A U E D R I: N Rban Arthquake Isaster ISK Ndex

Department of Civil and Environmental Engineering
Stanford University
AN URBAN EARTHQUAKE DISASTER RISK INDEX
by
Rachel A. Davidson
and
Haresh C. Shah
Report No. 121
June 1997
The John A. Blume Earthquake Engineering Center was established to promote
research and education in earthquake engineering. Through its activities our
understanding of earthquakes and their effects on mankind’s facilities and structures
is improving. The Center conducts research, provides instruction, publishes reports
and articles, conducts seminar and conferences, and provides financial support for
students. The Center is named for Dr. John A. Blume, a well-known consulting
engineer and Stanford alumnus.
Address:
The John A. Blume Earthquake Engineering Center

Department of Civil and Environmental Engineering
Stanford University
Stanford CA 94305-4020
(650) 723-4150
(650) 725-9755 (fax)
earthquake @ce. stanford.edu
http://blume.stanford.edu
©1997 The John A. Blume Earthquake Engineering Center

Abstract
The Earthquake Disaster Risk Index (EDRI) is a composite index that allows direct
comparison of the relative overall earthquake disaster risk of cities worldwide, and describes
the relative contributions of various factors to that overall risk. The development of the EDRI
brings together a body of knowledge about earthquake disasters from a wide range of
disciplines to provide three principal benefits.
First, the direct comparison of overall earthquake disaster risk provides a useful tool for
inter-city allocation of mitigation resources and effort. Most previous work in earthquake risk
assessment has focused on a single component of the risk, and/or on a single region. The EDRI
provides a systematic way to directly compare the overall earthquake disaster risk across a
large number of cities or regions. Second, the disaggregated EDRI will increase awareness of
the wide range of factors on which a city's earthquake disaster risk depends, from the expected
magnitude of ground shaking, to the number of structures, to a city’s current economic situation.
A comprehensive EDRI will highlight the fact that even in urban regions with low seismicity, an
earthquake may occur, and if it does, the other characteristics of the city could turn that single
event into a major disaster. Third, by reevaluating the index periodically, the EDRI may be used
to monitor trends in earthquake disaster risk over time.
The EDRI has been developed using the following six-step procedure: (1) create a
conceptual framework of all the factors that contribute to earthquake disaster risk—geological,
engineering, economic, social, political, and cultural factors; (2) identify simple, measurable
indicators to represent each of the factors in the framework (e.g., population, per capita Gross
Abstract iii
Domestic Product, percentage of the urbanized area that is soft soil); (3) combine the indicators
mathematically into the composite EDRI; (4) gather data and evaluate the EDRI for each of the
world’s major cities, (5) perform a sensitivity analysis to determine the robustness of the results,
(6) interpret the numerical findings to assess their reasonableness and implications, and present
the results in a variety of easily understandable graphical forms. A ten-city sample analysis was
conducted to explore the challenges associated with this process, and to illustrate its feasibility
and the usefulness of the results.
iv Abstract
Acknowledgments
This report is based on the doctoral dissertation of Rachel Davidson. The research
work was partially supported by a National Science Foundation Graduate Fellowship and by
the Stanford University Department of Civil Engineering. The author would like to acknowledge
some of the many individuals who helped to bring this project to fruition. Professor Anne S.
Kiremidjian, Professor George Mader, and Dr. Brian Tucker offered advice and support
consistently throughout the development of this work. Don Windeler, Mark Romer, Jawhar
Bouabid, David Carttar, and others at Risk Management Solutions, Inc. furnished data for many
of the indicators and cities in the sample analysis. Conversations with Dr. Aladdin Nasser,
Professor Barclay Jones, and Dr. Richard Eisner provided insight into issues associated with
Vulnerability, Exposure, and Emergency Response and Recovery, respectively. Finally,
earthquake engineering professionals from around the world supplied their expert opinion by
completing a questionnaire developed for this project. The author appreciates the valuable
assistance that these individuals and so many others provided.
Acknowledgments v
Table of Contents
Abstract ................................................................................................................................iii
Acknowledgments ................................................................................................................. v
Table of Contents .................................................................................................................vii
List of Tables........................................................................................................................ xi
List of Figures......................................................................................................................xiii
Chapter 1. Introduction.........................................................................................................1
1.1. Anticipated benefits of the EDRI .............................................................................2

1.2. Six-step development procedure .............................................................................5
1.3. Organization of the dissertation................................................................................8
Chapter 2. Key Features of the Earthquake Disaster Risk Index.............................................9
2.1. Definition of earthquake disaster risk .......................................................................9

2.2. Objectives of the EDRI.........................................................................................15
2.3. Holistic, multidisciplinary approach........................................................................16
2.4. Urban risk ............................................................................................................21
2.5. Composite index form...........................................................................................24
Chapter 3. Factor Identification and Conceptual Framework................................................27
3.1. Approach.............................................................................................................27
3.2. Challenges............................................................................................................28
3.3. Sample analysis solution........................................................................................30
3.3.1. Hazard factor............................................................................................31
3.3.1.1. Ground Shaking.........................................................................31
Table of Contents vii

3.3.1.2. Liquefaction...............................................................................32
3.3.1.3. Fire Following Earthquake .........................................................34
3.3.1.4. Landslide...................................................................................34
3.3.1.5. Tsunami and Seiche ...................................................................35
3.3.2. Exposure factor ........................................................................................36
3.3.2.1. Physical Infrastructure Exposure.................................................36
3.3.2.2. Population Exposure..................................................................38
3.3.2.3. Economy Exposure....................................................................39
3.3.2.4. Social-political System Exposure................................................39
3.3.3. Vulnerability factor....................................................................................40
3.3.3.1. Physical Infrastructure Vulnerability............................................40
3.3.3.2. Population Vulnerability..............................................................45
3.3.3.3. Economy Vulnerability...............................................................47
3.3.3.4. Social-political System Vulnerability...........................................52
3.3.4. External Context factor .............................................................................53
3.3.4.1. Economic External Context ........................................................54
3.3.4.2. Transportation External Context .................................................54
3.3.4.3. Political External Context ...........................................................55
3.3.4.4. Cultural External Context ...........................................................56
3.3.5. Emergency Response and Recovery Capability factor................................57
3.3.5.1. Planning.....................................................................................60
3.3.5.2. Resources Available Post-earthquake.........................................60
3.3.5.3. Mobility and Access Post-earthquake ........................................61
Chapter 4. Indicator Selection.............................................................................................63
4.1. Criteria for indicator selection................................................................................64

