-level data, control for team quality by including the team’s final league pos-
ition as an explanatory variable in its own attendance equation. While the team’s
final position is unknown at the time when decisions to attend are actually taken,
as the league competition unfolds potential and actual spectators are able to form
increasingly reliable estimates.
Incorporating team quality information into a match attendance model natur-
ally raises questions about the link between uncertainty of outcome and attend-
ance (see also Chapter 3, Section 3.1). In view of the emphasis that is attached to
competitive balance and uncertainty of outcome in the theoretical literature on
the economics of professional team sports leagues (see Chapters 1 and 2), it is per-
haps surprising to discover that convincing empirical evidence for a link between
these factors and spectator demand has been hard to pin down. Most empirical
studies have found either weak support, or in some cases no support, for the exist-
ence of any such link.
Hart, Hutton and Sharot (1975) include the absolute value of the difference
between the home and away teams’ league positions as an explanatory variable,
but fail to obtain the negative coefficient that would indicate that the more evenly
balanced are the teams and the more uncertain is the match outcome, the higher
is the attendance. Peel and Thomas (1988) use home win probabilities calculated
from bookmakers’ posted betting odds, and find that attendance is positively
related to the probability of a home win. However, by incorporating this vari-
able in linear (rather than quadratic) form, effectively they measure relative team
quality and not uncertainty of outcome directly. Matches at opposite ends of the
home-win probability scale (at 0.05 and 0.95 for example) should rate as roughly
equal (very low) in terms of uncertainty of outcome.
Forrest and Simmons (2002) also use bookmakers’ odds to measure match-level
uncertainty of outcome. Evidence is reported that bookmakers tend to offer more
generous odds for bets on wins by more heavily supported clubs (see also Chapter
12, Section 12.1). Accordingly, the uncertainty of outcome measure is based on
probabilities adjusted for this bias. For English T2, T3 and T4 matches in the 1998
season, there is some evidence that match attendances are positively related to an
uncertainty of outcome measure defined as the ratio of the (adjusted) home win
probability to the away win probability.
Benz, Brandes and Franck (2009) use quantile regression to investigate the
impact of uncertainty of match outcome on spectator demand for match attend-
ance in the German Bundesliga for the period 2001–4. Quantile regression allows
for variation in the factors influencing demand across different quantiles of the
distribution of demand. Consistent with Czarnitzki and Stadtmann (2002), match
uncertainty of outcome seems to be of secondary importance in influencing
attendance demand: the teams’ reputations (measured using performance indica-
tors from previous seasons) and current league standings are more important. The
impact of match uncertainty of outcome is strongest in the higher quantiles, in
matches for which the level of spectator demand is already relatively high.
Shifting the focus from uncertainty of match outcome towards the importance
of the match for end-of-season championship outcomes, Jennett (1984) constructs
dummy variables to indicate the importance of each match for the home and away
teams. While it remains mathematically possible for a team to achieve the points
total observed (after the event) to be required to win the championship, the dummy
is set equal to the reciprocal of the number of games that remain to be played. The
dummy therefore increases progressively for teams that remain in contention as
the prospect of winning the championship draws near, reflecting the heightened
importance of the final few matches for such teams. For teams that drop out of
contention after reaching a point from which the points required are no longer
attainable, the dummy is reset to zero for the season’s remaining matches. In the
estimations, these dummy variables are highly significant for both the home and
away teams, although similarly designed dummy variables for teams threatened
with relegation are insignificant.
Jennett’s approach has been criticised (Cairns, 1987; Peel and Thomas, 1988) on
the grounds that its computation requires knowledge of the ultimate champion-
ship winning points total. Of course, such knowledge is unavailable to spectators
at the time they take their decisions whether or not to attend. The Jennett measure
does nevertheless succeed in distinguishing between important and unimportant
matches. Provided the purpose is to model attendances retrospectively, rather than
forecast them prospectively, Jennett’s measure seems fit for purpose.4
For other sports, Borland (1987) examines uncertainty of outcome effects in an
annual attendance model for Australian rules football over the period 1950–87.
Uncertainty of championship outcome in each season is captured by various
measures of the dispersion of points across teams within the league at various
points in each season. Some limited evidence is found that annual attendances are
sensitive to uncertainty of championship outcome. The extent of long-run dom-
inance is reflected in the number of different teams that had qualified for places
in the end-of-season finals over the previous three seasons. There is no empir-
ical evidence, however, that long-run dominance, as measured, has any effect on
attendance.
Humphreys (2002) estimates a time-series regression for the aggregate attend-
ance for MLB over the period 1901–99, using a CBR (competitive balance ratio)
measure that takes account of both the time-series variation in each team’s win
percent calculated over a number of seasons, and the cross-sectional variation in
the teams’ win percents within each season (see Chapter 3, Section 3.1). A positive
association is found between the CBR and the aggregate attendance. This suggests
spectators value greater uncertainty of outcome, either in the form of high ‘churn-
ing’ in relative performance from season to season, or in the form of a closely bal-
anced league competition.
Live TV broadcasts, stadium attendances and TV audiences

Baimbridge, Cameron and Dawson (1996) investigate the effects of live television
broadcasts on English football league match attendances. 60 of the 462 English
T1 matches played during the 1994 season were televised live by BSkyB, mostly
on Sunday afternoons or Monday evenings. In addition, regular highlights of two
Saturday matches were shown on the BBC’s weekly Match of the Day programme.
Only the live transmissions on Monday evenings had a negative effect, of about
15 per cent on average, on attendance. The payment of about £73,000 received
by the home club for each live transmission was therefore more than adequate to
compensate for the average loss of gate income, estimated at about £34,000, for
Monday evening transmissions.
Using English data for T1 and T2 from seasons 1993 to 1998, Forrest, Simmons
and Szymanski (2004) report similar evidence that the negative effect of live TV
transmission on stadium attendance is rather small, and that any losses of gate and
stadium revenue incurred by the clubs are smaller in magnitude than the payments
they receive from the TV companies. Allan and Roy (2008) find that home sup-
porters who purchase individual tickets are less likely to attend Scottish Premier
League matches that are broadcast live and free-to-air by the BBC; but the attend-
ance of season ticket holders and away supporters is unaffected.
Using data on T2, T3 and T4 in England for the 2000, 2001 and 2002 seasons,
Forrest and Simmons (2006) find that attendances are reduced when midweek
matches are scheduled at the same time as a televised Champions League match,
and when a midweek match is played either immediately before or immediately
after (within three or four days of) a weekend fixture in which the home team was
the same in both fixtures. The negative effects on attendance are strongest in T3
and T4, whose clubs sell proportionately fewer season tickets and are more heavily
reliant on purchases of tickets for individual matches. Champions League matches
screened simultaneously on ITV’s free-to-air channel had a larger negative effect
on league attendances than matches screened on the ITV Digital pay-TV channel.
Buraimo, Forrest and Simmons (2009) report further evidence on the effect of
televised football on stadium attendances of matches played simultaneously.
Recently, several studies have shifted the emphasis away from the impact of
TV scheduling on stadium attendance, and towards the determinants of the size
of the TV audience itself. Forrest, Simmons and Buraimo (2005) report a probit
regression for the selection of matches for live broadcast by BSkyB, the UK’s lead-
ing pay-TV provider, between 1993 and 2002; and an OLS regression for the TV
audience size for those matches that were broadcast live. The selection of matches
favours local derbies, high-quality matches (with quality measured using the previ-
ous season’s combined wage expenditure of the two teams) and matches between
evenly balanced teams. The local derby effect on the TV audience is not statistic-
ally significant, but the TV audience is responsive to the quality of the match and
to uncertainty of match outcome. During the second half of the season, the size
of the TV audience is also responsive to matches involving teams currently occu-
pying the top two places in T1.
Buraimo (2008) analyses the determinants of stadium attendance and the size
of the TV audience jointly, for English T2 matches that were televised live between
seasons 1998 and 2004. The stadium attendance is reduced by the decision to
screen the match on TV, but the size of the TV audience is positively related to the
stadium attendance. This suggests that a match that attracts a full-capacity sta-
dium attendance offers a superior spectacle to the TV audience.
Alavy et al. (2006) use minute-by-minute TV audience ratings data to exam-
ine audience fluctuations over the course of televised Premier League matches.
TV audiences appear to value uncertainty of outcome: the audience tends to
decline when the difference between the win probabilities of the two teams is large.
However, TV audiences do not prefer draws: there is a tendency for viewers to
cease watching matches that remain scoreless for a long period, or matches for
which the probability of a draw increases as the match progresses. Audience sizes
are also affected by channel-hopping, and are prone to increase or decrease by
more on the half-hour and on the hour than at other times.
Other influences on attendance

Other variables to have appeared as covariates in some attendance demand studies
include time trends or dummy variables to allow for variations in average attend-
ance through the course of the season, or at the start and end of each season,
or between weekdays, weekends and Bank Holidays. Bird (1982), Cairns (1987)
and Baimbridge, Cameron and Dawson (1996) investigate the effects of the wea-
ther on attendance, but in general do not find evidence of any effect. Garcia and
Rodriguez (2002) find that rain depresses Spanish La Liga attendences, perhaps
because many Spanish stadia are open to the elements. Other issues to have been
investigated include: the effect on attendance of league restructuring; the effect
of team racial composition or the local demographic structure; ‘superstar’ effects
on attendance; and the effect on domestic club football attendances of national
teams’ successes in the international World Cup tournament.
Cairns (1987) investigates the effects of the change from a two- to a three-tier
league structure in Scotland from the 1976 season onwards, using dummy variables
to allow for changes in the coefficients of an attendance model after restructuring
took place. In a smaller top tier, a higher proportion of matches are significant
in influencing championship outcomes. The coefficients on dummies for matches
against Scotland’s two largest-market teams (Celtic and Rangers) became smaller
after the change, perhaps because in the new ten-club top tier teams met four
times rather than twice per season, reducing the importance of these matches for
supporters of other teams. Overall the downward trend in attendance became less
pronounced after the change.
Baranzini, Ramirez and Weber (2008) report a match-level attendance model
estimated using data for Switzerland for the period 2001–4, which includes a
change in league structure introduced in 2003. Until the 2003 season, after twen-
ty-two matches the twelve teams were split into two groups, with the top eight
playing fourteen further matches to determine the championship (with league
points from the first stage carrying a one-half weighting), and the bottom four
from the top division playing the top four from the second division in fourteen
matches to determine promotion and relegation. Since 2003, ten teams have con-
tested the Super League over thirty-six matches, with the bottom team relegated
and the second-bottom team entering a play-off against the second-placed team
from the second tier. The restructuring coincided with an increase in total attend-
ance of around 16 per cent, suggesting that simplification of the rules may have
had a positive impact on spectator demand.5
Following the launch of MLS in the US, marketing and promotion activity
targeted Hispanics, in the hope that the popularity of football in Mexico and else-
where in Latin America would help create a fan base for MLS. In an empirical
analysis of season attendance for individual teams during the first six seasons of
MLS, however, Jewell and Molina (2005) find an inverse relationship between the
proportion of Hispanics in the city population, and match attendances.6
Brandes, Franck and Nuesch (2008) investigate the impact on German
Bundesliga attendances of the presence of ‘superstar’ players in the home and
away teams. Superstars are defined as players above the upper 2 per cent quantile
by market value for the league as a whole, based on market values obtained from
listings published in the German football magazine Kicker (see also Chapter 7,
Section 7.4). Superstars are found to increase match attendances when playing
both at home and away.
Finally, Falter, Perignon and Vercruysse (2008) examine the impact of victory in
the World Cup on the demand for attendance at club football matches in the win-
ning country, focusing primarily on France’s 1998 World Cup success. Comparing
total league attendances during the two years before and the two years after World
Cup victories by European countries between 1966 and 1998, a positive effect is
observed in every case. The percentage increases in total attendance were 17.3 per
cent (England, 1966), 15.8 per cent (West Germany, 1974), 16.8 per cent (Italy,
1982), 14.5 per cent (West Germany, 1990) and 35.1 per cent (France, 1998). An
econometric analysis of the determinants of match-level attendances in France
shows that the positive impact on attendances was greatest for home teams based
in the stadia that were used for the 1998 World Cup.
11.2 Modelling the demand for attendance at English league football, 1947–1997
The remaining sections of this chapter describe the estimation of a model that
seeks to explain variations in club-level season average attendances at English
league matches during the post-Second World War period, from the 1947 to the
1997 seasons inclusive. Apart from a few minor changes, the model specification
and estimation procedure are the same as in Dobson and Goddard (1995), in
which the model is estimated using data to the end of the 1992 season.7
There are two stages to the empirical analysis. The first stage, reported in Section
11.2, involves the estimation of a model of the demand for attendance, using
Modelling the demand for attendance 337
pooled cross-section time-series data. The model’s coefficients reflect the influence
of four factors upon demand: loyalty in the short term (or persistence in attend-
ance from year to year); success (measured by league position); admission price;
and entertainment (proxied by goals scored). The model also produces rankings
of all clubs according to their estimated base levels of attendance: the attendances
they would expect to achieve after controlling for the short-term effects of loy-
alty, success, price and entertainment. The second stage, reported in Section 11.3,
examines the cross-sectional variation between clubs in their base levels of attend-
ance, and in their short-term loyalty, success, price and entertainment coefficients
obtained at the first stage. The explanatory variables include socioeconomic and
demographic characteristics of each club’s home town (population, occupational
structure and unemployment) and football-related characteristics (the club’s age
and the number of other clubs located in close geographical proximity).
At the first stage of the empirical analysis, we assume the average recorded
attendance for each club in each season is a true measure of demand, and we there-
fore ignore the impact of stadium capacity constraints on recorded attendances
(see Section 11.1). This assumption is reasonable for most of the period under
consideration, although inevitably it does lead to underestimation of demand in
some cases. The short-term loyalty effect is incorporated by using the previous
season’s average attendance as an explanatory variable in the regression model for
current attendance. The coefficient on this lagged dependent variable measures the
strength of persistence of attendance from one season to the next, and is therefore
interpreted as a measure of short-term loyalty.
Playing success is measured simply, by awarding points based on each team’s
finishing position in the league, with 92 points for the team finishing first in T1,
91 points for the team finishing second and so on. Preliminary inspection of the
data set indicated that the relationship between league position and attendance
is kinked towards the top end of each division. Other things being equal, a club
finishing in one of the top few positions in T2, T3 and T4 would expect to achieve
a higher attendance than if it finished at the bottom of the tier above. Three add-
itional dummy variables are therefore included as explanatory variables to allow
for this non-linear effect. Preliminary experimentation with the data also showed
that promotion or relegation at the end of the previous season’s campaign tend
to influence the current season’s attendance, to an extent that cannot be captured
fully either by the lagged dependent variable, or by the change in the team’s league
position variable. Accordingly, promotion and relegation dummy variables are
included.
From elementary microeconomic theory, the influence of admission price on
attendance is expected to be negative. Although admission represents only one
component of the total cost of attending a match, which may also include trans-
port, parking, refreshments and the spectator’s time costs, data that would permit
the measurement of any of the latter by club are not available. Lack of data also
precludes investigation of the effects of year-on-year variation in other possible
economic or demographic determinants of club-level attendances, such as popu-

lation or unemployment. Census of Population data are used, however, at the
second stage of the empirical analysis (see Section 11.3).
Finally, the total number of goals scored by each team (in home and away
matches) is used as a measure of the team’s entertainment value. Inclusion of
goals scored as an explanatory variable is a crude, but expedient way of captur-
ing a feature that does seem likely to influence demand, but is otherwise diffi-
cult or impossible to quantify. It is subject to several possible objections, however.
First, goal scoring does not necessarily reflect playing style, which may be a more
important determinant of whether a particular team is regarded as entertain-
ing. Second, goals scored at home, or goals scored and conceded at home, might
be a more important determinant of home attendances than goals scored in all
matches. Third, goal scoring is naturally correlated with league position. This cor-
relation, however, is by no means perfect, and on this basis we assume that goal
scoring reflects a separate and distinct attribute of each team.
As the main objective is to investigate and explain the distribution of attend-
ances between clubs, the effects of various trend or cyclical components in the
data that are common to all clubs are eliminated, by standardising the attend-
ance, admission price and goals series. The standardisation involves subtracting
from each observation on each variable the cross-sectional mean for the relevant
year, and dividing by the standard deviation. Because the model is intended to
show how a club’s attendance changes as its position changes throughout the
entire league, attendances are standardised using whole-league annual means and
standard deviations. Since average admission prices and (to a lesser extent) goal-
scoring records vary systematically between different tiers, however, clubs are
identified as cheap or expensive, entertaining or boring, by comparison with tier
norms. The admission price and goals scored variables are therefore standardised
using the annual means and standard deviations for each tier. Preliminary inspec-
tion of the data suggested that a semi-logarithmic functional form best describes
the relationship between attendance and the other variables. The attendance ser-
ies is therefore expressed in natural logarithmic form prior to standardisation.
Seventy-seven clubs were members of the league continuously during the fifty-one
seasons between 1947 and 1997 (inclusive). A further nineteen clubs that were league
members for at least eighteen of these seasons are also included in the data set. The
inclusion of lagged variables in the model necessitated the omission from the estima-
tions of the 1947 season (the first post-Second World War season) and, for clubs that
entered the league for the first time or re-entered, their first seasons of membership.
Initially the attendance model is estimated using fixed-effects estimation, with
a common set of slope coefficients denoted β1 to β10, reflecting the influences on
attendance of short-term loyalty, success, admission price, goals scored and the
dummy variables, and an individual (fixed) effect for each club, denoted αi and
referred to below as each club’s alpha coefficient. The procedure used to estimate
the alpha coefficients relies on the assumption that the slope coefficients (β1 to β10)
are the same for all clubs.
The full specification of the model is as follows:
a i,t = α i + β1a i,t-1 + β2 li,t + β3 li,t-1 + β 4 p i,t + β5 g i,t

+ β6 D1,i,t + β7 D2,i,t + β8 D3,i,t + β9 DPi,t-1 + β10 DRi,t-1 + ui,t [11.1]
The variable definitions for [11.1] are as follows:
ai,t = (ln(Ai,t) – μAt) / σAt

pi,t = (Pi,t – μPk,t) / σPk,t
gi,t = (Gi,t – μGk,t) / σGk,t
ln(Ai,t) = natural logarithm of team i’s average attendance in season t.
li,t = league position (on a scale of 92 for first place, 91 for second, and so on).
Pi,t = average admission price (= gate revenue ÷ attendance).
Gi,t = goals scored.
D1,i,t = 1 if team i finished in first place in its tier in season t, and 0 otherwise.
D2,i,t, D3,i,t are defined similarly for second and third place finishes.
DPi,t–1 = 1 if team i was promoted or elected to the league at the end of season t–1,
and 0 otherwise.
DRi,t–1 = 1 if team i was relegated at the end of season t–1, and 0 otherwise.
μAt,σAt = mean and standard deviation of ln(Ai,t) across all teams in league in
season t.
μPk,t,σPk,t = mean and standard deviation of Pi,t across all teams in k’th tier in
season t.
μGk,t,σGk,t = mean and standard deviation of Gi,t across all teams in k’th tier in
season t.
ui,t = error term.
The inclusion in [11.1] of the lagged terms in ai,t–1 and li,t–1, although required in
order to capture the dynamics of the short-term interactions between league pos-
ition and attendance, makes [11.1] difficult to interpret if the main interest is in the
long-term equilibrium or steady-state relationship between variables such as league
position, price and goals scored, and attendance. A corresponding steady-state
model can be derived by setting current and lagged values of each variable equal to
their steady-state values, and rearranging. The steady-state variables are denoted
a = a = … = a s ; l = l = … = ls ; p = p =… = ps ; g = g = … = g s
i,t i,t-1 i i,t i,t-1 i i,t i,t-1 i it it-1 i
Setting all dummy variables to zero for simplicity, the implied steady-state equa-
tion is
a si = γ i + δ1lsi + δ 2 psi + δ3 g si [11.2]
where γi = αi/(1–β1); δ1 = (β2+β3)/(1–β1); δ2 = β4/(1–β1); and δ3 = β5/(1–β1)

The estimates of αi, the club-specific alpha coefficients, are shown in column (3)
of Table 11.2,8 where the clubs are ranked in descending order of their estimated
alpha coefficients. The estimates of β1 to β10, together with the implied estimates
of δ1, δ2 and δ3, are shown below as equations [11.3] and [11.4]. z-statistics for the
Table 11.2 Attendance model: first-stage estimation results
Average Average
attendance position α̂i βˆ 1i δˆ 1i δˆ 2i δˆ 3i
Rank Team (1) (2) (3) (4) (5) (6) (7)
1 Manchester United 44,775 6.0 0.113 0.926*** −0.042 −0.910 0.211

2 Tottenham Hotspur 36,090 10.1 0.075 0.646*** 0.013 −0.166* 0.088
3 Everton 36,082 10.9 0.075 0.697*** 0.009 −0.044 −0.138
4 Arsenal 36,493 7.0 0.075 0.623*** 0.016 −0.051 0.107
5 Newcastle United 32,995 16.8 0.057 0.683*** −0.005 −0.056 0.232*
6 Liverpool 40,289 7.8 0.043 0.646*** 0.003 0.080 0.042
7 Manchester City 30,811 15.1 0.019 0.620*** 0.012 −0.212*** 0.004
8 Chelsea 30,271 15.2 0.007 0.354** 0.015*** −0.068*** −0.048
9 Sunderland 27,751 23.1 0.004 0.659*** 0.016* 0.117 0.145
10 Aston Villa 29,361 15.7 0.001 0.768*** 0.007 0.064 −0.204
11 Leeds United 27,226 17.1 −0.059 0.834*** −0.009 −0.155 0.207
12 Sheffield Wednesday 24,842 22.5 −0.062 0.692*** 0.011 −0.081 0.152
13 West Ham United 23,972 19.3 −0.066 0.846*** 0.036*** −0.071 0.027
14 Wolverhampton Wdrs 25,777 19.2 −0.103 0.557*** 0.007 −0.059 0.150*
15 Leicester City 21,479 21.6 −0.124 0.603*** 0.009 0.000 0.031
16 Birmingham City 22,112 25.6 −0.127 0.626*** 0.020** 0.096 0.238**
17 West Bromwich Albion 21,796 19.3 −0.136 0.528*** 0.010*** −0.030 0.068
18 Nottingham Forest 22,464 17.9 −0.136 0.589*** 0.027*** 0.041 0.043
19 Sheffield United 20,491 25.9 −0.144 0.555*** 0.015*** 0.018 0.134*
20 Coventry City 19,034 28.5 −0.149 0.496*** 0.019*** 0.090 −0.020
21 Norwich City 18,539 32.9 −0.151 0.641*** 0.008*** −0.300*** −0.018
22 Crystal Palace 16,337 43.4 −0.151 0.739*** 0.015*** −0.007 0.231
23 Middlesbrough 20,092 25.3 −0.157 0.575*** 0.024*** −0.058 0.157
24 Derby County 19,640 24.3 −0.172 0.876*** −0.001 −0.091 0.443
25 Southampton 18,220 24.2 −0.179 0.614*** 0.017* −0.027 0.118
26 Stoke City 19,230 23.5 −0.182 0.412*** 0.017*** −0.047 0.102
27 Portsmouth 17,499 33.0 −0.186 0.533*** 0.013** 0.091 0.189
28 Bristol City 14,689 44.5 −0.188 0.571*** 0.013*** −0.085 0.041
29 Brighton and Hove Albion 13,367 49.0 −0.209 0.533*** 0.019*** 0.056 0.024
30 Ipswich Town 17,028 26.7 −0.250 0.654*** 0.028*** −0.295*** −0.072
31 Burnley 16,667 30.6 −0.259 0.723*** 0.006 −0.205 0.043
32 Bolton Wdrs 17,063 30.1 −0.259 0.567*** 0.025*** 0.107 0.085
33 Preston North End 14,686 40.1 −0.272 0.627*** 0.013** −0.176 0.280**
34 Plymouth Argyle 12,500 47.4 −0.275 0.605*** 0.021** −0.086 0.064
35 Fulham 14,895 37.5 −0.278 0.581*** 0.019*** −0.211** 0.130*
36 Blackburn Rovers 16,068 27.2 −0.281 0.698*** 0.022*** 0.275* 0.187
37 Queens Park Rangers 14,240 32.8 −0.298 0.600*** 0.027*** −0.079 0.046
38 Swindon Town 10,656 53.8 −0.307 0.501*** 0.020*** −0.120* 0.134*
39 Watford 11,015 49.7 −0.312 0.738*** 0.018*** −0.060 0.152
40 Huddersfield Town 13,108 38.7 −0.326 0.017 0.025*** 0.077 0.217***
41 Millwall 11,495 47.8 −0.327 0.565*** 0.009** −0.174*** 0.016
42 Cardiff City 14,454 41.6 −0.329 0.604*** 0.028*** −0.613** 0.043
43 Charlton Athletic 14,689 30.3 −0.331 0.543*** 0.028*** −0.156*** −0.030
44 Hull City 12,549 46.5 −0.344 0.697*** 0.010 0.060 −0.023
45 Bristol Rovers 11,044 48.0 −0.352 0.859*** 0.016 −0.449 0.744**
46 Brentford 9,897 57.4 −0.353 0.503*** 0.015*** −0.058 −0.035
47 Blackpool 13,613 34.5 −0.356 0.413*** 0.022*** 0.032 −0.108*
48 Notts County 11,507 47.6 −0.364 0.741*** 0.007 −0.242 0.385*
49 Luton Town 12,382 32.0 −0.365 0.458*** 0.016*** −0.071** 0.013
50 Oldham Athletic 9,574 50.8 −0.391 0.512*** 0.011** −0.218* 0.188*
51 Swansea City 10,212 52.2 −0.400 0.676*** 0.025*** −0.188 0.199
52 Bradford City 7,475 64.1 −0.406 0.257 0.029*** −0.068 0.060
53 Peterborough United 6,215 66.1 −0.413 0.315* 0.018*** 0.054 0.054
54 Bournemouth 7,848 59.9 −0.421 0.528*** 0.012** −0.053 0.311***
55 Reading 8,200 57.6 −0.421 0.437*** 0.020*** 0.058 −0.012
56 Barnsley 9,560 49.2 −0.424 0.585*** 0.010 0.272 0.366**
57 Grimsby Town 8,881 50.6 −0.434 0.569*** 0.011*** 0.017 0.175**
Table 11.2 (cont.)
Average Average
attendance position α̂i βˆ 1i δˆ 1i δˆ 2i δˆ 3i
Rank Team (1) (2) (3) (4) (5) (6) (7)
58 Northampton Town 6,969 69.8 −0.437 0.455*** 0.015** −0.094 −0.011
59 Oxford United 7,521 45.3 −0.444 0.612*** 0.015*** 0.017 0.103
60 Southend United 7,187 61.1 −0.449 0.498*** 0.006 −0.181*** 0.145*
61 Port Vale 7,926 59.7 −0.453 0.517*** 0.022*** 0.113 −0.070
62 Leyton Orient 8,400 53.5 −0.460 0.533*** 0.013*** −0.121 0.086
63 Gillingham 6,298 67.1 −0.472 0.568*** 0.011* 0.024 0.109
64 Walsall 6,975 63.3 −0.473 0.497*** 0.011* −0.169** 0.055
65 Bradford Park Avenue 8,851 69.9 −0.476 0.206 0.019*** 0.048 0.000*
66 Rotherham United 8,414 49.6 −0.495 0.407* 0.023*** −0.162 −0.047
67 Lincoln City 6,770 65.5 −0.501 0.447*** 0.013*** 0.072 0.134*
68 Wrexham 6,572 65.5 −0.505 0.658*** 0.015 −0.210* 0.144
69 Chesterfield 6,658 63.1 −0.507 0.400*** 0.013*** −0.105 0.179*
70 Carlisle United 7,211 58.5 −0.507 0.463*** 0.018*** 0.086 0.029
71 Exeter City 5,391 75.1 −0.517 0.285* 0.007* −0.083 0.071
72 Doncaster Rovers 7,165 68.2 −0.525 0.464*** 0.024*** 0.198 0.233
73 Mansfield Town 6,194 66.7 −0.531 0.449*** 0.013** −0.061 0.124
74 Tranmere Rovers 6,188 64.4 −0.535 0.783*** 0.009 −0.115 0.277
75 Stockport County 5,580 72.4 −0.540 0.367*** 0.012** −0.188*** 0.122
76 Wimbledon 6,608 31.7 −0.553 0.830*** 0.075 −0.994 0.918
77 Hereford United 3,832 77.5 −0.553 0.328*** 0.022*** 0.196*** 0.046
78 Bury 7,276 54.9 −0.555 0.582*** 0.021*** 0.145** 0.027
79 York City 5,340 70.2 −0.566 0.308** 0.014*** −0.025 0.154**
80 Aldershot 4,638 77.8 −0.580 0.711*** 0.008 −0.061 0.259*
81 Colchester United 4,917 70.6 −0.588 0.463*** −0.006 −0.084 0.131
82 Scunthorpe United 5,024 68.9 −0.589 0.328** 0.015*** 0.114* 0.013
83 Cambridge United 4,190 61.2 −0.590 0.228 0.014*** −0.047 0.041
84 Shrewsbury Town 5,599 57.4 −0.599 0.610*** 0.014*** 0.050 −0.009
85 Newport County 5,435 73.0 −0.607 0.638*** 0.004 −0.173 0.238
86 Wigan Athletic 3,713 65.5 −0.614 0.753*** −0.028 −0.849 0.530
87 Torquay United 4,691 73.2 −0.617 0.604*** 0.022*** −0.093 0.100
88 Chester City 4,481 74.3 −0.621 0.651*** −0.002 −0.173 0.331**
89 Crewe Alexandra 4,292 76.5 −0.628 0.611*** 0.006 −0.090 0.132
90 Hartlepool United 4,283 79.5 −0.638 0.379** 0.019* 0.027 0.111
91 Darlington 4,022 80.0 −0.655 0.169 0.019*** −0.059 0.060
92 Halifax Town 4,010 76.9 −0.714 0.346** 0.011*** 0.004 0.030
93 Barrow 5,099 76.1 −0.729 0.184 0.007 −0.114** 0.099
94 Rochdale 3,665 77.5 −0.742 0.444*** 0.010 0.026 0.095
95 Southport 4,169 76.3 −0.801 0.164 0.001 0.115 0.212***
96 Workington Town 3,970 77.7 −0.845 0.564** 0.004 0.829*** −0.264
Note: *** = significantly different from zero, two-tail test, 1% level; ** = 5% level; * = 10% level.
Statistically significant α̂i’s are not indicated.
significance of the estimated coefficients are shown beneath in italics. ei,t is the
estimated error term or residual.
a i,t = α i + 0.5125***a i,t-1 + 0.0153***li,t − 0.0071*** li,t-1 − 0.0127***p i,t + 0.00024g i,t
38.5 38.3 14.2 3.02 1.20
+ 0.2676***D1,i,t + 0.1888***D2,i,t + 0.1356***D3,i,t
17.6 12.5 9.10
+ 0.0581***DPi,t-1 − 0.0285**DRi,t-1 + e i,t
4.88 2.50
no.of observations = 4,420 R2 = 0.95 [11.3]
a si = γ i + 0.0168*** lsi − 0.0260***psi + 0.00049 ***g si

[11.4]
33.6 2.76 2.04
** = 5% level;
* = 10% level.
z-statistics for significance of estimated coefficients are shown in italics.
Since the attendances are expressed in natural logarithms, and all variables are
standardised, the numerical values of the estimated alpha coefficients reported in
column (3) of Table 11.2 do not have any direct interpretation. As explained above,
however, the rankings of the clubs by their alpha coefficients can be interpreted as
corresponding to rankings by each club’s ‘base’ level of attendance, after control-
ling for the effects of short-term loyalty, league position, price and goals scored.
The top placing of Manchester United in Table 11.2 comes as no surprise. In
second to fourth places, the alpha coefficients of Tottenham Hotspur, Everton and
Arsenal are all virtually identical, and there is little to separate these three clubs
in terms of base attendance. The comparison between Everton and Liverpool is
instructive in illustrating how the alpha coefficient works. Although Liverpool’s
average post-war attendance (40,289) is higher than Everton’s (36,082), the differ-
ence is offset by Liverpool’s average finishing position of 7.8, about three places
higher than Everton’s of 10.9. After adjusting for the differences in the league pos-
ition variable (and in all other variables), Everton (ranked fourth) have a higher
alpha coefficient and therefore a higher base attendance than Liverpoool (ranked
sixth). Similarly, the high placings of clubs like Newcastle United (fifth), Manchester
City (seventh) and Sunderland (ninth) reflect the high levels of support these clubs
have maintained, despite enduring long barren periods starved of success.
Towards the top of the list, it is noticeable that all of the twenty highest-ranked
clubs are from cities with large populations (above 250,000 in the 1981 Census).
Only two of the top twenty (West Ham United and Coventry City) entered the
league after the First World War (in both cases in the 1920 season). The other
eighteen clubs were all pre-1914 entrants, and four of the top twenty (Everton,
Aston Villa, Wolverhampton Wanderers and West Bromwich Albion) were among
the league’s twelve founder-members in 1888. The highest-ranked clubs among

the twenty-two original members of Division 3 (South) in the 1921 season are
Norwich City (twenty-first) and Crystal Palace (twenty-second) while Wrexham
(sixty-eighth) are the highest placed of the twenty original Division 3 (North)
clubs admitted in the 1922 season.
Of the nine clubs included in the analysis that were admitted to the league for
the first time after the Second World War, seven can be found in the bottom twen-
ty-one places. Of the seven clubs that had left the league (and were not readmitted)
by the end of the 1997 season, three (Workington, Southport and Barrow) occupy
places in the bottom four. The alpha coefficients indicate that while clubs with low
base attendances (unsurprisingly) find it hard to survive, their replacements also
experience difficulty in building up their attendances after entering the league.
Early (pre-1914) entry to the league, on the other hand, together with a big-city
location, seems to have conferred significant first-mover advantages, allowing the
clubs concerned to achieve high base levels of attendance.
The estimates of the coefficients β1 to β10 reported in [11.3] all have the expected
signs. All are significantly different from zero at the 5 per cent level, except the
estimate of β5 (the coefficient on gi,t, goals scored), which is insignificant at any
conventional significance level. The estimates of β1 (the short-term loyalty coef-
ficient on ai,t–1, lagged attendance) and β2 and β3 (the coefficients on li,t and li,t–1,
currrent and lagged league position) are the strongest determinants of attendance
according to the t-statistics.
In the steady-state equation [11.4], the estimates of δ1 (the league position
coefficient) and δ2 (price) are significantly different from zero at the 1 per cent
level, and the estimate of δ3 (goals scored) is significantly different from zero at the
5 per cent level. Using data for the 1997 season for T1 clubs, [11.4] implies that an
improvement of one place in a club’s final league position would add 1.68 per cent
to its average attendance. A £1 increase in admission price would lead to a loss of
attendance of 0.81 per cent, implying a price elasticity of –0.114 for a club charg-
ing the divisional average admission price. These estimates are similar to those for
the period 1926–92 reported in Dobson and Goddard (1995).
It is also possible to estimate a set of club-specific attendance equations, in
which the slope coefficients (β1 to β10) are allowed to vary between clubs in the
same way as the intercept coefficient (αi) in [11.1] and [11.3].9 The model specifica-
tion is the same as [11.1], but with i-subscripts attached to each slope coefficient to
indicate that these coefficients are specific to each club. The steady-state model is
obtained using the same procedure as before.
Columns (4) to (7) of Table 11.2 report, for each club, the estimates of the coef-
ficients β1i (the short-term loyalty coefficient), δ1i (the steady-state league position
coefficient), δ2i (the steady-state price coefficient) and δ3i (the steady-state goals
scored coefficient).10 There is considerable variation between clubs in these coef-
ficient estimates, as would be expected given the relatively small number of time-
series observations for each club. The estimates of β1i are signed as expected in
all ninety-six cases, and are significantly different from zero at the 5 per cent level
on a two-tail test in eighty-six cases. The estimates of δ1i are signed as expected
in eighty-nine out of ninety-six cases, and significant in fifty-eight of these cases.
Price and goals scored are found to exert a less powerful effect on attendance, as
before. The estimates of δ2i and δ3i are signed as expected in fifty-nine and eighty
cases respectively. The numbers of these coefficients that are significant at the 5
per cent level are thirteen and ten respectively.
11.3 Explaining base attendances, and the loyalty, league position, price and goals
scored coefficients
The second stage of the empirical analysis investigates whether there are observ-
able characteristics of the clubs or of their home towns or cities that are important
in determining the estimated values of αi (the alpha coefficients), β1i (the short-
term loyalty coefficients), δ1i (the steady-state league position coefficients), δ2i (the
steady-state price coefficients) and δ3i (the steady-state goals scored coefficients)
reported in Table 11.2. At the second stage, estimations are carried out of five
cross-sectional regressions, in which the estimated coefficients listed above are
used as dependent variables, to be explained by a set of variables which measure
club or home-town characteristics. The full list of explanatory variables used in
the second-stage regressions is as follows:
Hi = natural logarithm of the population of the local authority district in which

club i is located, recorded in the 1981 Census of Population;
Ci = the number of other clubs that were members of the league for at least eight-
een seasons during the post-Second World War period, located within a 30 mile
radius of club i’s stadium;
Xi = the natural logarithm of the number of years since club i first entered the
league, measured at the end of the observation period in 1997;
Fi = percentage of employed residents working in agriculture in club i’s local
authority district, 1981 Census of Population;
Ii = percentage of employed residents working in energy and water, manufac-
turing and construction in club i’s local authority district, 1981 Census of
Population;
Ui = Males aged 16–64 out of employment as a percentage of those classed as eco-
nomically active in club i’s local authority district, 1981 Census of Population.
Where several clubs are located in the same town, all have the same Hi, Fi, Ii and
Ui based on the town’s census data. Inevitably, the use of data from a single cen-
sus introduces a large element of imprecision. There appears to be no alterna-
tive, however, since town-specific data are unavailable (even at ten-year intervals)
on a consistent definitional basis. In any event, the intention in the second-
stage estimations is to use variables that capture broadly defined demographic
and socioeconomic characteristics of the towns in which clubs are based. Even
Explaining base attendances 347
Table 11.3 Attendance model: second-stage estimation results
Dependent variable & interpretation

α̂i βˆ 1i δˆ 1i δˆ 2i δˆ 3i
Base Short-term League Price Goals
Independent
attendance loyalty position scored
variables &
interpretation (1) (2) (3) (4) (5)
Constant −2.4978*** −0.1791 0.01358 0.3460 0.3341

6.66 −0.47 0.88 1.43 1.56
Hi 0.0748*** 0.0168 0.00019 −0.0150 −0.0268*
Home-town 4.09 0.84 0.19 −1.02 −1.95
population
Xi 0.2841*** 0.1429** −0.00012 −0.0773* 0.0393
Duration of league 4.35 2.10 –0.06 1.74 1.47
membership
Ci −0.0095** −0.0008 0.00027 0.0037 −0.0002
Local −2.26 −0.16 1.13 0.95 −0.06
competition
Fi −0.0160 −0.0035 0.0001 0.0055 0.0021
Agricultural −0.81 −0.30 0.16 0.56 0.28
employment
Ii −0.0030 −0.0030 −0.00004 0.0009 −0.0018
Industrial −1.42 −1.15 −0.31 0.69 −1.13
employment
Ui 0.0132*** 0.0042 0.00008 0.0063** −0.0058**
Unemployment 3.00 0.74 0.47 2.25 −2.15
No. of obs. 96 96 96 96 96
R2 0.83 0.94 0.86 0.27 0.19
S.E. of regression 4.94 1.31 1.50 1.26 1.26
Note:
** = 5% level;
* = 10% level.
with a fifty-year observation period, it seems reasonable to assume that for most
towns, such characteristics are sufficiently stable for an exercise of this type to be
meaningful.
Table 11.3 reports the results of weighted least squares (WLS) estimations of
five equations, in which the estimated values of αi, β1i, δ1i, δ2i and δ3i obtained from
the previous section are used in turn as the dependent variable, and Hi, Xi, Ci, Fi,
Ii and Ui are used as explanatory variables. For each equation there are ninety-
six observations on each variable: one for each club included in the estimations
reported in Section 11.2.11
Column (1) of Table 11.3 reports the regression identifying the determinants of
base attendance, using each club’s estimated alpha coefficient, α̂i, as the explana-
tory variable. All of the explanatory variables, with the exception of the industrial
structure variables Fi and Ii, are significantly different from zero at the 5 per cent
level on two-tail tests. As expected, home-town population, Hi, and the number of
years since the club entered the league, Xi, both have positive and highly signifi-
cant estimated coefficients.
Hi is a significant determinant of base attendance, even though the propor-
tion of spectators who live in the immediate vicinity of their club’s stadium has
almost certainly declined over time (Bale, 1993; see also Chapter 6). The strength
of the coefficient on Xi in particular justifies the previous comments concerning
the first-mover advantages enjoyed by clubs that entered the league early. Bearing
in mind the fact that eighty-seven of the ninety-six clubs included in the data set
first entered the league before the Second World War (before the period covered
by the estimations) it seems remarkable that this effect is so strong. Clearly, this
finding testifies to the importance of tradition, and loyalties which persist over
generations, in shaping patterns of demand.
As expected from elementary economic theory, the number of clubs located
within close geographical proximity (a crude measure of the level of competition
facing each club), Ci, has a negative and significant association with base attend-
ance. On the other hand, the rate of unemployment, Ui, has a positive associ-
ation. This finding perhaps runs counter to theoretical expectations. As discussed
in Section 11.1, it might reflect the fact that for various sociological and historical
reasons, many of the most keenly supported clubs are located in cities or regions
that suffer from high structural unemployment.
In contrast to the results for the alpha coefficients, few of the explanatory varia-
bles are significant in the regressions using the short-term loyalty, league position,
price and goals scored coefficients (in turn) as the dependent variable, reported in
columns (2) to (5) of Table 11.3. To some extent this is expected. There are good
reasons to expect factors such as population, age or the strength of local com-
petition to influence the average or base level of a club’s attendance, but there
seems less reason to expect such factors to influence the sensitivity of attendances
to variations in league position, price and so on. Accordingly, we comment only
briefly on these estimations.
The only explanatory variable that is significant in the regression with the
short-term loyalty coefficient, β̂1i, as dependent variable, is the number of years
since the club entered the league, Xi. Clubs that entered early have a higher
short-term loyalty coefficient, a finding which seems to emphasise the import-
ance of this factor in shaping attendance patterns. None of the explanatory
variables is significant in the regression for the estimated steady-state league
position coefficient, δ̂1i. In the regression for the steady-state price coefficient,
δ̂2i, the coefficient on the rate of unemployment, Ui, is positive and significant at
the 5 per cent level, and the coefficient on the duration of league membership,
Xi, is negative and significant at the 10 per cent level. The former suggests that
attendances in towns with high unemployment are more sensitive to variations
in price than those elsewhere. The latter reinforces the notion that the support-
ers of first-movers (early entrants) are more loyal, as their attendance is less
price-sensitive.
Conclusion
Since the 1970s, sports economists have focused considerable attention on model-
ling match-by-match and season-by-season variations in attendances. This empir-
ical literature has been reviewed in some detail in this chapter. Economists are
naturally inclined to interpret attendance data as a measure of demand. Published
attendances, however, are aggregated over various spectator groups, including
season-ticket holders and purchasers of tickets (at various prices) for individual
matches. The nature of the aggregation tends to make the estimation of price
and other elasticities of demand problematic. The possibility of multi-directional
causality between variables such as price, team performance and attendance, and
the fact that published attendances increasingly tend to reflect stadium capacity
constraints, also present difficulties, both in the estimation of attendance equa-
tions, and in their interpretation as demand functions.
Despite definitions being somewhat arbitrary, market size is usually found to
be a significant determinant of match attendances, as is the geographical distance
separating the two teams contesting each match. Models of season attendances
are better equipped to detect price, income and unemployment effects, and a num-
ber of recent studies have quantified a relatively small, but statistically significant,
price elasticity. Some care needs to be exercised, however, in interpreting the results
of regressions estimated using non-stationary or trended time-series data. In any
event, team performance appears to exert a much greater influence on individual
club attendances than price, though there is less consensus among researchers as
to whether genuine uncertainty of match outcome, or simply a high probability
that the home team will win, is the more likely to generate a high match attend-
ance. Unsurprisingly, matches that are relevant for end-of-season championship
outcomes are shown consistently to attract higher attendances.
In the empirical sections of this chapter, a two-stage econometric model of vari-
ation in English annual average club-specific league attendances during the post-
Second World War period is presented. At the first stage, an attendance model
is estimated, which ranks all clubs according to their estimated base attendance,
and examines the effect on attendance of short-term loyalty, team performance,
admission price and goals scored. At the second stage, the cross-sectional variation
between clubs’ base levels of attendance, and their short-term loyalty, perform-
ance, price and entertainment coefficients, is explained in terms of socioeconomic,
demographic and football-related characteristics of each club and its home town.
Home-town population, the amount of competition from other local football
clubs, and the duration of each club’s league membership are all highly significant
determinants of base attendances. Duration of membership is also a significant
determinant of the short-term loyalty coefficient. These results suggest that clubs
that joined the league early continue to enjoy significant first-mover advantages, in
many cases more than a century after they first gained membership. It is evident
that sporting history and tradition remain important factors in shaping patterns
of demand for attendance at English football, even in the modern era.
Notes
1 Indeed, historically there was some flexibility in the capacity constraints themselves, with
spectators simply being packed into the terraces more tightly than usual for the most attract-
ive fixtures: a crowd management philosophy that led ultimately to catastrophe in the form
of the Hillsborough stadium disaster of 1989.
2 In a study of the demand for attendance at Australian rules football using annual data,
Borland (1987) addresses the problem of the endogeneity of the admission price variable,
under the assumption that clubs may adjust their admission prices in the light of the previous
season’s attendance. Current admission price is therefore correlated with whatever ‘random’
factors influenced last season’s attendance. If the effect of such ‘random’ factors tends to
persist over more than one season, current admission price is also correlated with the current
‘random’ element in attendance, creating a problem of simultaneity bias if single-equation
OLS is used to estimate the attendance equation. This issue is addressed by means of instru-
mental variables estimation, requiring the estimation of a separate equation for price,
dependent on attendances lagged by one and two years, real income and player payments.
3 A time series is stationary if its mean and variance are constant with respect to time, and
non-stationary if its mean and variance are changing with respect to time. An untrended ser-
ies can be stationary or non-stationary: a non-stationary untrended series is one that tends to
wander over time. A trended series is, by definition, non-stationary because the trend causes
its mean to change with respect to time. Some trended series, however, can be made station-
ary by de-trending. Other trended series remain non-stationary even after any deterministic
trend component is removed.
4 Data on changes throughout the season in bookmakers’ odds on the championship out-
come might also provide a useful measure, although to our knowledge no researchers have
exploited this possibility as yet.
5 For other sports, Burkitt and Cameron (1992) estimate a season attendance model for English
rugby league clubs over the period 1966–90. The period includes a major restructuring of the
league from a single to a two-division format in 1973. Attendances increased overall, but
while teams that moved into the higher division benefited, teams that moved into the lower
division suffered as a result of the split. Dobson, Goddard and Wilson (2001) use an empir-
ical match attendance model for rugby league to generate simulated match attendances under
various alternative league structures, including larger or smaller divisions, and regionalised
divisions. This approach permits quantification of the trade-offs for attendances between, for
example, less evenly balanced competition versus more local derbies, if divisional member-
ship is determined geographically rather than by playing strength via a system of promotion
and relegation.
6 For other sports, Schollaert and Smith (1987) find that the racial composition of NBA teams
had no discernible effect on attendances between 1969 and 1983. On the other hand, Hill,
Madura and Zuber (1982) find some evidence of discrimination among spectators: MLB
attendances in the 1970s were lower if the home team’s starting pitcher was black or Latin
American. In a season attendance model, Siegfried and Eisenberg (1980) find that the racial
composition of the home-town population of teams in minor league baseball also had a
discernible effect on attendance during the 1970s.
7 The data for seasons 1926 to 1939 inclusive (used in the earlier study) has been dropped
from the estimations reported here. The non-availability of the gate revenues series for
recent seasons has precluded the extension of the analysis beyond the 1997 season.
8 To correct for serial correlation, [11.1] is estimated using Hatanaka’s (1974) two-step pro-
cedure, assuming that the error term ui,t follows a first-order autoregressive process. To cor-
rect for heteroscedasticity, all standard errors and significance tests are based on White’s
(1980) adjustment.
9 The estimates of the coefficient on the lagged dependent variable ai,t-1 in [11.3] are down-
ward biased and inconsistent, due to the presence in the model of the individual club fixed
effects or alpha coefficients; although this problem is mitigated to some extent by the long
time dimension of the data set (Downward and Dawson, 2000). This issue does not affect
the club-specific estimations. The fact that the estimated β1 (equal to 0.5125) in [11.3] is
smaller than sixty-one of the ninety-six estimated β1i’s in Table 11.2 reflects the bias in the
pooled model.
10 With a maximum of fifty time-series observations per club, diagnostic tests show little evi-
dence of serial correlation in the residuals in the club-level estimations. These are carried
out using OLS, with no autoregressive error structure. There is, however, some evidence of
heteroscedasticity, and standard errors and significance tests are based on White’s (1980)
adjustment.
11 Following Saxonhouse (1976), in each equation each observation is weighted by the inverse
of the estimated variance of the dependent variable obtained from the earlier estimations.
Intuitively, the weights reflect the degree of confidence that can be placed in each observa-
tion. The weighting procedure takes account of the fact that the dependent variables are
themselves estimated coefficients obtained from the first-stage estimations, and not (as is
usually the case) hard data.
12 Gambling on football
Introduction
The theoretical and empirical economics literature contains numerous studies of
the informational efficiency of betting markets. Comprehensive reviews of the earl-
ier literature are provided by Sauer (1998) and Vaughan Williams (1999). Research
concerning the relationship between the market prices for bets on the outcomes
of sporting contests, and the probabilities associated with these outcomes, forms
a subfield within the literature on financial market efficiency. Although much of
this literature focuses on racetrack betting, the market for betting on match out-
comes in football has also attracted considerable attention. A divergence between
the market prices and the true probabilities implies a violation of the conditions
for the informational efficiency of the betting market. Any such divergence creates
the opportunity for a sophisticated bettor to extrapolate from historical data on
match outcomes (or other relevant information), in order to formulate a betting
strategy that yields either a positive return, or (at least) a loss smaller than would
be expected by an unsophisticated or indiscriminate bettor due to margins or com-
missions that are built into the odds.
Chapter 12 examines the economics of sports betting, with particular emphasis
on the markets for betting on the outcomes of football matches. Section 12.1
reviews the previous literature on this topic, including studies of the informational
efficiency of sports betting markets and the phenomenon of favourite-longshot
bias. Some attention is given to the impact on football betting of recent phenom-
ena such as the emergence of online betting exchanges and person-to-person bet-
ting, and the growth in popularity of football spread betting. Studies that search
for evidence of sentiment bias in the market prices for bets on football matches
are also reviewed.
Sections 12.2 to 12.4 present a retrospective analysis of the informational effi-
ciency of the fixed-odds betting market on half-time/full-time match outcomes
in English football. The analysis is based on an extension of the modelling
approaches used by Goddard and Asimakopoulos (2004) and Forrest, Goddard
and Simmons (2005) to test for the efficiency of the fixed-odds betting market on
352
Previous evidence on informational efficiency 353
full-time ‘home win/draw/away win’ match results. In these two previous studies,
an ordered probit regression, similar to the model presented in Chapter 4 of this
volume, is used to estimate a match results forecasting model, which generates
out-of-sample forecasts in probabilistic form. With nine bets available on each
match rather than three, the half-time/full-time betting market appears to offer
greater scope for detecting anomalies in the bookmakers’ odds, and for formulat-
ing profitable betting strategies.
The betting odds data set employed in the present study, comprising data
for more than 9,000 matches played over five football seasons, covers the same
matches as the Forrest, Goddard and Simmons (2005) study, but is consider-
ably larger than those used in most other football betting studies. We argue that
evidence from a relatively large holdout sample is required in order to validate
claims for the universal profitability of any specific betting strategy. Large-sample
evidence is necessary, because of the high level of small-sample variation in aver-
age returns. In this study, a simple betting strategy is shown to be profitable over
the entire five-season holdout sample period, and for each of the five seasons
individually. Accordingly, it is claimed that the present results constitute stronger
evidence than has been available previously that it is possible to use historical
data to formulate a betting strategy that is universally (rather than just locally)
profitable.
12.1 Previous evidence on the informational efficiency of football and other sports

betting markets
Betting market efficiency and the favourite-longshot bias
Many studies of the informational efficiency of betting markets focus on race-
track betting, and only a representative selection of studies is considered here.
Several early contributions focused on the relationship between the forecast prices
(FP), published in the sporting press on the race-day morning, and the starting
prices (SP), at which all bets are settled, compiled and published minutes before
the race takes place. Dowie (1976) interprets a finding of little or no difference
between the correlations between FP and race outcomes on the one hand, and SP
and race outcomes on the other, as evidence against the hypothesis that bets were
being placed on the basis of inside information between the publication of FP
and the start of the race. If inside information were being used, the SP correlation
should exceed the FP correlation. Crafts (1985), however, demonstrates that this
interpretation was not necessarily justified. Similar FP and SP correlations across
all outcomes could be consistent with substantial discrepancies between FP and
SP on individual outcomes, and with SP containing significantly superior infor-
mation. By betting at SP on outcomes whose SP were substantially shorter than
the longest prices that had previously been quoted, significantly smaller losses
than average could have been realised.
Much of the academic literature on sports betting markets has focused on a con-
sistent empirical regularity in racetrack betting odds known as favourite-longshot
bias, implying that the expected return to a bet placed at longer odds is lower than
at shorter odds. Several kinds of explanation have been put forward. One possi-
bility is that bettors may possess a positive preference for risk, and are therefore
willing to accept bets on longshots offering low expected returns but high variance
(Weitzman, 1965; Ali, 1977). Another possibility is that bettors’ subjective prob-
abilities of losing are systematically downward biased (Henerey, 1985). If so, it is
straightforward to show that SP are subject to favourite-longshot bias. Golec and
Tamarkin (1998) suggest incorporating skewness, as well as the mean and variance
of the return, into the bettor’s utility function. They demonstrate that it may be
rational for risk-averse bettors who love skewness (bets offering a high probability
of losing and a low probability of a very large return) to accept bets on longshots
despite the low expected returns.
An alternative type of explanation regards favourite-longshot bias as a rational
pricing response on the part of bookmakers to the presence in the market of insid-
ers trading on the basis of private information (Shin, 1991, 1992, 1993; Vaughan
Williams and Paton, 1997). Since both the average length of the odds and the
opportunities for insider trading are directly proportional to the number of run-
ners, bookmakers’ margins need to be higher (and bettors’ expected returns lower)
at longer odds, in order to protect bookmakers from the actions of the informed
traders. Terrell and Farmer (1996) propose a similar type of explanation that does
not rely on insider trading. In a world of imperfect information, there are two
distinct groups of bettors: pleasure bettors who place bets without knowing the
true probabilities of the various outcomes; and informed bettors who incur costs
in order to ascertain the true probabilities. Informed bettors study the actions of
pleasure bettors in order to identify and accept bets that offer a positive expected
return. The interaction between the two groups produces a distribution of odds
that fails to reflect the true probabilities. Moreover, since the intervention of the
informed bettors always lowers the odds on the outcomes with high expected
returns (and raises the odds on outcomes with low expected returns) there is a stat-
istical association between high odds and low expected returns. This may explain
the consistent tendency for empirical studies to identify favourite-longshot bias.
Efficiency of betting markets in North American professional team sports

A number of researchers have tested the efficiency of betting markets on the
outcomes of North American team sports. The market for betting on the NFL
operates on a points spread system. The bookmaker quotes a points spread,
reflecting an assessment of the expected points difference between the favourite
and the underdog. Bettors backing the favourite expect the latter to win by more
than the spread; those backing the underdog expect the favourite to win by less
than the spread, or expect the underdog to win. Once a bet is placed the payoffs are
fixed, but bookmakers can adjust the spread as the time of the match approaches,
in an attempt to equalise the volume of bets placed on either team. The book-
maker’s margin is provided by the ‘eleven-for-ten rule’, requiring a bettor to stake
$11 in order to win $10. To make a profit, the bettor therefore requires more than
52.4 per cent of bets to be correct.
Pankoff (1968) presented the first regression-based test of efficiency in the NFL
betting market, by regressing match outcomes (measured by the score differential)
on bookmakers’ spread using data for the period 1956–65. Intercept and slope
coefficients not significantly different from zero and one respectively suggest that
the spread was an unbiased predictor of the match outcome (see below). Obtaining
similar results using a 1980–85 NFL data set, Gandar et al. (1988) point to the low
power of regression-based tests to reject the null hypothesis of efficiency. They
propose a series of alternative economic tests, involving direct evaluation of the
returns that would have been earned from the implementation of various trading
rules. Two types of trading rule are considered: technical rules, which select bets
purely on the basis of the past performance of the teams; and behavioural rules,
which select bets in an attempt to exploit certain hypothesised behavioural pat-
terns of the general public. An example of a behavioural rule is to back underdogs
against favourites in cases where the favourite covered the spread by a large margin
the previous week. The behavioural hypothesis is that the public tends to overreact
in such cases, creating the opportunity for a profitable contrarian strategy. Gandar
et al. find that some behavioural rules are consistently profitable; technical rules, in
contrast, fail to produce a profit.
Golec and Tamarkin (1991) test the efficiency of the spreads posted on the out-
comes of NFL and college football matches, using data covering a fifteen-year
period from 1973 to 1987. For NFL betting they find evidence of inefficiencies
favouring bets on home wins and bets on underdogs. While the extent of the bias
against home teams in the bookmakers’ prices seems to have diminished over time,
the bias against underdogs has increased. No evidence of similar bias is found in
the college football betting spreads. Dare and MacDonald (1996) generalise the
empirical methodology for regression-based tests, and demonstrate that a number
of earlier tests were based on specifications that unknowingly imposed implicit
restrictions on a more general model. Tests that search only for evidence of home-
away team or favourite-longshot biases, without recognising the interdepend-
ence between these characteristics of teams, are capable of producing misleading
results. Little evidence of inefficiency is found when the general model is applied
to a 1980–93 NFL and college football data set.
In other NFL studies, Badarinathi and Kochmann (1996) use data for the
period 1984–93 to show that a strategy of betting regularly on underdogs when the
spread exceeded five points, and when variations in spreads among bookmakers
in different cities enabled the bettor to obtain an additional two-point advantage,
was systematically profitable. Similarly, using data for the period 1976–94, Gray
and Gray (1997) find that a strategy of betting on home-team underdogs pro-
duced average returns of over 4 per cent in excess of commissions; although there
were signs that this anomaly had tended to dissipate over time. There is evidence
of market overreaction to teams’ recent performance, suggesting that a contrarian
strategy of betting on strong teams that have performed poorly recently might be
profitable.
Investigating the question of market efficiency from a different perspective,
Gandar et al. (1998) find that the bookmakers’ closing prices provided a closer
approximation to match outcomes than their opening prices using an NBA bet-
ting market data set covering the period 1986–94. Movements in prices between
the opening and close of trade tended to eliminate biases in the opening prices,
suggesting that informed traders were both active and influential in this market. In
other NBA studies, Camerer (1989) finds that bets placed on teams that are cur-
rently experiencing a winning streak are more likely to lose than to win. Brown and
Sauer (1993) point out that bookmakers’ points spreads might adjust in response
to a winning streak. This might invalidate Camerer’s conclusion that the ‘hot-
hand’ phenomenon is a myth: the biases reported by Camerer might be consistent
with simultaneous and roughly parallel adjustments to the observation of win-
ning streaks by both bettors and bookmakers. More recently, Paul and Weinbach
(2005) report evidence of a reverse favourite-longshot bias in NBA betting and,
consistent with Camerer (1989), evidence that betting against the ‘hot-hand’ is
profitable.
The betting markets for MLB and the NHL operate on an odds system. For
example, if the bookmaker posts a line of (–150, +140), the bettor must stake
$1.50 in order to win $1 for a bet on the favourite; or the bettor stakes $1 in order
to win $1.40 for a bet on the underdog or longshot. Consistent with the findings of
several of the NFL betting studies cited above, Woodland and Woodland (1994,
2003) report evidence of a reverse favourite-longshot bias for the betting market
on MLB. This bias is longstanding, but it is insufficient to form the basis for a
profitable betting strategy. Woodland and Woodland (2001) find similar evidence
of a reverse favourite-longshot bias for betting on the NHL.
Informational efficiency of football betting markets

Testing the informational efficiency of the bookmaker’s prices for betting on the
results of English football matches raises a different set of issues. In contrast to
the situation with racetrack betting, prices are fixed by the bookmakers several
days before the match takes place, so both bookmaker and bettor know the price
and the payoffs when the bet is placed. And in contrast to North American team
sports betting, the bookmaker does not adjust the prices as bets are placed, even if
new information is received as the time of the match draws near.
For English football, Pope and Peel (1989) investigate the efficiency of the
prices set by four national high-street bookmakers for matches played during the
1982 season. A simple, though as before not necessarily a very powerful test of
the weak-form efficiency hypothesis is based on regressions of match outcomes

against implicit bookmaker’s probabilities. For the purposes of describing these
tests, the following variable definitions are used:
Hi,j = 1 if the match between team i and team j results in a home win, and 0
otherwise;
Di,j = 1 if the match between team i and team j results in a draw, and 0 otherwise;
Ai,j = 1 if the match between team i and team j results in an away win, and 0
otherwise.
In the linear probability model: Hi,j = ρ1 + ρ2φHi,j + vi,j, a necessary condition for
weak-form efficiency is {ρ1 = 0, ρ2 = 1}. Equivalent conditions apply in the cor-
responding regressions for Di,j and Ai,j. Any other combination of values for {ρ1,
ρ2} suggests that there may be inefficiencies. The combination {ρ1>0, ρ2<1}, for
example, means that bookmakers overestimate the home-win probabilities for the
matches least likely to produce home wins (with long odds), and underestimate the
probabilities for those most likely to produce home wins (with short odds). Noting
that ordinary least squares (OLS) estimation of the linear probability model gives
rise to a heteroscedastic error structure, Pope and Peel suggest using weighted
least squares (WLS) estimation, using Ĥ i,j(1 − Ĥ i,j) as weights, where Ĥi,j are the
fitted values of the dependent variable obtained from (preliminary) estimation of
the model using OLS. Alternatively, the model can be estimated as a logit regres-
sion, in which case different numerical estimates of the coefficients ρ1 and ρ2 are
expected.
In the linear probability model regressions, Pope and Peel find no significant
departures from the condition {ρ1 = 0, ρ2 = 1} for the odds on home and away win
outcomes. On the other hand, the draw odds have no significant predictive con-
tent for draw outcomes, suggesting a departure from efficiency conditions. The
returns achievable from two betting strategies are investigated. The first, using
only information derived from bookmakers’ odds, involves betting on outcomes
for which the predicted probability derived from the logit regression of match
outcome on bookmakers’ implicit probabilities exceeds the geometric mean of
the four bookmakers’ implicit probabilities. The second, also exploiting extrane-
ous information, involves betting on a consensus of the recommendations of six
national newspaper tipsters. Under both strategies, there is some evidence that
positive pre-tax returns could have been earned, although these would not have
been sufficient to generate a profit after allowing for payment of betting duty.
Cain, Law and Peel (2000) report evidence of favourite-longshot bias in the
fixed-odds betting market for match results and scores. The odds available for bets
on specific scores are dependent only on the odds posted for the three possible
match results (home win, draw or away win), and it is possible to bet on the score
of a single match. The Poisson and the negative binomial distributions are used to
model the home- and away-team scores independently. Estimates of the fair odds
for specific scores are computed, conditional on the bookmaker’s posted home
win odds. In comparisons of the estimated fair odds with the bookmaker’s actual
odds for specific scores, the former are generally found to be significantly longer
than the latter for longshot bets, indicating favourite-longshot bias. The fair odds
are sometimes shorter than the bookmaker’s odds for bets on strong favourites.
Direct calculations of average profitability for different categories of bet suggests
that bets on strong favourites may offer limited profitable betting opportunities.
Dixon and Coles (1997) use a scores-based forecasting model (see Chapter 4,
Section 4.1) to test the profitability of a strategy of betting on matches with the
highest expected return according to the match outcome probabilities derived from
the model. During the 1996 season, betting on outcomes for which the ratio of the
model’s probability to the bookmaker’s implicit probability exceeded about 20 per
cent would have produced a pre-tax return significantly greater than zero, after
allowing for the bookmaker’s margin. Dixon and Pope (2004) present further ana-
lysis of the forecasting performance of the Dixon and Coles model, finding that
the model’s probabilities could have been exploited to devise a profitable strategy
during seasons 1994 to 1996, inclusive. The positive expected return is borderline
statistically significant. In contrast to several other studies based on later data,
Dixon and Pope report evidence of a reverse favourite-longshot bias.
Using an early prototype of the ordered probit results-based forecasting model
that is presented in Chapter 4, Sections 4.4 and 4.5, Goddard and Asimakopoulos
(2004) compare probabilistic forecasts obtained from the model with the odds
quoted by the high-street bookmaker William Hill for fixed-odds betting on match
results. Regression-based tests indicate that the model contains information about
match outcomes that is not impounded into the bookmakers’ prices. A strategy
of selecting bets ranked in the top 15 per cent by expected return according to
the model’s probabilities would have generated a positive return of at least 4 per
cent in each of the four football seasons covered by the study (1999 to 2002, inclu-
sive). A strategy of betting on the match outcome for which the model’s ex ante
expected return is the highest would have generated positive or zero gross returns
in each of the four seasons.
Forrest, Goddard and Simmons (2005) present more extensive comparisons of
probabilistic forecasts obtained from an ordered probit results-based forecasting
model, and the prices quoted by five high-street bookmakers for fixed-odds bet-
ting for English league matches played during seasons 1999 to 2003, inclusive. In
terms of forecasting performance, the ordered probit model’s predictions prove to
be far from dominant over the implied probabilities derived from the bookmak-
ers’ odds; and the informational efficiency of the latter appears to have improved
over the course of the observation period. This improvement might reflect book-
makers’ responses to increasing competitive pressure on their margins during a
period of far-reaching technological and structural change in the betting industry.
Sections 12.2 to 12.4 below present a further, retrospective analysis of the market
for fixed-odds betting on English football during the same period, focusing on the
betting market for correct prediction of half-time and full-time match outcomes.
Graham and Stott (2008) find that a results-based forecasting model produced
probabilistic forecasts for match results that were of similar accuracy to those
obtained from the fixed odds quoted by the high-street bookmaker William Hill.
A forecasting model for the bookmaker’s odds was able to produce reasonable
predictions; however, the bookmaker’s odds contained a favourite-longshot bias.
Any anomalies in the bookmaker’s prices were insufficient to form the basis for
a consistently profitable betting strategy. Spann and Skiera (2009) and Vlastakis,
Dotsis and Markellos (2009) report further evidence on football betting market
efficiency for Germany, and for several European countries respectively.
Online betting exchanges and person-to-person betting

The betting industry is based on the processing and exchange of information,
and with hindsight it is unsurprising to observe that this industry has been trans-
formed by the emergence and growth of the Internet during the 1990s and 2000s.
The Internet has created technological opportunities for new entrants to com-
pete with established betting firms, since anyone can now transmit information
around the globe, either at low cost or (effectively) for free. In the betting industry,
the most successful business model to have emerged during the first decade of
the internet era is person-to-person betting transacted through an online betting
exchange. By the end of the 2000s the UK-based Betfair, launched in 2000 by the
entrepreneur Andrew Black, had established a position of dominance in this seg-
ment of the betting industry.
Betting exchanges exist to find a match between a backer, who wishes to bet
on an outcome at a specified price, and a layer, who is willing to offer the bet
at the price specified by the backer. Betfair publishes the three best prices cur-
rently being offered by members acting as backers, the three best prices offered by
members acting as layers, and the total stakes that are being offered or asked for
at those prices. If several members are offering the same price, the total stake is
pooled, so no direct one-to-one correspondence between a backer and a layer is
required for every bet placed. Any client is free to be a backer or a layer, and clients
can be both at different stages as a market evolves. Clients can either accept bets
that are currently being offered by other members, or post prices and stakes that
they themselves are willing to offer to other clients. As the market for bets on any
event develops, there is a tendency for the margin between the best available prices
at which clients can back or lay to converge towards the exchange’s commission
(Vaughan Williams, 2005).
It seems plausible to observe that there has been a self-perpetuating element to
Betfair’s exponential rise in popularity. This has been driven by a network exter-
nality effect: as the number of users of a service increases, the service becomes
more valuable to all users, and therefore more popular. In the case of Betfair,
the network externality operates through the increase in market liquidity as the
number of users increases, making it easier for any client to find a match for any
particular bet they wish to back or lay (Davies et al., 2005; Koning and van Velzen,
2010).
The betting exchange business model contains several innovative features.
Traditional betting firms or bookmakers always lay bets that are backed by their
clients, and it is illegal for an unlicensed client to lay a bet that is backed by another
client or by a bookmaker. In a series of legal challenges, the traditional betting
firms argued that by allowing clients to lay bets, the betting exchanges were enab-
ling those clients to act illegally as unlicensed bookmakers. The betting exchanges
responded, and the courts concurred, that the exchanges and not the clients were
the bookmakers. The betting exchanges’ rates of commission, typically between
2 per cent and 5 per cent, are lower than the margins built into the traditional
betting firms’ odds, enabling the exchanges to offer significantly more competi-
tive prices. Betting exchanges have pioneered in-the-running or in-play betting, in
which markets operate after the event has started. Finally, betting exchanges facili-
tate the development of trading strategies similar to those used by traders in other
financial markets. Clients of betting exchanges can trade in order to speculate
on movements in the price for a particular outcome, rather than on the outcome
itself. Clients can hedge exposures incurred in other transactions: in racetrack bet-
ting, for example, it is believed that on-course bookmakers make extensive use of
betting exchanges for hedging purposes. Clients can also use betting exchanges for
arbitrage, exploiting differences in prices in different markets to realise an instant-
aneous profit (Laffey, 2005).
For racetrack betting, Smith, Paton and Vaughan Williams (2006) find that the
favourite-longshot bias is lower in exchange betting than in traditional betting
markets. Since betting exchanges facilitate the flow of information, this finding
is consistent with the notion that the favourite-longshot bias is due to informa-
tional failure in the form of private or insider information that is not available to
all market participants. Accordingly, Smith, Paton and Vaughan Williams favour
information theories of favourite-longshot bias, rather than risk preference theor-
ies. Using data on an Irish online exchange that offers markets in both sports and
financial events, Tetlock (2004) finds that informational inefficiencies are greater in
the sports betting markets. Consistent with the notion that the favourite-longshot
bias widely observed in traditional betting markets is mitigated by online betting,
Tetlock finds a significant reverse favourite-longshot bias in the sports betting
markets, but no such bias in the financial markets.
Football spread betting

A spread bet is analagous to a forward or a future contract. In contrast to fixed-
odds betting or person-to-person betting via an online exchange, the possible pay-
offs to the bettor (both positive and negative) from an individual spread bet may
be unbounded. For each index for which a spread betting market exists, the book-
maker quotes a spread (B, T), and the bettor has a choice of either ‘buying the
spread at T’ or ‘selling the spread at B’. If the realised value of the index at the end
of the match is S, the payoffs to the bettor for a unit stake are S–T if the bettor
bought the spread, or B–S if the bettor sold the spread. If the payoff is positive
the bookmaker pays the bettor, and if the payoff is negative the bettor pays the
bookmaker. Spread betting is normally transacted online. The bettor maintains an
account with an online bookmaker, and the bookmaker accepts the bet only if the
bettor has sufficient credit to cover any potential liability that could arise from the
bet under realistic assumptions. Indices based on match results or goals for which
spread betting markets exist, quoted by Fitt (2009), include the following:
(i) Total home goals minus total away goals.
(ii) Total home goals plus total away goals.
(iii) Goal rush index (total goals: 0 = 0 points, 1 = 10 points, 2 = 20 points, 3 =
33 points, 4 = 50 points, 5 = 70 points, >5 = 100 points).
(iv) Total home goals.
(v) Home win index (25 points for a win, 10 points for a draw).
(vi) Home win handicap k (25 points for a win by more than k goals, 10 points
for a win by k goals, defined for various k).
(vii) Home-team score k and win index (25 points if correct based on half-time
score, 50 points if correct at full-time, 75 points maximum, defined for vari-
ous k).
(viii) k-k draw index (25 points if correct based on half-time score, 50 points if
correct at full-time, 75 points maximum, defined for k = 0,1 and k>1).
(ix) Home-team double result (25 points for a win, 10 points for a draw, based
on half-time and full-time scores added together).
(x) Sum of minutes in which all goals are scored.
(xi) Sum of minutes in which home-team goals are scored.
(xii) Number of minutes during the match in which home team was leading.
(xiii) Time (minutes) when k’th goal is scored (defined for various k, 90 for 0–0
draw).
(xiv) Time (minutes) when k’th home-team goal is scored (defined for various k,
90 if home team fails to score k goals).
(xv) Time (minutes) when last goal is scored (0 for 0–0 draw).
(xvi) Time (minutes) when winning goal is scored (0 for any draw).
Indices equivalent to (iv), (v), (vi), (vii), (ix), (xi), (xii), (xiv) are also defined for the
away team. Other football spread betting markets, based on corners, include total
corners, difference between the two teams’ corners, multicorners (the product of
the numbers of corners taken in each half of the match), and four flags (the time
in minutes that elapses before at least one corner has been taken from each of the
four corner positions).
Fitt (2009) derives formulae for the fair values of each of the match results
and goals-based spread bets listed above, based on an assumption that the total
numbers of goals scored by each team during each match follows a Poisson
distribution. In order to apply these formulae, the bettor needs to supply numer-
ical values for the Poisson means or expected goals scored by each team. These
values might be inferred from the centre spreads for bets (i) and (ii); or the bettor
might decide to take an independent view of the correct expected values for any
particular match.
In formulating a spread-betting strategy in a situation where multiple betting
opportunities are available, Fitt (2009) suggests the bettor faces a choice anala-
gous to the choice faced by the risk-averse investor in Markowitz’s (1952) portfolio
theory. The investor (bettor) seeks to create a portfolio of assets (bets) that either
maximises the expected return for any given target level of risk, or minimises the
risk for any given target expected return. Risk is measured by the variance of the
return on the portfolio of assets (bets). In fact, the analogy between spread bet-
ting on football and Markowitz’s portfolio theory is imperfect. In the latter the
solution to the risk-averse investor’s problem is to invest some proportion of his
or her total fund in a unique selection of risky assets known as the ‘optimum port-
folio’, and invest the remainder of the fund in a risk-free asset. The definition of
the optimum portfolio is the same for every investor, regardless of their individual
preferences concerning target expected return and acceptable levels of risk. In the
case of spread betting, all of the risk is concentrated into 90 minutes’ play, and
no alternative risk-free asset exists that would offer a meaningful return over such
a short period. Therefore the bettor’s sole decision is the selection of the unique
optimum portfolio of risky bets. For any array of bets for which the expected
returns and variances have been evaluated, the computation of the optimal port-
folio is a routine, mechanical task. Accordingly, Fitt (2009) concludes that there
is only one unique combination out of any available selection of spread bets that
any bettor need ever consider.
The optimum portfolio approach to the problem of selecting a portfolio of bets
offering different combinations of expected return and risk offers an alternative
betting strategy to the strategy based on the well-known ‘Kelly criterion’. Kelly
(1956) examines how a bettor with a fixed initial fund and an edge over the book-
maker, which gives a positive expected return for each bet placed, should place
bets over a sequence of betting opportunities in order to maximise his overall
expected return, in the case where no further bets are placed if the fund is wiped
out.1 Although the expected return for each bet is maximised by staking the entire
fund, this strategy would not maximise the bettor’s overall expected return, since
it would offer a low probability of a large win on the first bet and a high probabil-
ity of being wiped out by the first bet, and being unable to exploit the edge in any
future bets.
Kelly shows that the bettor’s optimal strategy is to stake a fraction of the remain-
ing fund on each bet, with the fraction chosen so as to maximise the expectation of
the natural logarithm of the fund’s value after the bet is settled. The optimal frac-
tion for each bet is a function of the odds, the edge, and any taxes or commission
payable on the bet, but it does not depend on the size of the fund. The optimal
stake for each bet, however, is a function of the size of the fund, and decreases pro
rata if the fund is depleted by a run of losing bets. In its original form, the Kelly
criterion is applicable for bets with a discrete (typically binary) distribution of
possible outcomes. Haigh (2000) and Chapman (2007) examine the adaptation of
the Kelly criterion for spread bets, for which the outcomes may be considered as
following a continuous distribution. The optimal fraction is shown to be a func-
tion of the distribution of payoffs.
Fitt, Howls and Kabelka (2006) examine the valuation of in-play spread betting
markets, which operate dynamically while matches are underway. Arrivals of goals
and corners are modelled as Poisson processes, with arrival rates that are assumed,
for simplicity but (as we have seen in Chapter 5) probably counterfactually, not to
change over the course of the match. As before, the bets are valued conditional
upon numerical values supplied by the user for the Poisson means or expected
goals scored by each team. Using simple probability theory, a fair value can be
calculated for each spread bet at any stage in the match, conditional on the state of
the match at that time (defined, for example, by the number of goals already scored
or corners already taken, their timings, and any other information that is relevant
to the bet in question). A variance can also be computed, to measure the riskiness
of the bet. Fitt, Howls and Kabelka explore the variation of the fair value and
the variance over the duration of the match, and the variation as relevant events
occur (for example, when a goal is scored or a corner is taken), for various spread
bets. For matches played during the 2004 European Championships (Euro 2004),
the spreads quoted by online bookmakers appear to be highly consistent with the
theoretical valuations.
Sentiment bias in betting markets

Sentiment bias in team sports betting markets is the subject of several recent
studies. Betting firms might offer more favourable odds for bets on wins by
popular teams, in an attempt to attract larger volumes of business by shading
margins. Alternatively, bettors might tend to form upwardly biased assessments
of the win probabilities for the teams they support, and betting firms might
attempt to exploit this tendency by offering downwardly biased odds to those
bettors. Forrest and Simmons (2008) test for sentiment bias using online bet-
ting odds data for top-tier football matches in Spain’s La Liga and the Scottish
Premier League. In both cases, there is a large disparity between the popularity
of the two largest-market teams (Barcelona and Real Madrid in Spain; Celtic
and Rangers in Scotland) and the rest, suggesting that these leagues should pro-
vide promising opportunities for the investigation of sentiment bias. Consistent
with a model of profit-maximising behaviour by bookmakers operating in a
competitive market, the odds offered for wins by the most popular teams are
found to be relatively favourable.
Similarly, in an analysis of betting odds for club football matches played in the
English Premier League and Football League between seasons 2001 and 2008,
Franck, Verbeek and Nuesch (2008) estimate probit regressions in which the
dependent variable is coded 1 for a winning bet and 0 for a losing bet, and the
explanatory variables are the betting odds, and proxies for the relative popular-
ity of the two teams based on attendance data and an indicator of the volume of
press coverage received by each team, both from the previous season. In the latter
part of the sample period, the odds for a win are found to be positively related to
the popularity of the team concerned. This suggests that betting firms may offer
more favourable prices in order to attract betting volume evoked by sentiment.
In an empirical analysis of betting odds data for international and European
club football matches, the prior of Page (2009) is that an ‘optimistic bias’ on the
part of supporters is exploited by bookmakers who offer less favourable odds.
Page examines whether betting firms located in various countries offer downward-
biased odds for wins by teams from their own country, and whether the odds
offered by all betting firms for wins by British teams are downward biased. If there
is an optimistic bias, the latter pattern might be explained by the fact that the bet-
ting market in the UK (relative to population) is substantially larger than betting
markets in other European countries. The empirical analysis, based on regressions
of the returns for individual bets against the odds and dummy variables for bets
on British teams and for bets on teams from the same country as the betting firm,
reveals no evidence of optimistic bias; on the contrary, there is some evidence that
British betting firms offer upwardly biased odds for wins by Wales, Scotland and
Northern Ireland in international matches.
Sentiment bias is investigated from a different angle by Palomino, Renneboog
and Zhang (2009), who use share price data on listed football clubs in the UK to
examine the stock market reaction to the publication of the betting odds for forth-
coming matches of the teams concerned, and to the results of the same matches.
The market does not react to the publication of betting odds, but share prices tend
to increase rapidly following match wins, and to decrease rather more slowly fol-
lowing defeats. The finding that football club share prices are sensitive to individ-
ual match results replicates results reported previously by Dobson and Goddard
(2001) and Gannon, Evans and Goddard (2006). The stock market response to vic-
tories seems to contain an element of overreaction: the positive share price effect
is apparent even for victories which, according to the betting odds, were strongly
anticipated. This feature is attributed to a form of optimistic bias, being driven
by investor sentiment rather than by rational expectations. However, the negative
share price reaction to defeats is weaker for defeats that were strongly anticipated,
in accordance with the rational expectations hypothesis. Overall, there is an elem-
ent of asymmetry between the stock market’s reactions to victories and defeats.
Similarly, Bernile and Lyandres (2008) analyse share price data on twenty stock
market-listed football clubs from eight European countries which participated
in European club competition between the 2000 and 2006 seasons. The ex ante
probabilities of match outcomes implied by prices of contracts traded on betting
A forecasting model for half-time/full-time outcomes 365
exchanges are biased, on average, both relative to the actual distribution of match
results and relative to the probabilities implied by bookmakers’ odds. It is argued
that betting exchange odds data provide a superior representation of bettor or
investor sentiment than betting firms’ odds data. The stock market reaction to
match results is asymmetric: on average, losses result in negative returns that are
larger in absolute terms than positive returns following victories. This pattern is
attributed to the presence of an optimistic bias in investors’ ex ante assessments
of the relevant win probabilities.
12.2 A forecasting model for half-time/full-time match outcomes