4.2. Strategy................................................................................................................66
4.3. Sample analysis solution........................................................................................67
4.3.1. Hazard factor............................................................................................69
4.3.1.1. Ground Shaking.........................................................................69
4.3.1.2. Liquefaction...............................................................................73
4.3.1.3. Fire Following Earthquake .........................................................75
4.3.1.4. Tsunami and Seiche ...................................................................76
4.3.2. Exposure factor ........................................................................................77
4.3.2.1. Physical Infrastructure Exposure.................................................78
4.3.2.2. Population Exposure..................................................................87
4.3.2.3. Economy Exposure....................................................................88
4.3.3. Vulnerability factor....................................................................................88
4.3.3.1. Physical Infrastructure Vulnerability............................................89
4.3.3.2. Population Vulnerability..............................................................96
4.3.4. External Context factor .............................................................................97
4.3.4.1. Economic External Context ........................................................97
viii Table of Contents

4.3.4.2. Political External Context ...........................................................98
4.3.5. Emergency Response and Recovery Capability factor................................99

4.3.5.1. Planning...................................................................................100
4.3.5.2. Resources Available Post-earthquake.......................................101
4.3.5.3. Mobility and Access Post-earthquake ......................................103
Chapter 5. Mathematical Combination...............................................................................107
5.1. General approach ...............................................................................................110

5.1.1. No combination......................................................................................110
5.1.2. Graphical methods..................................................................................110
5.1.3. Nonlinear combination............................................................................112
5.1.4. Linear combination..................................................................................113
5.1.5. Sample analysis solution..........................................................................117
5.2. Scaling................................................................................................................118
5.2.1. Contribution function...............................................................................118
5.2.2. Minimum and maximum observed............................................................122
5.2.3. Minimum and maximum possible .............................................................123
5.2.4. Base city and year...................................................................................124
5.2.5. Consistent zero to ten scale .....................................................................125
5.2.6. Mean......................................................................................................126
5.2.7. Mean minus two standard deviations .......................................................127
5.2.8. Inversely related indicators......................................................................128
5.3. Weighting ...........................................................................................................128
5.3.1. Regression..............................................................................................129
5.3.2. Principal components analysis..................................................................130
5.3.3. Subjective assessments ...........................................................................132
5.3.4. Sample analysis solution..........................................................................134
5.3.4.1. Physical Infrastructure Exposure weights...................................134
5.3.4.2. Questionnaire...........................................................................135
5.3.4.3. Sample analysis solution...........................................................139
5.4. Mathematical combination summary....................................................................142
Chapter 6. Data Gathering and Evaluation..........................................................................145
6.1. Data sources.......................................................................................................145

6.2. Evaluation procedure ..........................................................................................150
6.3. Sample analysis solution......................................................................................150
Chapter 7. Sensitivity Analysis...........................................................................................159
Table of Contents ix
7.1. Indicator definitions.............................................................................................160
7.2. Indicator selection...............................................................................................163
7.3. Scaling................................................................................................................169
7.4. Weights..............................................................................................................176
7.5. Data uncertainty..................................................................................................180
7.6. Sample city selection...........................................................................................184
7.7. City boundary definitions.....................................................................................188
7.8. Conclusions ........................................................................................................188
Chapter 8. Presentation and Interpretation.........................................................................193
8.1. Goals of the presentation.....................................................................................193

8.2. Sample analysis solution......................................................................................194
8.2.1. Overall EDRI and five main factor values.................................................196
8.2.2. Relative contributions of different factors .................................................205
8.2.3. Earthquake disaster risk map...................................................................207
8.2.4. City summaries .......................................................................................207
8.2.5. Group comparisons.................................................................................211
8.2.6. Hazard versus earthquake disaster risk ....................................................211
8.3. Reminders ..........................................................................................................213
Chapter 9. Future Work and Conclusions..........................................................................215
9.1 Future work .........................................................................................................215

9.2. Conclusions..........................................................................................................218
Appendices .......................................................................................................................221
A. Selected Previously Proposed Definitions of Disaster ..........................................221

B. Sample City Definitions .......................................................................................223
C. Emergency Response and Recovery Tasks..........................................................227
D. Physical Infrastructure Exposure Data Sets..........................................................237
E. Evaluation of “seismic code indicator” for Sample Analysis Cities.........................239
F. External Context Indicator Values for 319-City Data Set.....................................241
G. Principal Components Analysis of Physical Infrastructure Exposure Data Sets
..........................................................................................................................243
H. Questionnaire......................................................................................................245
I. Code for sens.c Sensitivity Analysis Computer Program......................................249
J. Minimum and Maximum Possible Values for Scaling Sensitivity Analysis...............251
K. Values of ∆xij for Data Uncertainty Sensitivity Analysis ........................................253
References.........................................................................................................................255
x Table of Contents
List of Tables
Table Page
2.1. Comparison of three earthquake risk assessment methods.........................................10
3.1. Physical infrastructure components ...........................................................................37
4.1. Soil classifications ....................................................................................................72
4.2. Liquefaction susceptibility of sedimentary deposits
(from Youd and Perkins 1978).................................................................................74
4.3. Conditional probability relationship for liquefaction susceptibility categories
(from Liao et al. 1988).............................................................................................75
4.4. "Tsunami potential indicator" definition .....................................................................77
4.5. Indicators in two physical infrastructure exposure data sets .......................................80
4.6. Principal components analysis results for Data Set One .............................................82
4.7. Principal components analysis results for Data Set Two.............................................82
4.8. Seismic code sophistication rating scale ....................................................................91
4.9. Seismic code enforcement rating scale ......................................................................91
4.10. Comparison of two methods for Physical Infrastructure Vulnerability factor assessment95
4.11. "Planning indicator" definition..................................................................................100
5.1. Schematic indicator-city table.................................................................................108
5.2. Comparison of scaling techniques...........................................................................119
5.3. Results of principal components analysis to determine Physical Infrastructure Exposure
weights (Statistics of World Large Cities data set)................................................134
List of Tables xi
Table Page
5.4. Results of principal components analysis to determine Physical Infrastructure Exposure
weights (Human Settlements Statistics data set) ...................................................135
5.5. Five main factor weights obtained from questionnaire survey...................................138
5.6. Summary of sample analysis weights.......................................................................140
6.1. Sample analysis data sources..................................................................................152
6.2. Unscaled indicator-city data...................................................................................154
6.3. Scaled indicator-city data.......................................................................................155
6.4. Factor component values .......................................................................................157
6.5. Relative contributions of each factor to the overall EDRI values...............................158
7.1. Summary of categorical groupings from sensitivity analysis.......................................168
7.2. Comparison of scaling techniques tested in sensitivity analysis..................................170
7.3. Comparison of average Spearman rank correlation coefficient (rs) values.................190
8.1. Sample analysis results for Figures 8.1 and 8.2 .......................................................203
xii List of Tables