The next three sections of Chapter 12 present a retrospective analysis of the
informational efficiency of the fixed-odds betting market on half-time/full-time
match outcomes in English League football. Section 12.2 begins by specifying a
model to generate probabilistic forecasts for permutations of half-time and full-
time match outcomes. Section 12.3 presents some comparisons of the forecasts
obtained from the model with the odds posted for fixed-odds betting by five
UK high-street bookmakers. Finally, Section 12.4 tests for the informational
efficiency of the half-time/full-time fixed-odds betting market, by examining
the prospects for the formulation of a profitable betting strategy based on the
forecasts of the model.
In the half-time/full-time match outcomes betting market, bettors choose
between nine bets on any match, corresponding to the nine possible permutations
of outcomes (states of the match) at half-time and at full-time. In order to model a
match outcomes dependent variable with this structure, two ordered probit regres-
sions are required: a ‘first-half’ model to generate unconditional probabilities for
the three possible outcomes (states of the match) at half-time; and a ‘second-half’
model to generate probabilities for the three possible outcomes at full-time, con-
ditioned on the half-time outcome. The model that is developed in this chapter
allows for a non-zero covariance between the first-half and second-half disturb-
ance terms for the same match.
Probabilistic forecasts for the nine possible permutations of half-time/full-
time outcomes can be generated from an extended version of the results-based
forecasting model (see Section 4.4). In the extended version of this model, it is
assumed that the half-time outcome of the match between teams i and j, denoted
y0,i,j, depends on the unobserved variable y*0,i,j, and a ‘first-half’ disturbance term
ε(F)
i,j, as follows:
Home team leading: R0,i,j = 1 if µ 0,2 < y*0,i,j + ε (F)

i,j
Teams level: R0,i,j = 2 if µ 0,1 < y*0,i,j + ε (F)

i,j < µ 0,2
Away team leading: R0,i,j = 3 if y 0,i,j + ε i,j < µ 0,1

* (F)
[12.1]
Similarly, the full-time outcome of the match between teams i and j, conditional
on the half-time outcome y0,i,j = k for k = 1,2,3, depends on the unobserved vari-
able y*k,i,j, and a ‘second-half’ disturbance term ε(s)i,j , as follows:
Home win: Rk,i,j = 1 if µ k,2 < y*k,i,j + ε (s)

i,j
Draw: Rk,i,j = 2 if µ k,1 < y k,i,j + ε i,j < µ k,2

* (s)
Away win: Rk,i,j = 3 if y*k,i,j + ε (s)

i,j < µ k,1 [12.2]
In [12.1] and [12.2], it is assumed that the latent variables y*k,i,j for k = 0,1,2,3
are linear functions of a set of covariates, which are defined using data that are
available before the match is played. The covariates are the same as those used
in the results-based forecasting model reported in Section 4.4. The magnitudes
and statistical significance of the estimated coefficients in the half-time/full-
time outcomes model are broadly similar to those in the full-time results-based
model (see Section 4.4 and Table 4.5). To conserve space the estimation results
for the half-time/full-time outcomes model are not reported.
In the estimations of [12.1] and [12.2], allowance is made for a separate pair
of values for the cut-off parameters μk,1 and μk,2 for each football season within
the estimation period. This allows for variation over time in the overall pro-
portions of matches producing each of the nine possible half-time/full-time
outcomes.
It is assumed that the joint distribution of the first-half and second-half
disturbance terms ε(F) i,j and ε i,j is bivariate normal, with E[ε i,j ] = E[ε i,j ] = 0,
(S) (F) (S)
var[ε i,j] = var[ε i,j] = 1, and cov[ε i,j, ε i,j] = corr[ε i,j, ε i,j] = σ, where σ is an ancil-
(F) (S) (F) (S) (F) (S)
lary parameter to be estimated. The prior is σ>0. ε(F) i,j and ε i,j allow for all fac-
(S)
tors other than those included among the defined model covariates that may
influence the match result. Some such factors, including good or bad luck with
refereeing decisions or the ‘run of the ball’, may operate quite randomly, with
no statistical association between the size and direction of the effects in the first
and second halves of the same match. Other factors, such as player absences
due to injury or suspension, or factors relating to playing conditions or the
crowd that tend to favour one team over the other, may well tend to operate in
the same direction during both halves of the same match, producing a positive
association between ε(F) i,j and ε i,j.
(S) 2
Let Φ1 and Φ2 denote the univariate and bivariate standard normal distribution
functions. The nine possible permutations of half-time and full-time outcomes are
m = {HH, HD, HA, DH, DD, DA, AH, AD, AA}, where HH denotes home team
leading at half time and home-team win, DH denotes match level at half time and
home-team win, AH denotes away team leading at half time and home-team win,
and so on. The nine (unconditional) probabilities for the match between team i
and team j, are defined as follows:

Comparing the model with betting odds 367
ˆ
i,j = 1 − Φ1 (µ 0,2 − y
pHH ˆ *0,i,j ) − Φ1 (µˆ 1,2 − yˆ 1,i,j
*
) + Φ 2 (µˆ 0,2 − yˆ *0,i,j ,µˆ 1,2 − yˆ 1,i,j
*
,σˆ )
p i,j = Φ1 (µˆ 0,2 − yˆ 0,i,j ) − Φ1 (µˆ 0,1 − yˆ 0,i,j )
DH * *
− Φ 2 (µˆ 0,2 − yˆ *0,i,j ,µˆ 2,2 − yˆ *2,i,j ,σˆ ) + Φ 2 (µˆ 0,1 − yˆ *0,i,j ,µˆ 2,2 − yˆ *2,i,j ,σˆ )
p i,j = Φ1 (µˆ 0,1 − yˆ 0,i,j ) − Φ 2 (µˆ 0,1 − yˆ *0,i,j ,µˆ 3,2 − yˆ *3,i,j ,σˆ )
AH *
pHD ˆ
i,j = Φ1 (µ1,2 − y ˆ 1,i,j
*
) − Φ 2 (µˆ 0,2 − yˆ *0,i,j ,µˆ 1,2 − yˆ 1,i,j
*
,σˆ ) − Φ1 (µˆ 1,1 − yˆ 1,i,j *
)
ˆ ˆ * ˆ
+ Φ 2 (µ 0,2 − y 0,i,j ,µ1,1 − y1,i,j ,σ )
ˆ * ˆ
pDD
i,j = Φ 2 (µˆ 0,2 − yˆ *
, ˆ
0,i,j µ 2,2 − y ˆ *2,i,j ,σˆ ) − Φ 2 (µˆ 0,2 − yˆ *0,i,j ,µˆ 2,1 − yˆ *2,i,j ,σˆ )
− Φ 2 (µˆ 0,1 − yˆ *0,i,j ,µˆ 2,2 − yˆ *2,i,j ,σˆ ) + Φ 2 (µˆ 0,1 − yˆ *0,i,j ,µˆ 2,1 − yˆ *2,i,j ,σˆ )
ˆ
i,j = Φ 2 (µ 0,1 − y
p AD ˆ *0,i,j ,µˆ 3,2 − yˆ *3,i,j ,σˆ ) − Φ 2 (µˆ 0,1 − yˆ *0,i,j ,µˆ 3,1 − yˆ *3,i,j ,σˆ )
p i,j = Φ1 (µˆ 1,1 − yˆ 1,i,j
HA *
) − Φ 2 (µˆ 0,2 − yˆ *0,i,j ,µˆ 1,1 − yˆ 1,i,j
*
,σˆ )
pi,j = Φ 2 (µ 0,2 − yˆ 0,i,j ,µ 2,1 − yˆ 2,i,j ,σ ) − Φ 2 (µ 0,1 − yˆ *0,i,j ,µˆ 2,1 − yˆ *2,i,j ,σˆ )
DA ˆ * ˆ * ˆ ˆ
ˆ
i,j = Φ 2 (µ 0,1 − y
p AA ˆ *0,i,j ,µˆ 3,1 − yˆ *3,i,j ,σˆ ) [12.3]
In order to generate probabilistic half-time/full-time forecasts for the 1999 sea-

son (for example), maximum likelihood estimation is used to obtain estimates of
the coefficients on the covariates in the linear equations for ŷ*k,i,j for k = 0,1,2,3, and
estimates of the additional parameters μk,1, μk,2 and σ, using data from the previ-
ous fifteen seasons (1984 to 1998 inclusive). The probabilities pmi,j are obtained by
substituting these coefficient and parameter estimates, together with the values
of the covariates for each match played in the 1999 season, into [12.3]. In order
to generate forecasts for subsequent seasons, the model is re-estimated using data
for successive (overlapping) fifteen-season windows. For example, forecasts for the
2000 season are generated using the version of the model estimated with data for
seasons 1985 to 1999 inclusive; forecasts for the 2001 season are generated using
the version of the model estimated with data for seasons 1986 to 2000 inclusive;
and so on. There are five estimated versions of the model in total. The five esti-
mated values of the parameter σ vary between 0.806 and 0.911. Each of these five
estimates is significantly different from zero at the 0.01 level, indicating that there
is (as expected) a strong positive association between ε(F)
i,j and εi,j .
(S)
12.3 Comparing the model’s probabilistic forecasts with betting odds

Section 12.3 presents some comparisons between the probabilistic half-time/full-
time forecasts generated by the model described in Section 12.2, and five sets of
odds posted by high-street bookmakers in the UK for bets on half-time/full-time
outcomes. The odds are those compiled by four major high-street bookmakers
that were trading throughout the period covered by this study: Coral, William
Hill, Ladbrokes and StanleyBet; and those of Super Soccer, a specialist agency
that supplies odds to independent bookmakers. The fixed-odds betting data cover
the five seasons from 1999 to 2003 inclusive. The data source is the electronic bet-
ting odds archive www.mabels-tables.com.
Bookmakers’ fixed odds for betting on the outcomes of English football matches
are published several days before the match is played, and are not adjusted as bets
are placed, even if new information is received. The odds are quoted in either frac-
tional form or decimal form. With fractional odds, a quoted price of a-to-b on
the outcome HH means that if b is staked on this outcome, the net payoffs to the
bettor are +a (the bookmaker pays the winnings and returns the stake) if the bet
wins; and –b (the bookmaker keeps the stake) if the bet loses. With decimal odds,
the quoted price for the same bet is (a+b)/b.
If the bet were ‘fair’, in the sense of producing an expected return of zero to
both the bookmaker and the bettor, the implicit probability for the outcome HH
would be θi,jHH = b/(a+b). A necessary condition for all nine possible bets on a sin-
gle match to be ‘fair’ is ∑ θmi,j =1 (where Θ is the set of nine possible outcomes, as
m∈Θ
before). However, in practice ∑ θmi,j > 1 , because the bookmakers’ odds contain
m ∈Θ
margins for costs and profit. The bookmaker’s overround is λ i,j = ∑ θmi,j − 1. An
m ∈Θ
implicit probability for the outcome HH (for example) is obtained by rescaling
i,j as follows: φi,j = θ i,j / ∑ θ i,j . There are analogous definitions for φ i,j for the
θHH HH HH m m
m ∈Θ
other eight outcomes. By construction ∑ φi,jm = 1 . φmi,j can be interpreted as prob-
m∈Θ
abilities on the assumption that the bookmakers set their odds to achieve the same
expected return from every bet.3
Different UK high-street bookmakers usually quote similar but not always iden-
tical odds for the same bet. There may therefore exist arbitrage opportunities for
a bettor who is willing to search for the bookmaker offering the most favourable
(longest) odds for each bet. In this study, the efficiency tests are based on two sets
of odds that are constructed from the five sets of bookmaker odds in the original
data set. The first constructed set, ‘best odds’, contains the longest odds available
from any of the five sets of bookmaker odds for each possible bet. The second
constructed set, ‘median odds’, contains the third-longest (or median) odds from
the five sets of bookmaker odds. ‘Best odds’ are the odds actually available to the
bettor who searches for the most favourable odds; and ‘median odds’ are represen-
tative of the odds typically available to the bettor who does not search.
Table 12.1 reports summary data on the mean values of various unconditional
and conditional probabilities for the 9,727 matches for which bookmaker odds and
probabilistic half-time/full-time forecasts are available.4 Panel 1 reports the propor-
tions of matches that produced each of the three possible half-time outcomes for (i)
the initial fifteen-year estimation period (seasons 1984 to 1998), and (ii) the five-year
holdout sample period over which the bookmakers’ odds are also observed (seasons
1999 to 2003). Panel 1 also reports (iii) the average values of the bookmakers’ implicit
probabilities based on ‘best odds’ for the holdout sample period, and (iv) the average
values of the forecasting model probabilities for the holdout sample period.
For example, the bookmakers’ implicit probability for the home team to be
leading at half-time is φHHi,j + φi,j + φi,j ; similarly, the forecasting model probability
HD HA
Comparing the model with betting odds 369
Table 12.1 Probabilities for half-time/full-time outcomes
1. Probabilities for half-time outcomes (unconditional)

Home team leading Level Away team leading
1984–1998: observed .3557 .4326 .2117
proportions
1999–2003: observed .3464 .4223 .2313
proportions
1999–2003: bookmakers’ .3449 .4282 .2269
implicit probabilities
1999–2003: forecasting .3466 .4280 .2254
model probabilities
2. Probabilities for full-time outcomes (unconditional)
Home win Draw Away win
1984–1998: observed .4750 .2716 .2534
proportions
1999–2003: observed .4552 .2723 .2724
proportions
1999–2003: bookmakers’ .4601 .2586 .2813
1999–2003: forecasting .4624 .2745 .2631
model probabilities
3. Half-time/full-time probabilities (unconditional)
HH DH AH HD DD AD HA DA AA
1984–1998: observed .2845 .1650 .0255 .0542 .1685 .0488 .0169 .0990 .1374
proportions
1999–2003: observed .2753 .1540 .0259 .0562 .1627 .0534 .0148 .1056 .1521
proportions
1999–2003: bookmakers’ .2685 .1619 .0297 .0513 .1545 .0528 .0252 .1118 .1444
1999–2003: forecasting .2779 .1583 .0261 .0550 .1683 .0512 .0137 .1013 .1482
model probabilities
4. Full-time probabilities (conditional on half-time outcome)
1984–1998: observed .8000 .3815 .1202 .1524 .3896 .2307 .0476 .2289 .6491
proportions
1999–2003: observed .7949 .3647 .1120 .1624 .3853 .2307 .0427 .2500 .6573
proportions
1999–2003: bookmakers’ .7783 .3781 .1311 .1487 .3609 .2327 .0730 .2611 .6362
1999–2003: forecasting .7885 .3725 .1282 .1678 .3924 .2359 .0437 .2359 .6358
model probabilities
is pi,jHH + pi,jHD + pi,jHA. The drop in the observed proportion of matches in which the
home team was leading between periods (i) and (ii), and the corresponding rise in
the proportion in which the away team was leading, reflects a weakening of the
home-field advantage effect (see Chapter 3, Section 3.2), which is also apparent in
Table 12.2 Mean returns based on bookmakers’ ‘best odds’
2. Bets on each half-time/

1. All bets ranked in order of decimal odds full-time outcome
Decimal Number Mean Mean Mean Mean

odds of bets odds return Bet odds return
All 87543 14.74 −.1529 HH 3.76 −.0758
1.4 – 4.33 8347 3.13 −.0953 DH 5.65 −.1613
4.5 – 5 9636 4.87 −.1023 AH 30.04 −.2346
5.5 9571 5.50 −.0978 HD 17.42 −.0232
6 – 6.5 8140 6.17 −.1283 DD 5.80 −.0691
7 – 16 14224 10.47 −.1000 AD 16.86 −.1047
17 15794 17.00 −.0765 HA 36.46 −.4976
19 – 29 12461 27.27 −.1944 DA 9.14 −.1470
34 – 101 9370 38.11 −.4885 AA 7.50 −.0634
the corresponding data for full-time outcomes shown in Panel 2. Here, the book-
makers’ probability for the home team to win the match is φi,jHH + φi,jDH + φAHi,j ; and
the forecasting model probability is pi,jHH + pDH
i,j + p AH
i,j .
Panel 3 reports the proportions of matches that produced each of the nine pos-
sible half-time/full-time outcomes, and the sample mean values of the correspond-
ing bookmakers’ implicit probabilities and the forecasting model probabilities. In
this case, the model generally comes closer than the bookmakers to replicating
the observed proportions. In particular, the bookmakers overstate the probabil-
ities for the two outcomes that are least likely on average, AH and HA (both of
which involve victory for the team that is trailing at half-time). Similar patterns are
also apparent in the full-time probabilities conditional on the half-time outcome,
shown in Panel 4.
Table 12.2 presents some further descriptive analysis of the bookmakers’ ‘best
odds’ data set. In Panel 1, the 87,543 available bets (= nine possible bets × 9,727
matches) are placed into eight bands based on a ranking in ascending order of the
decimal odds. The mean return is evaluated for the bets in each band. The over-
all mean return of –15.29 per cent is slightly worse than the return suggested by
the mean overround based on the ‘best odds’ of 0.1290.5 The calculation of mean
returns when the ‘best odds’ are classified by their length reveals clear evidence of
a favourite-longshot bias, in the form of mean returns for bets in the two longest
odds bands (decimal odds in the ranges 19 –29 and 34 –101) that are substantially
lower than the mean returns in any of the other six bands. Panel 2 presents a simi-
lar analysis for bets on each of the nine half-time/full-time outcomes individu-
ally. From the bettor’s perspective, bets on the outcomes AH and HA, with mean
returns of –23.5 per cent and – 49.8 per cent respectively, represent poor value on
average (as is also suggested by Table 12.1). Bets on AH and HA account for the
Informational efficiency of fixed-odds betting 371
majority of long-odds bets that are subject to the favourite-longshot bias. From
the bettor’s perspective, bets on the outcome HD, with a mean return of –2.32 per
cent, provide the best value on average.
12.4 Testing the informational efficiency of the half-time/full-time

fixed-odds betting market
Section 12.4 reports an empirical test for the informational efficiency of the half-
time/full-time fixed-odds betting market, in the form of an investigation of the
profitability of a simple betting strategy that selects the bet offering the highest
expected return according to the probabilistic forecasts generated from the model,
from the choice of nine bets that are available on each match. First, the mechanics
of the betting strategy are described, using as an example the ten Premier League
matches that were played on 11 May 2003, the final day of the 2003 season. Then
the results of applying the ‘highest expected return’ strategy across the entire hold-
out sample are presented.
The application of the ‘highest expected return’ strategy on 11 May 2003 is illus-
trated in Table 12.3. Panel 1 shows the forecasting model probabilities, pi,jm, for the
nine possible half-time/full-time outcomes (m = HH, DH, …) for each of the ten
matches. pi,jm are calculated from [12.3] using ŷ*k,i,j for k = 0,1,2,3, μ̂k,1 and μ̂k,2 calcu-
lated by applying the forecasting model estimated with data from seasons 1988 to
2002 to the covariates for each of the ten matches. Panel 2 shows the ‘best odds’
for each of the nine outcomes. Panel 3 shows the forecasting model’s expected
return for each bet placed at ‘best odds’. For example, a £1 bet on a HH outcome
in Birmingham versus West Ham at ‘best odds’ of 6.5 would provide a net return
of +5.5 if the bet wins, and –1 if the bet loses. The expected return is +5.5×.2862
+ –1×(1 – .2862) = 0.860. For each fixture, the highest of the nine expected returns
is shown in italics. On this basis, HH is the model’s recommended bet for three of
the ten matches (including Birmingham versus West Ham); HD is recommended for
four matches; and AA is recommended for three matches.
Panel 3 shows the actual half-time/full-time match outcomes, in the standard
format of ‘(half-time score) full-time score’ for each team. For two of the ten
matches highlighted in bold, Chelsea versus Liverpool (HH) and Manchester
City versus Southampton (AA), the half-time/full-time outcome corresponds to
the recommended bet. The bettor achieves positive net returns of +3.5 and +8
on these two bets respectively (shown in the final column of Panel 3). The other
eight bets are unsuccessful. Across all ten bets, the bettor’s overall net return
is +3.5. These ten bets turn out to be more profitable for the bettor than the
average for the entire holdout sample. As is shown below, however, the return
obtained by applying the same strategy across the entire holdout sample is also
positive.
Table 12.4 presents summary results for the application of the ‘highest expected
return’ strategy across the entire holdout sample. Panel 1 reports the outcome from
Table 12.3 Illustration of ‘highest expected return’ betting strategy: English Premier League, 11 May 2003
1. Fixtures, forecasting model probabilities

Birmingham West Ham .2862 .1401 .0229 .0457 .1693 .0525 .0069 .1080 .1684
Bolton Middlesbrough .3029 .1412 .0273 .0622 .1660 .0526 .0114 .1031 .1333
Charlton Fulham .2743 .1277 .0191 .0683 .1689 .0469 .0143 .1177 .1627
Chelsea Liverpool .2717 .1626 .0332 .0592 .1651 .0606 .0111 .0893 .1473
Everton Man Utd .1377 .0753 .0123 .0632 .1581 .0449 .0200 .1791 .3093
Leeds Aston Villa .2791 .1536 .0330 .0774 .1641 .0566 .0177 .0930 .1254
Man City Southampton .3046 .1639 .0260 .0608 .1615 .0516 .0109 .0850 .1358
Sunderland Arsenal .1002 .0510 .0057 .0758 .1409 .0285 .0383 .2186 .3410
Tottenham Blackburn .2772 .1476 .0262 .0630 .1673 .0542 .0122 .1004 .1520
WBA Newcastle .1326 .0744 .0187 .0634 .1573 .0568 .0208 .1796 .2964
2. Fixtures, best available bookmaker odds
Birmingham West Ham 6.5 8 34 17 6 19 29 4.5 3.4
Bolton Middlesbrough 2.4 4.5 29 19 6.5 17 41 11 10
Charlton Fulham 3.5 4.5 29 17 5.5 17 34 8 6
Chelsea Liverpool 4.5 5.5 29 17 5 17 34 6.5 5
Everton Man Utd 5 6 34 17 6 17 29 5.5 5
Leeds Aston Villa 3.2 4.5 29 17 5.5 17 34 8.5 7
Man City Southampton 2.75 4.5 29 17 6 17 41 10 9
Sunderland Arsenal 9 10 41 17 6 17 29 4.5 2.63
Tottenham Blackburn 5 7.5 34 17 6 17 29 5.5 3.6
WBA Newcastle 7 8.5 34 17 6 17 29 4.5 3.2
3. Match results, expected returns on each bet, selected bets (italics), winning bets (bold italics), actual returns
Birmingham (0) 2 West Ham (0) 2 .860 .121 −.221 −.222 .016 −.003 −.801 −.514 −.427 −1
Bolton (2) 2 Middlesbrough (0) 1 −.273 −.364 −.208 .181 .079 −.105 −.534 .134 .333 −1
Charlton (0) 0 Fulham (1) 1 −.040 −.425 −.445 .161 −.071 −.203 −.515 −.058 −.024 −1
Chelsea (2) 2 Liverpool (1) 1 .223 −.106 −.039 .007 −.175 .031 −.623 −.420 −.264 +3.5
Everton (1) 1 Man Utd (1) 2 −.312 −.548 −.582 .075 −.051 −.236 −.419 −.015 .547 −1
Leeds (1) 3 Aston Villa (1) 1 −.107 −.309 −.044 .315 −.098 −.038 −.397 −.209 −.122 −1
Man City (0) 0 Southampton (1) 1 −.162 −.263 −.247 .034 −.031 −.122 −.553 −.150 .222 +8
Sunderland (0) 0 Arsenal (2) 4 −.098 −.490 −.766 .289 −.155 −.516 .110 −.016 −.103 −1
Tottenham (0) 0 Blackburn (2) 4 .386 .107 −.110 .071 .004 −.078 −.645 −.448 −.453 −1
WBA (0) 2 Newcastle (1) 2 −.072 −.368 −.365 .078 −.056 −.034 −.398 −.192 −.052 −1
Note: Birmingham (0) 2 West Ham (0) 2 means half-time score of 0-0, full-time score of 2-2. Matches for which winning bets were selected,
and actual returns for winning bets, are shown in bold.
Table 12.4 Returns from ‘highest expected return’ and indiscriminate betting
strategies
Betting strategy:
‘Highest expected return’ Indiscriminate
Odds: Best Median Median
No. of Average Average Average Average
No. of winning winning rate of rate of rate of
bets bets odds return return return
1. Results across all matches

9727 1342 7.53 .0390 −.0223 −.2154
2. Results classified by selected bet
HH 2158 524 3.37 .0621 −.0180 −.1401
DH 732 142 4.29 .0260 −.0558 −.2158
AH 867 29 29.10 .0069 .0798 −.2948
HD 1674 99 16.26 .0209 −.0377 −.0937
DD 1082 190 4.77 .0139 −.0772 −.1274
AD 698 50 16.00 .2178 .1759 −.1633
HA 16 2 28.00 2.6250 2.7143 −.5420
DA 350 36 9.46 .0757 .0076 −.2171
AA 2150 270 6.78 −.0232 −.1024 −.1442
3. Results classified by season
1999 1944 328 5.98 .0095 −.0417 −.2265
2000 1946 290 6.85 .0210 −.0234 −.2256
2001 1946 226 9.00 .0458 .0470 −.1806
2002 1945 200 10.24 .0527 −.0320 −.2381
2003 1946 298 6.96 .0662 −.0614 −.2062
4. Results classified by stage of season
Aug. to Oct. 3148 450 7.72 .1037 −.0001 −.2113
Nov. to Dec. 1991 282 7.81 .1060 .0169 −.2267
Jan. to Feb. 1800 227 7.48 −.0570 −.0612 −.2252
Mar. to May 2788 401 7.20 .0349 −.0502 −.1900
5. Results classified by tier (division)
T1 1900 251 7.80 .0303 .0047 −.2325
T2 2760 406 7.40 .0887 −.0926 −.2142
T3 2714 379 7.18 .0022 −.0113 −.2063
T4 2353 324 7.95 .0951 .0257 −.2134
Source: Rothmans/Sky Sports Football Yearbook; www.football-data.co.uk

placing one bet on each of the 9,727 matches at ‘best odds’, on the half-time/full-
time outcome with the highest expected return according to the model. There are
1,342 winning bets (13.8 per cent of the total). The average decimal odds for the
winning bets are 7.53. If £1 is staked in each bet, the overall net return is £379.35
on a total stake of £9,727. Therefore the average rate of return is 379.35/9727 =
0.0390, or +3.9 per cent.
For purposes of comparison, the final two columns of Table 12.4 show the aver-
age rates of return under two alternative scenarios. In the first alternative, the
same betting strategy is applied using ‘median odds’ instead of ‘best odds’. This
produces an average return of –2.2 per cent. In the second alternative, nine bets
are placed on each of the 9,727 matches indiscriminately: one bet on each of the
nine possible outcomes. This produces an average return of –21.5 per cent (roughly
equivalent to the average of the overrounds implicit in the five sets of bookmaker
odds). Overall, a bettor who adopts the ‘highest expected return’ strategy using
‘best odds’ achieves a rate of return 25.4 per cent higher than a bettor who places
bets indiscriminately. Of this gain, 19.3 per cent can be attributed to the use of the
forecasting model to select bets with favourable expected returns, and 6.1 per cent
can be attributed to arbitrage between the five sets of bookmaker odds, using the
‘best odds’ rather than the ‘median odds’ for each bet.
The remaining panels of Table 12.4 show the equivalent results when the hold-
out sample of 9,727 matches is disaggregated in various ways: in Panel 2 by the
half-time/full-time outcome that was the selected bet according to the ‘highest
expected return’ criterion; in Panel 3 by the five football seasons within the hold-
out sample period chronologically; in Panel 4 by the stage of the season (identified
by calendar month) when the match was played; and in Panel 5 by the tier (or div-
ision) in which the match was played.
The classification by the half-time/full-time outcome that was the selected bet
indicates that the proportions in which the nine possible bets are selected do not
correspond closely either to the proportions in which the nine outcomes occur, or
to the average differences between the bookmakers’ implicit probabilities and the
forecasting model probabilities (see Table 12.1, Panel 3). For example, the rates of
occurrence of the two least common outcomes, AH and HA, are 2.6 per cent and
1.5 per cent; but AH and HA are the selected bets for 8.9 per cent and 0.2 per cent
of all matches. The bookmakers’ implicit probabilities for AH bets tend to over-
state the actual proportions, so AH bets offer poor value on average. However,
while AH bets would perform poorly if placed on all matches indiscriminately, it
is not unusual to find particular matches for which the forecasting model indicates
that an AH bet represents good value, in the form of a higher expected return than
for any of the other eight bets. Therefore the relatively high proportion of AH bets
chosen is due to the forecasting model’s selectivity capability.
The classification by season is of practical importance for bettors who might
realistically hope to achieve a positive return over a period perhaps considerably
shorter than five years. Although the average return is positive for all five seasons
Table 12.5 Average rate of return by tier (division): chronological analysis
T1 T2 T3 T4 All matches
08/98 to 10/98 .3623 .1151 .1854 −.1141 .1143

11/98 to 12/98 −.3069 .1387 −.2584 .1222 −.0596
01/99 to 02/99 −.0522 .4427 −.1861 .1170 .0831
03/99 to 05/99 −.2260 −.1420 −.1640 .1036 −.0999
08/99 to 10/99 .1655 .2457 −.4823 .0712 −.0067
11/99 to 12/99 .0988 −.2132 .5461 −.0819 −.0560
01/00 to 02/00 −.3333 −.1563 −.0009 −.2505 −.1641
03/00 to 05/00 −.0858 .1329 .0521 .7831 .2154
08/00 to 10/00 .3739 −.0821 −.2534 .5367 .1038
11/00 to 12/00 .5313 .1073 .2283 −.1107 .2030
01/01 to 02/01 .1957 .0733 −.4252 −.2451 −.1336
03/01 to 05/01 −.0090 .2211 −.1108 −.1496 −.0104
08/01 to 10/01 .1505 .4622 −.0559 −.3024 .0735
11/01 to 12/01 .6000 −.1852 .3220 −.2922 .1036
01/02 to 02/02 −.2887 .0564 .1420 .1115 .0371
03/02 to 05/02 −.0054 −.1000 .1793 −.1079 −.0061
08/02 to 10/02 −.1780 −.1024 .1528 .3914 .0832
11/02 to 12/02 .0216 −.1788 .1701 .4153 .0719
01/03 to 02/03 .3577 −.0301 −.3261 .6249 .1166
03/03 to 05/03 −.6420 .1136 .2041 .1430 .0120
individually, this result is borderline for the 1999 season. There appears to be suffi-
cient year-to-year variation in the results to suggest the probability of achieving a
negative average return in any individual season may be non-negligible. It is inter-
esting to note that there is no evidence of any improvement in the informational
efficiency of the half-time/full-time betting market within the five-year holdout
sample period. In fact, the ‘highest expected return’ betting strategy becomes pro-
gressively more profitable over time.
The classification by stage of the season and by tier (division) suggests the ‘high-
est expected return’ betting strategy was more effective over the earlier months of
the football season than over the later months; and more effective for bets on T2
and T4 matches than for bets on T1 and T3 matches. In an effort to investigate the
stability of these patterns, Table 12.5 reports a more detailed analysis of the aver-
age returns, disaggregated chronologically and by tier (division). The variations in
average returns by stage of the season and by tier do not appear to be at all stable
at the levels of aggregation that are used to construct Table 12.5. Accordingly, the
patterns in the average returns suggested by Panels 4 and 5 of Table 12.4 appear
more likely to be due to sampling error than to any factors that operate systemat-
ically. In fact, the results in Table 12.5 demonstrate that there is considerable small-
sample variation in the returns. This suggests that with the benefit of hindsight it
Table 12.6 Mean expected and actual returns: all possible bets ranked by expected
return
Band (ordered by Mean Mean expected Mean actual

expected return) odds return return
Best 9,727 bets 13.29 +.2223 +.0564

2nd-best 9,727 10.91 +.0487 −.0763
3rd-best 9,727 10.58 −.0222 −.0707
4th-best 9,727 10.23 −.0789 −.1207
5th-best 9,727 10.88 −.1329 −.1193
6th-best 9,727 11.60 −.1922 −.1656
7th-best 9,727 13.18 −.2649 −.2134
8th-best 9,727 18.50 −.3692 −.2208
Worst 9,727 bets 33.48 −.5906 −.4463
is not at all difficult to identify betting strategies that would have been locally
profitable, over limited numbers of bets placed over specific periods that are also
selected with the benefit of hindsight. An implication for future research is that it
is important for any claims concerning the out-of-sample or universal profitability
of particular betting strategies to be validated using large-sample evidence.
Finally, Table 12.6 reports the mean returns earned by ranking all 87,543 avail-
able bets (= nine possible bets × 9,727 matches) in descending order of their
expected return evaluated using the forecasting model probabilities. For conveni-
ence, the 87,543 bets are placed into nine equal-sized bands, each containing 9,727
bets. (It is possible, indeed common, for more than one bet on the same match to
appear in the same band.) While there is a clear association between the expected
and actual returns, the gradient (based on the change in values between successive
bands in Table 12.6) for expected returns is generally steeper than that for actual
returns. Bets in the top band by expected return produce a positive actual return
of +5.64 per cent, slightly higher than the mean return of +3.90 per cent produced
by the ‘highest expected return’ strategy that is analysed in Tables 12.4 and 12.5.
Although Table 12.6 could not have been compiled by a bettor operating in real
time (since knowledge of the outcomes over the entire five-season holdout sam-
ple period is required), a rule-of-thumb involving the selection of bets with an
expected return exceeding about +9 per cent or +10 per cent would have produced
a portfolio of bets approximating closely to the top band of Table 12.6.
Conclusion
The theoretical and empirical economics literature contains numerous studies of
the informational efficiency of betting markets. Stylised facts that emerge from this
literature include the following. The market prices or odds for racetrack betting,
and for some team sports betting, are subject to a consistent and longstanding
favourite-longshot bias, such that bets placed on favourites yield a higher average
return than bets placed on longshots or underdogs. There is some evidence of a
reverse favourite-longshot bias in the betting markets for North American major
league sports; but this pattern is not replicated for fixed-odds betting on football
match outcomes. Reasons for the existence of favourite-longshot bias, or its rever-
sal, have proven surprisingly difficult to pin down. Explanations popular in the
academic literature include risk-preference theories and information asymmetry
theories. There is evidence that a contrarian strategy of betting against teams cur-
rently undergoing a winning streak may be profitable in some cases, in defiance
of belief in the ‘hot-hand’ phenomenon. There is also some evidence that pricing
anomalies in betting markets have a tendency to correct themselves over time, as
bettors and bookmakers learn from past experience.
The growth of internet betting during the 1990s and 2000s has presented tre-
mendous new betting opportunities, based on the rapid information processing
and transmission capabilities of the new technology. It has also presented major
challenges to the traditional betting firms, which have experienced fierce new com-
petition and a squeeze on margins. Person-to-person betting, transacted through
an online betting exchange, has been the most successful new business model to
have emerged during the internet era. The increasing popularity of sophisticated
new spread betting markets for football (and other sports) has also been fuelled
by the development of internet technology. For all of the improvement in infor-
mation and communication technology, however, recent literature on sentiment
bias in sports betting suggests that bettors are not immune to a tendency for their
judgement to be sometimes coloured by their emotions.
This chapter has reported tests for the informational efficiency of the fixed-
odds betting market on the half-time/full-time match outcomes of English league
football matches. In this market, bettors choose between nine alternative bets on
any match, corresponding to nine possible permutations for the half-time and full-
time match outcome. With nine (rather than three) bets available on each match,
this betting market appears to provide more scope for detecting anomalies in the
bookmakers’ odds and for formulating profitable betting strategies than the fixed-
odds market for full-time match results, which has been the subject of previous
English football betting market efficiency studies.
Probabilistic forecasts for the nine possible half-time/full-time match outcomes
are obtained from a forecasting model that is estimated using data from the fif-
teen football seasons prior to the current season in which the match in question
takes place. The estimated model is updated at the start of each season. The model
combines ordered probit regressions for the half-time outcome, and the full-time
outcome conditional on each of the three possible half-time outcomes, using cov-
ariates based on information that would have been publicly available at the time
bets were being accepted immediately prior to the match being played.
The profitability of a relatively simple betting strategy is investigated. The strat-

egy involves placing a bet on every match on the half-time/full-time outcome for
which the best odds available from any of the five sets of bookmaker odds produce
the highest expected return according to the probabilities generated from the fore-
casting model. This strategy would have been profitable over a five-season holdout
sample period covering football seasons 1999 to 2003, inclusive, and for each of
these five seasons individually. It is claimed that these results constitute stronger
evidence than has been available previously that it is possible to use historical
data to formulate a betting strategy that is universally (rather than just locally)
profitable.
Notes
1 This type of constraint does not arise in portfolio theory, because an assumption of perfect
capital markets allows investors to either borrow or lend without any limits at an interest rate
equivalent to the risk-free rate of return.
2 The parameter σ plays a role in the model analogous to that of Heckman’s (1979) correction
for sample selection bias, exploiting the information from the estimation of the (uncondi-
tional) half-time outcome probabilities in order to correct for selection bias in the estima-
tions of the full-time probabilities conditional on the half-time outcome. For example, the
estimated probabilities for the full-time outcome conditional on the away team leading at
half-time are based on a non-random subsample of matches in which the away team was
leading at half-time. If in one particular match, the home team was expected to dominate,
but the away team was actually leading at half-time, ε(F) i,j must have been large and negative.
If σ>0, ε(S)
i,j is also expected to be large and negative. This information should be taken into
account in the estimation of the probability for the full-time outcome conditional on the
away team having been leading at half-time. Since σ is constrained to lie between –1 and
+1, it is convenient to estimate δ = atanh(σ), which is unconstrained. The estimated value
of the parameter σ is obtained using the transformation σ̂ = tanh( δ) ˆ = {exp( δ)−exp(−
ˆ ˆ
δ)}/
ˆ
{exp ( δ)+exp (− ˆ
δ)}.
3 This assumption need not necessarily be true. For example, bookmakers might set their
odds to exploit patterns of demand among bettors that are influenced by factors other
than expected return. If so, the probabilities used to set the odds might diverge from φmi,j;
and based on these probabilities some bets might yield higher expected returns than others.
Nevertheless, below we refer to φmi,j as the bookmakers’ implicit probabilities, invoking the
assumption of equal expected returns implicitly whenever we do so.
4 From 10,180 league matches played in total, the model is capable of generating predictions
for 9,728. The model does not generate predictions for matches involving teams that entered
the league up to two calendar years before the match is played, for which the full set of half-
time/full-time outcomes data over a two-year period prior to the match in question were not
available. In the 2002 season, one T4 fixture was rescheduled with insufficient notice for the
bookmakers to quote odds; this match is discarded from the sample.
5 The overround based on ‘median odds’, which corresponds closely to the margins built into
each individual set of bookmaker odds, is 0.2135.
13 Football around the world:
France, Germany, Brazil, Japan
and China
Introduction
Previous chapters of this volume have presented an extensive descriptive and
empirical analysis of English club football. Chapter 13 widens the perspective, by
providing a brief description of the historical development and present-day com-
petitive and commercial structure of football in five other countries from around
the world. Sections 13.1 and 13.2 examine two further major Western European
footballing powers: France and Germany. Section 13.3 examines Brazil, whose
national team’s highly skilled attacking style of play in several successive World
Cup tournaments, especially during the 1950s, 1960s and 1970s, gained plau-
dits and admirers throughout the world. Finally, Sections 13.4 and 13.5 examine
Japan and China, two east Asian countries in which attempts were made during
the 1990s and 2000s to establish professional football as a popular spectator sport
in territories with none of the longstanding traditions of participation and fanati-
cism that have characterised football in many other countries.
13.1 France
Football was first introduced into France, as in a number of other countries, by
travelling British businessmen during the late nineteenth century. The first club
was founded in the port city of Le Havre in 1872 by members of shipping and
transit companies who wished to play both football and rugby. A number of other
multi-sports clubs were formed in the 1870s and 1880s, before clubs devoted exclu-
sively to football began to appear during the 1890s. The first national cup tour-
nament was won by Standard AC in 1894, and an embryonic league competition
was won by Club Français in 1896. By the turn of the century, football was played
extensively throughout northern France, including Paris (Pickup, 1999).
At the start of the twentieth century, football spread rapidly to the east and south
of France. France’s first international was played against Belgium in 1904, but the
administration of the sport during the pre-war period was chaotic, due to rivalries
between a number of alternative governing bodies. This situation was resolved in
380
France 381
1919 by unification of the competing bodies under the Federation Française de

Football (FFF). A French official, Jules Rimet, was the first president of the newly
created international football federation FIFA when it was formed in 1920. Rimet
played a key part in the creation of the first international World Cup tournament,
played in Uruguay in 1930.
The legal framework for the organisation of professional football clubs in
France was established by the 1901 freedom of association law, which permitted
the formation of associations of citizens on a non-profit basis for any purpose.
Clubs such as Sochaux and RC Paris, dominated by local business personalities
seeking influence and prestige (rather than financial reward), became the norm
from the 1930s onwards. From the outset, there was also a strong tradition of local
government involvement in the financing and administration of football (includ-
ing municipal ownership of most major football stadia), deriving from a statutory
duty for municipalities to promote and develop sport.
In domestic football, a major advance came with the acceptance of profes-
sionalism in 1932, followed by the introduction of a national league competition
(currently named Ligue 1), won by Sête in its inaugural 1934 season. The 1935
winners were Sochaux, a club which had been formed in 1929 on a professional
basis with the backing of the Peugeot car manufacturer. This development had
been instrumental in prompting the authorities to adopt professionalism. Racing
Club (RC) Paris, formed by a Parisian estate agent in 1932, were champions in
1936. The 1930s also saw a trend towards the recruitment of foreign players. The
enforced transfer of players against their will encouraged the formation of a play-
ers’ mutual association in 1934, and a players’ union in 1936.
Before the introduction of fixed-term contracts in 1968, professional players
were tied contractually to a single club up to the age of 35. This draconian con-
tractual position may explain football’s relatively unattractive status as a work-
ing-class occupation following the introduction of professionalism: most players
were middle class and semi-professional. The proportion of imported players
was relatively high: 329 foreigners played in the top tier of French football before
1939. The pattern was similar after the Second World War (Lanfranchi, 1994).
Indigenous working-class involvement increased gradually, however, during the
post-war period (Eastham, 1999; Pickup, 1999; Mignon, 2000).
Since 1945 Ligue 1 has switched several times between an eighteen- and twenty-
team format. The current twenty-team Ligue 1 format has operated since the 2003
season. Promotion and relegation between Ligue 1 and Ligue 2 and between Ligue
2 and Ligue 3 is three-up three-down. Before automatic promotion and relega-
tion between Ligue 2 and Ligue 3 was introduced in 1970, membership of Ligue
3 included the reserve teams of Ligue 1 clubs. Four-up four-down promotion and
relegation operates between Ligue 3 and the Championnat de France Amateurs.
Between 1973 and 1976 a bonus league point was awarded if a team scored three
goals or more, regardless of the match result. Three points for a win was intro-
duced in the 1995 season.1
382 Football around the world
Table 13.1 Historical performance of top French teams in league and cup
competition
1946–1959 1960–1969 1970–1979 1980–1989 1990–1999 2000–2009
Championships
Stade Reims 4 St Etienne 4 St Etienne 4 Bordeaux 3 Marseille 4 Lyon 7
Nice 4 Stade Reims 2 Marseille 2 Monaco 2 Bordeaux 1 Bordeaux 1
Lille 2 Monaco 2 Nantes 2 Nantes 2 Lens 1 Monaco 1
St Etienne 1 Nantes 2 Strasbourg 1 St Etienne 1 Monaco 1 Nantes 1
Roubaix 1 Monaco 1 Marseille 1 Nantes 1
Bordeaux 1 Paris SG 1 Paris SG 1
Marseille 1 Auxerre 1
Championship – points for a top three finish
Stade Reims 19 St Etienne 12 St Etienne 15 Bordeaux 13 Marseille 16 Lyon 25
Lille 15 Stade Reims 9 Nantes 13 Nantes 12 Monaco 11 Bordeaux 7
Nice 12 Monaco 8 Marseille 11 Monaco 10 Paris SG 11 Monaco 7
Bordeaux 7 Nantes 8 Strasbourg 4 St Etienne 6 Bordeaux 5 Marseille 6
St Etienne 5 Bordeaux 6 Nice 4 Paris SG 6 Auxerre 5 Paris SG 4
Lens 5 RC Paris 5 Monaco 3 Marseille 5 Nantes 4 Lille 4
Roubaix 4 Nimes 3 Sochaux 2 Sochaux 3 Lens 3 Lens 2
Marseille 4 Nice 2 Lens 2 Roubaix 2 Lyon 3 Toulouse 1
Nimes 4 Valenciennes 2 Nimes 2 Auxerre 1 Metz 2 Auxerre 1
Sochaux 2 Sochaux 1 Lyon 2 Montpellier 1
Toulouse 2 Lens 1 Sedan 1 Toulouse 1
Strasbourg 1 Angers 1 Bastia 1
Le Havre 1 Sedan 1
Monaco 1 Metz 1
Angers 1
RC Paris 1
French Cup wins
Lille 5 Monaco 2 St Etienne 4 Monaco 2 Paris SG 3 Auxerre 2
Stade Reims 2 St Etienne 2 Marseille 2 Paris SG 2 Auxerre 2 Paris SG 2
Nice 2 Lyon 2 St Rennais 1 Metz 2 Nice 1 Sochaux 1
Strasbourg 1 CS Sedan 1 Lyon 1 Bordeaux 2 Nantes 1 Lyon 1
RC Paris 1 Stade Rennais 1 Nantes 1 Marseille 1 Montpellier 1 Guingamp 1
CS Sedan 1 Strasbourg 1 Nancy Lorr. 1 Bastia 1 Monaco 1 Nantes 1
Toulouse 1 Marseille 1 Strasbourg 1
Le Havre 1 Lorient 1
League Cup wins (from 1995)
Paris SG 2 Bordeaux 3
Metz 1 Lyon 1
Strasbourg 1 Monaco 1
Lens 1 Sochaux 1
Strasbourg 1
Nancy 1
Paris SG 1
Gueugnon 1
France 383
Table 13.1 (cont.)
1946–1959 1960–1969 1970–1979 1980–1989 1990–1999 2000–2009
Eurpean Competition wins (from 1956)