List of Figures
Figure Page
1.1. Six-step EDRI development procedure ......................................................................6
2.1. Impact parameters ...................................................................................................13
2.2. Basic steps in regional seismic hazard and risk analysis..............................................19
3.1. Conceptual framework of earthquake disaster risk....................................................31
3.2. Schematic example of a damage curve......................................................................41
3.3. Schematic example of a set of fragility curves............................................................41
4.1. Sample analysis indicators........................................................................................68
4.2. Schematic example of a hazard curve .......................................................................71
4.3. Distribution of GDP in China, France, India, Indonesia, and USA.............................85
4.4. Example of a loss estimation model damage curve.....................................................94
4.5. Comparison of Physical Infrastructure Vulnerability factor assessment using EDRI and
loss estimation models..............................................................................................94
4.6. Graphical representation of “extreme weather indicator”..........................................105
5.1. “Distance from ideal” graphical method of combination...........................................111
5.2. “Similar groups” graphical method of combination...................................................112
5.3. Graphical comparison of scaling, weighting, and contribution functions.....................116
5.4. Graphical comparison of scaling techniques.............................................................120
5.5. Hypothetical contribution function for the indicator “population” ..............................121
5.6. Graphical comparison of regression and principal components analysis ....................130
5.7. Summary of questionnaire survey results.................................................................138
List of Figures xiii

5.8. Factor component weights .....................................................................................141
Figure Page
6.1. Sources cited in Table 6.1......................................................................................153
7.1. Results of sensitivity analysis for indicator definitions................................................161
7.2. Results of sensitivity analysis for indicator selection..................................................165
7.3. Results of sensitivity analysis for scaling functions ....................................................171
7.4. Illustration of implicit weighting effect of base city scaling technique..........................175
7.5. Results of sensitivity analysis for weights .................................................................177
7.6. Results of sensitivity analysis for data uncertainty.....................................................181
7.7. Results of sensitivity analysis for city sample............................................................185
8.1. Sample analysis results: Relative values of EDRI and five main factors .....................197
8.2. Sample analysis results: Relative values of components of five main factors ..............198
8.3. Sample analysis results: Relative contributions of five main factors to EDRI for each of
ten cities in the sample analysis ...............................................................................206
8.4. Earthquake disaster risk map..................................................................................208
8.5. Example of a city summary for Tokyo, Japan..........................................................209
8.6. Comparison of EDRI for cities with high and medium human development...............211
8.7. Comparison of EDRI and Hazard factor.................................................................212
8.8. Equal Hazard, unequal EDRI..................................................................................213
xiv List of Figures

Chapter One
Introduction
1. Introduction
Imagine a landscape without people, buildings, or any human influence. Suppose an

earthquake occurs in that setting. The geology of the area will determine the characteristics of
the ground shaking. Now imagine a city, with all its infrastructure and residents, is
superimposed on the same landscape, and the same earthquake occurs. The characteristics of
the city will determine the sequence of events that follow the ground shaking. Finally, if that city
is removed and replaced with another, the same earthquake will precipitate a different sequence
of events because the second city exhibits a different set of characteristics.
This study aims to identify and understand the factors (i.e., city characteristics) that
together determine the sequence of events surrounding an urban earthquake, and to represent
those factors and the overall risk in an easily accessible way, so that the earthquake disaster risk
of two cities, like those described above, can be compared directly. These objectives are
pursued through the development of a comprehensive urban Earthquake Disaster Risk Index
(EDRI). The EDRI is a composite index that allows direct comparison of the relative overall
earthquake disaster risk of cities worldwide, and describes the relative contributions of various
factors to that overall risk. The index demonstrates that the risk in Jakarta is about the same as
the risk in San Francisco, but significantly less than the risk in Tokyo. Examining the
components of the EDRI indicates that while the risk in Jakarta is mostly the result of the high
Chapter One: Introduction 1

vulnerability and insufficient emergency response and recovery capability, the risk in San
Francisco is due primarily to the high frequency of earthquakes, and the risk in Tokyo is driven
by the incredible number of people and structures exposed.
The EDRI represents a novel approach to earthquake risk assessment. It is more

comprehensive than traditional earthquake loss estimation models in that it aims to measure not
the risk of a certain amount of physical damage or economic loss, but the risk of an earthquake
disaster. Assessing the risk of a disaster requires that the implications of the expected impact
be interpreted in the economic, political, and social context in which that impact will occur. The
EDRI directly assesses risk for a city as whole, not for a specific site. By using the composite
index as a means to synthesize the many determinants of risk, the EDRI enables comparative
analyses that are more succinct and easily understandable than earthquake damage scenarios
can offer. Finally, the EDRI data requirements are far less restrictive than those for either loss
estimation models or earthquake damage scenarios. (See Table 2.1 for a summary comparison
of these three methods).
1.1. Anticipated benefits of the EDRI
The development of the Earthquake Disaster Risk Index brings together a body of
knowledge about earthquake disasters from a wide range of disciplines to provide three
principal benefits. First, the EDRI allows direct comparison of the relative overall earthquake
disaster risk of different cities throughout the world. Many factors (e.g., frequency of
earthquakes, structural vulnerability, quality of emergency response and recovery planning)
contribute to a city's overall risk. A city may have a relatively high risk with respect to certain
factors, and low with respect to others. By creating a composite index that combines all the
pertinent factors, the overall disaster risk of different cities can be compared directly. Such a
comparison could be useful for governments and international aid organizations as they allocate
resources among various cities. Multinational companies will be able to refer to the index in
locating manufacturing facilities. Insurance and reinsurance companies could employ it as they
2 Chapter One: Introduction