Marseille 1
Paris SG 1
Note: Marseille’s 1993 domestic championship and European Cup victories were subsequently
rescinded, but are included here. The French Cup was cancelled at the semi-final stage in 1992,
following the collapse of a stand at the Armand Césari Stadium in Bastia (Corsica), during a
match between Bastia and Marseille. Eighteen people were killed in the incident.
Source: French Professional League, www.ligue1.com/indexSite.asp
Table 13.1 presents data on the performance of French clubs in domes-

tic and European competition. In comparison with many other European
countries, success in French domestic football has not been highly concen-
trated. Stade Reims, the most successful team of the 1950s and early 1960s,
were eventually relegated in 1967 and have never returned to the top tier. The
dominance of St Etienne, champions ten times between 1957 and 1981 and
a major force in European competition during the 1970s, had run its course
by the early 1980s. The club was relegated in 1996, but returned to Ligue 1
in 2004. Spells of success for Bordeaux and Marseilles in the 1980s and
1990s ended abruptly in scandal and enforced relegation in both cases. Since
2000 Lyon have been the dominant force and the wealthiest club in French
football.
French officials were highly influential in the development of the first
European club tournaments during the 1950s, as they had been in establishing
the World Cup thirty years earlier. Stade Reims were runners up to Real Madrid
in the first European (Champions) Cup tournament in 1956, and again in 1959.
In domestic football, however, the 1960s and 1970s were a period of retrench-
ment. The successes of St Etienne in domestic competition took place against a
background of declining spectator interest and financial crisis on the domestic
scene. Population decline in many smaller industrial towns led to reductions in
municipal funding, and several clubs reverted to amateur status (Pickup, 1999;
Mignon, 2000).
The French tradition of state involvement in the organisation of sport takes
some of the credit for steps taken during the 1970s that have subsequently con-
tributed to a renaissance in the fortunes of French football, both at club and
(especially) at international level. In 1974 the National Football Institute was
established to provide training to forty of the most promising young players. At
the same time, all professional clubs were required to set up their own training
centres with links to local schools and other community organisations (Eastham,
1999). At international level, many of the products of this system went on to
achieve unprecedented success. France were World Cup semi-finalists in 1982

and 1986, World Cup winners in 1998, World Cup finalists in 2006 and European
Champions in 1984 and 2000. Domestically the benefits were more ambiguous,
with many of the leading players choosing to enhance their earnings by playing for
Italian, Spanish, German or British clubs, rather than remaining with their (rela-
tively impoverished) French counterparts. Naturally, the Bosman ruling has made
this ambition easier to achieve.
Weaknesses in the administrative and financial control of French club foot-
ball were apparent following the implication of officials of St Etienne, Paris Saint
Germaine and Bordeaux in a series of financial scandals during the 1970s and
1980s. In 1984, a change to the legal constitution of the larger professional clubs
permitted greater involvement of private investors, while still preventing prof-
its from being paid to chairmen and directors. Several early experiences of the
involvement of the 1980s breed of entrepreneur were far from happy, however.
Most notorious was Bernard Tapie’s flamboyant reign as chairman of Olympique
Marseilles, which won five successive domestic championships between 1989 and
1993 (the last of which was subsequently rescinded). Marseille were also stripped
of their victory in the 1993 European Cup. This phase ended in enforced relega-
tion for the club in 1994, and in Tapie (and several other club officials) being jailed
in 1995 for a variety of bribery and corruption offences including match-fixing
during the 1993 season.
In 1991 the National Directorate for the Control of Management of Sporting
Bodies (DNCG) was set up, comprising independent members nominated by the
clubs and other organisations with interests in football, and with representatives from
the accountancy profession. The DNCG has powers to intervene in cases of alleged
illegal payments, accounting fraud or tax evasion (Amara et al., 2005). As part of its
supervisory role, it also has powers to block any spending that may cause a financial
imbalance. The DNCG is authorised to place a club under temporary suspension,
and ban all transfer activity, if it is dissatisfied with a club’s finances. If DNCG dir-
ectives are ignored, a club can be relegated to a lower tier (Eastham, 1999).
The commercial development of French football has lagged behind several other
major Western European countries. Mignon (2000) suggests there is an underlying
difference between the football culture in France (based on Jacobin, republican
values) and the more financially driven ethos that exists elsewhere. Nevertheless, the
1990s witnessed a strong revival in the fortunes of French football at club level, which
has been sustained during the 2000s. Table 13.2 reports the trend in Ligue 1 average
attendances for seasons 1981 to 2009 (inclusive). Ligue 2 recorded an increase in
average attendances during the 2000s, from 5,000 in 2001 to 8,600 in 2009.
Table 13.3 reports data on the principal categories of revenue and expenditure of
Ligue 1 clubs since the late 1990s. Between 1999 and 2008 total revenue increased
by 151 per cent. In 1999 matchday income and revenue from sponsorship and
advertising accounted for 41.8 per cent of total revenue; in 2008 the corresponding
Germany 385
Table 13.2 Average league attendances, France Ligue 1 and German Bundesliga 1,
1981–2010 (’000)
Season Ligue 1 Bundesliga 1 Season Ligue 1 Bundesliga 1
1981 11.4 24.5 1996 16.0 30.1

1982 10.8 22.1 1997 16.4 30.5
1983 11.5 21.8 1998 17.9 30.4
1984 11.2 20.7 1999 19.3 31.9
1985 10.9 19.9 2000 22.3 31.2
1986 10.9 19.2 2001 22.9 30.8
1987 12.6 21.4 2002 21.8 32.7
1988 13.5 20.0 2003 19.6 33.8
1989 11.8 19.0 2004 20.1 37.5
1990 14.3 21.8 2005 21.3 37.8
1991 12.2 21.8 2006 21.5 40.7
1992 12.4 24.7 2007 21.8 40.0
1993 15.1 26.8 2008 21.8 39.4
1994 13.1 27.1 2009 21.1 42.6
1995 14.6 29.6 2010 20.1 42.5
Source: http://rernes.free.fr./; www.european-football-statistics.co.uk/attn.htm
figure was 33.3 per cent. The share of broadcast revenue in total revenue increased
from 41.8 per cent to 56.3 per cent over the same period. In 2009 the average rev-
enue per Ligue 1 club was €52.4 million. Lyon and Marseille are regularly listed
among Europe’s twenty wealthiest clubs, with revenues in 2009 of €139.6 million
and €133.2 million, respectively (Deloitte, 2010). Between 1999 and 2008, total
costs increased by 123 per cent. The share of wages and salary expenditure in
total costs increased from 45.0 per cent in 1999 to 54.5 per cent in 2008. Taking
transfer expenditure and receipts into account, Ligue 1 clubs recorded aggre-
gate losses in most of the seasons between 1998 and 2008. The deficit would have
been larger without a healthy positive contribution from net transfer expenditure.
Ligue 1 clubs carry a significant debt burden, although the ratio of assets to debts
improved markedly towards the end of the 2000s.
13.2 Germany
The official history of football in Germany began with the formation of the German
Football Association (DFB) in Leipzig on 28 January 1900, with eighty-six mem-
ber clubs. Prior to the formation of the Bundesliga in the 1960s, German football
was played at an amateur level in a large number of subregional leagues (Oberliga).
Regional champions and, from 1925 onwards, runners-up played a series of play-off
matches for the right to compete in a final game for the national championship.
Table 13.3 Revenue, costs and profitability, aggregates and breakdown, France, all
Ligue 1 clubs, selected years, € thousands
1999 2002 2005 2008
Matchday income 84,877 97,552 131,264 136,971

Sponsorship and advertising 79,583 126,541 142,735 191,832
Broadcast rights 164,173 332,810 343,874 556,737
Other 64,549 86,186 78,522 103,048
Total revenue 393,183 643,089 696,394 988,588
Wages and salaries 208,203 340,505 339,978 562,346
Social security 65,298 101,038 98,811 140,901
Other 189,220 299,830 274,316 328,211
Total costs 462,721 741,373 713,105 1,031,458
Transfer fee balance 65,885 −68,080 3,018 133,231
Pre-tax accounting balance −7,081 −46,324 −32,470 24,944
Note: Eighteen clubs in 1999 and 2002, twenty clubs in 2005 and 2008.
Source: French Professional League, www.lfp.fr/dncg/index.asp
After the Second World War, Oberliga play resumed in 1946 on a regional basis
in the south and south-west of West Germany. Berlin and the other regions fol-
lowed. FC Nuremberg defeated FC Kaiserslautern to become the first post-war
national champions in 1948. There were two fundamental problems with the
Oberliga system. Over time, each regional Oberliga became increasingly domi-
nated by a small number of teams. A lack of real competition caused German
clubs to lag behind their European counterparts, with teams from countries with
professional national leagues, such as Italy and Spain, dominating the early years
of European club competition. Germany’s disappointing quarter-final exit at the
1962 World Cup in Chile strengthened the case for a national league; and in July
1962 an agreement was reached to launch a league of sixteen clubs in the 1964
season. The original Bundesliga membership included five clubs each from the
Oberliga South and Oberliga West, three from the Oberliga North, two from the
Oberliga Southwest, and one from the Stadtliga Berlin. As well as sporting merit,
each club’s economic situation was taken into account; and initially no city was to
be represented by more than one club.2
Presently, eighteen teams compete in each of the top two tiers of the Bundesliga.
Automatic promotion and relegation between tiers 1 and 2 is two-up two-
down, with a possible third promotion and relegation place decided in a play-
off between the third-bottom team in Bundesliga 1 and the third-placed team
from Bundesliga 2. In 2009 a new third tier, Liga 3, was launched, to sit between
Bundesliga 2 and the Regionalliga in the league pyramid. Unlike Bundesliga 1 and
2, Liga 3 is run directly by the DFB. The rules governing promotion and relegation
between Bundesliga 2 and Liga 3 are the same as those for Bundesliga 1 and 2.
Three-points-for-a-win was adopted in 1996.
Germany 387
In 2001, control of the Bundesliga was removed from the DFB and awarded to
the newly formed German Football League (DFL). The DFL, while subordinate
to the DFB, manages Germany’s professional leagues, and is responsible for issu-
ing licences to clubs, fiscal oversight of the Bundesliga and selling the marketing
rights for the two top tiers. In 2005, German football was overshadowed by a
match-fixing scandal involving referee Robert Hoyzer, who confessed to fixing and
betting on matches in Bundeliga 2, the German Cup and the Regionalliga. Hoyzer
was banned for life and received a 29-month prison sentence. Other officials, play-
ers and a group of Croatian-based gamblers were implicated. Despite the scandal,
Bundesliga 1 has remained popular with fans. A new attendance record was set
in 2009, with a total attendance of 12.8 million and a per-game average of 42,600
(see Table 13.2). Bundesliga 1 is the world’s most popular national football league
by average attendance.3
Table 13.4 presents data on the performance of German clubs in domestic and
European competition. Although seven different teams won the first seven cham-
pionships following the establishment of Bundesliga 1 as a national competition in
the 1964 season, one of these seven teams, Bayern München, subsequently emerged
as the dominant force in German club football. Borussia Moenchengladbach were
also a powerful force during the 1970s, winning five championships (in 1970, 1971,
1975, 1976 and 1977) to Bayern’s three (1972, 1973 and 1974). During the 1980s,
1990s and 2000s, however, Bayern’s ascendancy has been challenged only briefly,
by Hamburg during the 1980s, by Borussia Dortmund, Kaiserslauten and Werder
Bremen during the 1990s and 2000s, and by Stuttgart who have won the cham-
pionship on three occasions, one in each decade. Former German winners of the
European Cup or Champions League are Bayern München (1973, 1974, 1975,
2001), Hamburg (1983) and Borussia Dortmund (1997).
German clubs were traditionally constituted as registered associations (eingetra-
gener Verein). The eingetragener Verein is a non-profit member organisation, man-
aged by representatives democratically elected by the members. Liability is limited
to the assets of the Verein, and individual members are not personally liable. Assets
may not be distributed to members. Neither the Verein nor individual membership
rights can be sold. Dietl and Franck (2007) argue that this legal structure creates
problems. There is no profit incentive, because the proceeds of the Verein cannot
be distributed. Club members have a strong incentive to reinvest any surpluses, and
borrow against the club’s future, because competitive success is the only personal
reward. Decision-making authority rests with the members; but because support-
ers are numerous and heterogeneous, there is no simple and reliable mechanism
for the aggregation of preferences. The outcome is a governance vacuum. Elected
representatives tend to seize control, and derive utility from competitive success
without bearing any financial responsibility. This creates a tendency to discount
future liabilities heavily, in an attempt to maximise current performance.
Since 1998 professional clubs have been able to operate as incorporated busi-
nesses. So far only a minority of the thirty-six Bundesliga clubs (first and second
Table 13.4 Historical performance of top German teams in league and cup competition
1964–1969 1970–1979 1980–1989 1990–1999 2000–2009
Championships
Köln 1 Bor Moench’ng’bach 5 Bayern München 6 Bayern München 4 Bayern München 6
Werder Bremen 1 Bayern München 3 Hamburg 2 Borussia Dortmund 2 Borussia Dortmund 1
Bayern München 1 Köln 1 Werder Bremen 1 Kaiserslauten 2 Werder Bremen 1
1860 München1 Hamburg 1 Stuttgart 1 Werder Bremen 1 Stuttgart 1
Eintr Braunschweig 1 Stuttgart 1 Wolfsburg 1
Nurnberg 1
Köln 5 Bor Moench’ng’bach 20 Bayern München 21 Bayern München 20 Bayern München 22
Werder Bremen 5 Bayern München 14 Hamburg 14 Borussia Dortmund 9 Werder Bremen 9
1860 München 5 Köln 5 Werder Bremen 10 Kaiserslauten 8 Schalke 04 7
Borussia Dortmund 4 Hertha Berlin 5 Stuttgart 6 Werder Bremen 6 Stuttgart 5
Bayern München 4 Hamburg 5 Köln 6 Bayer Leverkusen 6 Bayer Leverkusen 5
Eintr Braunschweig 3 Schalke 04 4 Bor Moench’ng’bach 2 Eintracht Frankfurt 3 Borussia Dortmund 5
Nurnberg 3 Fortuna Düsseldorf 2 Bayer Uerdingen 1 Stuttgart 2 Wolfsburg 3
Duisberg 2 Stuttgart 2 Köln 2 Hamburg 2
Bor Moench’ng’bach 2 Eintracht Frankfurt 1 Hertha Berlin 1
Alemannia Aachen 2 Eintr Braunschweig 1 Schalke 04 1
Eintracht Frankfurt 1 Kaiserslauten 1 Freiburg 1
German Cup wins
Bayern München 3 Eintracht Frankfurt 2 Bayern München 3 Werder Bremen 3 Bayern München 5
Borussia Dortmund 1 Köln 2 Eintracht Frankfurt 2 Kaiserslauten 2 Schalke 04 2
1860 München 1 Schalke 04 1 Bayer Uerdingen 1 Bayer Leverkusen 1 Werder Bremen 2
Köln 1 Bor Moench’ng’bach 1 Borussia Dortmund 1 Hannover 96 1 FC Nuremburg 1
Kickers Offenbach 1 Köln 1 Bor Moench’ng’bach 1
Bayern München 1 Fortuna Düsseldorf 1 Stuttgart 1
Fortuna Düsseldorf 1 Hamburg 1 Bayern München 1
Hamburg 1
European Competition wins
Borussia Dortmund 1 Bayern München 3 Eintracht Frankfurt 1 Borussia Dortmund 1 Bayern München 1
Bayern München 1 Bor Moench’ng’bach 2 Hamburg 1 Bayern München 1
Hamburg 1 Bayer Leverkusen 1 Schalke 04 1
Werder Bremen 1
Note: Data from 1964 to 1991 are for West Germany.

Source: www.rsssf.com/histdom.html; www.rsssf.com/ec/ecomp.html
tiers) have become incorporated; and for those that have, the governance mecha-
nisms of regular corporations do not apply. German football regulations require
that in floated clubs the Verein must hold 50 per cent plus one vote of the corpor-
ation (the ‘50+1 rule’). This restriction means that even if one individual acquired
all of the share capital, a majority of the votes would remain under control of the
Verein.
Borussia Dortmund provides a prominent example of this governance model.
In April 2000 Borussia Dortmund hived off its football team by creating Borussia
Dortmund GmbH & Co. KGaA, a hybrid corporate form combining elements
of a German stock corporation and a limited partnership.4 The general partner,
Borussia Dortmund GmbH, is completely owned by Borussia Dortmund’s Verein
and manages the company’s business as a permanently functioning executive. The
shareholders are not allowed to exercise control through the supervisory board of
the hybrid with respect to personnel or substantive management decisions (Dietl
and Franck, 2007).
German football operates a licensing system, operated by the national foot-
ball association, but without any form of external control and with the football
association not permitted to challenge the data provided by the clubs. In 2002 the
national association refused to issue a licence to Eintracht Frankfurt, because of
doubts over a bank guarantee to cover a €4 million shortfall. An internal court of
arbitration subsequently ruled in the club’s favour, and concluded that licensing
decisions must be based only on the data provided by the clubs. External scru-
tiny of the financial reporting of German football clubs is weak. Pure member
clubs are not required to publish accounts. Some incorporated clubs have not been
punished for failing to report, even though they are required to do so (Brase and
Reichart, 2005). German society tolerates weak law enforcement because football
is perceived to be other than a business. FC Kaiserslautern, for example, received
a licence between 1997 and 2002 because it succeeded in hiding large salary pay-
ments (over €20 million) to star players. The national association fined the club
€125,000 and deducted three points for the 2004 season. The club later received
financial support from its home state, even though the hidden payments consti-
tuted tax fraud (Dietl and Franck, 2007).
Table 13.5 indicates that strong growth in revenues for Bundesliga 1 clubs since
the late 1990s has been accompanied by relative restraint in total expenditure on
wages and salaries. Total revenue increased by 149 per cent between seasons 1999
and 2008. Average revenue per club in 2008 was €79.9 million. According to fig-
ures reported by Deloitte (2010), Bayern Munchen’s revenue of €295 million in
2008 was around twice that of Schalke 04 (€148 million), its nearest rival, with
Hamburg (€128 million) in third place. Matchday income accounts for around
one-quarter of total revenue of the leading German clubs; broadcast revenue con-
tributes around one-quarter; and other commercial activities contribute around
one-half. Expenditure on wages and salaries increased by 129 per cent between
Brazil 391
Table 13.5 Revenue, costs and profitability, aggregates, Germany, all Bundesliga 1
clubs, selected years, €m
1999 2002 2005 2008
Total revenue 577 1043 1236 1438

Wages and salaries 317 553 576 725
Operating profit 47 100 183 136
Source: Deloitte
1999 and 2008, less than the rate of growth in revenue. Consequently profitability
has improved, and throughout the 2000s the financial condition of German club
football has been relatively healthy.
13.3 Brazil
According to FIFA, there are over 2 million registered footballers in Brazil, and
over 29,000 football clubs. Tradition has it that football first came to Brazil in
1894 when Charles Miller, a Brazilian-born Englishman, returned to Sao Paulo
from England carrying a ball and rulebook. The game was initially taken up by
the British community in Sao Paulo and the cosmopolitan upper classes. Soon
the game spread across the city’s social landscape. Clubs were founded and neigh-
bourhood and city-wide leagues were organised. Apart from Miller’s club (Sao
Paulo Athletic Club), many other clubs were organised on the basis of ethnic and
national identity (Bellos, 2002). For example, the first city league in 1901 consisted
of two Brazilian clubs, one club of European immigrants and one club from the
German population, as well as the British Sao Paulo Athletic. In the years that fol-
lowed, other ethnic/national clubs emerged. Rio de Janeiro, the other major cen-
tre of early Brazilian football, was home to British clubs, such as the Payssandu
Cricket Club, and a Portuguese club, Club de Regattas Vasco de Gama. Clubs based
upon ethnicity and nationality could also be found in Salvador, Belo Horizonte
and elsewhere, as football proliferated throughout Brazil (Bocketti, 2008). Not
all football during this early period was dominated by ethnic clubs. Clubs such as
Corinthians and Paulistano in Sao Paulo, and Fluminense and Flamengo in Rio
de Janeiro, originated in neighbourhoods, schools or workplaces.
In the early days of Brazilian club football, explicit barriers were constructed
to deny participation to non-whites and working-class players. In Sao Paulo and
elsewhere, the amateur status of clubs and leagues was a major barrier, ensuring
that the vast majority of players were of European ethnicity and could draw on
personal financial resources. By the 1920s, however, working-class representation
had increased, and the practice of paying players under the table was widespread
(Bocketti, 2008). Many clubs embraced professionalism from 1933 onwards, and
professional and amateur leagues emerged in Rio de Janeiro and Sao Paulo. The
Brazilian Confederation (CBF) recognised professionalism in 1937.
Brazilian club football is organised in a series of interconnected leagues, com-
prising a national pyramid and state pyramids. The leading teams compete in the
national and state pyramids simultaneously. The best-placed teams in the state
championships, and the highest-ranked teams in CBF’s rankings, compete in the
Copa do Brasil. National competition is organised by CBF, and the state cham-
pionships are organised by the football federations in each state. The national
pyramid competitions start in April and end in December. The state pyramid has
a different duration and schedule in each state, but in states with teams competing
in the national first and second divisions, the main state championships run from
January/February to April/May. Most states have at least one secondary tourna-
ment, which runs from July to December, involving teams that do not participate
in the top two leagues of the national championship. Competition in states whose
teams do not participate in the national competitions usually runs from April to
October.5
The national pyramid has four tiers, named Serie A to Serie D. Serie A, B and
C have twenty teams each, and Serie D has forty teams. Before 2009 there were
three tiers (A, B and C), with sixty-four teams in Serie C. Promotion and relega-
tion between each tier is four-up four-down. The membership of Serie D consists
of the highest-placed state championship clubs that do not participate in Serie A,
B and C. Teams that are successful in their state leagues can be promoted to Serie
D. State championships sometimes operate experimental formats. For example,
in the 2008 state championship in Rio de Janeiro, the four most popular teams,
Botafogo, Flamengo, Fluminense and Vasco da Gama, played all of their matches
at home.6
National competition in Brazil at club level has a relatively short tradition, due
partly to the large geographical size of the country. The modern national cham-
pionship began in 1971. Previously, the most prestigious competitions were the
state championships of Sao Paulo and Rio de Janeiro. Prior to 2003, the national
championship was decided in a play-off format, most commonly with the top eight
regular-season teams competing in a single elimination tournament.7 Since 2003,
Serie A has been contested in a double round-robin format, with no play-offs. In
2006 the number of teams in Serie A was reduced to twenty and the number of
foreign players fielded as players or substitutes in any match was limited to three.
Table 13.6 presents data on the performance of Brazilian clubs in the national
championship and in cup competition (domestic and South American). The con-
centration of success in the national championship is relatively low, with no club
winning the championship more than three times within any decade. Seven differ-
ent Brazilian clubs have been successful in the Copa Libertadores, which is con-
tested by the leading clubs from South American countries.
Brazil 393
Table 13.6 Historical performance of top Brazilian teams in league and cup
competition
1971–1979 1980–1989 1990–1999 2000–2009
Championships
Internacional 3 Flamengo 3 Corinthians 3 Sao Paulo 3
Palmeiras 2 Bahia 1 Palmeiras 2 Santos 2
Atletico – MG 1 Coritibia 1 Botafogo 1 Atletico – PR 1
Guarani 1 Fluminese 1 Flamengo 1 Flamengo 1
Sao Paulo 1 Gremio 1 Gremio 1 Corinthians 1
Vasco de Gama 1 Sao Paulo 1 Sao Paulo 1 Cruzeiro 1
Sport 1 Vasca de Gama 1 Vasca de Gama 1
Vasco de Gama 1
Internacional 11 Flamengo 9 Corinthians 12 Sao Paulo 12
Palmeiras 8 Gremio 6 Palmeiras 8 Santos 10
Sao Paulo 7 Sao Paulo 5 Botafogo 5 Internacional 6
Atletico – MG 6 Vasco de Gama 5 Sao Paulo 5 Atletico – PR 5
Vasco de Gama 5 Guarani 5 Gremio 4 Corinthians 5
Cruzeiro 5 Atletico – MG 4 Vasca de Gama 4 Gremio 5
Botafogo 3 Fluminese 4 Flamengo 3 Flamengo 4
Guarani 3 Sport 3 Vitória 3 Sao Caetano 4
Santos 1 Bahia 3 Santos 3 Cruzeiro 3
Fluminese 1 Internacional 3 Cruzeiro 3 Vasca de Gama 3
Coritiba 1 Coritibia 3 Atlético Mineiro 3 Goias 1
Operario 1 Bangu 3 Bragantino 2 Fluminese 1
Corinthians 1 Santos 2 Portuguesa 2
Ponte Preta 1 Atlético – MG 1
Cruzeiro 1 Guarani 1
Brasil (Pel.) 1 Internacional 1
Brazilian Cup wins (from 1989)
Gremio 1 Cruzeiro 2 Cruzeiro 2
Gremio 2 Gremio 1
Internacional 1 Corinthians 1
Flamengo 1 Santo Andre 1
Criciúma 1 Paulista 1
Corinthians 1 Flamengo 1
Palmeiras 1 Fluminese 1
Juventude 1 Sport 1
Corinthians 1
Copa Libertadores wins
Sao Paulo 1 Internacional 1 Sao Paulo 1 Atletico – PR 1
Cruzeiro 1 Flamengo 1 Gremio 1 Sao Paulo 1
Gremio 1 Vasco de Gama 1
Source: www.rsssf.com/histdom.html; www.rsssf.com/intclub.html#sam

Table 13.7 Percentage breakdown of total revenue, twenty-one Brazilian clubs,

2004–2007
2004 2005 2006 2007
Gate revenue 7 6 8 8
Broadcast revenue 29 26 29 22
Transfer proceeds 30 31 23 34
Sponsorship and advertising 11 15 16 11
Other 23 22 24 25
Source: Casual Auditores Independentes Annual Surveys, 2006–8.
Brazil’s national team has enjoyed more success than any other. Brazil has won
the World Cup more than any other nation and won more World Cup games.
They have also been ever-present at the Finals, never once failing to negotiate their
region’s qualifying tournament. Brazil’s success at the World Cup began in 1958 in
Sweden when a 17-year-old Pele guided Brazil to victory. His skills, along with those
of Garrincha and others enabled Brazil to add further titles in Chile 1962 and, per-
haps most spectacularly of all, at Mexico 1970. Following Pele’s retirement, there
was a downturn in Brazil’s fortunes and it was not until 1994 that Brazil lifted the
World Cup trophy again, beating Italy on penalties after the first goalless Final.
Though tipped to win again in 1998 the team lost to France in the Final. Although
Luiz Felipe Scolari’s unheralded team arrived at Korea/Japan 2002 without the tag
of favourites, they played some dazzling football to claim a fifth FIFA World Cup.
In 2006, Brazil lost at the quarter-final stage to a Zidane-inspired France; and in
2010 Brazil lost at the same stage to Holland.8
The availability of published financial data on Brazil’s football clubs is limited,
but the accounting firm Casual Auditores Independentes publishes a financial
survey (in English) of Brazil’s top twenty-one clubs. The figures quoted below
are drawn from this source. The 2007 survey (published in 2008) was the most
recent available at the time of writing. The picture that emerges of the financial
structure of Brazilian club football in the mid-2000s is one of rapid growth in
revenue, but even faster growth in expenditure and a rapidly deteriorating finan-
cial position. The clubs covered by the 2004 survey reported total revenue of
$311.6 million. The 2007 figure was $703 million. Increased proceeds from player
transfers, an increase in gate revenues and a favourable movement in the Brazilian
currency against the US dollar all contributed to the growth in total revenue
measured in dollars. Sao Paulo, with reported revenue of $100 million, had the
largest turnover of any Brazilian club in 2007. Ticket prices in Brazil are rela-
tively low, and gate revenue makes a relatively small contribution to the clubs’
total revenue. Table 13.7 reports a percentage breakdown of total revenue for the
period 2004–7.
Japan 395
Total reported costs in 2007 were $861 million. Six of the twenty-one clubs
recorded profits in 2007, and the other fifteen clubs reported losses. The aggre-
gate loss was $158 million, up from $107 million in 2006 and $5 million in
2005. The largest single loss-makers in 2007 were Fluminese ($73.4 million) and
Flamengo ($31.2 million). Combined assets totalled $822.7 million in 2004, and
$1.67 billion in 2007. Total liabilities grew from $682 million in 2004 to $1.36
billion in 2007.
For many decades, one of Brazilian football’s defining features has been its cap-
acity for exporting its most talented players overseas. As long ago as 1930, Italian
recruiting agents travelled to Brazil and within a year had recruited thirty-nine
players to play in Italy (Bocketti, 2008). In 2008 1,776 players left Brazil (an 8.3
per cent increase on 2007). Europe was the destination for more than 40 per cent
of these players, with 209 bound for Portugal alone.9 Brazilian footballers can
be found in many countries throughout the world: Vietnam, Azerbaijan, Japan,
the United Arab Emirates, South Africa and Australia, to mention only a few.
The trend has accelerated in more recent times, however. Every player in Brazil’s
squad for the 1970 World Cup played in the Brazilian league; but in 1994 less than
half the squad (ten players) were based in Brazil. The numbers of home-based
players in the 1998, 2002 and 2006 World Cup squads were nine, ten and two,
respectively.10
The drain of footballing talent assists the clubs in balancing the books or ser-
vicing their debts. Often, however, the Brazilian clubs are not the main financial
beneficiaries. Kaka, elected FIFA’s best player in the world in 2007, was sold by
Sao Paulo to AC Milan in 2003 for $8.25 million. Six years later Kaka was sold
by Milan to Real Madrid for $100 million. In recent times, there has been an
increase in the number of players returning to Brazil later in their careers. High-
profile examples include Roberto Carlos joining Corinthians, Vagner Love joining
Flamengo, and Robinho rejoining former club Santos on loan. Returning stars
tend to be highly marketable. Adriano’s goal-scoring achievements for Flamengo
in 2009 assisted the club in selling more than one million replica shirts, a record
figure.11
13.4 Japan
The first football match to be played in Japan is reported to have taken place just
one year after the sport first appeared in France. In 1873, Lieutenant Commander
Archibald Douglas of the British navy organised a game at the Naval Academy
in Tokyo Bay. Trade links with Britain ensured that Japan received continued
exposure to football during the rest of the nineteenth century, and by the turn
of the century football featured on the curricula of a number of teacher train-
ing schools. A schools championship was introduced in 1918, and the Japanese
Football Association (JFA) was formed in 1921 to administer the first national
football tournament. The JFA joined FIFA in 1929 (Horne, 1996, 2000; Sugden
and Tomlinson, 1998).
Despite these early developments, the growth in football’s popularity was
severely constrained by the pre-eminence of baseball as Japan’s leading national
sport. The initial introduction and subsequent development of baseball was due
to the strong American influence, which extended throughout the Pacific Rim
and was paramount in Japan between the late nineteenth century and the 1930s,
and again after 1945. Academics following Whiting (1977) have attempted to
explain baseball’s popularity in terms of its compatability with certain attributes
of the Japanese national character, most clearly identified (in the west at least)
with the samurai warrior class. According to this interpretation, the distinctive
Japanese variant of baseball shares many characteristics with the martial arts,
including a rigorous training and disciplinary regime, an emphasis on the repe-
tition of set moves, a focus on the psychological duel between pitcher and hitter,
and the suppression of individuality for the benefit of the team (Horne, 1996,
2000).
Whether such insights are helpful, or whether they merely reflect the stereo-
typical cultural prejudices of outsiders, has become a matter for debate among
sociologists. There is no doubt, however, that baseball remained Japan’s most
popular national sport throughout the twentieth century. Professional baseball
was introduced in 1936, and the present-day competitive structure comprising the
Central League and the Pacific League was introduced, along US lines, in 1950.
Baseball rather than football is played at the elite universities, which are the main
recruiting grounds for the top positions in industry and commerce, the media and
the professions. Ownership of the leading professional baseball teams rests with
large corporations. Japan’s most popular team, Tokyo’s Giants (of the Central
League) regularly attract crowds of 50,000, and in general the level of television
exposure is high.
Despite its tendency to celebrate rather than suppress individuality, football
maintained its subsidiary position within the schools curricula throughout the
twentieth century. The Japan Football League (JFL), a national league com-
petition comprising non-professional company teams (whose players typically
worked in the mornings and trained in the afternoons) was first launched in
1965. Foreign players were admitted from 1967, but had to become employees
of the sponsoring company. The principle of professionalism was not accepted
by the JSL until 1985 (Nogawa and Maeda, 1999). At international level Japan
entered the World Cup qualifying tournament for the first time in 1954, but prior
to the 1990s success for the national team was confined mainly to the Olympics.
As host nation Japan were quarter-finalists in 1964 and then semi-finalists in
Mexico four years later. Japan hosted the FIFA World Youth Championships
in 1979, and staged a challenge match between the top European and South
American club teams a number of times during the 1980s. Eventually, how-
ever, repeated failure by the national team to achieve World Cup qualification
Japan 397
Table 13.8 J. League winners and runners-up, Japan, 1993–2009
1st stage 2nd stage 1st stage 2nd stage
1993 Kashima Antlers Verdy Kawasaki* 2003 Yokohama F Marinos+

1994 Sanfreece Hiroshima Verdy Kawasaki* 2004 Yokohama F Marinos* Urawa Red Diamonds
1995 Yokohama Marinos* Verdy Kawasaki
1996 Kashima Antlersx
1997 Kashima Antlers Jubilo Iwata* Champion Runner-up
1998 Jubilo Iwata Kashima Antlers* 2005 Gamba Osaka Kashima Antlers
1999 Jubilo Iwata* Shimizu S-Pulse 2006 Urawa Red Diamonds Gamba Osaka
2000 Yokohama F Marinos Kashima Antlers* 2007 Kashima Antlers Gamba Osaka
2001 Jubilo Iwata Kashima Antlers* 2008 Kashima Antlers Nagoya Grampus
2002 Jubilo Iwata+ 2009 Kashima Antlers Gamba Osaka
Notes: * Play-off winners; x J. League was played in a single-season format in 1996, before
reverting to the split-season format in 1997; + Winners of both stages in the split-season format.
Source: http://en.wikipedia.org/wiki/J._League#An_era_after_the_boom_.281996–1999.29
convinced officials of the need to introduce professionalism. In 1986 a commit-

tee was set up to investigate the issue, and in 1990 plans were announced for the
launch of the professional J. League in May 1993 (Sugden and Tomlinson, 1998;
Horne, 2000).
Part of the launch strategy was for the J. League to assume a role in urban
redevelopment, so no team was allowed to locate in central Tokyo. A stadium
capacity of at least 15,000 and a youth policy linked to local schools were
among the criteria for league membership. All teams were required to employ
qualified coaches. Broadcast rights and sponsorship were to be sold collect-
ively by the J. League. The clubs were registered corporations, with football
as their main business concern, ensuring that the management of clubs would
be professional and the clubs would not be subsidiaries dominated by other
interests.
The initial J. League membership of ten was increased progressively to eight-
een teams by 1998. The addition of a second tier in 1999 increased the total to
twenty-six (sixteen in J1, ten in J2). By 2010 the number of teams had increased
to thirty-seven (eighteen in J1, nineteen in J2). In most seasons between 1993 and
2004, the competition in the top tier of the J. League was divided into two stages,
with a play-off between the first-stage and second-stage winners held at the end
of the season to determine the overall champions. From 2005 the competition was
reorganised so as to operate on a round-robin basis along European lines, with the
championship winner determined by league points and no play-off.12 Table 13.8
reports the winners and runners-up in each season from 1993 to 2009.
Initially, automatic promotion and relegation between J1 and J2 was two-up
two-down. Between 2004 to 2008, the third-placed J2 club entered a play-off
series against the sixteenth-placed J1 club. From 2009, automatic promotion

and relegation is three-up three-down. The Japan Football League (JFL) is a
semi-professional league, from which promotion to the J. League is possible but
not automatic. In 2000, 2001 and 2006 the JFL champion was promoted to J2;
in 2005 two teams were promoted. Since 2007 J. League Associate Membership
and at least a fourth-place finish in JFL are requirements for promotion to J2.
Currently, there is no relegation from J2 to JFL.13
An important part of football’s ‘rebranding’ was the creation of team identities
associated with places of residence rather than employment, and with parts of
team names borrowed from Europe and North America; for example, Furukawa
Electric from the JSL became JEF United Ichihara, Mitsubuishi Motors became
Urawa Red Diamonds, Sumitomo Metals became Kashima Antlers and Yomiuro
FC became Verdy Kawasaki (the first J. League champions). Several foreign
coaches, including Arsène Wenger and Osvaldo Ardiles, were employed in the
early years of the J League, and a number of high-profile foreign players were
recruited, including England’s Gary Lineker, Germany’s Pierre Litbarski and
Brazil’s Zico and Dunga. Links with Brazil are especially strong due to Brazil’s
large ethnic Japanese community (Nogawa and Maeda, 1999; Birchall, 2000;
Horne, 2000).
It would be hard to argue, as in the case of England in the 1880s or France
in the 1930s, that the introduction of professionalism in Japanese football in
the 1990s was a natural and inevitable stage in the sport’s organic develop-
ment. Instead Birchall (2000) has described the creation of the J. League as
‘perhaps one of the greatest mass-marketing events even Japan, and probably
the world, has ever seen’. All aspects of the launch were subject to detailed and
carefully coordinated planning. While the commercialisation of football has
evolved gradually elsewhere, in Japan commercial concerns were paramount
from the outset. For example, Japan’s second-largest advertising company,
Hakuhodo, were given responsibility for marketing and publicity, while Sony
Creative Products took charge of all aspects of product design, including the
design of team shirts and other items of merchandise (Birchall, 2000). Initial
three-year J. League sponsorship deals were arranged with drinks manufacturer
Suntory and consumer credit firm Nippon Shinpan (Nicos) for ¥400 million
($4.4 million at 2010 exchange rates) each, while the sale of broadcast rights
raised more than ¥1,000 million ($11.0 million) in 1993, and twice as much in
1994 (Horne, 1996, 2000).
Table 13.9 indicates that the average revenue of the J. League J1 clubs increased
from ¥2,349 million ($25.8 million) in 1999 to ¥3,328 million ($36.6 million) in 2008.
Average attendances increased from 11,658 to 19,278 over the same period, but in
2008 ticket sales accounted for only around 22 per cent of revenues. Advertising
made the largest single contribution, of around 45 per cent. Expenditure on the
wages and salaries of players and coaches increased at a similar rate from an aver-
age of ¥1,144 million ($12.6 million) in 1999 to ¥1,593 million ($17.5 million) in
China 399
Table 13.9 Average attendance, revenue and salary expenditure, Japan, J1 clubs,
1999–2008, ¥m
Wages and salaries expenditure

Operating Home Foreign
Attendance revenue Coaches players players Total
1999 11658 2349 241 555 349 1144