plan their portfolio diversification and establish premiums for earthquake insurance policies.
Any interested person will be able to use the EDRI to improve his or her understanding of the
risk in different cities worldwide by comparing those cities to ones for which he has an intuitive
feel. While risk assessment studies have been performed for many regions around the world,
for the most part, each project focuses on a single component of the risk, and/or on a single
region. Relatively little work has been done in systematically comparing the overall risk across a
large number of cities or regions. The EDRI aims to provide this service.
There have been many appeals to create a method for interregional comparison of the
risk of earthquake disasters (or more generally, the risk due to all natural disasters). Throughout
Urban Scale Vulnerability: Proceedings of the U.S.-Italy Colloquium on Urban Design
and Earthquake Hazard Mitigation (Heikkala 1982), there are repeated calls for the
development of a holistic, multidisciplinary method to compare the vulnerability of different
urban regions to earthquake disasters. The Editor’s Overview of Environmental
Management and Urban Vulnerability (Kreimer and Munasinghe 1992) lists among the
important topics for future research the development of an “index of vulnerability” to measure a
city’s vulnerability to natural catastrophes. In fact, the need to create a method of interregional
comparison is not just an academic notion. The United States Federal Emergency Management
Agency currently is developing a method to compare the multi-hazard natural disaster risk of
each state in the country. The agency hopes to use the resulting technique to help determine the
most effective and efficient way to allocate its mitigation resources (Nishenko 1997).
Second, closer examination of the disaggregated EDRI conveys information about the
various factors that comprise the overall risk. One can focus on a single factor, and explore
how the values of that factor vary among cities. For example, a structural engineer interested
primarily in the vulnerability of the infrastructure could see that the Physical Infrastructure
Vulnerability in Boston is about 1½ times that in Tokyo and two-thirds that in Manila.
Alternatively, one could focus on one city at a time, and determine the relative contributions of
different factors to the overall risk within each. The analysis might show, for example, that in

Lima, Hazard contributes twice as much to the overall risk as Exposure does; while in Mexico
City, the reverse is true, i.e., Hazard is about half as important as Exposure. Understanding the
causal factors of earthquake disaster risk constitutes the first step in designing the most effective
and efficient strategies to mitigate that risk. In a given city, those factors that are contributing
most substantially to the risk, and that are most easily “controlled” through structural or non-
structural means will emerge as the most promising targets for mitigation efforts in that city.
Since every city is unique—in its natural setting, its infrastructure, the activities it supports, and
its relationships with other regions, the relative contributions of these causal factors will not be
the same in every city, and therefore, the optimal mitigation strategy will not be the same across
cities either. Looking at the components of the EDRI will help to illustrate, at the broadest level,
which factors are the most important in each city.
The disaggregated EDRI also should increase awareness of the wide range of factors on
which a city's earthquake disaster risk depends, from the expected magnitude of ground
shaking, to the number of structures, to the current economic situation in the city. The frequency
of historical earthquakes is not the only determinant of a city’s earthquake disaster risk, and a
comprehensive EDRI will highlight that important point. This contribution will be particularly
valuable in urban regions with infrequent earthquakes and negligible awareness. The EDRI will
convey the critical message that although they are rare in those cities, an earthquake may occur,
and if it does, the other characteristics of the city could turn that single event into a major
disaster. Further, the index can serve as a vehicle to help change the way in which earthquake
disasters are commonly conceived—from events due only to uncontrollable natural forces, to
events whose severity are determined in large part by the economic, social, and political
characteristics of the city on which those natural forces are unleashed. While people may not be
able to stop the occurrence of earthquakes, they will be able to affect some of the other
contributing factors (e.g., resources available for emergency response and recovery,
vulnerability of the infrastructure), thereby reducing the frequency and severity of earthquake
disasters.

Third, by reevaluating the index periodically, trends in earthquake disaster risk over time
could be monitored with the EDRI. The predominant feeling among earthquake professionals
today suggests that worldwide earthquake disaster risk has grown significantly over the past
several decades due to exploding populations in seismic regions, unimpeded urbanization, and
the increasingly interconnected and internally complex nature of cities. Currently, however, this
is little more than a feeling. There exists no systematic way to track the earthquake disaster risk
over time. The EDRI can fill this gap and demonstrate, not only if the risk is growing as
believed, but also by how much each year, and why. It may indicate that the risk is increasing
over time because the population is growing, because the infrastructure is becoming more
vulnerable as it ages, or due to some other reason.
1.2. Six-step development procedure
Outlining the basic six-step procedure used to create the Earthquake Disaster Risk
Index should help both to describe the index further and to provide an outline for the remainder
of the dissertation. The following is a brief introduction to the approach (Fig. 1.1). Each of the
steps will be elaborated on at length in the chapters that follow.
1. Factor identification and conceptual framework development

A systematic investigation identifies all the factors—geological, engineering, social,
economic, political, and cultural—that contribute to a city’s earthquake disaster
risk, and a conceptual framework is created to organize these factors and facilitate
understanding of how they relate to each other and to the overall disaster risk. A
contributing factor can be any characteristic of a city’s physical makeup, location,
residents, or activities that can significantly affect the expected impact from future
earthquakes.

INDICATORS CITIES
A B C D E
EARTHQUAKE DISASTER RISK Indicator 1 X 1A X 1B X1C X 1D X 1E
Indicator 2 X2A X 2B X 2C X2D X 2E
HAZARD HA HB HC HD HE
Emerg. Resp.& Indicator 3 X 3A X3B X 3C X 3D X3E
Hazard Exposure Vulnerability External Indicator 4 X 4A X4B X 4C X 4D X4E
Context Recovery Cap.
EXPOSURE EA EB EC ED EE
Indicator 5 X 5A X 5B X 5C X 5D X5E
Ground Physical Physical Planning
shaking infrastructure infrastructure Economy Indicator 6 X 6A X 6B X6C X 6D X 6E
VULNERABILITY VA VB VC VD VE
Collateral Population Population Resources Indicator 7 X 7A X7B X 7C X7D X7E
hazards Transportation
Indicator 8 X 8A X8B X 8C X8D X8E
Mobility EXT. CONTEXT CA CB CC CD CE
Economy Economy Politics X9B X 9C X9D X9E
& Access Indicator 9 X 9A
Indicator 10 X10A X 10B X 10C X 10D X 10E
Social-political Social-political RESPONSE RA RB RC RD RE
system system Culture
EDRI EDRIA EDRI B EDRIC EDRID EDRI E
Step 1. Factor identification and conceptual framework Step 4. Data gathering and evaluation
FIVE MAIN
FACTORS INDICATORS
XH : exp(MMI w/50-year return per.)
XH
1 : exp(MMI w/500-yr. return per.) Sensitivity of Hazard factor to data uncertainty
XH
2 : Percent of urb’d area w/soft soil
Hazard XH4
3 : Percent of urb’d area with high 3.5
liquefaction susceptibility
XH5: Percent of buildings that are wood
XH6: Population density 3.0
XH7: Tsunami potential indicator
XE1: Population
XE2: Per capita GDP, const. 1990 US$ 2.5
Exposure XE3: Number of housing units
XE : Urbanized land area
XE4 :Population 2.0
XE6
5 : Per capita GDP, const. 1990 US$
XV1: Seismic code indicator 1.5