2000 11065 2370 258 555 333 1146
2001 16547 2727 267 559 410 1236
2002 16368 2724 285 586 399 1269
2003 17351 2756 285 576 349 1210
2004 18965 2956 288 661 421 1370
2005 18765 3084 303 710 506 1518
2006 18292 3019 323 698 424 1445
2007 19081 3267 345 753 436 1533
2008 19278 3328 347 808 438 1593
Source: J. League, www.j-league.or.jp/eng/data/index_03.html
2008. The share of home-based players in total wages and salaries expenditure has
increased gradually over this period.
13.5 China
When modern football was introduced to China in the late nineteenth century
it was played mainly by university and high school students. The development
of the game was slow until the formation of the Peoples’ Republic of China on
1 October 1949. On that day the Shengyan football team was invited to Beijing to
play as part of the celebration of the establishment of the new communist regime.
‘Football was seen as a symbol of modernity; a talisman of topicality; a state-
ment of intent. The intention was clear: modern sport would characterize modern
China’ (Jinxia and Mangan, 2001, p79).
The first National Football Championships took place in Tianjin in December
1951, contested by teams from the six administrative areas, the Army and the
Railway Association. Shortly afterwards, thirty-one players were selected to form
the national football squad. This marked the beginning of state-organised foot-
ball in China. Subsequently, ‘specialised’ (professional) teams were established in
virtually every province, municipality and autonomous region, as Chinese football
became institutionalised, professionalised and politicised (Jinxia and Mangan,
2001). In the early 1950s China established close football contacts with other
socialist countries. The most influential contact was with Hungary, whose national
team at the time was world-class. Hungarian experts were invited to China to give
lectures, and the Hungarian national team travelled to China to demonstrate its
Table 13.10 Football League and Super League winners and runners-up, China,
1994–2009
Champion Runner-up Champion Runner-up
1994 Dalian Wanda Guangzhou Apollo 2003 Shanghai Shenhua Inter Shanghai
1995 Shanghai Shenhua Beijing Guoan 2004 Shenzhen Jianlibao Shandong Luneng
1996 Dalian Wanda Shanghai Shenhua Taishan
1997 Dalian Wanda Shanghai Shenhua 2005 Dalian Shide Shanghai Shenhua
1998 Dalian Wanda Shanghai Shenhua 2006 Shandong Luneng Shanghai Shenhua
1999 Shandong Luneng Liaoning Taishan
Taishan 2007 Chanchung Yatai Beijing Guoan
2000 Dalian Shide Shanghai Shenhua 2008 Shandong Luneng Shanghai Shenhua
2001 Dalian Shide Shanghai Shenhua Taishan
2002 Dalian Shide Shenzhen Ping An 2009 Beijing Guoan Changchun Yatai
Source: http://en.wikipedia.org/wiki/Chinese_football_champions
skills. China also sent a squad of twenty-five players to Hungary for eighteen
months’ training.
As a means of improving playing quality, a national league system was set up
in 1956 comprising two divisions. This seemed to have the desired result: play-
ing standards improved, and by the end of the 1950s China was a footballing
power in Asia. As the Chinese economy entered a more turbulent phase in the
early 1960s, followed by the Cultural Revolution which began in 1966, organised
football in China was hard hit. The number of ‘centralised’ footballers and teams
was reduced, and the national football championship was not staged for several
years.
Following Chairman Mao Tse Tung’s death in 1976, the ruling party embarked
on a period of economic and social reform. The promotion of football through-
out the country was given higher priority, and football was permitted to become
more ‘commercial’. In 1988, ‘professional’ clubs were formed and, as a conse-
quence, players were rewarded financially and became more professional in their
attitude (Jinxia and Mangan, 2001).
The Chinese Football League was established in 1994 with two tiers: Division
1A (Jia A) and Division 1B (Jia B). Initially there were twelve clubs in each tier,
until 1998 when two more clubs were added to Division 1A. Table 13.10 reports
the winners and runners-up of the Football League (1994–2003) and its successor
the Super League (2004–9).
In 1998, the annual income of each player in Jia A was at least 100,000 yuan
($14,700 at 2010 exchange rates), twenty times the average personal income. A
star player could earn more than one or two million yuan per year. Most ‘pro-
fessional’ clubs were run by national enterprises, and were underwritten finan-
cially by the state. Consequently, rising debt was not seen as a problem (Jinxia and
China 401
Mangan, 2001). While footballers were being highly paid, the national team was
underachieving. This prompted calls for wages to be reduced. In January 2001 the
Chinese Football Association introduced maximum salaries and transfer fees. The
monthly salary for a Jia A player was capped at 12,000 yuan ($1,765), and each
team’s win bonus was capped at 400,000 yuan ($58,800) (Jinxia and Mangan,
2001). Transfer fees were set by the Chinese Football Association, and the transfer
market was largely controlled. A player seeking a transfer requires the approval
of his club, even if his contract has already expired, for up to thirty months from
when he last represented the club. This rule appears to contravene a FIFA regu-
lation that any professional footballer is free to conclude a contract with another
club, if his contract with his present club has expired or is due to expire within six
months.
Clubs are allowed to set their own ticket prices, subject to state agreement
under the auspices of the local Price Management Department. Jia A clubs
were required to pay 5 per cent of their ticket income to the Chinese Football
Association and Jia B clubs 2.5 per cent (Amara et al., 2005). The Chinese
Football Association is responsible for negotiating and controlling the sale of
broadcast rights at national level, though local broadcast rights are owned by
clubs and may be negotiated locally. The income for Jia A clubs was, on average,
3.08 million yuan ($450,000) in 2000, of which approximately 1.2 million yuan
($175,000) came from naming rights and sponsorship (Amara et al., 2005). To
promote greater professionalism, enterprises were encouraged to provide spon-
sorship and other forms of financial support, and major investors are identi-
fied in the names of most clubs. The number of investors in each Jia A club
varies from one to more than twenty. In practice the majority of the investing
enterprises are state-owned, and company sponsorship is simply another form
of public subsidy by a different name.
Even so, there has been pressure for football clubs to operate on a more com-
mercial footing. The Chinese Football Association, for example, established a set
of eighteen benchmarks for professional clubs wishing to join the Chinese Super
League (CSL), formed in 2004 when Jia A was replaced by the CSL and Jia B was
renamed the China League (CL). Requirements include a stadium with a seat-
ing capacity of 30,000, an income in excess of 3.3m yuan (excluding transfers), a
financial record that does not show losses for three consecutive years, and a healthy
balance sheet in the year of entry into the Premier League. The clubs argued that
these requirements were too stringent, but the Chinese Football Association felt
it would rather have fewer clubs participating than lower the standards for entry
(Amara et al., 2005). Soon after the CSL started, play was suspended temporarily
over issues of match-fixing and corruption, and the financial instability of several
clubs became apparent.
In its inaugural year the CSL had twelve member teams. Initially it was planned
to increase the number of teams to sixteen by 2006, with no teams relegated at the
end of the first two campaigns. A membership of sixteen was not achieved until
2008, however, due to the withdrawal of Sichuan Guancheng in 2006, and the
merger between Shanghai United and Shanghai Shenhua in 2007. The CSL starts
in February–March and ends in November–December. From 2008, two-up two-
down promotion and relegation operated between the CSL and the CL. The first
sponsor of the CSL was Siemens. Following a controversial first season, Siemens
did not renew its sponsorship. The start of the second season in 2005 was delayed
for a month while new sponsors were sought, unsuccessfully. At the time of writ-
ing, Pirelli are the sponsors for the period 2009–11.14
Before the start of the 2010 season, the Chinese Football Association relegated
two teams from the CSL to the CL following a corruption scandal involving accu-
sations of match-fixing and illicit gambling. The two lowest-placed teams at the
end of the 2009 season retained their places in the CSL. One of the relegated
teams, Chengdu Blades, was owned by the English Football League club Sheffield
United, which acquired a majority stake in 2006 with the aim of developing new
playing talent. Chengdu Blades were promoted to the CSL in 2007, but were sub-
sequently accused of bribing an opposing team to lose a crucial game in order
to secure promotion. Guangzhou GPC, also promoted in 2007, were similarly
accused. These actions are part of a widening crackdown on corruption. Twenty
senior officials and players were implicated, including the former head of the
Chinese Football Association.
Conclusion
Chapter 13 has presented a brief description of the historical development and
present-day competitive and commercial structure of football in France, Germany,
Brazil, Japan and China. The histories of professional football in these coun-
tries provide many fascinating comparisons and contrasts, both with each other
and with the story in England, the main subject of the previous chapters of this
volume.
In no other country does football’s top tier come close to matching the English
Premier League financially, in terms of the revenue the Premier League gener-
ates and the sums its member clubs spend, primarily on players’ remuneration.
Nevertheless, the existence of a somewhat less rampant commercial ethos in
French professional football has not precluded the development of a national
team capable of competing successfully at the very highest level, in a manner
that has proven notoriously elusive for England in successive World Cups and
European Championships since the national team’s sole World Cup triumph in
1966. Likewise the German national team, and (especially) its West German pre-
decessor before unification, has been a highly successful international competi-
tor. The commercial organisation of German club football also presents some
stark contrasts with the English model. Although German clubs tend to feature
China 403
less prominently at the very highest levels of European club football than they
once did, the club licensing system that operates in the Bundesliga appears to
be effective in preventing some of the more reckless styles of financial manage-
ment at club level that have occasionally embarrassed or disfigured the Premier
League.
The Brazilian national teams that triumphed in the 1994 and 2006 World Cups
were more prosaic in terms of their style of play than their predecessors of 1958,
1962 and 1970, but Brazil still retains a special place in the affections of many foot-
ball supporters worldwide. Rather like France, a tradition of exporting its most
talented players has, over the years, acted as a drain on the competitive strength
of Brazilian club football. Brazil is renowned for the fanatical enthusiasm of its
population for participation in football, from the grassroots level upwards. In con-
trast, the east Asian economic powerhouses of Japan and China have little or no
historical footballing tradition, at either the grassroots or the professional level.
Recent attempts in these countries to establish a market for professional football
from the top down present an interesting contrast to the bottom-up evolution of
the professional sport elsewhere, from its early historical origins in England and
other countries.
Notes
1 See http://en.wikipedia.org/wiki/Ligue_1#Foundation
2 See www.bundesliga.de/en/liga/news/2008/index.php?f = 0000128159.php
3 See http://en.wikipedia.org/wiki/Fu%C3%9Fball-Bundesliga
4 A KgaA is a separate legal entity whose share capital is divided into shares which are held
by at least one shareholder (the general partner) with unlimited liability and limited liability
shareholders (Kommanditaktionäre) that are not personally liable for the debts of the com-
pany. Borussia Dortmund GmbH & Co. KgaA is different to a German stock corporation
in that it has no management board. Instead, Borussia Dortmund GmbH, as the general
partner, is solely responsible for its management and representation. Borussia Dortmund’s
Verein is the sole shareholder of GmbH. The supervisory board of KgaA has no authority
to appoint and dismiss managing directors at Borussia Dortmund GmbH or to regulate
the terms of their contracts. See http://eng.borussia-aktie.de/?%9FY%1B%E4%F4%9D.
5 See http://en.wikipedia.org/wiki/Brazilian_football_league_system
6 See http://en.wikipedia.org/wiki/Brazilian_football_league_system
7 Bellos (2002) describes the format as ‘unfeasibly complicated’. For example, in 1978 there
were seventy-four teams divided into six groups (A–F). Each team in each group played
each other once and all teams qualified for the second round. The six top clubs in groups
A–F formed four groups of nine (G–J). In these groups the six best teams moved to the third
round. The other thirty-eight teams from round one formed six groups (K–P) and the win-
ners of each qualified for the third round. Together with the best-placed loser from G–J and
K–P, these thirty teams formed four groups of eight (Q–T). The top two in each of Q–T went
through to the next phase of eight teams who played a knockout of quarter final, semi-final
and then a final.
8 See www.fifa.com/worldcup/preliminaries/southamerica/teams/team = 43924/profile.html
9 See www.insidefutbol.com/2010/02/05/economic-crisis-changes-brazilian-game/16983/
10 See www.planetworldcup.com/atwc/atwc_squads.html
11 See www.insidefutbol.com/2010/02/05/economic-crisis-changes-brazilian-game/16983/
12 See www.j-league.or.jp/eng/history/
13 See http://en.wikipedia.org/wiki/J._League
14 See http://en.wikipedia.org/wiki/Chinese_Super_League
14 The economics of the World Cup
Introduction
Countries compete fiercely for the right to host mega sporting events such as
football’s World Cup, and the Olympic Games. In order to host a mega event,
large volumes of public subsidy are usually required. In this respect, the 1984 Los
Angeles and 1996 Atlanta Olympics were exceptional, in having been funded pri-
marily from the private sector. Those who advocate the use of public funding to
secure and stage a mega event invariably promise a wide range of benefits for host
cities and the host nation, including increased employment and per capita income,
as well as less tangible social and cultural benefits deriving from an enhanced
local or national reputation. Since the economic benefits are the more tangible,
these have tended to receive the most attention from academic and non-academic
economists. Indeed, bids to host the World Cup have in recent times been ‘sold’ to
the public on the basis of extravagant predictions about the large gains in employ-
ment and income that can be anticipated if the bid is successful.
In recent decades, academic economists have gained some experience in evaluat-
ing the credibility of claims of this kind, due to a remarkable late twentieth-century
boom in the construction of sports stadia for major league sports franchises in
North America. Crudely speaking, the public sports strategies of many cities are
based on offering taxpayer-funded financial subsidies for investment in sports sta-
dium construction or renovation projects, in an attempt to attract or retain major
league franchises. Taking advantage of tight restrictions on the supply of major
league sports that are imposed by the franchise system, team owners have been com-
plicit in encouraging a competitive bidding war to emerge between municipalities.
Team owners can and sometimes do make threats to move to another city offering
a better deal, unless the present host city makes an acceptable counter-offer. Many
local politicians have participated willingly in this process, mindful of the possible
kudos to be gained by attracting a new major league franchise, or fearful of the
opprobrium the departure of an established franchise might bring. Consequently,
a curious market has developed in economic impact studies produced by consult-
ancies, showing that the economic benefits of particular stadium infrastructure
405
projects justify the proposed public subsidy. Academic economists, however, tend
to view such claims with extreme scepticism. Some economists believe the net eco-
nomic impact of sports stadium infrastructure projects is negligible at best, or
perhaps even significantly negative at worst. Noll and Zimbalist (1997) present a
collection of detailed case study material, and Siegfried and Zimbalist (2000) pro-
vide a concise review.
In view of the enormous expenditures involved in hosting the World Cup,
Chapter 14 examines whether a successful World Cup bid is likely to yield eco-
nomic benefits to the host nation that are capable of justifying the enormous finan-
cial outlays. Hosting the World Cup certainly carries heavy costs. FIFA requires
the World Cup host country to provide at least eight and preferably ten modern
stadia capable of seating 40,000 spectators or more. For the jointly hosted 2002
World Cup in Japan and South Korea, each country offered to provide ten separ-
ate stadia. As neither country had any large pre-existing football infrastructure,
South Korea constructed ten new stadia at an estimated cost of nearly $2 billion,
and Japan built seven new stadia and refurbished three others at a cost of at least
$4 billion (Baade, 2006).
As in the case of city-level stadium infrastructure projects for US major league
sports, when the question is asked whether the massive expenditures required to
host a mega event are justified by corresponding anticipated gains in employment
or per capita income, the answer is typically an unequivocal ‘yes’ when given by
those advocating the use of public funds in order to secure and stage the event.
The answer tends to be an equally resounding ‘no’ when given by academic econo-
mists. For example, the 1994 World Cup Organizing Committee in the US pre-
dicted that around one million international visitors would come to the US,
making the event one of the most significant tourist attractions in American his-
tory (Baade and Matheson, 2004). Prior to the event, it was estimated that hosting
the 2010 World Cup would boost South Africa’s national income by more than $6
billion (Grant Thornton, 2008). Academics, however, have been quick to point out
the failings of prospective economic impact studies. It is claimed that many such
studies are based on poor methodology, fail to distinguish between gross and net
effects, fail to take account of substitution and crowding-out effects (see below),
and are based on over-optimistic assumptions concerning the propensity for any
additional expenditures to deliver economic benefits at the local level.
Section 14.1 provides a description of the origins and competitive structure
of the World Cup, and highlights several of its unique economic characteristics.
Section 14.2 discusses some of the key issues and principles involved in evaluat-
ing the economic benefits derived from hosting a mega sporting event. Section
14.3 reviews a number of prospective economic impact studies, most of which use
either input-output models or computable general equilibrium (CGE) models to
generate predictions for the employment and income effects. While offering some
improvements over the relatively simplistic input-output approach, CGE models
are still based upon some strong assumptions about the structure of the economy.
The World Cup 407
For this reason, most academic economists prefer to rely upon retrospective econo-
metric analysis of the effects of mega events on indicators such as employment or
per capita income, either at city, regional or national level.
Section 14.4 reviews several retrospective studies, which seek to establish
whether the experience of hosting a World Cup conformed to the prospectively
advertised benefits. Both the prospective and retrospective analyses focus primar-
ily on the pecuniary costs and benefits of hosting the World Cup. There is also a
smaller literature that takes a broader perspective, examining the impact of the
mega event on non-pecuniary (or intangible) factors, such as happiness or the
‘feelgood’ effect. Section 14.5 considers the nature of these intangible benefits, and
discusses ways in which they might be measured and assessed.
14.1 The World Cup

The World Cup has been contested by the men’s national football teams of the
member associations of the Federation of International Football Associations
(FIFA) once every four years since the inaugural tournament was held in Uruguay
in 1930, with the exceptions of 1942 and 1946 when the tournament was not staged
during and immediately after the Second World War. Table 14.1 provides summary
historical details of all previous World Cup tournaments, up to and including the
South Africa World Cup Finals staged in 2010. The current format of the tourna-
ment involves thirty-two teams competing for the title at venues located in the host
nation(s), over a period of around a month’s duration (the Finals phase) in June
and July. A qualification phase, which takes place over the course of the preceding
three years, determines which teams qualify for the Finals phase, together with
the host nation(s) who receive a bye. Prior to the 2006 tournament, the winning
team from the previous World Cup also received a bye, but currently the previous
winner is required to participate in the qualification phase. The World Cup is the
most widely viewed sporting event in the world: the estimated cumulative televi-
sion audience for the 2006 World Cup in Germany was 26.2 billion, an average of
409 million viewers per match. It is estimated that 715.1 million people watched
the final between Italy and France (FIFA, 2008).1
The World Cup generates estimated worldwide revenue of around $4 billion for
FIFA, making it the world’s most popular and probably the world’s most prof-
itable sporting event (Szymanski, 2003; Kurscheidt, 2006). FIFA funds most of
its operations from the profits it generates by staging this single event once every
four years. On the cost side of the profit equation, the players’ salaries continue to
be paid by their clubs as their ultimate employers, and any additional payments
for appearing in the tournament are borne by the national football associations.
Selection to participate in the World Cup Finals phase is considered to be the pin-
nacle of a player’s career, and the accolade of selection may yield huge benefits to
either the player or his employing club or both, in the form of enhanced bargain-
ing power for the player in salary negotiations, the player’s enhanced marketability
Table 14.1 History of the World Cup
Teams Teams
in in Number
qualifying Finals of Runner-up
Year Host nation phase phase cities/stadia Winning team team
1930 Uruguay — 13 1/3 Uruguay Argentina

1934 Italy 31 16 8/8 Italy Czechoslovakia
1938 France 36 15 9/10 Italy Hungary
1950 Brazil 33 13 6/7 Uruguay Brazil
1954 Switzerland 38 16 6/6 West Germany Hungary
1958 Sweden 53 16 12/12 Brazil Sweden
1962 Chile 56 16 4/4 Brazil Czechoslovakia
1966 England 71 16 7/8 England West Germany
1970 Mexico 71 16 5/5 Brazil Italy
1974 West Germany 98 16 9/9 West Germany Holland
1978 Argentina 106 16 5/6 Argentina Holland
1982 Spain 109 24 14/17 Italy West Germany
1986 Mexico 121 24 9/12 Argentina West Germany
1990 Italy 112 24 12/12 West Germany Argentina
1994 US 144 24 9/9 Brazil Italy
1998 France 172 32 9/10 France Brazil
2002 South Korea/Japan 193 32 20/20 Brazil Germany
2006 Germany 195 32 12/12 Italy France
2010 South Africa 204 32 9/10 Spain Holland
2014 Brazil — 32 — — —
Source: Kurscheidt (2006), Sky Sports Football Yearbook
in football’s player transfer market, as well as lucrative opportunities for involve-

ment in advertising and commercial sponsorship.
Many of the other costs of staging the tournament, including security, commu-
nications, and investment in the construction of stadia, other physical infrastruc-
ture and transport infrastructure, are borne by the local organising committee
and are usually heavily subsidised from public funds. Despite the heavy costs it
entails, competition among bidding nations to stage the event is intense, and the
bidding process is highly politicised. By awarding the 1994 Finals to the US, the
2002 Finals jointly to two Asian nations, Japan and South Korea, and the 2010
Finals to South Africa, FIFA has attempted to use the tournament as a means
to stimulate awareness and interest in football in countries and continents where
the sport’s popularity and commercial potential was hitherto underdeveloped.
Beginning with the decision to award the 2010 Finals to South Africa, which was
taken in 2004, FIFA announced that it would in future operate a rotation system
between continents for the award of the Finals among its member federations
(Kurscheidt, 2006).
Costs and benefits of hosting a mega sporting event 409
The qualification phase comprises a set of six separate qualifying tournaments

held within the six FIFA continental zones (Africa, Asia, North and Central
America and Caribbean, South America, Oceania, Europe), and overseen by their
respective confederations. For each tournament, FIFA decides the number of
places awarded to each of the continental zones beforehand, generally based on the
relative strength of the confederations’ teams, but also subject to lobbying from the
confederations. The formats of the qualification tournaments differ between con-
federations, and in some cases the qualification process begins almost three years
before the Finals, and lasts for around two years. One or two places in the Finals
are awarded to winners of intercontinental play-offs. For example, the winner from
the Oceania zone (considered to be the weakest of the six) and the fifth-placed team
from the Asia zone competed in a play-off for a place in the 2010 World Cup.
Since the 2002 World Cup, the Finals phase has featured thirty-two teams. There
are two stages: a group stage followed by a knockout stage. In the group stage, the
teams compete within eight groups of four teams each. Eight of the thirty-two
teams are seeded (including the hosts, with the other teams selected using a for-
mula based on both the FIFA World Rankings and performances in recent World
Cups) and assigned to separate groups. The other twenty-four teams are assigned
to three different ‘pots’, based on team quality and on geographical criteria; and
one team from each pot is assigned randomly to each of the eight groups. Since
1998 constraints have been built into the draw to ensure that no group contains
more than two European teams, and no group contains more than one team from
any other confederation.
Each group plays a round-robin tournament, giving every team three group
matches. The two final matches in each group are scheduled simultaneously. The
top two teams from each group advance to the knockout stage. Points are used
to rank the teams within each group. Since 1994, three points have been awarded
for a win, one point for a draw and none for a loss. If two or more teams have the
same number of points after three matches, a series of tiebreakers is applied: first
goal difference; then total goals scored; then head-to-head results; then finally
drawing lots at random. The knockout stage is a single-elimination tournament,
in which teams compete on a pairwise basis in one-off matches, with extra time
and penalty shoot-outs used to decide the winner if necessary. The knockout stage
begins with the ‘round of sixteen’, in which the winner of each group plays against
the runner-up of another group. This is followed by the quarter-finals, the semi-
finals, the third-place match (contested by the losing semi-finalists) and the final.
14.2 Costs and benefits of hosting a mega sporting event

The appropriate methodology for the evaluation of the economic impact of
mega events such as the World Cup or Olympic Games, and for the evaluation of
the impact of smaller-scale and more localised projects such as the construction
of a new stadium in a particular US city for a major league sports franchise, has
been the subject of vigorous debate. As noted in the introduction, a chasm has
emerged between the hyperbolic claims of many prospective economic impact
studies carried out by consultants, and the sober assessments obtained from
retrospective analysis of historical income or employment data at local level
before, during and after the construction of a facility or the staging of a mega
event.
A typical prospective economic impact study prepared by consultants estimates
the number of visitors an event is expected to draw, the number of days each
spectator is expected to stay, and the average expenditure of each visitor on each
day. By combining these figures, an estimate of the ‘direct economic impact’ is
obtained. A multiplier is then applied to the direct economic impact, to account
for the recirculation of the initial direct expenditure through the local economy.
Ticket-holders for the event buy hot dogs from a vendor located outside the sta-
dium, who spends some of the proceeds on a round of drinks in the local pub,
and the publican celebrates by purchasing an additional lottery ticket from the
local newsagent, and so on. The additional expenditure is known as the ‘indir-
ect economic impact’. Multipliers close to two are commonly used, making the
total economic impact approximately twice the size of the initial direct expend-
iture. Matheson (2006) uses a three-way classification for the possible sources of
bias in this type of study, arising from incorrect treatment of substitution effects,
crowding-out or displacement effects, and leakages.
A focus on gross rather than net expenditure is believed to be a principal cause
of the tendency for many prospective studies to present exaggerated or hyper-
bolic assessments of economic impact. Substitution effects are incorrectly treated
if the economic impact is presented as the aggregation of all gross expenditures or
receipts associated with the event. The substitution effect refers to the likelihood
that money spent by local residents at an event would have been spent elsewhere in
the local economy if the event had not been staged. Therefore expenditure by local
residents should not count in the calculation of the economic impact of the event.
Incorrect treatment of substitution effects can also arise in respect of investment
expenditure. For example, the construction of new transport infrastructure might
have gone ahead anyway, even if the event had not been staged.
Similar issues may arise with expenditures by visitors from outside the city or
region. Visitors who have travelled specifically in order to attend the mega event,
and who would not have visited either at the time of the event or at some other
time if the event had not been staged, bring new expenditure that should count
towards the economic impact. The expenditure of visitors who just happen to be
in town and decide to attend, and who would have spent their free time and money
on other leisure activities in town had the event not been staged, should not count.
Neither should the expenditure of visitors who rearrange a trip they would have
made anyway in order to attend the event. In this case, the question as to what
constitutes gross and net expenditure is heavily dependent on market definition.
An event that diverts tourists from one city to another within the same country
Costs and benefits of hosting a mega sporting event 411
might generate a positive net benefit if the measurement is at city level, but no net
effect if the measurement is at national level.
Crowding out refers to the displacement of expenditures that would have taken
place if the event had not been staged, but which are diverted elsewhere as a result
of the decision to stage the event. During the preparation phase, the construction
of the stadium and infrastructure might divert productive resources from alter-
native projects that would otherwise have gone ahead. Problems of noise, pollu-
tion or traffic congestion in the vicinity of the site might result in the relocation
of some residents or the diversion of some forms of economic activity to other
localities. During the mega event itself, tourists who would otherwise have visited
the city might decide not to visit, in order to avoid exorbitant hotel bills or traffic
congestion. If the city’s hotels were already operating close to full capacity, then
the mega event might simply displace one type of visitor with another, yielding no
net effect.
Issues relating to crowding out also arise in respect of the opportunity costs of
the public expenditures deployed to subsidise a mega event. For example, expend-
iture on the construction of a sports stadium might have generated larger eco-
nomic benefits if the same money had been used to build new hospitals or schools.
If the public investment is debt-funded, the opportunity costs in terms of higher
taxes or reduced public services might produce a drain on the local economy that
lasts for many years. This effect might be mitigated to some extent by longstanding
benefits from ongoing local usage of the stadia and other facilities, if the danger
that these infrastructures assume ‘white elephant’ status following the departure
of athletes and the media circus can be averted.
Leakages occur when expenditure associated with an event takes place in the local
economy, but is remitted elsewhere by its immediate recipients. Leakages reduce
the size of the indirect economic impact on the local economy. For example, the
construction workers hired to build a stadium might be brought in from outside
the city, region or country, and they might remit a large proportion of their earn-
ings to their families who are located elsewhere. The hotel which accommodates
visitors to the event might form part of a national or international chain. The
revenues it generates contribute to the chain’s profits, which accrue to sharehold-
ers who could be located anywhere. In measuring the indirect economic impact,
the effects of leakages should be reflected in an appropriate choice of multipliers.
The multipliers used in prospective economic impact studies are often calculated
using input-output models, which trace the impact of changes in expenditure by
consumers, investors or governments on all sectors of an economy, by quantifying
the linkages between sectors created by the use in each sector of inputs that are the
outputs of other sectors.
A model that has been widely used in economic impact studies for the construc-
tion of sports facilities in the US is the Bureau of Economic Analysis’ Regional
Industrial Multiplier System (RIMS II). This model provides final-demand out-
put multipliers for 473 detailed industries, including hotel accommodation, eating
and drinking establishments, and arts, entertainment and recreation. A commonly

voiced criticism of prospective studies of the economic impact of the construc-
tion of new stadia for North American major league sports franchises is that the
assumed multipliers tend to be too high. With major league sports, much of the
consumer additional expenditure that the franchise generates, for example through
ticket sales, contributes to athletes’ salaries, but many athletes live outside the
local area in which they play. Expenditure on tickets to watch major league sports
is therefore less likely to be recycled through the local economy than expenditure
in local restaurants, theatres or cinemas, to which the local consumer expenditure
on tickets might otherwise have gone, if the stadium had not been built (Siegfried
and Zimbalist, 2000).
More generally, input-output models are based on highly restrictive assump-
tions of linearity in production technology; in other words, any increase in final
expenditure always generates a pro rata increase in the production of all inter-
mediate and final outputs. Accordingly, it is assumed that the structure of the
economy is invariant to shifts in demand and supply, there is no price mechanism
and there are no capacity constraints. The use of incorrect multipliers compounds
errors that may already have been introduced through the incorrect treatment of
substitution and crowding-out effects. Despite the use of multipliers in macroeco-
nomics having been widely discredited for various theoretical reasons, their use in
economic impact studies is commonplace and largely unquestioned by practition-
ers (Humphreys, 2006).
The issue of inflated multipliers may become even more problematic in the case
of mega events. The multipliers in RIMS II (and those from other input-output
tables) are based upon inter-industry relationships within regions based upon an
economic area’s normal production patterns. During a mega event, however, pat-
terns of economic activity within a region or a city are expected to differ substan-
tially from those that are observed normally. Therefore the inter-industry linkages
on which the model is based may not be relevant. If there is no reason to believe
that the usual economic multipliers apply, any economic analysis based upon these
multipliers is questionable (Baade, 2006; Porter and Fletcher, 2008).
Besides the methodological questions associated with assessing the economic
effects of mega events, there are also serious problems in the implementation.
Most prospective (ex ante) economic impact studies anticipate large gains in
income and employment from hosting a mega sporting event. Many such studies
are carried out by consulting agencies hired by the sport’s governing body, or by
local or national government. Consultancies face obvious financial incentives to
produce conclusions tailored to suit the wishes of commissioning body or organ-
isation, and tend to produce reports that are biased towards the overstatement of
benefits and understatement of costs. Ideally, the results of prospective studies
should be compared with those of retrospective (ex post) studies in order to deter-
mine whether, in hindsight, they were correct. However, governments, sports gov-
erning bodies or organising committees have little or no incentive to commission
Prospective economic impact studies 413
retrospective studies for mega events that have taken place in the past (Coates and
Humphreys, 2003b).
Political backing for the use of tax revenues to subsidise sports events tends to
gain traction from the distribution of support and opposition among electorates.
Sports enthusiasts might constitute only a minority of the population, but might be
highly motivated and highly vocal in their advocacy of public subsidy. The major-
ity of the population might be either indifferent or only mildly hostile, but poorly
organised in order to oppose proposals that might only translate into a small or
negligible additional tax burden per person (Siegfried and Zimbalist, 2000).
14.3 Prospective economic impact studies

In the past 15–20 years many prospective impact studies of mega sporting events
have been produced, although until recently such studies for the World Cup were
relatively few in number.2 Rahmann et al. (1998) report a cost-benefit and scenario
analysis that was conducted on behalf of the German Soccer Association ahead
of the 2006 World Cup. Estimates are presented of the economic impact over the
next ten years following the World Cup, based on the number of venues and the
spending of World Cup tourists. The estimated expected benefit was around €1.5
billion, and the top estimate was €3.4 billion. Ahlert (2000) uses an input-output
model to estimate the economic impact of the 2006 World Cup on the German
economy. Under a variety of financing schemes for the construction of the neces-
sary infrastructure, the predicted increase in employment over the period 2003–10
was more than 2,400 jobs annually, with a peak employment gain of 7,300 jobs in
2006. The predicted increase in GDP was €1.5 billion in 2006, and the cumulative
gain in GDP over 2003–10 was €5.3 billion.
South Africa’s (unsuccessful) bid for the 2006 World Cup was based, in part,
on the premise that it would boost the economy by approximately $6 billion, and
create 129,000 new jobs (Khoza, 2000). Prior to the 2002 World Cup, a study
by the Dentsu Institute for Human Studies estimated a $24.8 billion economic
impact for Japan, and a $8.9 billion economic impact for South Korea. These
figures represent 0.6 and 2.2 per cent of the total Japanese and South Korean
GDP, respectively (Finer, 2002). In a report on the benefits of the 2010 World Cup,
hosted by South Africa, Grant Thornton (2003) estimate that the event would
generate additional direct expenditure of $1.7 billion (R12.7 billion), producing
an overall economic gain to the South African economy of $2.8 billion. This esti-
mate was based on an assumption that 230,000 foreign tourists would visit dur-
ing the tournament, staying for an average of fifteen days. The estimated total
costs of construction amounted to $1.8 billion. The event would create 159,000
new employment opportunities. The overall economic impact was calculated using
‘appropriate income and employment multipliers and taxation rates’.
In an update to the earlier report, Grant Thornton (2008) report a revised esti-
mate that the 2010 World Cup would contribute at least $6.8 billion (R51.1 billion)
to South Africa’s GDP, up from $2.8 billion in the earlier report. More optimistic
estimates of the event’s capacity to attract foreign tourists account for $2.1 billion
of the increase. The estimated direct economic impact is $4.1 billion, up from $1.7
billion. The revised estimates are based on assumptions of 289,000 overseas tour-
ists watching an average of between three and four matches each, 48,000 African
tourists watching an average of three matches each, and 115,000 domestic tourists
watching an average of two matches each.
Mabugu and Mohamed (2008) also report a prospective study for the 2010
World Cup. A specially designed fiscal Social Accounting Matrix (SAM) multi-
plier model is used to model the macroeconomic, sectoral and household
impacts of the R17.4 billion allocated by the government for World Cup capital
projects.3 At the macroeconomic level, the model predicts an increase in GDP
of R163 million, equivalent to a 1.2 per cent increase relative to the base year
(2004). The increase in GDP is driven largely by consumer expenditure, which
can be traced back, in part, to higher real household income. The $2.3 billion
increase in government expenditure is estimated to yield a 1.28 per cent increase
in domestic production, while the multiplier analysis suggests domestic supply
would increase by 1.23 per cent relative to 2004. At the household level, low-
income households are expected to benefit more than higher-income households
from the labour market and employment effects, due to the low-skill bias in
construction sector employment. Middle- and higher-income households should
benefit from a savings effect, attributed to the fact that the project is wholly
government-financed.
With specific reference to the predictions of the Grant Thornton (2003) study
for South Africa, Swinnen and Vandermoortele (2008) argue that the economic
impact may be overstated, for several reasons. First, the study includes domestic
residents’ expenditures at the event as direct benefits. However, this is merely a
reallocation of existing expenditure, and not a net addition to GDP (Baade, 2006;
Johnson and Sack, 1996). Second, according to Bohlmann (2006), the report’s use
of multipliers is questionable and overly optimistic. Third, the report estimated
that expenditures of $240 million and $66 million would be required for upgrades
to stadia and infrastructure, respectively. The International Marketing Council of
South Africa (2008) reports much higher investment costs, however: $1.1 billion
for ten stadia (five to be renovated and five newly constructed). The cost esti-
mates for the construction of new stadia in Durban and Cape Town were $350
million and $380 million, respectively. The cost of infrastructure upgrades, for
example, upgrades of airports and improvements to the country’s road and rail
network, was estimated at $1.2 billion. Fourth, the Local Organizing Committee
(LOC) planned to recruit unpaid volunteers, shedding a different light on the
interpretation of the estimated 159,000 ‘employment opportunities’. Many of the
jobs would only be temporary. Due to the economic situation in neighbouring
Zimbabwe, and in response to announcements of numerous job vacancies, a huge
inward migration of skilled and semi-skilled construction workers to South Africa
Retrospective economic impact studies 415
was anticipated and considered likely to account for a considerable share of the
newly created employment.
In recognition of the theoretical and methodological difficulties with input-
output models, there has been increasing use recently of computable general
equilibrium (CGE) models. CGE models offer several advantages in conducting
prospective economic impact analysis in comparison with input-output models.
CGE models still require several heroic assumptions, however, concerning the
structure of the economy and its responsiveness to the hosting of a mega event.
By modelling household expenditure as income-constrained, CGE models are
capable of handling displacement effects that arise from sports-related spending.
Similarly, scarcity and diversion of resources from other activities is taken into
account. Furthermore, the researcher is able to run simulations, in which govern-
ments neutralise the effect the event has on their own budgetary position, and, in
the case of a national government, on the country’s external debt (Giesecke and
Madden, 2007).
While CGE modellers have paid considerable attention to the Olympic Games,
less attention has been devoted to the World Cup.4 As far as we are aware,
Bohlmann and van Heerden (2008) is (at the time of writing) the only prospective
CGE study of the World Cup. A 32-sector CGE model of the South African econ-
omy is used to predict the economic impact of hosting the event. The model simu-
lates the effect of various shocks, including an increase in government expenditure
on construction and increased tourism. A short-term increase in real GDP of 0.61
per cent is estimated, together with a 1.11 per cent increase in employment. The
employment gains are driven mainly by unskilled unemployed resources drawn
into economic activity by the demand injection. The impact of financing by the
government is considered, assuming most of the expenditure is financed through
higher taxes. Under a high-tax scenario, there is a negligible increase in GDP,
and lower employment. In this case, the economic costs of the event outweigh
the benefits. Under a low-tax scenario, with future investment, GDP growth and
taxes generating the remaining required resources, both GDP and employment are
increased in the short term.
14.4 Retrospective economic impact studies

A very different approach to the same question of economic impact involves retro-
spective analysis of whether the experience of hosting a mega event, and trends in
economic indicators for the cities, regions or nations concerned before, during or
after the event, correspond to the claimed or advertised prospective benefits. An
obvious difficulty arises in disentangling changes in income, employment, infla-
tion, tourism, and other possible effects caused by the mega event from changes
caused by other factors (currency movements, changes in fiscal and monetary pol-
icy, and so on). Major sporting events tend to generate levels of media publi-
city and public awareness that are disproportionate to their economic significance
measured in terms of contribution to gross domestic product. Consequently, any

retrospective analysis of aggregated income or employment data may encounter
difficulties in identifying any economic effect which, even if the claims of propo-
nents of the event are accurate, might represent only a tiny fraction of the city or
region’s total economic activity. Generally speaking, the greater is the level of spa-
tial or temporal disaggregation of the data employed, the higher is the likelihood
of identifying a significant economic impact.
Retrospective studies typically employ the tools of multivariate econometric
analysis, and most report that the hosting of sporting events has little or no sig-
nificant impact on employment or income (Baade, 1987; Baade and Dye, 1990;
Baade, 1994; Baade and Sanderson, 1997; Baade and Matheson, 2001, 2002).
Some studies even arrive at negative and significant estimated effects (Coates
and Humphreys, 1999, 2000a, 2000b, 2002, 2003a, 2003b; Teigland, 1999). The
unanimity of these findings is startling, and seems almost unprecedented for a
profession that is notorious for its capacity for disagreement among its leading
practitioners. The sources of the divergence between the prospective claims and
the retrospective evidence are explored by Crompton (1995) and Késenne (1999).
Matheson (2006) reviews a number of retrospective impact studies.5
Baade and Matheson (2004) presented the first major econometric study of the
economic impact of the World Cup. Their analysis of the 1994 World Cup, hosted
by the US, suggests that the economic impact of the event did not justify the mag-
nitude of the expenditures. The host cities experienced an aggregate loss estimated
at $9.255 billion, in stark contrast to the projected windfalls of the order of $4
billion that were being touted in advance by advocates of the event. Baade and
Matheson examine data on growth in real personal incomes for the seventy-three
largest metropolitan statistical areas (MSAs) in the US by population, for the time
period 1970–2000. A model of the determinants of growth is fitted, for which the
explanatory variables include measures of relative nominal wages and local taxes
for each MSA, a time trend and various dummy variables. The differences between
the actual values for growth in real personal income for each MSA in 1994 are
compared with the fitted values obtained from the model, in order to obtain a
monetary estimate of the impact of the World Cup for each city. The city-level
estimates vary rather widely, from a positive estimated impact for Chicago of $992
million to negative estimates of just over $3.5 billion for both Los Angeles and
Fort Worth. Overall, nine of the thirteen host MSAs experienced actual growth
in personal disposable income that was lower than the predicted growth derived
from the model.
Hagn and Maennig (2008) analyse the long-term employment effects of the
1974 World Cup, hosted by West Germany, on the nine German host cities (Berlin,
Dortmund, Düsseldorf, Frankfurt, Gelsenkirchen, Hamburg, Hannover, Munich
and Stuttgart), using multivariate analysis. The study uses data for the period
1961–88 on the seventy-five most densely populated municipalities in Germany
Intangible benefits of mega events 417
in 1974, including all of the host cities. Three methodological approaches are
adopted and compared. First, employment in the municipalities is modelled along
the same lines as in Baade and Matheson (2004), in order to obtain estimates of the
short-term impact on employment in each case. Second, a fixed-effects regression
is estimated, in which employment in each municipality in each year is modelled
using a set of year dummy variables to register effects acting equally on all cities,
city-specific time trends to register specific developments in individual cities, and
a World Cup dummy variable for the host cities. Third, a difference-in-difference
estimator is used to register changes in the ‘intercept’ (employment and wage lev-
els) and ‘slope’ (growth of these two variables) before and after the World Cup, by
comparing host municipalities with others that were unaffected by the event. This
approach assumes that the affected municipalities would have performed the same
as the unaffected municipalities if the event had not been staged. The results sug-
gest that the 1974 World Cup did not generate any statistically significant employ-
ment effects, in either the short term or the long term.
Hagn and Maennig (2009) report an analysis of the short-term effects on
unemployment of the 2006 World Cup, also hosted by Germany, using monthly
data for the seventy-five largest urban districts in Germany including the twelve
match venues for the period from January 1998 to March 2007. The dependent
variable is the monthly numbers of the unemployed for the urban districts and
the independent variables include lagged unemployment, population, dummy
variables for East German districts, seasonal dummy variables, gross value-
added sector dummies and a World Cup dummy variable for the host cities. The
empirical methodology is essentially the same as in Hagn and Maennig (2008).
As before, the results suggest that the pattern of unemployment in the twelve
match venues cities was not significantly different from its pattern in the non-
venue cities.6
14.5 Intangible benefits of mega events

Empirical evidence from retrospective studies suggests that the economic ben-
efits of hosting mega sporting events like the World Cup are doubtful at best.
However, one limitation of these studies is that they tend to overlook the pos-
sible intangible benefits (or ‘feelgood’ effects) that derive from the hosting of a
mega sporting event. Adopting a broader view of the potential benefits a mega
sporting event might deliver may help explain the fierce competition to host.
One of the intangible benefits most widely quoted is civic pride: mega events
bring intangible psychological benefits to the communities that host them.
Another key intangible benefit is that of national and international exposure.
Sports fans might enjoy their visit to the city and decide to pay a return visit later
on, raising future tourist revenues. Corporate visitors might decide to relocate
manufacturing or administrative facilities or company headquarters to the city.
Television viewers might decide to take a trip to the host city at some time in
the future, based on what they see of the city during the broadcast of the mega
event. Finally, hosting a major event might raise the profile or general aware-
ness of a city, so that it becomes a ‘major league’ or ‘world class’ city and travel
destination.
Of course, intangible effects of this kind are extremely difficult to quantify, and
there is a paucity of empirical research that can conclusively demonstrate any long-
run connections between the hosting of a mega event and future tourism demand
or inward investment. Using survey methods, Kim and Patrick (2005) investigate
Seoul residents’ perceptions of the impact of the 2002 World Cup. A factor ana-
lysis indicates that residents’ perceptions of the factors ‘tourism resource devel-
opment and urban revitalisation’, ‘image enhancement and consolidation’ and
‘interest in foreign countries or their cultures’ were positive. However, negative
perceptions were identified for several factors including ‘disorder and conflict’,
‘traffic problems and congestion’ and ‘negative economic perception’. The latter
may well reflect public concern over the levels of public expenditure that were
required to underwrite the event, and over the utilisation of the ten brand-new
soccer stadia after the World Cup was complete. In general, females, especially
housewives, had a more positive perception than males. Perceptions were rather
variable over time, and were less favourable three months after the event than they
were at the time.
A different approach to the measurement of intangible benefits is taken by
Kavetsos and Szymanski (2010). Drawing upon the economics of happiness litera-
ture, the impact of hosting a major sporting event (Olympic Games, World Cup
and European Football Championships) and the effect of athletic success in the
same event for the host nation, on happiness among the citizens of the host nation
is measured using data from the Eurobarometer Survey Series. The data cover
twelve countries – Belgium, Britain, Denmark, France, Germany, Greece, Ireland,
Italy, Luxembourg, Netherlands, Portugal and Spain – for the period 1974–2004.
The Surveys are conducted on behalf of the European Commission, in order to
collect information on the social, health and political aspects of the quality of life
for the populations of EU member states. The surveys are conducted twice per
year (spring and autumn) and each time around 1,000 randomly selected individ-
uals are interviewed in each country. A key question concerns satisfaction with
life, with interviewees invited to respond (on a four-point scale) to the follow-
ing: ‘On the whole, are you very satisfied, fairly satisfied, not very satisfied, or not
at all satisfied with the life you lead?’
In view of the ordinal structure of the survey responses, Kavetsos and
Szymanski use an ordered logit model, in which the independent variables cap-
ture effects emanating from a recently staged sports tournament (host-nation
dummy variables and indicators of medals or tournament rankings). For the
Olympic Games, the key performance measure is the difference between actual
and predicted medals; and for the two football tournaments, performance is
measured using the difference between each country’s pre- and post-tourna-
ment FIFA rankings. Controls are included for macroeconomic conditions
(GDP per capita, unemployment rate and inflation rate), personal charac-
teristics (employment status, sex, age, marital status, household income and
education), and individual country and time effects. The results contain little
evidence to suggest that happiness is correlated with better-than-expected ath-
letic performance; and for the Olympic Games, host-nation status appears to be
negatively associated with reported happiness. Host-nation status for the two
football events is positively associated with reported happiness, however, sug-
gesting that hosting an international football tournament may indeed produce
a tangible feelgood effect.
As seen in Chapter 12, Section 12.1, financial markets present opportunities for
the identification of sentiment effects. Ashton, Gerrard and Hudson (2003) track
changes in the FTSE 100 index on the days following matches played by England’s
national team, including World Cup matches. There is a tendency for the London
stock market to react positively to wins and negatively to losses. Edmans, Garcia
and Norli (2007) document the stock market reaction to match results in major
sporting events, including World Cup football, and international cricket, rugby
and basketball. A negative and statistically significant reaction to elimination
from the World Cup on the national stock market of the country whose team has
been eliminated is interpreted as evidence that sports results have a strong effect
on mood. The effect is interpreted as a reflection of sentiment, rather than as a
rational adjustment of stock prices to new information, for several reasons. The
effect does not appear to be impounded into stock prices in advance in cases where
elimination from the World Cup is strongly anticipated. The effect is strongest in
the stock markets of developed Western European countries, with strong football
traditions and national teams that always feature prominently among the tourna-
ment favourites. The effect is stronger in small stocks, whose prices are believed
to be sensitive to investor mood, because they are held predominantly by local
investors.
Heyne, Maennig and Suessmuth (2007) use contingent valuation methods to
evaluate the feelgood factor in Germany, following the hosting of the 2006 World
Cup. A survey asked 500 respondents about their willingness to pay to stage the
World Cup.7 Three months prior to the start of the tournament, the respond-
ents were informed that FIFA was (hypothetically) leaning towards relocating the
Finals to Switzerland (host of the European Football Championships in 2008) for
safety reasons. There was still a chance that the tournament would take place in
Germany, if a series of costly safety measures was implemented, to be financed
by means of immediate voluntary contributions from the population. Each per-
son was asked if they would be willing to contribute financially to ensure that the
Finals could be hosted in Germany. Three months after the Finals were staged, the
same respondents were asked to imagine being faced with the same pre-tournament
decision, and were offered the opportunity to change their original decision or the
size of their offered financial contribution, in the light of the experience of having
seen the tournament actually staged.
The average prospective offered contribution of respondents whose willingness
to pay was non-zero was $30.4; but less than 20 per cent of respondents reported
a non-zero willingness to pay. The average prospective willingness to pay for all
respondents was therefore only $5.66. Extrapolated over a total population of
82 million, the estimated total prospective willingness to pay was $467 million.
After the tournament was staged, however, more than 40 per cent of respondents
reported a positive willingness to pay. The average retrospective willingness to pay
for all respondents was $13.38, yielding an estimated total retrospective willing-
ness to pay of $1.1 billion.
The average retrospective offered contribution of respondents whose willing-
ness to pay was non-zero was $31.4, only marginally higher than the correspond-
ing prospective contribution. Accordingly, the increase in total willingness to pay
was driven almost entirely by an increase in the numbers who were willing to pay
anything, and not by any increase in the average amount each individual was will-
ing to contribute. Respondents from Eastern Germany, and respondents with
lower levels of educational attainment, were among the groups most likely to have
altered their decision as a result of having witnessed or experienced the hosting
of the tournament. In the language of consumer theory, the interpretation placed
on these findings is that large sports events have the characteristics of experience
goods: goods for which consumers cannot assess user value in advance, but only
upon consumption or from past experience (Nelson, 1970). The estimated total
retrospective willingness to pay of $1.1 billion suggests the value of the feelgood
effect is non-negligible, especially when compared to the doubtful or non-existent
direct economic effects identified by the retrospective studies that are reviewed in
Section 14.4.8
Conclusion
Advocates of the allocation of large publicly funded subsidies in order to secure
host nation status for mega events like the Olympic Games or football’s World
Cup routinely claim that hosting the event will deliver large economic benefits.
However, the overwhelming majority of independent academic studies sug-
gest that the economic impact of the hosting of mega events, or other sports
infrastructure projects, is limited. While the gross consumption and investment
expenditures that can be identified with mega events are undoubtedly large,
attracting tens or hundreds of thousands of live spectators as well as massive
TV audiences, most of the retrospective evidence suggests that the net impact of
mega events on economic indicators such as employment and per capita income
in host cities or nations is either small or negligible at best, and possibly even
negative at worst.
Academic economists have been quick to point out the failings of prospect-
ive economic impact studies that make extravagant claims of large benefits. Such
studies sometimes fail to distinguish between gross and net expenditures, because
they do not take account of substitution and crowding-out effects. Substitution
occurs if the expenditure of spectators at the event has been diverted from other
leisure activities based in the same locality that would have been patronised if
the event had not been staged. Crowding out occurs when economic activities
that are not associated with the event are diverted elsewhere, due to congestion or
constraints on the supply capacity of the local economy. Some studies prophesy-
ing large economic gains are based on overoptimistic assumptions concerning the
propensity for additional expenditures associated with the event to deliver eco-
nomic gains at the local level. Unrealistic assumptions concerning the size of local
multipliers may be incorporated, leading to the underestimation of leakages that
limit the benefits to local businesses of expenditures accruing during the prepar-
ation and the staging phases.
One way of squaring the enthusiasm with which cities or nations compete to
attract mega events with the lack of tangible evidence of any discernible economic
impact is to search for evidence of intangible benefits or ‘feelgood’ effects. Some
recent research suggests that intangible benefits might be rather important, and
perhaps even sufficient to provide a justification for the public subsidy of major
sporting events.
Notes
1 According to FIFA the 2006 World Cup was aired in a total of 43,600 broadcasts across 214
countries and territories, generating total coverage of 73,072 hours.
2 Matheson (2006) discusses a number of prospective (ex ante) economic impact studies
for other sports, and for mega events. Kasimati (2003) reviews prospective studies for the
Olympic Games.
3 The SAM, created by South Africa’s Financial and Fiscal Commission, includes forty-eight
economic activities and forty-eight household types, and disaggregates government accounts
according to the hierarchy of tax/spending authorities (Central, Provincial and Local/
Municipal) and in respect of the major revenue sources and major categories of expenditure.
The SAM is similar in nature to the RIMS II model that has been used for impact assess-
ments of the Olympic Games and World Cup when hosted by the USA.
4 Prospective CGE studies of the Olympic Games include Blake (2005) and New South Wales
Treasury (1997). Blake estimates that the 2012 London Olympic Games would lead to an
increase in GDP of £1.96 billion between 2005 and 2016, and create 8,164 permanent jobs.
5 Few retrospective studies report significant economic benefits from the hosting of mega events.
Most of those that do so are for the Olympic Games (Kang and Perdue, 1994; Hotchkiss,
Moore and Zobay, 2003; Jasmand and Maennig, 2008). Recent retrospective CGE studies
of the 2000 Sydney Olympic Games conclude that prospective assessments overestimated
the economic benefits of the event, because the assumptions relating to induced tourism and
the responsiveness of the labour market were too optimistic (Madden, 2006; Giesecke and
Madden, 2007).
6 In a cost-benefit analysis of the 2006 World Cup, Maennig and du Plessis (2007) obtain simi-
lar results for the impact on employment and sales.
7 Contingent valuation methods have been used to estimate willingness to pay for a number
of sports infrastructure projects in the USA. For example, Johnson and Whitehead (2000),
Johnson, Groothuis and Whitehead (2001) and Owen (2006) report that total willingness to
pay was insufficient to cover projected or realised costs.
8 Using contingent valuation methods, Atkinson et al. (2008) estimate that Britons would be
willing to pay £2 billion for the intangible benefits associated with hosting the 2012 Olympic
Games in London.
References
Ahlert, G. (2000). The economic effects of the soccer World Cup 2006 in Germany with regard
to different financing. Economic System Research 13, 109–127.
Aidt, T. S., Leong, B., Saslaw, W. C. and Sgroi, D. (2006). A power-law distribution for tenure
lengths of sports managers. Physica A 370, 697–703.
Akerlof, G. A. (1980). A theory of social custom of which unemployment may be one conse-
quence. Quarterly Journal of Economics 94, 749–775.
(1997). Social distance and social decisions. Econometrica 65, 1005–1027.
Akerlof, G. A. and Kranton, R. E. (2000). Economics and identity. Quarterly Journal of
Economics 115, 715–753.
Alavy, K., Gaskell, A., Leach, S. and Szymanski, S. (2010). On the edge of your seat: demand
for football on television and the uncertainty of outcome hypothesis. International Journal
of Sport Finance 5, 79–95.
Alchian, A. A. and Demsetz, H. (1972). Production, information costs and economic organisa-
tion. American Economic Review 62, 777–795.
Ali, M. M. (1977). Probability and utility estimates for racetrack bettors. Journal of Political
Economy 85, 803–815.
Allan, G. and Roy, G. (2008). Does television crowd out spectators? New evidence from the
Scottish Premier League. Journal of Sports Economics 9, 592–605.
Allen, M. P., Panian, S. K. and Lotz, R. E. (1979). Managerial succession and organizational per-
formance: a recalcitrant problem revisited. Administrative Science Quarterly 24, 167–180.
Amara, M., Henry, I., Liang, J. and Uchiumi, K. (2005). The governance of professional soc-
cer: five case studies – Algeria, China, England, France and Japan. European Journal of
Sport Science 5, 189–206.
Anderson, T. R. and Sharp, G. P. (1997). A new measure of baseball batters using DEA. Annals
of Operations Research 73, 141–155.
Antonioni, P. and Cubbin, J. (2000). The Bosman Ruling and the emergence of a single market
in soccer talent. European Journal of Law and Economics 9, 157–173.
Ascari, G. and Gagnepain, P. (2006). Spanish football. Journal of Sports Economics 6, 76–89.
Ashton, J. K., Gerrard, B. and Hudson, R. (2003). Economic impact of national sporting suc-
cess: evidence from the London stock exchange. Applied Economics Letters 10, 783–785.
Ashworth, J. and Heyndels, B. (2007). Selection bias and peer effects in team sports: the effect
of age grouping on earnings of German soccer players. Journal of Sports Economics 8,
355–377.
Atkinson, G., Mourato, S., Szymanski, S. and Ozdemiroglu, E. (2008). London’s bid to host the
2012 summer Olympic Games. Urban Studies 45, 419–444.
423
424 References
Atkinson, S. E., Stanley, L. R. and Tschirhart, J. (1988). Revenue sharing as an incentive in

an agency problem: an example from the National Football League. RAND Journal of
Economics 19, 27–43.
Attrill, M. J., Gresty, K. A., Hill, R. A. and Barton, R. A. (2008). Red shirt colour is associated
with long-term team success in English football. Journal of Sports Sciences 26, 577–582.
Audas, R., Dobson, S. and Goddard, J. (1997). Team performance and managerial change in the
English Football League. Economic Affairs 17, 30–36.
(1999). Organizational performance and managerial turnover. Managerial and Decision
(2002). The impact of managerial change on team performance in professional sports. Journal
of Economics and Business 54, 633–650.
Azar, O. H. and Bar-Eli, M. (2010). Do soccer players play the mixed-strategy Nash equilib-
rium? Applied Economics doi: 10.1080/00036841003670747.
Baade, R. A. (1987). Is there an economic rationale for subsidizing sports stadiums? The
Heartland Institute Policy Study 13.
(1994). Stadiums, professional sports, and economic development: assessing the reality. The
Heartland Institute Policy Study 62.
(2006). The economic impact of mega sports events, in: Andreff, W. and Szymanski, S. (eds.)
Handbook on the economics of sport. Cheltenham: Edward Elgar.
Baade, R. A. and Dye, R. (1990). The impact of stadiums and professional sports on metropol-
itan area development. Development Growth and Change 21, 1–14.
Baade, R. A. and Matheson, V. A. (2001). Home run or wild pitch? Assessing the economic impact
of Major League Baseball’s all-star game. Journal of Sports Economics 2, 307–327.
(2002). Bidding for the Olympics: fool’s gold?, in: Barros, C. P. and Ibrahim, M. (eds.)
Transatlantic sport. Cheltenham: Edward Elgar.
(2004). The quest for the cup: assessing the economic impact of the World Cup. Regional
Studies 38, 343–354.
Baade, R. A. and Sanderson, R. A. (1997). The employment effect of teams and sports facilities,
in: Noll, R. G. and Zimbalist, A. (eds.) Sports, jobs, and taxes. Washington, DC: Brookings
Institution.
Bachan, R., Reilly, B. and Witt, R. (2008). The hazard of being an English football league
manager: empirical estimates for three recent league seasons. Journal of the Operational
Research Society 59, 884–891.
Badarinathi, R. and Kochman, L. (1996). Football betting and the efficient market hypothesis.
American Economist 40, 52–55.
Baimbridge, M., Cameron, S. and Dawson, P. (1996). Satellite television and the demand for
football: a whole new ball game? Scottish Journal of Political Economy 43, 317–333.
Balduck, A. and Buelens, M. (2007). Does sacking the coach help or hinder the team in the
short term? Evidence from Belgian soccer. Working Paper 2007/430, University of Ghent.
Bale, J. (1993). Sport, space and the city. London: Routledge.
Balmer, N. J., Nevill, A. M. and Williams, M. (2001). Home advantage in the Winter Olympics
(1908–1998). Journal of Sports Sciences 19, 129–139.
Banerjee, A. N., Swinnen, J. F. N. and Weersink, A. (2007). Skating on thin ice: rule changes and
team strategies in the NHL. Canadian Journal of Economics 40, 493–514.
Bar-Eli, M., Azar, O. H., Ritov, I., Keidar-Levin, I. and Schein, G. (2007). Action bias among
elite soccer goalkeepers. Journal of Economic Psychology 28, 606–621.
Baranzini, A., Ramirez, J. and Weber, S. (2008). The demand for football in Switzerland: an
empirical estimation. Working Paper, University of Applied Sciences of Western
Switzerland, Geneva School of Business Administration.
Barnett, V. and Hilditch, S. (1993). The effect of an artificial pitch surface on home team per-
formance in football (soccer). Journal of the Royal Statistical Society Series A 156, 39–50.
References 425
Baroncelli, A. and Lago, U. (2006). Italian football. Journal of Sports Economics 7, 13–28.
Barros, C. P., Assaf, A. and Earp, F. (2010). Brazilian football league technical efficiency: a
Simar and Wilson approach. Journal of Sports Economics 11, 641–651.
Barros, C. P., del Corral, J. and Garcia-del-Barrio, P. (2008). Identification of segments of soc-
cer clubs in the Spanish League First Division with a latent class model. Journal of Sports
Barros, C. P. and Garcia-del-Barrio, P. (2008). Efficiency measurement of the English football
Premier League with a random frontier model. Economic Modelling 25, 994–1002.
Barros, C. P., Garcia-del-Barrio, P. and Leach, S. (2009). Analysing the technical efficiency of the
Spanish Football League First Division with a random frontier model. Applied Economics
41, 3239–3247.
Barros, C. P. and Leach, S. (2006). Analyzing the performance of the English F.A. Premier
League with an econometric frontier model. Journal of Sports Economics 7, 391–407.
(2007). Technical efficiency in the English Football Association Premier League with a sto-
chastic cost frontier. Applied Economics Letters 14, 731–741.
Battese, G. and Coelli, T. (1992). Frontier production functions, technical efficiency and panel
data: with application to paddy farmers in India. Journal of Productivity Analysis 3, 153–169.
Battré, M., Deutscher, C. and Frick, B. (2009). Salary determination in the German Bundesliga: a
panel study. University of Paderborn. Mimeo.
Baumann, F., Friehe, T. and Wedow, M. (2010). General ability and specialization: evidence from
penalty kicks in soccer. Journal of Sports Economics doi: 10.1177/1527002510371194.
Baur, D. and Lehmann, S. (2008). Does the mobility of football players influence the success of
the national team? Institute for International Integration Studies, Trinity College, Dublin.
Mimeo.
Becker, G. S. (1971). The economics of discrimination, 2nd edn. University of Chicago Press.
Becker, G. S. and Murphy, K. (2000). Social economics: market behavior in a social environment.
Cambridge, MA: Harvard University Press.
Becker, M. A. and Suls, J. (1983). Take me out to the ballgame: the effects of objective, social
and temporal performance information on attendance at Major League baseball games.
Journal of Sport Psychology 5, 302–313.
Bellemore, F. A. (2001). Racial and ethnic employment discrimination: promotion in major
league baseball. Journal of Sports Economics 2, 356–368.
Bellos, A. (2002). Futebol: the Brazilian way of life. London: Bloomsbury.
Benz, M. A., Brandes, L. and Franck, E. (2009). Do soccer associations really spend on a good
thing? Empirical evidence on heterogeneity in the consumer response to match uncertainty
of outcome. Contemporary Economic Policy 27, 216–235.
Berle, A. A. and Means, G. C. (1932). The modern corporation and private property. New
York: Macmillan.
Bernheim, D. (1994). A theory of conformity. Journal of Political Economy 102, 841–877.
Bernile, G. and Lyandres, E. (2008). Understanding investor sentiment: the case of soccer.
University of Miami, School of Business. Mimeo.
Berri, D. J. and Jewell, T. (2004). Wage inequality and firm performance: professional basket-
ball’s natural experiment. Atlantic Economic Journal 32, 130–139.
Berri, D. J., Schmidt, M. B. and Brook, S. L. (2004). Stars at the gate. Journal of Sports Economics
5, 33–50.
Berri, D. J. and Simmons, R. (2009). Race and the evaluation of signal callers in the National
Football League. Journal of Sports Economics 10, 23–43.
Birchall, J. (2000). Ultra Nippon: how Japan reinvented football. London: Headline.
Bird, P. (1982). The demand for league football. Applied Economics 14, 637–649.
Bittner, E., Nussbaumer, A., Janke, W. and Weigel, M. (2007). Self-affirmation model for foot-
ball goal distributions. Europhysics Letters 78, 58002.
426 References
Bittner, E., Nußbaumer, A., Janke, W. and Weigel, M. (2009). Football fever: goal distributions
and non-Gaussian statistics. The European Physics Journal B 67, 459–471.
Blake, A. (2005). The economic impact of the London 2012 Olympics. Discussion Paper 2005/05,
Christel deHaan Tourism and Travel Research Institute, Nottingham University.
Bocketti, G. (2008). Italian immigrants, Brazilian football, and the dilemma of national identity.
Journal of Latin American Studies 40, 275–302.
Bodvarsson, Ö. and Brastow, R. T. (1998) Do employers pay for consistent performance?
Evidence from the NBA. Economic Inquiry 36, 145–160.
Bohlmann, H. R. (2006). Predicting the economic impact of the 2010 FIFA World Cup on
South Africa. Department of Economics Working Paper Series 2006–11, University of
Pretoria. www.up.ac.za/up/web/en/academic/economics/index.html.
Bohlmann, H. R. and van Heerden, J. H. (2008). Predicting the economic impact of the
2010 FIFA World Cup on South Africa. International Journal of Sport Management and
Marketing 3, 383–396.
Borland, J. (1987). The demand for Australian rules football. Economic Record 63, 220–230.
Borland, J. and MacDonald, R. (2003). Demand for sport. Oxford Review of Economic Policy
19, 478–502.
Boscá, J. E., Liern, V., Martínez, A. and Sala, R. (2009). Increasing offensive or defensive effi-
ciency? An analysis of Italian and Spanish football. Omega 37, 63–78.
Bourg, J. F. (2004). Professional team sports in Europe: which economic model? in: Fort, R. and
Fizel, J. (eds.) International sports economics comparisons. London: Praeger.
Boyko, R., Boyko, A. and Boyko, M. (2007). Referee bias contributes to home advantage in
English Premiership football. Journal of Sports Sciences 26, 577–582.
Brandes, L. and Franck, E. (2007). Who made who: an empirical analysis of competitive bal-
ance in European soccer leagues. Eastern Economic Journal 33, 379–403.
Brandes, L., Franck, E. and Nuesch, S. (2008). Local heroes and superstars: an empirical ana-
lysis of star attraction in German soccer. Journal of Sports Economics 9, 266–286.
Brase, J. and Reichart, T. (2005). Expensive goals – the sluggish financial statements of the
Bundesliga. www.zdf.de/ZDFde/Inhalt/28/0,1872,2289628,00.html.
Brocas, I. and Carrillo, J. D. (2004). Do the ‘three-point victory’ and ‘golden goal’ make soccer
more exciting? Journal of Sports Economics 5, 169–185.
Brown, E., Spiro, R. and Keenan, D. (1991). Wage and nonwage discrimination in professional
basketball: do fans affect it? American Journal of Economics and Sociology 50, 333–345.
Brown, M. (1982). Administrative succession and organizational performance: the succession
effect. Administrative Science Quarterly 27, 1–16.
Brown, T. D., Van Raalte, J. L., Brewer, B. W., Winter, C. R. and Cornelius, A. A. (2002). World
cup soccer home advantage. Journal of Sport Behavior 25, 134–144.
Brown, W. O. and Sauer, R. D. (1993). Does the baseball market believe in the hot hand: a com-
ment. American Economic Review 83, 1377–1386.
Bruinshoofd, A. and ter Weel, B. (2003). Manager to go? Performance dips reconsidered with
evidence from Dutch football. European Journal of Operational Research 148, 233–246.
Bryson, A., Frick, B. and Simmons, R. (2009). The returns to scarce talent: footedness and
player remuneration in European soccer. CEP Discussion Paper No. 948.
Buraimo, B. (2008). Stadium attendance and television audience demand in English league foot-
ball. Managerial and Decision Economics 29, 513–523.
Buraimo, B., Forrest, D. and Simmons, R. (2009). Insights for clubs from modelling match
attendance in football. Journal of the Operational Research Society 60, 147–155.
Buraimo, B., Forrest, D. and Simmons, R. (2010). The twelfth man? Refereeing bias in English
and German soccer. Journal of the Royal Statistical Society Series A 173, 431–449.
Burdekin, R. C., Hossfeld, R. T. and Kilholm-Smith, J. (2005). Are NBA fans becoming indif-
ferent to race? Evidence from the 1990s. Journal of Sports Economics 6, 144–159.
References 427
Burkitt, B. and Cameron, S. (1992). Impact of league restructring on team sport attendances: the
case of rugby league. Applied Economics 24, 265–271.
Butler, B. (1987). The Oxford illustrated history of the Football League. London: Macdonald
Queen Anne Press.
Buzzacchi, L., Szymanski, S. and Valletti, T. (2003). Equality of opportunity and equality
of outcome: open leagues, closed leagues and competitive balance. Journal of Industry,
Competition and Trade 3, 167–186.
Cain, L. P. and Haddock, D. D. (2006). Measuring parity: tying into the idealized standard devi-
ation. Journal of Sports Economics 7, 330–338.
Cain, M., Law, D. and Peel, D. (2000). The favourite-longshot bias and market efficiency in UK
football betting. Scottish Journal of Political Economy 47, 25–36.
Cairns, J. A. (1987). Evaluating changes in league structure: the reorganisation of the Scottish
Football League. Applied Economics 19, 259–275.
(1990). The demand for professional team sports. British Review of Economic Issues 12,
1–20.
Cairns, J. A., Jennett, N. and Sloane, P. J. (1986). The economics of professional team sports: a
survey of theory and evidence. Journal of Economic Studies 13, 1–80.
Camerer, C. (1989). Does the baseball market believe in the hot hand? American Economic
Review 79, 1257–1261.
Carillo, J. D. (2007). Penalty shoot-outs: before or after extra time? Journal of Sports Economics
8, 505–518.
Carmichael, F. and Thomas, D. (1995). Production and efficiency in team sports: an investiga-
tion of rugby league football. Applied Economics 27, 859–869.
(2005). Home-field effect and team performance: evidence from English Premiership football.
Journal of Sports Economics 6, 264–281.
Carmichael, F., Thomas, D. and Ward, R. (2000). Team performance: the case of English
Premiership football. Managerial and Decision Economics 21, 31–45.
(2001). Production and efficiency in association football. Journal of Sports Economics 2,
228–243.
Carron, A., Loughhead, T. and Bray, S. (2005). The home advantage in sport competi-
tions: Courneya and Carron’s (1992) conceptual framework a decade later. Journal of
Sports Sciences 23, 395–407.
Carter, N. (2006). The football manager: a history. London: Routledge.
Casual Auditores Independentes (2008). Results of the casual list of Brazilian football clubs.
Sao Paulo: Casual Auditores Independentes.
Cave, M. (2000). Football rights and competition in broadcasting, in: Hamil, S., Michie, J.,
Oughton, C. and Warby, S. (eds.) Football in the digital age: whose game is it anyway?
Edinburgh: Mainstream.
Cave, M. and Crandall, R. (2001). Sports rights and the broadcast industry. Economic Journal
111, 4–26.
Chapman, S. J. (2007). The Kelly criterion for spread bets. IMA Journal of Applied Mathematics
72, 43–51.
Chiappori, P. A., Levitt, S. and Groseclose, T. (2002). Testing mixed-strategy equilibria when
players are heterogeneous: the case of penalty kicks in soccer. American Economic Review
92, 1138–1151.
Christiansen, N. A. and Sievertsen, H. H. (2008). The exploitation of talent. Nationaløkonomisk
Tidsskrift 146, 319–326.
Clarke, S. R. and Norman, J. M. (1995). Home ground advantage of individual clubs in English
soccer. The Statistician 44, 509–521.
Clowes, J. and Clements, N. (2003). An explanation of discriminatory ticket pricing practice in
the English football Premier League. Managing Leisure 8, 105–120.
428 References
Coase, R. H. (1937). The nature of the firm. Economica 4, 386–405.