XV2: City wealth indicator
Vulnerability XV3: City age indicator
XV4: Population density 1.0
XV5: City development speed indicator Bos. Ist. Jak. Lima Man. Mex. SF Sant. St. L. Tok.
XV6: Percent of pop’n. aged 0-4 or 65+
External
XC1: Economic context indicator Step 5. Sensitivity analysis
XC2: Political country context indicator
Context XC3: Political world context indicator
XR1: Planning indicator

XR2: Per capita GDP, const. 1990 US$
XR3: Ten-year average of annual real
Emergency growth in per capita GDP
Response & XR4: Housing vacancy rate
Recovery XR5: Num. hospitals per 100,000 2.6 EDRI 23%
values Factor5
Planning XR6: Num. physicians per 100,000 2.2 values 14%
XR7: Extreme weather indicator 4 26%
XR8: Population density 1.8 3
XR9: City layout indicator 17%
1.4 2
1.0 1 18%
SF Bos. Tok. Lima Jak. 0
Step 2. Indicator selection SF Bos. Tok. Lima Jak. 25% Jakarta
8% 17%
EDRI
EDRI
RESULTS
RESULTS 34%
[x − (x -2s )]
ij i i
• Scaling:
EDRI
si 9% 32%
subjective Tokyo
• Weighting: assessment
Hazard Exposure Vulnerability External Context Emer. Resp. & Recov.

EDRI = w HH + w EE + wV V + wCC + wRR Time
Step 3. Mathematical combination Step 6. Presentation and interpretation
Figure 1.1. Six-step EDRI development procedure
2. Indicator selection
One or more simple, measurable indicators (e.g., population, per capita gross
domestic product, number of hospitals) are selected to represent each of the broad,

abstract factors in the framework. Operationalizing the factors, and thus the
concept of earthquake disaster risk that they collectively define, enables the
performance of objective, quantitative analysis.
3. Mathematical combination
A mathematical model is developed to combine the indicators into the composite
EDRI that best represents the concept of earthquake disaster risk.
4. Data gathering and evaluation

Data is gathered for each indicator and each of the world's major cities. The values
of the main contributing factors and of the EDRI are evaluated for each city using
the mathematical model developed in Step 3.
5. Sensitivity analysis
A sensitivity analysis is performed to determine the robustness of the results, given
the many uncertainties involved in the analysis.
6. Presentation and interpretation

The results are presented using a variety of graphical forms (e.g., charts, maps) to
make them as easily accessible as possible, and the numerical findings are
interpreted to assess their reasonableness and their implications.
The EDRI development begins theoretically and gradually incorporates the restrictions
associated with gathering and comparing international data. The conceptual framework
constructed at the outset provides a solid foundation and rationale for the choice of indicators
that are included in the EDRI and the way in which they are combined. Operationalizing the
framework’s ideas moves the framework from theory to practice, transforming it into a working
model that can produce objective, directly usable results. As a final note, the EDRI
development process is not actually as linear as this six-step method suggests. Rather, it is an

iterative process in which the steps often cannot be performed strictly independently of each
other.
1.3. Organization of the dissertation
In this first chapter, after motivating the development of the EDRI through a discussion
of its anticipated benefits, the basic six-step approach used to create the EDRI was outlined.
Chapter 2 explicitly highlights a few key features of the EDRI to distinguish it from traditional
probabilistic earthquake loss estimation models and earthquake damage scenarios, and to clarify
what the EDRI aims to achieve and how. Proceeding sequentially, each of the Chapters 3 to 8
expounds on one of the steps in the six-step procedure outlined above: Factor identification
and conceptual framework development (Chap. 3), Indicator selection (Chap. 4),
Mathematical combination (Chap. 5), Data gathering and evaluation (Chap. 6),
Sensitivity analysis (Chap. 7), and Presentation and interpretation (Chap. 8). In the course
of this project, a sample analysis was conducted both to produce the EDRI proposed herein,
and to illustrate the EDRI development process so that future researchers could repeat the
process, altering it as they see fit. Each of the Chapters 3 to 8 describes the basic procedure
necessary to complete the associated step, discusses the issues and challenges that arise in
doing so, and presents the results of the sample analysis for that step. Chapter 9 concludes by
suggesting future work to improve and build on the EDRI.

Chapter Two
Key Features of the Earthquake Disaster Risk Index
2. Introduction
Earthquake risk assessment efforts to date generally have taken one of two forms—loss
estimation models or earthquake damage scenarios. The Earthquake Disaster Risk Index is
fundamentally different from both. The remainder of this chapter highlights a few key features of
the EDRI to distinguish it from the more familiar types of earthquake risk assessment, and to
elucidate and elaborate on what the EDRI aims to achieve and how. First, since the concepts
that the three techniques assess are somewhat different, Section 2.1 discusses at length the
definition of earthquake disaster risk, the idea which the EDRI aims to describe. The objectives
of the index are laid out explicitly in Section 2.2. Section 2.3 describes the holistic,
multidisciplinary approach that has been adopted in creating the EDRI. Section 2.4 discusses
the EDRI’s use of the greater metropolitan area as a unit of study, and Section 2.5 explains why
the composite index form was selected as a means of presenting the risk assessment findings.
Table 2.1 offers a summary comparison of the EDRI, loss estimation, and earthquake damage
scenario methods.
2.1. Definition of earthquake disaster risk
While loss estimation models estimate the expected impact of future earthquakes (e.g.,
deaths, injuries, damaged buildings, economic loss), the EDRI aims to assess the risk of
earthquake disaster. The economic, social, political, and cultural context of the
Chapter Two: Key Features 9

Table 2.1. Comparison of three earthquake risk assessment methods
EDRI Loss Estimation Models Earthquake Damage
Scenarios
Input Simple, measurable, Detailed technical Detailed qualitative and
scalar indicators information quantitative info.
Output Composite index. Expected structural Qualitative description
Relative EDRI values of damage and economic of expected series of
cities. Relative loss in absolute terms events following an
contributions of factors earthquake
to EDRI.
Unit of study Greater metropolitan Specific site (aggregate Greater metropolitan
area to region or portfolio) area
Data needs Low High High
earthquake hazard, therefore, plays a critical role. As it is considered here, disaster is a function
of not only the physical impact of an earthquake, but also the capacity of the affected city to
sustain that impact, and the implications of the impact to the city and to world affairs.
Many attempts have been made to define the term earthquake disaster by setting forth
the set of conditions that must be satisfied to make an event a disaster (Appendix A). These
past efforts can be divided into two groups. In the first, the conditions are quantitative and
relate to the resulting number of deaths and injuries, economic loss, and/or duration of
functionality loss of various services. For example, World Disasters Report 1994 defined as a
disaster any event that caused ten deaths, and/or affected one hundred people, and/or led to an
appeal for international assistance (International Federation of Red Cross and Red Crescent
Societies, and the Centre for Research on the Epidemiology of Disasters 1994).
In the second camp, the conditions are more qualitative and incorporate the importance
of the context in which the damage and losses occur. Wenger, for example, asserts that a
disaster is an “event that creates demands upon the community system that cannot be met by its
traditional, institutional structure” (Wenger 1978). A tradeoff exists between the quantitative
and qualitative types of conditions. The former neglect the importance of the context in which
10 Chapter Two: Key Features

the damage occurs and involve somewhat arbitrarilydefined cutoff values, but are objective and
operational; the latter are more comprehensive and do not include arbitrary threshold values, but
are subjective and require judgment to apply. Ideally, a definition of earthquake disaster could
be both operational and comprehensive.
While most people have some intuition about what constitutes a disaster, the plethora of
definitions of disaster that have been proposed demonstrates that “there cannot be one all-
purpose term with a single referent which can meet all needs—legal, operational, or scientific—
and be equally useful to all users. What is important is not consensus on one definition—an
impossible goal—but clarity of the term and its referent on the part of various users” (Quarantelli
1987). With that understanding, it is clear that the treatment of the term earthquake disaster
risk suggested here will not be satisfactory to all users any more than those presented
previously. The following approach to earthquake disaster risk has been tailored to the issues
of earthquake-related risk assessment and the EDRI project in particular. It aims to be both
operational, like the quantitative definitions, and comprehensive, like the qualitative ones.
In the development of the EDRI, earthquake disaster risk is explored, not by directly
examining the expected consequences of earthquake disaster risk (e.g., deaths, economic loss),
but by considering the factors that contribute to earthquake disaster risk (e.g., frequent
earthquakes, vulnerable structures). In the first approach, loss estimation studies could be
conducted in every city, and the cities with higher expected losses would be considered to have
higher risk. In the EDRI approach, a city is considered to have a relatively high risk if it exhibits
a lot of characteristics that typically cause more severe or more frequent earthquake losses.
This latter approach requires identifying the factors that contribute to earthquake disaster risk
and assuming either a direct or inverse relation between each factor and the risk. If a factor has
a direct relationship with risk and the factor value is higher in City A than City B, then the
earthquake disaster risk in City A will be higher too. If the same factor had an inverse
relationship with risk, the earthquake disaster risk will be lower in City A than in City B. Once

it is decided how Cities A and B compare with respect to each contributing factor, that
information can be integrated to determine how they compare with respect to the overall risk.
In this project, the main factors that have been identified as contributing to a city's earthquake
disaster risk are:
• Hazard
Severity, extent, and frequency of the geological trigger phenomena to which the city
may be subjected.
• Exposure
Size of the city. Quantity of people and physical objects, and the amount and type of
activities they support.
• Vulnerability
How easily the exposed people, physical objects, and activities may be affected in the
short- or long-term.
• External Context
How impact within a city affects people and activities outside the city.
• Emergency Response and Recovery Capability
How effectively and efficiently a city can reduce the impact of an earthquake through
formal, organized efforts made specifically for that purpose.
Evaluating a city with respect to all these contributing factors and combining the resulting
information provides a portrait of the city's earthquake disaster risk. This new approach
circumvents many of the difficulties associated with consequence-based disaster definitions.
The particular types of consequences, or impact parameters (Fig. 2.1) are not addressed
individually, because the definition of earthquake disaster risk proposed here does not rely on
enumerating the expected consequences of risk. Instead, they are all

II. Terms of assessment
I. Type of losses • Total for the city overall

• Total for the city as a percentage
In physical terms: • Distribution within city
• Damaged buildings and • Degree to which it is felt outside city
infrastructure facilities
• Destroyed buildings and
III. Examples of impact parameters
infrastructure facilities
An impact parameter is a unit of measure
In economic terms:
that could be used to assess the severity
• Direct economic loss
of earthquake impact. Impact parameters
• Indirect economic loss can be formed by choosing one item from
each of boxes I and II. The following are
In human terms: some examples of possible impact
• Deaths parameters:
• Injuries
• Homelessness • Total number of damaged or destroyed
structures in the city
In terms of disruption: • Direct economic loss as a percentage of
• Disruption to lifeline services the city’s total wealth
• Disruption to political processes • Percentage of people homeless in the
• Disruption to social processes most severely impacted group
• Degree to which political disruption is
felt outside the city
Figure 2.1. Impact parameters
included by defining a contributing factor as any characteristic of the city that increases or
reduces the consequences of an earthquake, as measured by any one or more types of losses
(Fig. 2.1; Box I), using any of the four terms of assessment (Fig. 2.1; Box II). With the
proposed definition, it is not necessary to decide which of the impact parameters should be
combined and how (Should the definition depend only on the expected number of deaths? Is
economic loss important too? Are they equally important?); what threshold values should be
used for each parameter (Do ten deaths constitute a disaster? One hundred?); or if the
definition should focus on the city as a whole, on its most severely impacted neighborhoods,
and/or on the extent to which the impact is felt outside the city. Unfortunately, the problem of
weighing the relative importance of different impact parameters (e.g., number of deaths versus
dollars of economic loss) is not entirely avoided, but rather replaced by the task of weighing the