(1960). The problem of social cost. Journal of Law and Economics 3, 1–44.
Coates, D. and Humphreys, B. R. (1999). The growth effects of sport franchises, stadia, and
arenas. Journal of Policy Analysis and Management 18, 601–624.
(2000a). The stadium gambit and local economic development. Cato Review of Business and
Government 23, 15–20.
(2000b). The economic consequences of professional sports strikes and lockouts. Southern
Economic Journal 67, 737–747.
(2002). The economic impact of postseason play in professional sports. Journal of Sports
(2003a). The effect of professional sports on earnings and employment in the services and
retail sectors in US cities. Regional Science and Urban Economics 33, 175–198.
(2003b). Professional sports facilities, franchises and urban economic development. Public
Finance and Management 3, 335–357.
Coloma, G. (2007). Penalty kicks in soccer: an alternative methodology for testing mixed-
strategy equilibria. Journal of Sports Economics 8, 530–545.
Conlin, M. and Emerson, P. M. (2006). Discrimination in hiring v. retention and promo-
tion: an empirical analysis of within firm treatment of workers in the NFL. Journal of Law,
Economics and Organization 22, 115–136.
Conn, D. (1997). The football business: fair game in the 90s? Edinburgh: Mainstream.
Coon, L. (2010). Larry Coon’s NBA salary cap FAQ. Available at: http://members.cox.net/
lmcoon/salarycap.htm
Cornwell, C., Schmidt, P. and Sickles, R. C. (1990). Production frontiers with cross-sectional
and time-series variation in efficiency levels. Journal of Econometrics 46, 185–200.
Courneya, A. and Carron, A. (1992). The home advantage in sport competitions: a literature
review. Journal of Sport and Exercise Psychology 14, 28–39.
Cox, D. R. (1972). Regression models and life tables. Journal of the Royal Statistical Society
Series B 34, 187–202.
Coyne, C., Isaacs, J. and Schwartz, J. (2010). Entrepreneurship and the taste for discrimination.
Journal of Evolutionary Economics 20, 609–627.
Crafts, N. F. R. (1985). Some evidence of insider knowledge in horse racing betting in Britain.
Economica 52, 295–304.
Crompton, J. L. (1995). Economic impact analysis of sports facilities and events: eleven sources
of misapplication. Journal of Sport Management 9, 14–35.
Crowder, M., Dixon, M., Ledford, A. and Robinson, M. (2002). Dynamic modelling and pre-
diction of English Football League matches for betting. The Statistician 51, 157–168.
Curme, M. A. and Daugherty, G. (2004). Competition and pay for National Hockey League
players born in Quebec. Journal of Sports Economics 5, 186–205.
Curran, J., Jennings, I. and Sedgwick, J. (2009). Competitive balance in the top level of English
football, 1948–2008: an absent principle and a forgotten ideal. International Journal of the
History of Sport 26, 1668–1680.
Cyrenne, P. (2009). Modelling professional sports leagues: an industrial organization approach.
Review of Industrial Organization 34, 193–215.
Czarnitzki, D. and Stadtmann, G. (2002). Uncertainty of outcome versus reputation: empirical
evidence for the first German football division. Empirical Economics 27, 101–112.
Darby, P. and Solberg, E. (2010). Differing trajectories: football development and patterns of
player migration in South Africa and Ghana. Soccer and Society 11, 118–130.
Dare, W. H. and MacDonald, S. (1996). A generalized model for testing the home and favorite
team advantage in point spread markets. Journal of Financial Economics 40, 295–318.
Davies, B., Downward, P. and Jackson, I. (1995). The demand for rugby league: evidence from
causality tests. Applied Economics 27, 1003–1007.
References 429
Davies, M., Pitt, L., Shapiro, D. and Watson, R. (2005). Betfair.com: five technology forces
revolutionize worldwide wagering. European Management Journal 23, 533–541.
Dawson, P. and Dobson, S. (2002). Managerial efficiency and human capital: an application to
English association football. Managerial and Decision Economics 23, 471–486.
(2010). The influence of social pressure and nationality on individual decisions: evidence
from the behaviour of referees. Journal of Economic Psychology 31, 181–191.
Dawson, P., Dobson, S. and Gerrard, B. (2000a). Stochastic frontiers and the temporal structure
of managerial efficiency in English soccer. Journal of Sports Economics 1, 341–362.
(2000b). Estimating coaching efficiency in professional team sports: evidence from English
Association football. Scottish Journal of Political Economy 47, 399–421.
Dawson, P., Dobson, S., Goddard, J. and Wilson, J. (2007). Are football referees really biased
and inconsistent? Evidence on the incidence of disciplinary sanction in the English Premier
League. Journal of the Royal Statistical Society Series A 170, 231–250.
de Dios Tena, J. and Forrest, D. (2007). Within-season dismissal of football coaches: statis-
tical analysis of causes and consequences. European Journal of Operational Research 181,
362–373.
de Paola, M. and Scoppa, V. (2008). The effects of managerial turnover: evidence from coach
dismissals in Italian soccer teams. MPRA Paper No. 11030.
Deloitte. (2009). Annual review of football finance. Manchester: Deloitte.
(2010). Football money league. Manchester: Deloitte.
Demmert, H. H. (1973). The economics of professional team sports. Lexington: DC Heath.
Department of Labour and Productivity (1971). British labour statistics, historical abstract
1886–1968. London: Department of Labour and Productivity.
Depken II, C. A. (1999). Free-agency and the competitiveness of Major League Baseball. Review
of Industrial Organization 14, 205–217.
Depken II, C. A. (2000). Wage disparity and team productivity: evidence from Major League
Baseball. Economics Letters 67, 87–92.
Depken II, C. A. and Ford, J. M. (2006). Customer-based discrimination against major league
baseball players: additional evidence from all-star ballots. Journal of Socio-Economics 35,
1061–1077.
Dey, M. S. (1997). Racial differences in National Basketball Association players’ salaries: another
look. American Economist 41, 84–90.
Dietl, H. and Franck, E. (2007). Governance failure and financial crisis in German football.
Dietl, H., Franck, E. and Lang, M. (2008). Overinvestment in team sports leagues: a contest
theory model. Scottish Journal of Political Economy 55, 353–368.
Dietl, H., Franck, E. and Nuesch, S. (2006). Are voluntary salary cap agreements self-enforcing?
European Sport Management Quarterly 6, 23–34.
Dietl, H., Franck, E., Lang, M. and Rathke, A. (2008). Welfare effects of salary caps in sports
leagues with win-maximising clubs. Working Paper 08–25, International Association of
Sports Economists.
Dietl, H. and Lang, M. (2008). The effect of gate revenue sharing on social welfare. Contemporary
Economic Policy 26, 448–459.
Dietl, H., Lang, M. and Rathke, A. (2009). The effect of salary caps in professional team sports
on social welfare. The B.E. Journal of Economic Analysis and Policy 9, Article 17.
Dietl, H., Lang, M. and Werner, S. (2009). Social welfare in sports leagues with profit-maximiz-
ing and/or win-maximizing clubs. Southern Economic Journal 76, 375–396.
Dilger, A. and Geyer, H. (2009). Are three points for a win really better than two? A compari-
son of German soccer league and cup games. Journal of Sports Economics 10, 305–318.
Dixon, M. J. and Coles, S. G. (1997). Modelling association football scores and inefficiencies in
the football betting market. Applied Statistics 46, 265–280.
430 References
Dixon, M. J. and Pope, P. F. (2004). The value of statistical forecasts in the UK association foot-
ball betting market. International Journal of Forecasting 20, 697–711.
Dixon, M. J. and Robinson, M. E. (1998). A birth process model for association football
matches. The Statistician 47, 523–538.
Dobson, S. and Goddard, J. (1992). The demand for standing and seating viewing accommoda-
tion in the English Football League. Applied Economics 24, 1155–1163.
(1995). The demand for professional league football in England and Wales 1925–1992. The
Statistician 44, 259–277.
(1996). The demand for football in the regions of England and Wales. Regional Studies 30,
443–453.
(1998a). Performance and revenue in professional league football: evidence from Granger
causality tests. Applied Economics 30, 1641–1651.
(1998b). Performance, revenue and cross subsidisation in the English Football League,
1927–94. Economic History Review 51, 763–785.
(2001). The economics of football, 1st edn. Cambridge University Press.
(2003). Persistence in sequences of football match results: a Monte Carlo analysis. European
Journal of Operational Research 148, 247–256.
(2004). Revenue divergence and competitive balance in a divisional sports league. Scottish
Journal of Political Economy 51, 359–376.
(2010). Optimizing strategic behaviour in a dynamic setting in professional team sports.
European Journal of Operational Research 205, 661–669.
Dobson, S., Goddard, J. and Wilson, J. (2001). League structure and match attendances in
English rugby league. International Review of Applied Economics 15, 335–351.
Dohmen, T. J. (2008a). The influence of social forces: evidence from the behavior of football
referees. Economic Inquiry 46, 411–424.
(2008b). Do professionals choke under pressure? Journal of Economic Behaviour and
Organization 65, 636–653.
Dowie, J. (1976). On the efficiency and equity of betting markets. Economica 43, 139–150.
Downward, P. and Dawson, A. (2000). The economics of professional team sports.
London: Routledge.
Downward, P., Dawson, A. and Dejonghe, T. (2009). Sports economics: theory, evidence and
policy. Oxford: Butterworth-Heinemann.
Downward, P. and Jones. M. (2007). Effects of crowd size on referee decisions: analysis of the
FA Cup. Journal of Sport Sciences 25, 1541–1545.
Dunning, E., Murphy, P. and Williams, J. (1988). The roots of football hooliganism: an historical
and sociological study. London and New York: Routledge and Kegan Paul.
Eastham, J. (1999). The organisation of French football today, in: Dauncey, H. and Hare, G.
(eds.) France and the 1998 World Cup. London: Frank Cass.
Easton, S. T. and Rockerbie, D. W. (2005). Revenue sharing, conjectures and scarce talent in a
sports league model. Journal of Sports Economics 6, 359–378.
Eckard, E. W. (1998). The NCAA cartel and competitive balance in college football. Review of
Economic Organization 13, 347–369.
(2001). Free agency, competitive balance, and diminishing returns to pennant contention.
Economic Inquiry 39, 430–443.
Edmans, A., Garcia, D. and Norli, O. (2007). Sports sentiment and stock returns. Journal of
Finance 62, 1967–1998.
El-Hodiri, M. and Quirk, J. (1971). An economic model of a professional sports league. Journal
of Political Economy 79, 1302–1319.
Einolf, K. W. (2004). Is winning everything? A data envelopment analysis of Major League
Baseball and the National Football League. Journal of Sports Economics 5, 127–151.
References 431
Ericson, T. (2000). The Bosman Case: effects of the abolition of the transfer fee. Journal of
Sports Economics 1, 203–218.
Eschker, E., Perez, S. J. and Siegler, M. V. (2004). The NBA and the influx of international bas-
ketball players. Applied Economics 36, 1009–1020.
Espita-Escuer M. and García-Cebrián, L. (2004). Measuring the efficiency of Spanish first div-
ision soccer teams. Journal of Sports Economics 5, 329–346.
Falter, J. M., Perignon, C. and Vercruysse, O. (2008). Impact of overwhelming joy on consumer
demand: the case of a soccer World Cup victory. Journal of Sports Economics 9, 20–42.
Feess, E., Frick, B. and Muehlheusser, G. (2004). Legal restrictions on buyout fees: theory and
evidence from German soccer. IZA Discussion Paper No. 1180, Bonn.
Feess, E. and Muehlheusser, G. (2003a). Transfer fee regulations in European football. European
Economic Review 47, 645–668.
Feess E. and Muehlheusser, G. (2003b). The impact of transfer fees on professional sports: an
analysis of the new transfer system for European football. Scandinavian Journal of
Economics 105, 139–154.
Feess, E. and Stahler, F. (2009). Revenue sharing in professional sports leagues. Scottish Journal
FIFA (2008). FIFA World Cup and television. www.fifa.com/mm/document/fifafacts/ffprojects/
ip-401_06e_tv_2658.pdf
Fikentscher, A. (2006). Joint purchasing of sports rights: a legal viewpoint, in: Jeanrenaud, C.
and Késenne, S. (eds.) The economics of sport and the media. Cheltenham: Edward Elgar.
Finer, J. (2002). The grand illusion. Far Eastern Economic Review 7, 32–36.
Fishwick, N. (1989). English football and society, 1910–1950. Manchester University Press.
Fitt, A. D. (2009). Markowitz portfolio theory for soccer spread betting. IMA Journal of
Management Mathematics 20, 167–184.
Fitt, A. D., Howls, C. J. and Kabelka, M. (2006). Valuation of soccer spread bets. Journal of the
Operational Research Society 57, 975–985.
Fizel, J. L. and D ’ Itri, M. (1996). Estimating managerial efficiency: the case of college basket-
ball coaches. Journal of Sport Management 10, 435–445.
(1997). Managerial efficiency, managerial succession and organizational performance.
Managerial and Decision Economics 18, 295–308.
Flores, R., Forrest, D. and de Dios Tena, J. (2008). Decision taking with external pressure: evi-
dence on football manager dismissals in Argentina and their consequences. Fundacion de
las Cajas de Ahorros, Documento de Trabajo, No. 411/2008.
Foley, M. and Smith, F. H. (2007). Consumer discrimination in professional sports: new evi-
dence from major league baseball. Applied Economics Letters 14, 951–955.
Forrest, D., Goddard, J. and Simmons, R. (2005). Odds-setters as forecasters: the case of English
football. International Journal of Forecasting 21, 551–564.
Forrest, D. and Simmons, R. (2002). Outcome uncertainty and attendance demand in sport: the
case of English soccer. The Statistician 51, 229–241.
(2006). New issues in attendance demand: the case of the English Football League. Journal of
(2008). Sentiment in the betting market on Spanish football. Applied Economics 40,
119–126.
Forrest, D., Simmons, R. and Buraimo, B. (2005). Outcome uncertainty and the couch potato
audience. Scottish Journal of Political Economy 52, 641–661.
Forrest, D., Simmons, R. and Feehan, P. (2002). A spatial cross-sectional analysis of the elasti-
city of demand for soccer. Scottish Journal of Political Economy 49, 336–355.
Forrest, D., Simmons, R. and Szymanski, S. (2004). Broadcasting, attendance and the ineffi-
ciency of cartels. Review of Industrial Organization 24, 243–265.
432 References
Fort, R. (2000). European and North American sports differences(?) Scottish Journal of Political
Economy 47, 431–455.
(2006). Competitive balance in North American sports, in: Fizel, J. (ed.) Handbook of sports
economics research. Armonk, NY: M.E. Sharpe.
(2007). Comments on ‘measuring parity’. Journal of Sports Economics 8, 642–651.
Fort, R. and Lee, Y. H. (2007). Structural change, competitive balance, and the rest of the major
leagues. Economic Inquiry 45, 519–532.
Fort, R., Lee, Y. H. and Berri, D. (2008). Race, technical efficiency, and retention: the case of
NBA coaches. International Journal of Sport Finance 3, 84–97.
Fort, R. and Quirk, J. (1995). Cross-subsidization, incentives, and outcomes in professional
team sports leagues. Journal of Economic Literature 33, 1265–1299.
(2004). Owner objectives and competitive balance. Journal of Sports Economics 5, 20–32.
Fort, R. and Rosenman, R. (1999). Winning and management for streaks. Proceedings of
the Joint Statistical Meetings: Section on Statistics in Sports. Alexandria, VA: American
Statistical Association.
Fort, R. and Winfree, J. (2009). Sports really are different: the contest success function and the
supply of talent. Review of Industrial Organization 34, 69–80.
Franck, E. and Nüesch, S. (2010). The effect of wage dispersion on team outcome and the way
team outcome is produced. Applied Economics doi: 10.1080/00036840903427224.
Franck, E., Verbeek, E. and Nuesch, S. (2008). The sentiment bias in English soccer bet-
ting. Working Paper No. 89, University of Zurich, Institute for Strategy and Business
Economics.
Franck, E., Verbeek, E. and Nüesch, S. (2010). Prediction accuracy of different market struc-
tures - bookmakers versus a betting exchange. International Journal of Forecasting 26,
448–459.
Frick, B. (2003). Contest theory and sport. Oxford Review of Economic Policy 19, 512–529.
(2006). Salary determination and the pay-performance relationship in professional soc-
cer: Evidence from Germany, in: Rodriguez, P., Késenne, S. and Garcia, J. (eds.) Sports
economics after 50 years: essays in honor of Simon Rottenberg. Oviedo: Ediciones de la
Universidad de Oviedo.
(2007). The football players’ labour market: empirical evidence from the major European
leagues. Scottish Journal of Political Economy 54, 422–446.
(2009). Globalization and factor mobility: the impact of the ‘Bosman-Ruling’ on player
migration in professional soccer. Journal of Sports Economics 10, 88–106.
Frick, B. and Deutscher, C. (2009). Salary determination in the German Bundesliga: a panel
study. University of Paderborn. Mimeo.
Frick, B. and Lee, Y.H. (2010). Temporal variations in technical efficiency: evidence from
German soccer. Journal of Productivity Analysis doi: 10.1007/s11123-010-0175-1.
Frick, B., Barros, C. P. and Passos, J. (2009). Coaching for survival: the hazards of head coach
careers in the German ‘Bundesliga’. Applied Economics 41, 3303–3311.
Frick, B., Barros, C. P. and Prinz, J. (2010). Analysing head coach dismissals in the German
‘Bundesliga’ with a mixed logit approach. European Journal of Operational Research 200,
151–159.
Frick, B., Prinz, J. and Winkelmann, K. (2003). Pay inequalities and team performance. Empirical
evidence from the North American major leagues. International Journal of Manpower 24,
472–488.
Gamson, W. A. and Scotch, N. A. (1964). Scapegoating in baseball. American Journal of
Sociology 70, 69–72.
Gandar, J. M., Zuber, R. A., O ’ Brien, T. and Russo, B. (1988). Testing rationality in the point
spread betting market. Journal of Finance 43, 995–1008.
References 433
Gandar, J. M., Dare, W. H., Brown, C. R. and Zuber, R. A. (1998). Informed traders and price
variations in the betting market for professional basketball games. Journal of Finance 53,
385–401.
Gannon, J., Evans, K. and Goddard, J. (2006). The stock market effects of the sale of live broad-
casting rights for English Premiership football. Journal of Sports Economics 7, 168–185.
Garcia, J. and Rodriguez, P. (2002). The determinants of football match attendance revis-
ited: empirical evidence from the Spanish football league. Journal of Sports Economics 3,
18–38.
Garcia-del-Barrio, P. and Pujol, F. (2009). The rationality of under-employing the best-
performing soccer players. Labour 23, 397–419.
Garcia-Sanchez, I. M. (2007). Efficiency and effectiveness of Spanish football teams: a three-
stage-DEA approach. Central European Journal of Operations Research 15, 21–45.
Garicano, L., Palacios-Huerta, I. and Prendergast, C. (2005). Favoritism under social pressure.
Review of Economics and Statistics 87, 208–216.
Gerrard, B. (2006). Financial innovation in professional team sports: the case of English
Premiership soccer, in: Andreff, W. and Szymanski, S. (eds.) Handbook on the economics of
sport. Cheltenham: Edward Elgar.
Giesecke, J. A. and Madden, J. R. (2007). Regional adjustment to globalization: developing a
CGE analytical framework, in: Cooper, R. J., Donaghy, K. P. and Hewings, G. J. D. (eds.)
Globalization and economic modelling. Berlin: Springer-Verlag.
Gius, M. and Johnson, D. (1998). An empirical investigation of wage discrimination in profes-
sional basketball. Applied Economics Letters 5, 703–705.
Goddard, J. (2005). Regression models for forecasting goals and match results in English foot-
ball. International Journal of Forecasting 21, 331–340.
Goddard, J. and Asimakopoulos, I. (2004). Modelling football match results, and the efficiency
of fixed-odds betting. Journal of Forecasting 23, 51–66.
Goddard, J., Sloane, P. and Wilson, J. (2010). The Bosman ruling and labour mobility in foot-
ball, in: Kahane, L. and Schmanske, S. (eds.) Handbook of sports economics. Oxford
University Press.
Goddard, J. and Wilson, J. (2009). Racial discrimination in English professional football: evi-
dence from an empirical analysis of players’ career progression. Cambridge Journal of
Goldberg, A. and Wragg, S. (1991). It’s not a knock-out: English football and globalisation,
in: Williams, J. and Wragg, S. (eds.) British football and social change: getting into Europe.
Leicester University Press.
Golec, J. and Tamarkin, M. (1991). The degree of inefficiency in the football betting mar-
ket: statistical tests. Journal of Financial Economics 30, 311–323.
(1998). Bettors love skewness, not risk, at the horse track. Journal of Political Economy 106,
205–225.
Goossens, J. (2006). Competitive balance in European football: comparison by adapting meas-
ures: national measure of seasonal imbalance and top 3. Revista Di Dirrito Ed Economia
dello Sport, 2, 78–122.
Graham, I. and Stott, H. (2008). Predicting bookmaker odds and efficiency for UK football.
Applied Economics 40, 99–109.
Grant Thornton (2003). South Africa 2010 Soccer World Cup Bid Executive Summary. www.
polity.org.za.
(2008). 2010 Soccer World Cup Facts You Should Know. www.gauteng.net/research/pdf/
soccer1.pdf.
Gray, P. K. and Gray, S. F. (1997). Testing market efficiency: evidence from the NFL sports bet-
ting market. Journal of Finance 52, 1729–1737.
434 References
Greene, W. (2005). Fixed and random effects in stochastic frontier models. Journal of Productivity
Analysis 23, 7–32.
Groot, L. (2008). Economics, uncertainty and European football. Cheltenham: Edward Elgar.
Groothuis, P.A. and Hill, J.R. (2008). Exit discrimination in baseball 1990–2004. Southern
Grossmann, M., Dietl, H. and Lang, M. (2010). Revenue sharing and competitive balance in a
dynamic contest model. Review of Industrial Organization 36, 7–36.
Grusky, O. (1963). Managerial succession and organizational effectiveness. American Journal of
Sociology 69, 21–31.
(1964). Reply to scapegoating in baseball. American Journal of Sociology 70, 72–76.
Gustafson, E. (2006). The luxury tax in professional sports, in: Andreff, W. and Szymanski, S.
(eds.) Handbook on the economics of sport. Cheltenham: Edward Elgar.
Gustafson, E. and Hadley, L. (1995). Arbitration and salary gaps in Major League Baseball.
Quarterly Journal of Business and Economics 34, 32–46.
(1996). The luxury tax proposal for major league baseball: a partial equilibrium analysis,
in: Fizel, J., Gustafson, E. and Hadley, L. (eds.) Baseball economics: current research.
Westport, CT: Praeger.
Guzmán, I. and Morrow, S. (2007). Measuring efficiency and productivity in professional
football teams: evidence from the English Premier League. Central European Journal of
Operations Research 15, 309–328.
Haas, D. J. (2003). Productivity efficiency of English football teams: a data envelopment ana-
lysis approach. Managerial and Decision Economics 24, 403–410.
(2004). Technical efficiency in the Major League Soccer. Journal of Sports Economics 4,
203–215.
Haas, D. J., Kocher, M. G. and Sutter, M. (2004). Measuring efficiency of German football
teams by data envelopment analysis. Central European Journal of Operations Research 12,
251–268.
Hadley, L., Ciecka, J. and Krautmann, A. C. (2005). Competitive balance in the aftermath of
the 1994 players’ strike. Journal of Sports Economics 6, 379–389.
Hagn, F. and Maennig, W. (2008). Employment effects of the football World Cup 1974 in
Germany. Labour Economics 15, 1062–1075.
(2009). Large sport events and unemployment: the case of the 2006 soccer World Cup in
Germany. Applied Economics 41, 3295–3302.
Haigh, J. (2000). The Kelly criterion and bet comparisons in spread betting. The Statistician 49,
531–539.
Hamilton, B. H. (1997). Racial discrimination and professional basketball salaries in the 1990s.
Applied Economics 29, 287–296.
Hanssen, A. F. and Andersen, T. (1999). Has discrimination lessened over time? A test using
baseball’s all-star vote. Economic Inquiry 37, 326–352.
Hart, R. A., Hutton, J. and Sharot, T. (1975). A statistical analysis of association football
attendances. Journal of the Royal Statistical Society C 24, 17–27.
Hatanaka, M. (1974). An efficient estimator for the dynamic adjustment model with autocor-
related errors. Journal of Econometrics 2, 199–220.
Hautsch, N., Frick, B., Lehmann, E. and Warning, S. (2001). Could I stay or must I go?
Team-coach-separations in professional soccer in Germany. Faculty of Management and
Economics, Witten/Herdecke University. Mimeo.
Heckman, J. (1979). Sample selection bias as a specification error. Econometrica 47, 153–161.
Helsen, W. and Bultynck, J. B. (2004). Physical and perceptual cognitive demands of top-class
refereeing in association football. Journal of Sports Sciences 22, 179–189.
Henerey, R. J. (1985). On the average probability of losing bets on horses with given starting
price odds. Journal of the Royal Statistical Society Series A 148, 342–349.
References 435
Heuer, A., Müller, C. and Rubner, O. (2010). Soccer: is scoring goals a predictable Poissonian
process? Europhysics Letters 89, 38007.
Heyne, M., Maennig, W. and Suessmuth, B. (2007). Experience goods and biases in CVM – the
case of the soccer World Cup 2006. Hamburg Contemporary Economic Discussions No.
5, University of Hamburg.
Hill, G. (2009). The effect of frequent managerial turnover on organizational performance: a
study of professional baseball managers. The Social Science Journal 46, 557–570.
Hill, I. D. (1974). Association football and statistical inference. Applied Statistics 23, 203–208.
Hill, J. R., Madura, J. and Zuber, R. A. (1982). The short run demand for major league baseball.
Atlantic Economic Journal 10, 31–35.
Hoang, H. and Rascher, D. (1999). The NBA, exit discrimination and career earnings. Industrial
Relations 38, 69–91.
Hoehn, T. and Lancefield, D. (2003). Broadcasting and sport. Oxford Review of Economic Policy
19, 552–568.
Hoehn, T. and Szymanski, S. (1999). The Americanization of European football. Economic
Policy, 205–240.
Hofler, R. A. and Payne, J. E. (1996). How close to their offensive potential do National Football
League teams play? Applied Economics Letters 3, 743–747.
(1997). Measuring efficiency in the National Basketball Association. Economics Letters 55,
293–299.
Holgate, P. (1964). Estimation for the bivariate Poisson distribution. Biometrika 51, 241–245.
Horne, J. (1996). ‘Sakka’ in Japan. Media, Culture and Society 18, 527–547.
(2000). Soccer in Japan: is wa all you need? in: Finn, G. P. T. and Giulianotti, R. (eds.) Football
culture: local contests, global visions. London: Frank Cass.
Horrie, C. (2002). Premiership: lifting the lid on a national obsession. London: Pocket Books/
Simon and Schuster.
Hotchkiss, J., Moore, R. E. and Zobay, S. M. (2003). Impact of the 1996 Summer Olympic
Games on employment and wages in Georgia. Southern Economic Journal 69, 691–704.
Huebl, L. and Swieter, D. (2002). Der spielermarkt in der fußball-Bundesliga, zeitschrift für
betriebswirtschaft, ergänzungsheft 4. Sportökonomie 72, 105–125.
Hughes, M., Hughes, P., Mellahi, K. and Guermat, C. (2010). Short-term versus long-term
impact of managers: evidence from the football industry. British Journal of Management
21, 571–589.
Hugman, B. (1998). The PFA Premier and Football League players’ records, 1946–1998.
Harpenden: Queen Anne Press.
Humphreys, B. R. (2002). Alternative measures of competitive balance in sports leagues. Journal
of Sports Economics 3, 133–148.
(2006). The economic impact of sporting facilities, in: Andreff, W. and Szymanski, S. (eds.)
Handbook on the economics of sport. Cheltenham: Edward Elgar.
International Marketing Council of South Africa (2008). www.southafrica.info.
Jacklin, P. (2005). Temporal changes in home advantage in English football since the Second
World War: what explains improved away performance? Journal of Sports Sciences 23,
669–679.
Jacobs, D. and Singell, L. (1993). Leadership and organizational performance: isolating links
between managers and collective success. Social Science Research 22, 165–189.
Jasmand, S. and Maennig, W. (2008). Regional income and employment effects of the 1972
Munich Olympic Summer Games. Regional Studies 42, 991–1002.
Jenkins, J. A. (1996). A reexamination of salary discrimination in professional basketball. Social
Science Quarterly 77, 594–608.
Jeanrenaud, C. and Késenne, S. (2006). Sport and the media: an overview, in: Jeanrenaud, C.
and Késenne, S. (eds.) The economics of sport and the media. Cheltenham: Edward Elgar.
436 References
Jennett, N. (1984). Attendances, uncertainty of outcome and policy in Scottish League Football.
Scottish Journal of Political Economy 31, 176–198.
Jensen, M. C. and Meckling, W. H. (1976). Theory of the firm: managerial behaviour, agency
costs and ownership structure. Journal of Financial Economics 3, 305–360.
Jewell, R. T. and Molina, D. J. (2004). Productive efficiency and salary distribution: the case of
US Major League Baseball. Scottish Journal of Political Economy 51, 127–142.
Jewell, R. T. and Molina, D. J. (2005). An evaluation of the relationship between Hispanics and
Major League Soccer. Journal of Sports Economics 6, 160–177.
Jinxia, D. and Mangan, J. (2001). Football in the new China: political statement, entrepreneur-
ial enticement and patriotic passion. Soccer and Society 2, 79–100.
Jiobu, R. M. (1988). Racial inequality in a public arena: the case of professional baseball. Social
Forces 67, 524–534.
Johnson, A. T. and Sack, A. (1996). Assessing the value of sports facilities: the importance of
non-economic factors. Economic Development Quarterly 10, 369–381.
Johnson, B. K., Groothuis, P. A. and Whitehead, J. C. (2001). The value of public goods gener-
ated by a major league sports team: the CVM approach. Journal of Sports Economics 2,
6–21.
Johnson, B. K. and Whitehead, J. C. (2000). Value of public goods from sports stadiums: the
CVM approach. Contemporary Economic Policy 18, 48–58.
Johnston, R. (2008). On referee bias, crowd size and home advantage in the English soccer
Premiership. Journal of Sports Sciences 26, 563–568.
Jones, C. J., Nadeau, S. and Walsh, W. (1999). Ethnicity, productivity, and salary: player com-
pensation and discrimination in the National Hockey League. Applied Economics 31,
593–608.
Jordet, G., Hartman, E., and Sigmundstad, E. (2009). Temporal links to performing under
pressure in international soccer penalty shootouts. Psychology of Sport and Exercise 10,
621–627.
Jordet, G., Hartman, E., Visscher, C., and Lemmink, K. (2007). Kicks from the penalty mark
in soccer: the roles of stress, skill, and fatigue for kick outcomes. Journal of Sports Sciences
25, 121–129.
Kahn, L. M. (1991). Discrimination in professional sports: a survey of the literature. Industrial
and Labor Relations Review 44, 395–418.
(1993). Free agency, long term contracts and compensation in major league baseball. Review
of Economics and Statistics 75, 157–164.
(2000). The sports business as a labor market laboratory. Journal of Economic Perspectives
14, 75–94.
(2006). Race, performance, pay and retention among National Basketball Association head
coaches. Journal of Sports Economics 7, 119–149.
Kahn, L. M. and Sherer, P. O. (1988). Racial differences in professional basketball players’ com-
pensation. Journal of Labour Economics 6, 40–61.
Kanazawa, M. T. and Funk, J. P. (2001). Racial discrimination in professional basketball: evi-
dence from Nielsen Ratings. Economic Inquiry 39, 599–608.
Kang, Y. S. and Perdue, R. (1994). Long-term impacts of a mega-event on international tour-
ism to the host country: a conceptual model and the case of the 1988 Seoul Olympics,
in: Uysal, M. (ed.) Global tourism behaviour. New York.
Karlis, D. and Ntzoufras, I. (2003). Analysis of sports data using bivariate Poisson models. The
Statistician 52, 381–393.
Kasimati, E. (2003). Economic aspects and the Summer Olympics: a review of related research.
International Journal of Tourism Research 5, 433–444.
Kavetsos, G. and Szymanski, S. (2010). National wellbeing and international sports events.
Journal of Economic Psychology 31, 158–171.
References 437
Keifer, N. M. (1988). Economic duration data and hazard functions. Journal of Economic
Literature 26, 646–679.
Kelly, J. L. (1956). A new interpretation of information rate. Bell System Technological Journal
35, 917–926.
Kern, M. and Suessmuth, B. (2005). Managerial efficiency in German top league soccer: an
econometric analysis of club performances on and off the pitch. German Economic Review
6, 485–506.
Késenne, S. (1996). League management in professional team sports with win maximising clubs.
European Journal for Sport Management 2, 14–22.
(1999). Miscalculations and misinterpretations in economic impact analysis, in: Jeanreaud, C.
(ed.) The economic impact of sports events. Neuchatel.
(2000). The impact of salary caps in professional team sports. Scottish Journal of Political
Economy 47, 431–455.
(2003). The salary cap proposal of the G14 in European football. European Sports Management
Quarterly 3, 120–128.
(2007a). The peculiar economics of professional football in Europe. Scottish Journal of
Political Economy 54, 388–399.
(2007b). The economic theory of professional sports leagues. Cheltenham: Edward Elgar.
(2009). The impact of pooling and sharing broadcast rights in professional team sports.
International Journal of Sport Finance 4, 211–218.
Khoza, I. (2000). World Cup bid speech, South African Football Association. www.safa.org.za/
html/bid_news.html on 2 July 2002.
Kim, S. and Patrick, J. F. (2005). Residents’ perceptions on impacts of the FIFA 2002 World
Cup: the case of Seoul as a host city. Tourism Management 26, 25–38.
King, J. and Kelly, D. (1997). The cult of the manager: do they really make a difference?
London: Virgin Books.
Koch, J. V. and Vander Hill, C. W. (1988). Is there discrimination in the ‘black man’s game’?
Social Science Quarterly 69, 83–94.
Kocherlakota, S. and Kocherlakota, K. (1992). Bivariate discrete distributions. New York: Marcel
Dekker.
Koning, R. H. (2000). Balance in competition in Dutch soccer. The Statistician 49, 419–431.
(2003). An econometric evaluation of the firing of a coach on team performance. Applied
Koning, R. H. and van Velzen, B. (2010). Betting exchanges: the future of sports betting?
International Journal of Sports Finance 4, 42–62.
Koyama, M. and Reade, J. (2010). Playing like the home team: an economic investigation into
home advantage in football. International Journal of Sports Finance 4, 16–41.
Krautmann, A. C. (1999). What’s wrong with Scully-estimates of a player’s marginal revenue
product? Economic Inquiry 37, 369–381.
Kringstad, M. and Gerrard, B. (2007). Beyond competitive balance, in: Parent, M. M. and
Slack, T. (eds.) International perspectives on the management of sport. Burlington,
MA: Butterworth-Heinemann.
Kuethe, T.H. and Motamed, M. (2010). Returns to stardom: evidence from US Major League
Soccer. Journal of Sports Economics 11, 567–579.
Kurscheidt, M. (2006). The World Cup, in: Andreff, W. and Szymanski, S. (eds.) Handbook on
the economics of sport. Cheltenham: Edward Elgar.
Kuss, O., Kluttig, A. and Stoll, O. (2007). ‘The fouled player should not take the penalty him-
self’: an empirical investigation of an old German football myth. Journal of Sports Sciences
25, 963–967.
Laffey, D. (2005). Entrepreneurship and innovation in the UK betting industry: the rise of
person-to-person betting. European Management Journal 23, 351–359.
438 References
Laffont, J. J. and Mortimont, D. (2002). The theory of incentives: the principal-agent model.
Princeton University Press.
Lambert, C. (1997). The boss. London: Vista.
Lanfranchi, P. (1994). The migration of footballers: the case of France, 1932–1982, in: Bale, J.
and Maguire, J. (eds.) The global sports arena: athletic talent migration in an interdependent
world. London: Frank Cass.
Larsen, A., Fenn, A. J. and Spenner, E. (2006). The impact of free agency and the salary cap
on competitive balance in the National Football League. Journal of Sports Economics 7,
374–390.
Lavoie, M. (2000). The location of pay discrimination in the NHL. Journal of Sports Economics
1, 401–411.
Lazear, E. P. and Rosen, S. (1981). Rank-order tournaments as optimum labor contracts. Journal
Lee, A. (1999). Modelling rugby league data via bivariate negative binomial regression.
Australian and New Zealand Journal of Statistics 41, 153–171.
Lee, S. (1999). The BSkyB bid for Manchester United PLC, in: Hamil, S., Michie, J. and
Oughton, C. (eds.) The business of football: a game of two halves. Edinburgh: Mainstream.
Lee, Y. H. and Berri, D. (2008). A re-examination of production functions and efficiency estimates
for the National Basketball Association. Scottish Journal of Political Economy 55, 51–66.
Lee, Y. H. and Fort, R. (2005). Structural change in baseball’s competitive balance: the depres-
sion, team location, and integration. Economic Inquiry 43, 158–169.
(2009). Competitive balance: time series lessons from the English Premier League. Mimeo.
Lee, Y. H. and Schmidt, P. (1993). A production frontier model with temporal variation in
technical efficiency, in: Fried, H. O., Lovell, C. A. K. and Schmidt, S. (eds.) The measure-
ment of productive efficiency: techniques and applications. New York: Oxford University
Press.
Lehmann, E. (2000). Verdienen fußballspieler was sie verdienen? in: Schellhaaß, H. M. (ed.)
Sportveranstaltungen zwischen liga-und medieninteressen. Schorndorf: Karl Hofmann.
Lehmann, E. and Schulze, G. (2008). What does it take to be a star? The roles of performance
and the media for German soccer players. Applied Economics Quarterly 54, 59–70.
Lehmann, E. and Weigand, J. (1999). Determinanten der entlohnung von profifußballspielern –
eine empirische analyse für die deutsche Bundesliga. Betriebswirtschaftliche Forschung und
Praxis 51, 124–135.
Lenten, L. J. (2008). Unbalanced schedules and the estimation of competitive balance in the
Scottish Premier league. Scottish Journal of Political Economy 55, 488–508.
(2009a). Towards a new dynamic measure of competitive balance: a study applied to Australia’s
two major professional ‘football’ leagues. Economic Analysis and Policy 39, 407–428.
(2009b). Unobserved components in competitive balance and match attendances in the
Australian Football League, 1945–2005: where is all the action happening? Economic
Record 85, 181–196.
Lucey, B. M. and Power, D. (2004). Do soccer referees display home team favouritism? Working
Paper, School of Business Studies and Institute for International Integration Studies,
Trinity College Dublin.
Lucifora, C. and Simmons, R. (2003). Superstar effects in sport: evidence from Italian soccer.
Mabugu, R. and Mohamed, A. (2008). The economic impacts of government financing of
the 2010 Fifa World Cup. Stellenbosch Economic Working Papers: 08/08, University of
Stellenbosch.
MacDonald, D. N. and Reynolds, M. O. (1994). Are baseball players paid their marginal prod-
ucts? Managerial and Decision Economics 15, 443–457.
References 439
MacDonald, G. (1988). The economics of rising stars. American Economic Review 78,
155–166.
Madden, J. R. (2006). Economic and fiscal impacts of mega sporting events: a general equilib-
rium assessment. Public Finance and Management 6, 346–394.
Maennig, W. and du Plessis, S. (2007). World Cup 2010: South African economic perspec-
tives and policy challenges informed by the experience of Germany 2006. Contemporary
Maguire, J. (1994). Preliminary observations on globalisation and the migration of sport labour.
Sociological Review 42, 452–480.
Maguire, J (2004). Sport labour migration research revisited. Journal of Sport and Social Issues
28, 477–482.
Maguire, J. and Pearton, R. (2000). Global sport and the migration patterns of France ’98 World
Cup finals players: some preliminary observations, in: Garland, J., Malcolm, D. and Rowe,
M. (eds.) The future of football: challenges for the twenty-first century. London: Frank
Cass.
Maguire, J. and Stead, D. (1998). Border crossings: soccer labour migration and the European
Union. International Review for the Sociology of Sport 33, 59–73.
Maher, M. J. (1982). Modelling association football scores. Statistica Neerlandica 36, 109–118.
Malcolm, D., Jones, I. and Waddington, I. (2000). The people’s game? Football spectatorship
and demographic change, in: Garland, J., Malcolm, D. and Rowe, M. (eds.) The future of
football: challenges for the twenty-first century. London: Frank Cass.
Marburger, D. (1997). Gate revenue sharing and luxury taxes in professional sports. Contemporary
(2006). Chasing the elusive salary cap, in: Andreff, W. and Szymanski, S. (eds.) Handbook on
the economics of sport. Cheltenham: Edward Elgar.
Markowitz, H. (1952). Portfolio selection. Journal of Finance 7, 77–91.
Marques, A. (2002). Competitive balance in the Portuguese premier league of professional soc-
cer. Mimeo.
Matheson, V. A. (2006). Mega-events: the effect of the world’s biggest sporting events on local,
regional, and national economies. Paper No. 06–10, College of the Holy Cross, Department
of Economics, Faculty Research Series.
Mazur, M. (1994). Evaluating the relative efficiency of baseball players, in: Charnes, A., Cooper,
W., Lewin, A. and Seiford, L. (eds.) Data envelopment analysis: theory, methodology, and
applications. Boston: Kluwer.
McCormick, R. E. and Tollison, R. D. (2001). Why do black basketball players work more for
less money? Journal of Economic Behaviour and Organization 44, 201–219.
McGarry, T. and Franks, I. M. (2000). On winning the penalty shoot-out in soccer. Journal of
McGovern, P. (2002). Globalisation or internationalisation? Foreign footballers in the English
league, 1946–95. Sociology 36, 23–42.
Medoff, M. H. (1976). On monopolistic exploitation in professional baseball. Quarterly Review
of Economics and Business 16, 113–121.
Michie, J. and Oughton, C. (1999). Football and broadcasting and the MMC case, in: Hamil,
S., Michie, J. and Oughton, C. (eds.) The business of football: a game of two halves.
(2004). Competitive balance in football: trends and effects. Football Governance Research
Centre Research Paper No. 2, Birkbeck University.
Mignon, P. (2000). French football after the 1998 World Cup: the state and the modernity of
football, in: Finn, G. P. T. and Giulianotti, R. (eds.) Football culture: local contests, global
visions. London: Frank Cass.
440 References
Mixon, F. G. and Trevino, L. J. (2004). How race affects dismissals of college football coaches.
Journal of Labor Research 25, 645–656.
Mizak, D., Neral, J. and Stair, A. (2007). The adjusted churn: an index of competitive balance
for sports leagues based on changes in team standings over time. Economics Bulletin 26,
1–7.
Monopolies and Mergers Commission (1999). British Sky Broadcasting Group PLC and
Manchester United PLC: A report on the proposed merger. London: TSO, Cm 4305.
Mood, A. (1940). The distribution theory of runs. Annals of Mathematical Statistics 11,
367–392.
Moroney, M. J. (1956). Facts from figures, 3rd edn. London: Penguin.
Morrow, S. (1999). The new business of football: accountability and finance in football.
London: Macmillan.
Moschini, G. (2010). Incentives and outcomes in a strategic setting: the 3-points-for-a-win sys-
tem in soccer. Economic Inquiry 48, 65–79.
Mullin, C. J. and Dunn, L. F. (2002). Using baseball card prices to measure star quality and
monopsony. Economic Inquiry 40, 620–632.
Neale, W. (1964). The peculiar economics of professional sports. Quarterly Journal of Economics
78, 1–14.
Neave, N. and Wolfson, S. (2003). Testosterone, territoriality, and the ‘home advantage’.
Physiology and Behavior 78, 269–275.
Nelson, A. C. (1970). Information and consumer behaviour. Journal of Political Economy 78,
311–329.
Nevill, A., Balmer, N. J. and Williams, A. M. (2002). The influence of crowd noise and experi-
ence upon refereeing decisions in football. Psychology of Sport and Exercise 3, 261–272.
Nevill, A. and Holder, R. (1999). Home advantage in sport: an overview of studies on the
advantage of playing at home. Sports Medicine 28, 221–236.
Nevill, A., Newell, S. and Gale, S. (1996). Factors associated with home advantage in English
and Scottish soccer matches. Journal of Sports Sciences 14, 181–186.
New South Wales Treasury (1997). The economic impact of the Sydney Olympic Games. New
South Wales Treasury Research and Information Paper TRP97–10. Sydney: NSW Office
of Financial Management. www.treasury.nsw.gov.au/pubs/trp97_10/index.htm.
Nogawa, H. and Maeda, H. (1999). The Japanese dream: soccer culture towards the new
millenium, in: Armstrong, G. and Giulianotti, R. (eds.) Football cultures and identities.
Basingstoke: Macmillan.
Noll, R. G. (1974). Attendance and price setting, in: Noll, R.G. (ed.) Government and the sports
business. Washington, DC: Brookings Institution.
(1988). Professional basketball. Studies in Industrial Economics Paper No. 144. Stanford
University.
(2002). The economics of promotion and relegation in sports leagues: the case of English
football. Journal of Sports Economics 3, 169–203.
(2003). The organization of sports leagues. Oxford Review of Economic Policy 19, 530–551.
Noll, R. G. and Zimbalist, A. (1997). Sports, jobs and taxes. Washington, DC: Brookings
Institution.
Owen, J. G. (2006). The intangible benefits of sports teams. Public Finance and Management 6,
321–345.
Owen, P. D. (2010). Limitations of the relative standard deviation of win percentages for meas-
uring competitive balance in sports leagues. Economics Letters 109, 38–41.
Pace, A. and Carron, A. V. (1992). Travel and the National Hockey League. Canadian Journal
of Sport Sciences 17, 60–64.
Palomino, F., Renneboog, L. and Zhang, C. (2009). Information salience, investor sentiment,
and stock returns: the case of British soccer betting. Journal of Corporate Finance 15,
368–387.
References 441
Palomino, F., Rigotti, L. and Rustichini, A. (2000). Skill, strategy, and passion: an empirical
analysis of soccer. Discussion Paper, No. 129, Tilberg University.
Page, L. (2009). Is there an optimistic bias on betting markets? Economics Letters 102,
70–72.
Page, L. and Page, K. (2007). The second home leg advantage: evidence from European football
cup competitions. Journal of Sports Sciences 25, 1547–1556.
(2010). Alone against the crowd: individual differences in referees’ ability to cope under pres-
sure. Journal of Economic Psychology 31, 192–199.
Palacios-Huerta, I. (2003). Professionals play minimax. Review of Economic Studies 70,
395–415.
Pankoff, L. D. (1968). Market efficiency and football betting. Journal of Business 41, 203–214.
Parlasca, S. (2006). Collective selling of broadcast rights in team sports, in: Andreff, W. and
Szymanski, S. (eds.) Handbook on the economics of sport. Cheltenham: Edward Elgar.
Paul, R. J. and Weinbach, A. P. (2005). Bettor misperceptions in the NBA: the overbetting of
large favorites and the ‘hot hand’. Journal of Sports Economics 6, 390–400.
Pedace, R. (2008). Earnings, performance, and nationality discrimination in a highly competi-
tive labor market as an analysis of the English professional soccer league. Journal of Sports
Peel, D. A. and Thomas, D. (1988). Outcome uncertainty and the demand for football. Scottish
(1992). The demand for football: some evidence on outcome uncertainty. Empirical Economics
17, 323–331.
Pelnar, G. (2007). Antitrust analysis of sports leagues. MPRA Paper No. 5382.
PEP (1966). English professional football. Planning 32, No. 496.
Pettersson-Lindbom, P. and Priks, M. (2010). Behaviour under social pressure: empty Italian
stadiums and referee bias. Economics Letters 108, 212–214.
Pickup, I. (1999). French football from its origins to 1984, in: Dauncey, H. and Hare, G. (eds.)
France and the 1998 World Cup. London: Frank Cass.
Poli, R. (2006). Africans’ status in the European football players’ labour market. Soccer and
Society 7, 278–291.
(2009). Labour market migration to the five major leagues in European football: the impact
on national team selection, in: Walters, G. and Rossi, G. (eds.) Labour market migration
in European football: key issues and challenges. London: Birkbeck Sport Business Centre
Research Paper Series.
(2010). African migrants in Asian and European football: hopes and realities. Sport in Society
13, 1001–1011.
Pollard, R. (2002). Evidence of a reduced home advantage when a team moves to a new sta-
dium. Journal of Sports Sciences 20, 969–973.
(2006). Worldwide regional variations in home advantage in association football. Journal of
Pollard, R. and Gomez, M. (2009). Home advantage in football in South-West Europe: long-
term trends, regional variation, and team differences. European Journal of Sport Science 9,
341–352.
Pollard, R. and Pollard, G. (2005). Long-term trends in home advantage in professional
team sports in North America and England (1876–2003). Journal of Sports Sciences 23,
337–350.
Pope, P. F. and Peel, D. A. (1989). Information, prices and efficiency in a fixed odds betting mar-
ket. Economica 56, 323–341.
Porter, P. and Fletcher, D. (2008). The economic impact of the Olympic Games: ex ante predic-
tions and ex post reality. Journal of Sport Management 22, 470–486.
Porter, R. and Scully, G. W. (1982). Measuring managerial efficiency: the case of baseball.
Southern Economic Journal 48, 642–650.
442 References
Poulter, D. R. (2009). Home advantage and player nationality in international club football.
Journal of Sports Sciences 27, 795–805.
Prendergast, C. (1999). The provision of incentives in firms. Journal of Economic Literature 37,
7–63.
Prendergast, C. and Topel, R. (1996). Favoritism in organisations. Journal of Political Economy
104, 958–978.
Preston, I. and Szymanski, S. (2000). Racial discrimination in English football. Scottish Journal
Quirk, J. and El-Hodiri, M. (1974). The economic theory of a professional league, in: Noll, R.
G. (ed.) Government and the sports business. Washington, DC: Brookings Institution.
Quirk, J. and Fort, R. (1992). Pay dirt: the business of professional team sports. Princeton
University Press.
(1999). Hard ball: the abuse of power in pro team sports. Princeton University Press.
Rahmann, B., Weber, W., Groening, Y., Kurscheidt, M., Napp, H. G. and Pauli, M. (1998).
Socio-economic analysis of the football World Cup 2006 in Germany. Cologne: Sport und
Buch Strauss.
Raimondo, H. J. (1983). Free agents’ impact on the labor market for baseball players. Journal of
Labor Research 4, 183–193.
Rascher, D. A. (1997). A model of a professional sports league, in: Hendricks, W. (ed.) Advances
in the economics of sport, vol. 2. Greenwich: JAI Press.
Reep, C., Pollard, R. and Benjamin, B. (1971). Skill and chance in ball games. Journal of the
Royal Statistical Society Series A 131, 581–585.
Reilly, B. and Witt, R. (2007). The determinants of base pay and the role of race in Major
League Soccer: evidence from the 2007 league season. Discussion Papers in Economics DP
19/07, University of Surrey.
Ribeiro, H., Mendes, R., Malacarne, L., Picoli, S. and Santoro, P. (2010). Dynamics of tourna-
ments: the soccer case. The European Physical Journal B - Condensed Matter and Complex
Systems 75, 327–334.
Richards, D. G. and Guell, R. C. (1998). Baseball success and the structure of salaries. Applied
Economics Letters 5, 291–296.
Rickman, N. and Witt, R. (2008). Favouritism and financial incentives: a natural experiment.
Economica 75, 296–309.
Ronay, B. (2009). The manager: the absurd ascent of the most important man in football.
London: Sphere.
Rosen, S. (1981). The economics of superstars. American Economic Review 71, 845–858.
Rosen, S. and Sanderson, A. (2001). Labour markets in professional sports. Economic Journal
111, F47–F68.
Rottenberg, S. (1956). The baseball players’ labour market. Journal of Political Economy 64,
242–258.
Rue, H. and Salvesen, O. (2000). Prediction and retrospective analysis of soccer matches in a
league. The Statistician 49, 399–418.
Ruggerio, J., Hadley, L. and Gustafson, E. (1996). Technical efficiency in Major League
Baseball, in: Fizel, J., Gustafson, E. and Hadley, L. (eds.) Baseball economics: current
research. Westport: Praeger.
Russell, D. (1997). Football and the English: a social history of association football in England,
1863–1995. Preston: Carnegie.
(1999). Associating with football: social identity in England 1863–1998, in: Armstrong, G.
and Giulianotti, R. (eds.) Football cultures and identities. Basingstoke: Macmillan.
Salomo, S. and Teichmann, K. (2000). The relationship of performance and managerial suc-
cession in the German Premier Soccer League. European Journal for Sport Management 7,
99–119.
References 443
Sauer, R. D. (1998). The economics of wagering markets. Journal of Economic Literature 36,
2021–2064.
Saxonhouse, G. R. (1976). Estimated parameters as dependent variables. American Economic
Review 66, 178–183.
Schmidt, M. B. (2001). Competition in Major League Baseball: the impact of expansion. Applied
Economics Letters 8, 21–26.
Schmidt, M. B. and Berri, D. J. (2001). Competitive balance and attendance: the case of Major
League Baseball. Journal of Sports Economics 2, 145–167.
(2003). On the evolution of competitive balance: the impact of an increasing global search.
Economic Inquiry 41, 692–704.
Schofield, J. A. (1988). Production functions in the sports industry: an empirical analysis of
professional cricket. Applied Economics 20, 177–193.
Schollaert, P. T. and Smith, D. H. (1987). Team racial composition and sports attendance.
Sociological Quarterly 28, 71–87.
Scoppa, V. (2008). Are subjective evaluations biased by social factors or connections? An econo-
metric analysis of soccer referee decisions. Empirical Economics 35, 123–140.
Scott, F. A., Long, J. E. and Somppi, K. (1985). Salary vs marginal revenue product under mon-
opsony and competition: the case of professional basketball. Atlantic Economic Journal
13, 50–59.
Scully, G. W. (1974). Pay and performance in Major League Baseball. American Economic
Review 64, 915–930.
(1989). The business of major league baseball. University of Chicago Press.
(1992). Is managerial turnover rational? Evidence from professional team sports, in: G. W.
Scully (ed.) Advances in the economics of sport. Greenwich, CT: JAI Press.
(1994). Managerial efficiency and survivability in professional team sports. Managerial and
Decision Economics 15, 403–411.
(1995). The market structure of sports. University of Chicago Press.
Shin, H. S. (1991). Optimal betting odds against insider traders. Economic Journal 101,
1179–1185.
(1992). Prices of state contingent claims with insider traders, and the favourite-longshot bias.
(1993). Measuring the incidence of insider trading in a market for state-contingent claims.
Siegfried, J. and Eisenberg, J. (1980). The demand for minor league baseball. Atlantic Economic
Journal 8, 59–69.
Siegfried, J. and Zimbalist, A. (2000). The economics of sports facilities and their communities.
Journal of Economic Perspectives 14, 95–114.
Simar, L. and Wilson, P. W. (1999). Estimating and bootstrapping Malmquist indices. European
Journal of Operational Research 115, 459–471.
(2007). Estimation and inference in two-stage, semi-parametric models of production proc-
esses. Journal of Econometrics 136, 31–64.
Simmons, R. (1996). The demand for English league football: a club level analysis. Applied
(1997). Implications of the Bosman ruling for football transfer markets. Economic Affairs 17,
13–18.
Singell, L. D. (1993). Managers, specific human capital, and productivity in Major League
Baseball. Atlantic Economic Journal 21, 47–59.
Sir Norman Chester Centre for Football Research (2002). Racism and football, fact sheet 6. Sir
Norman Chester Centre for Football Research, University of Leicester.
Sloane, P. (1969). The labour market in professional football. British Journal of Industrial
Relations 7, 181–199.
444 References
(1971). The economics of professional football: the football club as a utility maximizer.
Scottish Journal of Political Economy 7, 121–146.
Smailes, G. (1992). The Breedon book of Football League records. Derby: Breedon.
Smith, M., Paton, D. and Vaughan Williams, L. (2006). Market efficiency in person-to-person
betting. Economica 73, 673–689.
Smith, R. (2003). The home advantage revisited: winning and crowd support in an era of
national publics. Journal of Sport and Social Issues 27, 346–371.
Solberg, H. and Turner, P. (2010). Exporting sports rights to overseas markets: the case of
European football. Sport in Society 13, 354–366.
Spann, M. And Skiera, B. (2009). Sports forecasting: a comparison of the forecast accuracy of
prediction markets, betting odds and tipsters. Journal of Forecasting 28, 55–72.
Staudohar, P. (1999). Salary caps in professional team sports, in: Jeanrenaud, C. and
Késenne, S. (eds.) Competition policy in professional sports: Europe after the Bosman case.
Antwerpen: Standaard Edition.
Sugden, J. and Tomlinson, A. (1998). FIFA and the contest for world football.
Cambridge: Polity.
Sutter, M. and Kocher, M. G. (2004). Favoritism of agents – the case of referees’ home bias.
Journal of Economic Psychology 25, 461–469.
Swinnen, J. and Vandemoortele, T. (2008). Sports and development: an economic perspective
on the impact of the 2010 World Cup in South Africa. LICOS Centre for Institutions and
Economic Performance, Katholieke Universiteit Leuven.
Szymanski, S. (2000). A market test for discrimination in the English professional soccer leagues.
(2001). Income inequality, competitive balance and the attractiveness of team sports.
Economic Journal 111, F69–F84.
(2003). The economic design of sporting contests: a review. Journal of Economic Literature
41, 1137–1187.
(2004). Professional team sports are only a game: Walrasian fixed-supply conjecture model,
contest Nash equilibrium and the invariance principle. Journal of Sports Economics 5,
111–126.
(2009). Playbooks and checkbooks: an introduction to the economics of modern sports. Princeton
University Press.
Szymanski, S. and Késenne, S. (2004). Competitive balance and gate revenue sharing in team
sports. Journal of Industrial Economics 52, 165–177.
Szymanski, S. and Kuypers, T. (1999). Winners and losers: the business strategy of football.
London: Viking.
Szymanski, S. and Smith, R. (1997). The English football industry: profit, performance and
industrial structure. International Review of Applied Economics 11, 135–153.
Szymanski, S. and Zimbalist, A. (2006). National pastime: how Americans play baseball and the
rest of the world plays soccer. Washington, DC: Brookings Institution.
Tabner, B. (1992). Through the turnstiles. Harefield: Yore.
Tainsky, S. and Winfree, J. (2010). Discrimination and demand: the effect of international play-
ers on attendance in Major League Baseball. Social Science Quarterly 91, 117–128.
Tapp, A. (2004). The loyalty of football fans: we’ll support you evermore? Journal of Database
Marketing and Customer Strategy Management 11, 203–215.
Tapp, A. and Clowes, J. (2002). From ‘carefree casuals’ to ‘professional wanderers’: seg-
mentation possibilities for football supporters. European Journal of Marketing 36,
1248–1269.
Taylor, A. J. P. (1965). English history: 1914–1945. Oxford University Press.
Taylor, Lord Justice (1990). The Hillsborough stadium disaster (15 April 1989), inquiry by the
Rt. Hon. Lord Justice Taylor, Final Report. Cmd. 962. London: HMSO.
References 445
Taylor, M. (2006). Global players? Football migration and globalization, 1930–2000. Historical
Social Research 31, 7–30.
Taylor, R. and Ward, A. (1995). Kicking and screaming: an oral history of football in England.
London: Robson.
Teigland, J. (1999). Mega-events and impacts on tourism: the predictions and realities of the
Lillehammer Olympics. Impact Assessment and Project Appraisal 17, 305–317.
ter Weel, B. (2006). Does manager turnover improve performance? New evidence using infor-
mation from Dutch soccer, 1986–2004. IZA Discussion Paper 2483, Institute for the Study
of Labor.
Terrell, D. and Farmer, A. (1996). Optimal betting and efficiency in parimutuel betting markets
with information costs. Economic Journal 106, 846–868.
Tervio, M. (2006). Transfer fee regulations and player development. Journal of the European
Economic Association 4, 957–987.
Tetlock, P. C. (2004). How efficient are information markets? Evidence from an online exchange.
Harvard University, Department of Economics.
Todd, J. R. (2003). Voting for the baseball hall of fame: the effect of race on election date.
Industrial Relations 12, 87–100.
Todd, J. R., Brown, R. W. and Miles, S. E. (2002). Measuring discrimination in major league
baseball: evidence from the baseball hall of fame. Applied Economics 34, 167–178.
Torgler, B., Schmidt, S. and Frey, B. (2006). Relative income position and performance: an
empirical panel analysis. FEEM Working Paper 39.
Trandel, G. A. and Maxcy, J. (2009). Adjusting winning-percentage standard deviations and a
measure of competitive balance for home advantage. Mimeo.
Turner, D. and White, A. (1993). The Breedon book of football managers. Derby: Breedon
Books.
Utt, J. and Fort, R. (2002). Pitfalls to measuring competitive balance with Gini coefficients.
Vamplew, W. (1988). Pay up and play the game: professional sport in Britain, 1875–1914.
Cambridge University Press.
Van de Burg, T. and Prinz, A. (2005). Progressive taxation as a measure for improving competi-
tive balance. Scottish Journal of Political Economy 52, 65–74.
Vasili, P. (2000). Colouring over the white line: the history of black footballers in Britain.
Vaughan Williams, L. (1999). Information efficiency in betting markets: a survey. Bulletin of
Economic Research 51, 1–30.
(2005). Information efficiency in financial and betting markets. Cambridge University Press.
Vaughan Williams, L. and Paton, D. (1997). Why is there a favourite longshot bias in British
racetrack betting markets? Economic Journal 107, 150–158.
Vlastakis, N., Dotsis, G. and Markellos, R. (2009). How efficient is the European football bet-
ting market? Evidence from arbitrage and trading strategies. Journal of Forecasting 28,
426–444.
Volz, B. (2009). Minority status and managerial survival in Major League Baseball. Journal of
Vrooman, J. (1995). A general theory of professional sports leagues. Southern Economic Journal
61, 971–990.
(2000). The economics of American sports leagues. Scottish Journal of Political Economy 47,
594–619.
(2007). Theory of the beautiful game: the unification of European football. Scottish Journal
(2009). Theory of the perfect game: competitive balance in monopoly sports leagues. Review
of Industrial Organization 34, 5–44.
446 References
Waddington, I. and Malcolm, D. (1998). The social composition of football crowds in Western
Europe. International Review for the Sociology of Sport 33, 155–169.
Wagner, S. (2010). Managerial succession and organizational performance – evidence from the
German soccer league. Managerial and Decision Economics 31, 415–430.
Walker, B. (1986). The demand for professional league football and the success of football
league teams: some city size effects. Urban Studies 23, 209–219.
Walker, M. and Wooders, J. (2001). Minimax play at Wimbledon. American Economic Review
91, 1521–1538.
Walvin, J. (1994). The people’s game: the history of football revisited. Edinburgh: Mainstream.
Waters, A. and Lovell, G. (2002). An examination of the homefield advantage in a professional
English soccer team from a psychological standpoint. Football Studies 5, 46–59.
Weitzman, M. (1965). Utility analysis and group behaviour: an empirical study. Journal of
Political Economy 73, 18–26.
White, H. (1980). A heteroscedasticity-consistent covariance matrix estimator and a direct test
for heteroscedasticity. Econometrica 48, 817–838.
White, P., Persad, S. and Gee, C. J. (2007). The effect of mid-season coach turnover on team
performance: the case of the National Hockey League (1989–2003). International Journal
of Sports Science and Coaching 2, 143–152.
Whiting, R. (1977). The chrysamthemum and the bat. New York: Avon.
Williams, J. (1997). The ‘new football’ in England and Sir John Hall’s ‘new geordie nation’,
in: Gehrmann, S. (ed.). Football and regional identity in Europe. Munster: Lit Verlag.
Wilson, D. P. and Ying, Y. H. (2003). Nationality preferences for labour in the international
football industry. Applied Economics 35, 1551–1560.
Wirl, F. and Sagmeister, S. (2008). Changing of the guards: new coaches in Austria’s premier
football league. Empirica 35, 267–278.
Witt, R. (2005). Do players react to anticipated sanction changes? Evidence from the English
Premier League. Scottish Journal of Political Economy 52, 623–640.
Woodland, L. M. and Woodland, B. M. (1994). Market efficiency and the favourite-longshot
bias: the baseball betting market. Journal of Finance 49, 269–279.
(2001). Market efficiency and profitable wagering in the National Hockey League: can bettors
score on longshots? Southern Economic Journal 67, 983–995.
(2003). The reverse favourite-longshot bias and market efficiency in major league baseball: an
update. Bulletin of Economic Research 55, 113–123.
Zak, T. A., Huang, C. J. and Siegfried, J. J. (1979). Production efficiency: the case of profes-
sional basketball. Journal of Business 52, 379–392.
Zech, C. E. (1981). An empirical estimation of a production function: the case of Major League
baseball. American Economist 25, 19–23.
Zimbalist, A. S. (1992). Salaries and performance: beyond the Scully model, in: Sommers,
P. M. (ed.) Diamonds are forever: the business of baseball. Washington, DC: Brookings
Institution.
(2002). Competitive balance in sports leagues: an introduction. Journal of Sports Economics
3, 111–121.
Index
Abramovich, Roman, 192 baseball, 2, 3, 4, 5, 12, 14, 16, 51, 78, 205, 212, 213,
AC Milan, 395 236, 239, 255, 256, 270, 272, 274, 275, 322,
Adams, Micky, 262 396
admission price, 163, 165, 167, 180, 181, 194, basketball, 2, 14, 15, 17, 18, 121, 213, 236, 256,
254, 322, 324, 328, 329, 337, 338, 339, 345, 272, 273, 287, 294, 322, 419
349, 350 Bates, Ted, 262
Adriano, 395 Bateson, Mike, 262
AFC Telford United, 194 Bayern Munchen, 387, 390
AFC Wimbledon, 194 Beckham, David, 198, 210
Africa, 230, 231, 232, 236, 409 behavioural theory of the firm, 249
Albania, 231 Belgium, 45, 47, 196, 220, 231, 232, 380, 418
Alcock, Paul, 315 Benitez, Rafa, 262, 264
Alternative Investment Market, 189 best-response functions, 31
American Basketball Association, 18 Betfair, 359
American Broadcasting Company (ABC), 18 betting markets
American football, 2, 14, 16, 18, 256, 275, 322 favourite-longshot bias, 352, 354, 356, 357, 358,
American Football League, 18 359, 360, 370, 371, 378
American League, 4, 5, 14, 18, 49, 212 fixed-odds betting, 79, 81, 100, 101, 352, 357,
American Professional Football Association, 18 358, 360, 365, 367, 371, 378
Ardiles, Osvaldo, 220, 398 informational efficiency, 101, 352, 353, 356, 358,
Argentina, 51, 198, 231, 288 365, 371, 376, 377, 378
Arnold, Sir Thomas, 140 online betting exchanges, 352, 359, 360
Arsenal, 47, 92, 105, 143, 146, 153, 157, 158, 162, racetrack betting, 352, 353, 354, 356, 360, 377
171, 172, 189, 190, 191, 198, 220, 251, 252, sentiment bias, 352, 363, 364, 378
254, 310, 326, 344 spread betting, football, 352, 360, 361, 362, 363,
Asia, 230, 236, 400, 409 378
Aston Villa, 101, 142, 155, 189, 190, 344 Birmingham City, 142, 187, 201, 371
Atkinson, Ron, 239 Bjornebye, Stig Inge, 220
attendances, 3, 9, 10, 52, 79, 88, 92, 104, 139, 150, Black, Andrew, 359
153, 154, 155, 156, 157, 165, 169, 171, 178, Blackburn Olympic, 140
180, 195, 239, 240, 254, 258, 319, 321, 322, Blackburn Rovers, 92, 101, 140, 146, 149,
323, 326, 327, 328, 329, 330, 331, 332, 333, 264, 266
334, 335, 336, 338, 344, 345, 348, 349, 350, Bolivia, 51
384, 398 Bolton Wanderers, 149, 155, 157
Australia, 395 bookmakers, see betting markets
Australian Football League, 48 Bordeaux, 384
Australian rules football, 333, 350 Borussia Dortmund, 387, 390, 403
Austria, 196, 231, 232 Borussia Moenchengladbach, 387
Azerbaijan, 395 Bosman, Jean-Marc, 182, 197
Bosman ruling, 159, 182, 183, 184, 195, 217, 220,
Barcelona, 198, 363 235, 384
Barnet, 196, 264 Boston United, 196
Barnsley, 313 Botafogo, 392
Barrow, 345 Bradford City, 157
447
448 Index
Brazil, 11, 231, 236, 259, 319, 380, 391, 394, 395, competitive equality, see competitive balance
402, 403 competitive inequality, see competitive balance
Brazilian Confederation, 392 computable general equilibrium (CGE) models,
Brentford United, 194 406, 415
Brighton and Hove Albion, 264 concentration ratio, 47, 48, 49, 77
British Broadcasting Corporation (BBC), 171, 172, Copa do Brasil, 392
173, 174, 175, 204, 254, 334 Copa Libertadores, 392
British Satellite Broadcasting (BSB), 171, 172, Coral, 367
173 Corinthians, 391, 395
British Sky Broadcasting (BSkyB), 171, 172, 173, count data, 304
174, 175, 177, 178, 179, 190, 334 Coventry City, 157, 344
broadcast revenue, 18, 139, 159, 161, 169, 175, 191, Crewe Alexandra, 260
268, 385, 390 cricket, 11, 296, 419
broadcast rights, 9, 18, 159, 173, 176, 177, 195, Croatia, 231
397, 401 Crystal Palace, 149, 345
Brooklyn Dodgers, 18 Cultural Revolution, 400
Buckley, Major Frank, 251, 252, 293, 294 Czech Republic, 232, 235
Bundesliga, see German Bundesliga
Burge, Keith, 313, 315 Dagenham and Redbridge, 196
Burnley, 146 Dalglish, Kenny, 143, 264
Busby, Matt, 252, 253, 254, 293 Dean, Mike, 315
Denmark, 45, 47, 220, 418
Cambridge United, 196 Derby County, 146, 157, 253
Cambridge University, 140 Desailly, Marcel, 239
Cameroon, 231 Di Matteo, Roberto, 266
Cantona, Eric, 183, 220 diminishing returns, 3, 7
Carlisle United, 52 DirecTV, 18
Carlos, Roberto, 395 discrimination
cartel, 6, 171, 177, 179 co-worker, 237, 238
Catterick, Harry, 270 customer, 237, 238, 239, 240, 247
CBS, 18 employer, 237, 239, 240
Celtic, 253, 266, 335, 363 hiring, 10, 216, 236, 238, 241, 246, 248
Central League, 396 price, 169, 196
Championnat de France Amateurs, 381 racial, 10, 212, 216, 236, 237, 238, 239, 247,
Champions League, 87, 142, 146, 162, 177, 191, 248
196, 257, 264, 302, 334, 387 retention, 236, 239, 241
Channel 4, 171 salary, 236, 237, 238, 241
Channel 5, 171 diseconomies of scale, 202, 203, 204
Chapman, Herbert, 143, 251, 293 Doncaster Rovers, 157
Charlton Athletic, 149, 161 Douglas, Lieutenant Commander Archibald,
Chelsea, 47, 101, 146, 158, 161, 189, 190, 192, 193, 395
198, 239, 264, 266, 326, 371 Dowd, Phil, 315
Cheltenham Town, 196 draft, 8, 13, 16
Chengdu Blades, 402 Dunga, 398
Chesterfield, 194 duration analysis, see hazard function
Chile, 394 Durkin, Paul, 315
China, 11, 380, 399, 400, 402, 403
China League, 401 Eastern Europe, 230, 231, 232, 236, 241
Chinese Football Association, 401, 402 Eastham, George, 181
Chinese Football League, 400 Ebbsfleet United, 194
Chinese Super League, 400, 401, 401 economies of scale, 201, 203, 237
churn index, 49 Ecuador, 51
Clattenburg, Mark, 315 eingetragener Verein, 387, 390
Clough, Brian, 253, 262, 293 Eintracht Frankfurt, 390
Club Français, 380 elasticity of demand
cointegration, 330, 331 income, 329
Colchester United, 196 price, 323, 329, 330, 331, 345, 349
collusion, 3 England, 1, 11, 14, 45, 47, 48, 49, 56, 59, 140, 142,
Colombia, 51 153, 155, 157, 158, 171, 177, 178, 181, 195,
competitive balance, 3, 8, 12, 15, 25, 27, 29, 36, 37, 198, 215, 216, 224, 231, 232, 235, 240, 248,
42, 43, 44, 59, 77, 143, 309, 332 253, 257, 275, 288, 293, 294, 296, 297, 334,
competitive balance ratio, 44, 333 336, 391, 398, 402, 403
Index 449
English football home team bias, 10, 52, 295, 296, 298, 299, 300,
competitive structure, 9, 139, 141 301, 305, 308, 309, 315, 319
financial structure, 159, 161, 180, 181 professionalism, 298, 302
governance, 9, 139, 189, 195 social pressure, 295, 300, 301, 302
ownership, 9, 139, 187, 188, 190, 192, 194, football teams
195 strategic behaviour, 107, 111, 122, 136, 137
ESPN, 18, 173, 174 Fox, 18
Europa League, see UEFA Cup France, 11, 14, 45, 47, 48, 49, 51, 56, 59, 196, 215,
European Championship, 176, 216, 232, 235, 248, 220, 231, 232, 235, 275, 319, 336, 380, 381,
254, 296, 363, 402, 418, 419 384, 394, 395, 398, 402, 403, 407, 418
European Commission, 173, 177, 178, 418 franchise, 4, 8, 13, 14, 15, 18, 19, 193, 254, 405,
European Court of Justice, 184 409, 412
European Cup, 142, 157, 253, 384, 387 Frank copula, 63, 305, 320
European Cup Winners’ Cup, 142 Frechet distribution, 82
Everton, 100, 135, 136, 140, 155, 172, 191, 196, free agency, 3, 4, 12, 16, 44, 184, 195, 200, 204, 208,
262, 328, 344 209, 212, 213, 214, 217, 238
exponential distribution, 122, 131 free agent, see free agency
externality Fry, Barry, 264
network, 359 Fulham, 100, 135, 181, 264
positive, 176, 177 Furukawa Electric, 398
FA Cup, 50, 79, 84, 87, 92, 104, 105, 140, 141, 143, G14 payroll cap, 33, 38, 39, 192
146, 155, 157, 170, 171, 186, 291, 294, 313 game theory, 9, 106, 107, 137
Fagan, Joe, 143 dominant strategy, 115
Fairs Cup, see UEFA Cup mixed-strategy equilibrium, 108, 109, 110
FC Kaiserslautern, 386, 387, 390 non-cooperative equilibrium, 115
FC Nuremberg, 386 penalty kick, 9, 106, 107, 109, 110, 137
FC United of Manchester, 194 prisoner’s dilemma, 115, 116, 137
Federal League, 18 Garrincha, 394
Federation Française de Football, 381 gate revenue, 9, 18, 139, 159, 162, 163, 165, 168,
Federation of International Football Associations 170, 184, 187, 268, 324, 351, 394
(FIFA), 182, 236, 296, 381, 391, 394, 396, 401, German Bundesliga, 111, 210, 211, 241, 257, 258,
406, 407, 408, 409, 419, 421 274, 275, 289, 299, 300, 301, 332, 336, 386
Ferguson, Alex, 143, 254, 262, 264, 294 German Football Association, 385, 387
fixed effects, 351 German Football League, 387
Flamengo, 391, 392, 395 Germany, 11, 15, 45, 47, 48, 49, 56, 59, 176, 183,
Fluminese, 391, 392, 395 184, 215, 231, 232, 235, 241, 275, 380, 385,
Football Association (FA), 140, 173, 180, 296, 297 402, 407, 417, 418, 419
Football League, 1, 9, 49, 46, 93, 119, 140, 141, Gillett, George, 193
157, 159, 160, 161, 163, 170, 171, 172, 173, Gillingham, 264
174, 175, 179, 184, 185, 187, 193, 194, 195, Gini coefficient, 45, 46, 47, 48, 77
198, 201, 216, 217, 226, 239, 241, 358, 363, Glazer, Malcolm, 192
402 goals scored
football managers, 10, 251, 266, 293, 297, 298, 313 forecasting model, 79, 80, 82, 86, 96, 98, 100,
managerial change, 250, 260, 270, 271, 285, 286, 104, 122, 135, 358
287, 288, 290, 291, 293, 294 statistical properties, 63, 77
managerial contribution, 10, 143, 255, 258 timings of, 107, 127, 137
managerial efficiency, 250, 255, 256, 257, 273, golden goal rule, 112, 138
275, 287 Gradi, Dario, 260
managerial succession effect, 251, 270, 286, 289, Granada Media, 190
293 Granger causality tests, 324
football match results Grant, Avram, 264
forecasting model, 79, 80, 82, 86, 95, 96, 98, 99, Grayson, Simon, 264
100, 104, 289, 290, 307, 353, 358, 359, 365, Greece, 47, 231, 302, 418
366 Greenwood, Ron, 262
persistence, 8, 42, 67, 71, 73, 74, 76, 77, 78, 98 Guangzhou GPC, 402
statistical properties, 42, 59 Gumbel distribution, 82
football referees
disciplinary sanction, 118, 124, 296, 299, 300, Hakuhodo, 398
301, 302, 303, 307, 308, 309, 310, 313, 315, Hamburg, 387, 390
317, 318, 319, 320 Harris, Rob, 313
favouritism, see home team bias Haynes, Johnny, 181
450 Index
hazard function, 122, 124, 137, 250, 274, 275, 276, Limpar, Anders, 220
277, 278, 281, 285 Lincoln City, 157
Hereford United, 196 linear probability model, 109, 273, 357
Herfindahl index, 47, 48, 77 Lineker, Gary, 398
Heysel, 142, 157, 195 Liverpool, 47, 140, 143, 146, 155, 157, 158, 161,
Hicks, Tom, 193 172, 177, 181, 190, 193, 220, 253, 262, 264,
Hiddink, Guus, 264, 266 294, 301, 313, 328, 344, 371
Hillsborough, 157, 195, 326, 350 London Stock Exchange, 188, 189
hockey, 2, 51, 322 London Weekend Television, 172
home-field advantage, 42, 43, 50, 51, 52, 53, 56, 59, Lorenz curve, 45, 46, 48, 77
73, 77, 78, 112, 119, 123, 131, 260, 288, 291, Louis Schmelling Paradox, 4
296, 300, 301, 309, 319, 369 Love, Vagner, 395
hooliganism, 154, 155, 156, 157, 189, 195, 326 Luton Town, 52
Hoyzer, Robert, 387 Luxembourg, 418
Huddersfield Town, 143, 146, 251 luxury tax, 16, 17, 33, 34, 35, 41
Hull City, 88, 91, 92, 93, 94, 98, 99, 100, 105, 157 Lyon, 383, 385
human capital, 257, 277, 281, 287, 294
Hungary, 399 Macclesfield Town, 196
Maidstone United, 196
imperfect substitution, 201, 202 Major League Baseball, see baseball
Ince, Paul, 266 Major League Soccer, 209, 210, 212, 258, 336
Independent Television (ITV), 171, 172, 173, 174, Malta, 231
175, 177, 190, 204, 334 managers, see football managers
input-output models, 406, 411, 412, 415 Manchester City, 153, 157, 190, 191, 192, 193, 198,
invariance proposition, 29, 37 344, 371
Ipswich Town, 146, 149, 191, 310 Manchester United, 47, 101, 105, 140, 142, 143,
Ireland 146, 153, 156, 158, 161, 170, 172, 177, 178,
Northern, 220, 221, 242, 364 181, 183, 188, 189, 190, 192, 193, 194, 198,
Republic of, 220, 221, 242, 418 201, 220, 252, 254, 264, 310, 326, 344
Israel, 220 Mansour bin Zayed Al Nahyan, Sheik, 192
Italy, 15, 45, 47, 48, 49, 82, 175, 212, 215, 231, 232, marginal revenue product, 12, 200, 204, 212, 213
235, 236, 259, 289, 299, 302, 319, 336, 386, Marseille, 383, 384, 385
394, 395, 407, 418 maximum likelihood, 81, 278, 367
ITV, see Independent Television maximum wage, 165, 179, 180, 181, 195, 197
ITV Digital, 159, 175, 186, 193, 268, 334 mean-reversion, 285, 287, 290, 294
Messi, Lionel, 198
J. League, 397, 398 Mexican League, 18
Japan, 11, 275, 380, 394, 395, 398, 402, 403, 406, Mexico, 319, 336, 394, 396
408, 413 Middlesbrough, 100, 149, 157, 190
Japan Football League, 396, 398 migration, 9, 10, 216, 217, 227, 228, 232, 235,
Japanese Football Association (JFA), 395 248
JEF United Ichihara, 398 Miller, Charles, 391
Jensen, John, 220 Millwall, 157, 188, 189
Juventus, 157 minimum admission price, 163
Minor League Baseball, see baseball
Kaka, 395 Mitsubuishi Motors, 398
Kanchelskis, Andrei, 220 MK Dons, 157, 266
Kashima Antlers, 398 Mobray, Tony, 266
Keegan, Kevin, 264 Molby, Jan, 220
Kendall’s tau statistic, 49 Monopolies and Mergers Commission (MMC),
Kidderminster Harriers, 196 178, 190
Kroenke, Stan, 190 monopoly, 3, 4, 5, 14, 15, 18, 19, 177, 179
natural, 5, 6, 202
LA Galaxy, 198 rent, 213
Ladbrokes, 367 monopsony, 2, 3, 5, 208, 209, 213, 238
Lampard, Frank, 198 moral hazard, 250
Latin America, 51, 230, 231, 232, 241, 336 Morecambe, 196
League Cup, 87, 141, 143, 170, 175 Morocco, 220
Leeds United, 146, 183, 190, 191, 253, 262, 264, Mourinho, Jose, 264
313 Moyes, David, 262
Leicester City, 157, 189, 190, 191, 264, 301 multi-plant firm, 5
Libarski, Pierre, 398 Murdoch, Rupert, 18, 171
Index 451
Nash equilibrium, 23, 31, 41, 115, 116, 117, 118, Paraguay, 51
207 Paris Saint Germain, 384
National Basketball Association (NBA), 14, 16, 17, Paulistano, 391
18, 41, 43, 51, 53, 237, 238, 239, 256, 272, 274, payroll cap, 8, 13, 14, 16, 17, 20, 33, 35, 36, 37, 40
350, 356 Payssandu Cricket Club, 391
National Collegiate Athletic Association (NCAA), Pearson, Nigel, 264
43 Pele, 394
National Football Institute, 383 Peru, 51
National Football League (NFL), 14, 16, 17, 18, Peterborough United, 196
43, 51, 53, 238, 239, 256, 257, 272, 275, 286, Peugeot, 381
354, 355, 356 Pirelli, 402
National Hockey League (NHL), 43, 51, 53, 112, player dismissals
238, 287, 356 timings of, 107, 124, 126, 137
National League, 5, 14, 18, 19, 212 player sale-and-leaseback, 191, 192
National Measure of Seasonal Imbalance, 45 player transfers, 160, 182, 194, 253, 394
National Rugby League, 48 player wages and salaries, 3, 27, 33, 35, 40, 139,
Nayim, 220 159, 160, 180, 181, 189, 192, 194, 197, 208,
NBA, see National Basketball Association 209, 210, 212, 213, 215, 240, 390, 398, 407
NBC, 18 determination of, 197, 200, 205, 209, 210, 212
NCAA, see National Collegiate Athletic Plus Market, 189
Association Poisson distribution, 42, 59, 63, 64, 77, 80, 304,
NFL, see National Football League 306, 357, 362
NHL, see National Hockey League bivariate, 78, 81, 82, 299, 306, 320
negative binomial distribution, 42, 59, 63, 64, 77, Port Vale, 262
80, 82, 304, 306, 357 Portsmouth, 100, 146, 149, 193
bivariate, 306, 320 Portugal, 47, 198, 231, 235, 395, 418
neoclassical theory, 249 Premier League
Netherlands, 45, 47, 232, 302, 319, 418 Austrian, 288
New York Giants, 18 Dutch, 288
New York Yankees, 4, 17 English, 1, 9, 32, 49, 46, 93, 104, 119, 141, 157,
Newcastle United, 100, 135, 136, 146, 149, 159, 160, 162, 170, 173, 174, 177, 179, 183,
153, 181, 189, 190, 191, 251, 264, 294, 328, 344 184, 186, 189, 190, 191, 193, 194, 195, 196,
Newton Heath, see Manchester United 198, 201, 204, 206, 216, 217, 220, 223, 226,
Nicholson, Bill, 262, 270 235, 241, 297, 299, 323, 328, 335, 358, 363, 402
Nigeria, 231 Scottish, 45, 47, 323, 328, 334, 363
Nimes, 183 Premiership, see Premier League: English
Nippon Shinpan, 398 Preston North End, 52, 140, 171, 189
Nolan, Kevin, 136 production frontier, 250, 255, 256
non-stationary, 328, 330, 349, 350 data envelopment analysis (DEA), 256, 258, 258,
normal distribution, 71, 73, 95, 100, 242, 366 259
Norway, 220, 302 random frontier model, 257
Norwich City, 149, 191, 310, 345 stochastic frontier analysis, 256, 257, 273
Nottingham Forest, 146, 157, 253, 262 Professional Footballers Association, 181
Notts County, 194 professional sports leagues, organisation
NTL, 190 European model, 14
North American model, 13, 14
Oceania, 230, 409 professional sports leagues, theory
Ofex Market, see Plus Market closed model, 13, 21, 22, 24, 25, 27, 28, 32, 34,
Office of Fair Trading, 172, 179 36, 37, 38, 39, 40
Oldham Athletic, 52, 262 open model, 13, 22, 26, 27, 29, 32, 34, 35, 36, 37,
Olympic Games, 78, 176, 396, 405, 409, 415, 418, 39, 40
420, 421, 422 profit maximisation, 6, 15, 21, 22, 23, 24, 25, 26,
ONDigital, 175 27, 28, 29, 31, 32, 33, 37, 40, 41, 200, 249, 254
Opta Index, 259 Pulis, Tony, 262
ordered probit, 45, 73, 82, 241, 358
overdispersion, 64, 67, 77, 304, 307 Queens Park Rangers, 52, 149, 189
Oxford United, 196
Oxford University, 140 Rangers, 335, 363
rank-order tournament model, 9, 198, 205, 206,
Pacific Coast League, 18 209, 213, 215
Pacific League, 396 RC Paris, 381
Paisley, Bob, 143 Reading, 155, 157
452 Index
Real Madrid, 198, 363, 383, 395 Sao Paulo, 391, 394, 395
Redknapp, Harry, 294 Scarborough, 196
Reed, Mike, 313, 315 Schalke 04, 390
referees, see football referees Schmeichel, Peter, 220
Regional Industrial Multiplier System, 411, 412 Scholar, Irving, 188
regression Scolari, Phil, 264, 266, 394
binary logit, 273, 274, 275, 357 Scotland, 11, 56, 59, 158, 171, 220, 225, 242, 248,
binary probit, 241, 273, 274, 275, 334, 364 253, 279, 302, 335, 363, 364
bivariate Poisson, 309 Scunthorpe United, 196
bivariate probit, 300 securitisation, 190, 191
cross-sectional, 84, 93, 237, 346 Serbia-Montenegro, 231
discrete-time logit, 274 Setanta, 173
fixed effects, 289, 338, 417 Sete, 381
instrumental variables, 328, 350 Shakhtar Donetsk, 105
mixed logit, 275 Shanghai Shenhua, 402
OLS, 210, 211, 272, 324, 328, 334, 350, 351, 357 Shanghai United, 402
ordered logit, 80, 418 Shankly, Bill, 143, 253, 254, 262, 270, 293
ordered probit, 79, 80, 81, 95, 104, 241, 242, 288, share prices, 189, 364
353, 365, 378 Sheffield United, 105, 161, 171, 402
Poisson, 79, 83, 104 Sheffield Wednesday, 157
quantile, 210, 211, 237, 332 Shrewsbury Town, 196
random effects, 211 Sichuan Guancheng, 402
random effects logit, 111, 236 Siemens, 402
simultaneous equations, 109 simulation
time-series, 333 Monte Carlo, 42, 73, 74, 76, 81
tobit, 324, 326 numerical, 119
Weibull, 274 stochastic, 107, 131, 137
Reith, John, 171 Sky, 171, 172, 173
relative competitive inequality measure, 43 Small Heath, see Birmingham City
Rennie, Uriah, 315 Sochaux, 381
reservation wage, 3, 4, 12, 200 Sony Creative Products, 398
reserve clause, 2, 3, 4, 7, 8, 12, 13, 15, 200, 212, 213 South Africa, 395, 406, 407, 408, 413, 414, 415
Restrictive Practices Court (RPC), 179 South America, 236, 240, 392, 396, 409
retain-and-transfer system, 179, 180, 181, 195 South Korea, 394, 406, 408, 413
returns to scale Southampton, 149, 157, 191, 262, 371
constant, 258 Southend United, 264
variable, 259 Southern League, 141
revenue functions, 12, 20, 24, 28, 29 Southport, 345
revenue maximisation, 169 Spain, 15, 45, 47, 48, 49, 56, 59, 176, 196, 215, 231,
revenue sharing, 4, 7, 8, 13, 14, 16, 18, 20, 27, 28, 232, 235, 259, 275, 288, 319, 328, 363, 386, 418
29, 30, 31, 32, 40, 170 Spearman rank correlation coefficient, 49
Revie, Don, 253, 262, 270, 293 sponsorship, 162, 170, 198, 214, 236, 254, 384, 397,
Riley, Mike, 315 398, 401, 402, 408
Rimet, Jules, 381 St Etienne, 384
Robinho, 395 Stade Reims, 383
Romania, 231, 302 stadium capacity, 153, 168, 202, 240, 258, 301, 321,
Ronaldo, Christiano, 198 324, 326, 337, 349, 350, 397, 412
Rosenior, Leroy, 262 Standard AC, 380
Rosenthal, Ronnie, 220 StanleyBet, 367
Rotherham United, 142 Stein, Jock, 253, 254
Royal Engineers, 140 Stockport County, 194
Royle, Joe, 262 Stoke City, 100, 157, 310
Rudge, John, 262 Stuttgart, 387
rugby league, 11, 176, 324, 350 Sumitomo Metals, 398
Rugby school, 140 Sunderland, 149, 153, 157, 171, 190, 328, 344
rugby union, 11, 176 Suntory, 398
runs test, 73, 110 superstars
Rushden and Diamonds, 196 economics of, 9, 201, 203, 204, 212, 213, 214
Russia, 232 Supporters Direct, 194
surprise index, 48
salary cap, see payroll cap survivor function, 273, 277, 279
Santos, 395 Swansea City, 157, 262
Index 453
Sweden, 47, 220, 302, 394 Wales, 364