relative importance of the factors that contribute to earthquake (e.g., Hazard versus Exposure;
see Section 5.3). The latter approach does not require “value of life” judgments, or the
estimation of the expected number of deaths, economic loss, and other impact parameters in
absolute terms. The EDRI approach is also more amenable to assessment of a comprehensive
concept earthquake disaster risk that addresses all impact parameters, rather than one focused
on the more quantitative aspects of earthquake impact.
In the EDRI approach, disaster risk is viewed from a global perspective. An observer
may consider a specified impact level to be more or less serious depending on his point-of-
view. For example, if the impact level is considered from the perspective of an individual, a
single fatality would be a disaster if it was the individual’s parent who died, but fifty deaths may
not be a disaster for that individual if he had no personal connection to any of those killed.
Similarly, an earthquake might be a disaster for the Japanese government, but not for the Cuban
government. Therefore, to interpret the impact level from a global perspective implies not that
the impact must shake the entire global civilization, but that all the effects of an earthquake must
be considered, whether they occur within the area of ground shaking or elsewhere in the world.
This global point-of-view gives significance to a city’s prominence on the world stage, and is
represented by the inclusion of the External Context factor.
The EDRI considers relative risk only, not absolute risk. Definitions that propose an
absolute scale with which to measure earthquake disaster risk (e.g., number of deaths, injuries,
buildings damaged, economic loss, or some combination thereof) neglect part of what it takes to
have a disaster. Since no units exist with which to measure the broad concept of earthquake
disaster risk used in the EDRI, the index presents a relative measure of risk, which is entirely
adequate for comparing the risk in different cities. The level of earthquake disaster risk is
conceived of as a continuous, unitless, open-ended scale, not a binary categorization of events
with only two possible values: Disaster or Not a disaster.

The EDRI addresses disaster risk directly; no definition of a disaster is provided. To
make the transition from a definition of disaster to disaster risk, the concept of probability must
be incorporated. With a definition of disaster in hand, disaster risk can be defined simply as
the probability that a disaster will occur in a specified future time period. In this project, with no
definition of exactly which events are classified as disasters and which are not, that approach is
not possible. Instead, probability is incorporated in the Hazard factor, which includes the
probabilistically expected levels of ground shaking for two different time periods. The choice of
time period implicitly defines a time horizon for the EDRI. If the level of ground shaking with a
one-year return period was used in the Hazard factor, for example, the hazard for many cities
would be zero. However, there is an enormous difference between a small chance of an
earthquake and no chance. By using the levels of ground shaking with fifty-year and five-
hundred-year return periods, the EDRI addresses the small, but real hazard in areas of low
seismicity. Return periods of fifty and five hundred years correspond to shaking with a ten
percent chance of being exceeded in five and fifty years, respectively. These are reasonable
time horizons for planners and individual citizens.
2.2. Objectives of the EDRI
It is useful to set forth the EDRI’s objectives up front so that they may guide the
decisions encountered during the index’s development, and so that they may provide a basis for
assessing the success of the final index in achieving its aims. The broad goal of the EDRI is
similar to that of the Consumer Price Index or the Physical Quality of Life Index, but instead of
rating the relative levels of prices in different years, or the relative quality of life in various
countries, the EDRI rates the relative levels of earthquake disaster risk in different cities. Like
the other composite indexes, the EDRI aims to establish a yardstick with which to measure an
unobservable concept, in this case, earthquake disaster risk. Specifically, the EDRI is intended
to:

1. Convey the relative overall earthquake disaster risk (as defined in Section 2.1) at a
point in time for each major city in the world.
2. Convey the relative contribution1 of each component factor to a city’s overall
earthquake disaster risk.
3. Convey the change in a city’s relative overall earthquake disaster risk over time.
4. Be straightforward to develop, evaluate, and interpret so that the information it
conveys is easily accessible to the general public.
The EDRI does not attempt to measure the earthquake disaster risk on any absolute
scale (see Section 2.1). While that goal is appealing, as yet no units exist to measure
earthquake disaster risk as it is defined in this study. Instead, a city’s earthquake disaster risk is
assessed only relative to that of other cities. The EDRI also does not attempt to provide
information about the distribution of risk within a city. While this would be another worthwhile
goal, it is not one that the EDRI endeavors to achieve. The EDRI is oriented toward providing
inter-city, not intra-city comparisons of risk. Finally, the EDRI is fundamentally concerned with
providing information about the factors that come together to create earthquake risk (see
Objective 2), not only in providing a way to describe the varying degrees of expected impact.
2.3. Holistic, multidisciplinary approach
The development of the EDRI represents a holistic, multidisciplinary approach to

earthquake risk assessment. Taking a fresh look at urban earthquake risk, the process begins
by asking, without bias to any single discipline, “What characteristics of a city help determine the
impact that an earthquake will have?” A characteristic could be of any type—geological,
structural, economic, social, political, cultural, or anything else. As long as it contributes to
earthquake disaster risk (as defined in Section 2.1), the characteristic should be considered a
factor.

To get a handle on the expansiveness of the problem, the challenges of earthquake risk
assessment have been divided into smaller, focused tasks, and groups from each discipline have
undertaken to address the tasks that relate to their area of expertise. Geologists study the
characteristics of fault rupture. Structural engineers explore the vulnerability of the built
environment. Emergency response planners investigate the requirements of an effective
response effort. Economists try to understand how damage to the physical infrastructure
translates into disruption in a region’s economy. Each group has delved deeply into the details
of their task, and the understanding of what happens following an earthquake has improved
markedly. As a holistic, multidisciplinary effort, the EDRI aims to put the pieces of the puzzle
back together; to integrate the findings of the detailed studies in every field, and to interpret their
implications for the ‘big picture.’
The EDRI is, of course, not the first study to exhibit a holistic orientation. The following
brief literature survey will offer some background on the history of holistic, multidisciplinary
earthquake risk assessment work, and will demonstrate how the EDRI aims to contribute
something new and valuable to the collection. The surveyed work can be separated into the
engineering and social science camps.
The efforts of the engineering community to put the pieces together are primarily
centered on earthquake loss estimation modeling. The development of probabilistic seismic
hazard analysis (PSHA) in the 1960s (Cornell 1968) provided a methodology for mapping the
probabilistically expected levels of earthquake ground shaking for a region. In determining the
expected magnitude of ground motion for a specified site and future time period, the framework
integrates data about earthquake sources (e.g., locations, recurrence relationships, and
maximum possible magnitudes), the attenuation of ground motion as it travels from the source to
the site, and the ability of the soil at the site to amplify or reduce the ground motion. Typically,
the hazard is presented in terms of the probability of exceedence of a certain peak ground
1
Refers to the city-specific relative contribution (Section 5.1.4).