Swindon Town, 149, 155 Walker, Jack, 146
Switzerland, 231, 232, 235, 275, 335, 419 Wanderers, The, 140
Watford, 161, 189, 262
Tapie, Bernard, 384 Weibull distribution, 82
Taylor, Graham, 262 weighted least squares (WLS), 347, 357
Taylor, Lord Justice, 157 Wenger, Arsène, 254, 262, 264, 398
Taylor, Peter, 264 Werder Bremen, 387
TBS, 18 West Bromwich Albion, 71, 100, 135, 155, 161,
television, 5, 139, 142, 154, 155, 157, 171, 172, 266, 344
179, 195, 204, 253, 297, 321, 333, 396, West Ham United, 262, 313, 344, 371
407, 418 Western Europe, 11, 231, 232, 236, 241
Tevez, Carlos, 198 Wigan Athletic, 100, 150, 157, 196
Thatcher, Margaret, 156 Willard, Gary, 313
Thorstvedt, Eric, 220 William Hill, 358, 359, 367
Tokyo Giants, 396 Wimbledon, 194, 196
Torquay United, 262 win percent, see win ratio
Tottenham Hotspur, 88, 91, 92, 93, 94, 98, 99, 100, win ratio, 19, 20, 41, 42, 43, 44, 45, 46, 47, 53, 59,
105, 172, 188, 189, 191, 220, 262, 344 77, 96, 250, 255, 262, 269, 272, 274, 286,
transactions costs, 249, 265 310
transfer expenditure, 139, 159, 184, 186, 193, 385 win-percent maximisation, 21, 22, 23, 24, 25, 26,
transfer fees, 8, 160, 182, 189, 401 27, 29, 30, 32, 33, 35, 37, 39, 40
transfer market, 10, 139, 143, 184, 185, 210, 217, Wolverhampton Wanderers, 143, 149, 251, 294, 344
228, 229, 252, 255, 293, 401, 408 Woolwich Arsenal, see Arsenal
transfers, see player transfers Workington, 345
two-part tariff, see discrimination: price World Cup, 11, 51, 153, 176, 232, 235, 236, 319,
335, 336, 380, 381, 384, 386, 394, 395, 396,
UEFA, see Union of European Football Associ- 402, 405, 406, 407, 409, 413, 415, 416, 417,
ations 418, 419, 420, 421
UEFA Cup, 87, 92, 105, 142, 146, 257, 302, 313 economic benefits, 11, 405, 406, 411, 417, 420,
UEFA Pro-Licence, 251, 294 421
Ukraine, 220, 231 intangible benefits, 407, 417, 418, 421, 422
uncertainty of outcome, 3, 8, 42, 45, 50, 332, 333, prospective economic impact studies, 405, 406,
335 410, 411, 412, 413, 421
Union of European Football Associations retrospective economic impact studies, 415, 416,
(UEFA), 182, 231, 251 421
United Arab Emirates, 395 World Series, 4, 5
Urawa Red Diamonds, 398 World Youth Championship, 396
Uruguay, 51, 381, 407 Wrexham, 345
utility function, 7, 27, 113, 200, 254, 298, 354 Wycombe Wanderers, 196, 264
utility maximisation, 7, 200
Yeovil Town, 196
Van Den Hauwe, Pat, 220 Yomiuro FC, 398
Vasco da Gama, 391, 392 York City, 194
Verdy Kawasaki, 398
vertical integration, 178, 190 zero conjectural variations, 23, 31
Vietnam, 395 Zico, 398
Villa, Ricardo, 220 Zidane, Zinedine, 394
Virgin Media, 179 Zimbabwe, 414

The Economics of Football, 2nd Edition

Uploaded by

Copyright:

Available Formats

The Economics of Football, 2nd Edition

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Economics of Football, 2nd Edition

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics does the book cover?

What topics does the book cover?

This page intentionally left blank

The Economics of Football

The second edition of this popular book presents a detailed

stephen d o b s o n is Professor of Economics at Hull

john g o d d a r d is Professor of Financial Economics at

cambridge university press

Published in the United States of America by Cambridge University Press,

© Stephen Dobson and John Goddard 2011

This publication is in copyright. Subject to statutory exception

First published 2011

Printed in the United Kingdom at the University Press, Cambridge

Library of Congress Cataloguing in Publication data

ISBN 978-0-521-51714-0 Hardback

Cambridge University Press has no responsibility for the persistence or

To: My daughters Imogen and Hannah

To: My family and friends

2 The economic theory of professional sports leagues 12

3 Competitive balance, uncertainty of outcome and home-field advantage 42

4 Forecasting models for football match results 79

4.4 A results-based forecasting model 95

5 Game theory and football games 106

6 English professional football: historical development and commercial

7 Determinants of professional footballers’ salaries 197

8 Professional footballers: employment patterns and racial discrimination 216

9 The football manager 249

9.3 Patterns of managerial change in English football 260

10 The football referee 295

11 Spectator demand for football 321

12 Gambling on football 352

14 The economics of the World Cup 405

We would like to thank colleagues and students at the Department of Economics,

2.1 Profit-maximisation and win-percent-maximisation

8.14 Cross-tabulation of squad players’ home countries and countries of

focus on the traditional major league sports of baseball, basketball, (American)

1.1 The economics of professional team sports: three seminal contributions

simply threaten to withdraw his services. If a player’s reservation wage is $10,000,

competition is not the same as economic competition. Similarly, although in law

maximisation subject to a financial solvency constraint. The financial solvency

1.2 Outline of this volume

Game theory is defined as the study of decision-making in situations of conflict

outcomes, forms a subfield within the literature on financial market efficiency.

indicates that in no professional team sport under current rules of operation is

teams are algebraically complex. In the sports economics literature, it is common

2.1 The North American professional team sports model

In theory, by equalising expenditure on players’ compensation throughout

operating together with the NL as a de facto joint monopoly supplier. In some

2.2 A model of an n-team league

Second, note that τi = ti / t .8 The partial derivative with respect to ti is

In most variants of the theoretical model, the equilibrium allocation of talent

The n-team closed model

The n-team open model

In an open model, playing talent is purchased from an international market at a

2.3 A model of a two-team league

special case where n = 2. The notion of a two-team league is obviously counter-

The two-team closed model

If the team owner’s objective is win-percent maximisation, the equilibrium salary

2.1 Profit-maximisation and win-percent-maximisation

∑ ∑ (w i,t − 0.5)2 / nT = ∑ ∑ (w i,t − w i )2 / nT + ∑ (w i − 0.5)2 / n [3.3]