acceleration in a specified time period. Other local site effects, like liquefaction, landslides, and
surface fault rupture are portrayed on separate maps.
In probabilistic earthquake loss estimation modeling, engineers combine the hazard

information that is the output of the PSHA with information about the quantity, value, location,
and vulnerability of the exposed infrastructure to determine quantitative estimates of the potential
damage and economic losses that a region will suffer in a specified future time period. Figure
2.2, reprinted from King and Kiremidjian (1994), shows the basic steps in regional seismic
hazard and risk analysis. Other models have been developed to estimate economic losses
arising from business interruption and damage to non-structural components and contents; and
non-monetary losses, including casualties, injuries, unemployment, and homelessness. King and
Kiremidjian (1994) provides a clear overview of the seismic hazard and risk analysis
methodology.
Recently, this holistic engineering work has consisted primarily of automating the loss
estimation methodology, applying it to an increasing number of geographical regions and hazard
types, and refining the modules that comprise it. Earthquake loss estimation software has been
developed using expert systems and Geographic Information Systems (GIS) technologies
(Shah, Boyle, and Dong 1991). The software extends the possibilities of this methodology
enormously by enabling the maintenance of the large databases required, increasing the speed of
analyses, and providing user-friendly spatial representation of the data and analysis results with
GIS. Mention of a few studies will suffice to illustrate the variety of projects that explore
specific components of loss estimation modeling. Coburn, Spence, and Pomonis (1992)
estimate levels of human life loss due to building collapse in earthquakes. Scawthorn (1986)
estimates losses from fire following an earthquake. Cret and Katayama (1992) estimate
damage to lifeline systems during earthquakes. Assessment of the State-of-the-Art
Earthquake Loss Estimation Methodologies (RMS and CUREe 1994), which provides an
extensive review and comparison of available earthquake loss estimation studies, offers a helpful
summary of recent developments in this area.

frequency
point source
x STUDY x
REGION
magnitude
line
source
area source
a. Identify seismic sources b. Model earthquake occurrences
liquefaction zone
x STUDY fault rupture STUDY
REGION zone REGION
landslide zone
attenuation
bedrock surface
models
motion motion
soil profile
c. Determine motion at bedrock d. Model local site effects
damage STUDY loss STUDY

REGION REGION
% damage % loss
motion damage
tables of * direct losses
inventory * indirect losses
information * casualties
e. Estimate damage distributions f. Estimate loss distributions
Figure 2.2. Basic steps of regional seismic hazard and risk analysis
Reprinted from King and Kiremidjian (1994)
Working from an entirely different perspective and knowledge base, geographers and
other social scientists in disaster studies also have endeavored to put the pieces of earthquake

risk together. The body of mostly conceptual work they have produced over the past twenty
years chronicles the evolution of the way in which they conceive of disasters. At least until the
1970s, the prevailing view described natural disasters as being entirely the result of the
externallyimposed extremes of natural processes. Accordingly, research traditionally has
focused on understanding the geophysical or climatic mechanisms of different hazards, and on
the development of technical methods to reduce their impact on society. The traditional view
also tries to encapsulate natural disasters as a special problem distinct from “ordinary” life,
completely separate from the rest of society’s relationship with the environment. This “extreme
narrowing of the range of interpretation and acceptable evidence” (Hewitt 1983) makes the
complex problem of natural disaster mitigation more manageable, but neglects the
interconnectedness of everyday life and disasters.
The new paradigm that has been evolving over the past twenty years highlights the
relevance of the societal context in determining an event’s impact. A geophysical event alone
cannot create a disaster. Disasters occur only when a natural phenomenon intersects in space
and time with a vulnerable population. There is a new awareness of the interaction between
nature and society. Cities are no longer only at the receiving end of the disaster, innocently
suffering the damage nature inflicts. In the new paradigm, a city’s vulnerability and capacity to
handle impact are causal factors as significant as the earthquake itself. As part of this emerging
viewpoint, a broader concept of vulnerability has appeared that goes beyond the engineering
notion of structural vulnerability. Social, political, economic, and cultural characteristics of a city
are also important ingredients of the city’s vulnerability. For more discussion of this broad view
of natural disasters, see the following: Albala-Bertrand (1993); Blaikie et al. (1994); Burton,
Kates, and White (1993); Hewitt (1983); Kreimer and Munasinghe (1992); and Varley (1994).
The earthquake loss estimation modeling of the engineering community offers a

quantitative, objective approach to holistic work in earthquake risk assessment. It is focused on
delivering a product to its users, usually an estimate of the expected damage and economic loss
as it is distributed throughout a region. The engineering work is increasingly incorporating

social, economic, and political considerations, but mostly in the types of impacts that are
estimated (the output), not in the factors that create those impacts (the input). Engineering work
continues to exhibit a bias toward the technical and quantitative, largely ignoring the contextual
factors that contribute to earthquake risk. The social science community, on the other hand, has
taken a much more comprehensive approach to putting the pieces of earthquake risk together,
including all types of factors that contribute to the risk—geological, engineering, social,
economic, political, cultural—on equal footing. Nevertheless, its work has been mostly
conceptual, often remaining unknown, or at least unusable, to the decision-makers charged with
mitigating the assessed risk.
The EDRI aims to bring together the parallel research paths laid down by the
engineering and social science communities. The project to create the EDRI has been designed
to merge the comprehensiveness and broad perspective of the social sciences with the
engineering emphasis on remaining objective and quantitative, operationalizing, and delivering a
helpful product to those involved in managing earthquake risk.
2.4. Urban risk
The EDRI assesses a single value of earthquake disaster risk for each greater
metropolitan area (hereafter referred to as a city), providing no direct information about the
distribution of risk within that area. The index uses the city as the unit of study for a few
reasons. First, focusing on urban risk reflects the author’s recognition of the growing concern
among earthquake professionals about the potential impact of earthquakes in the world’s largest
urban areas. Recent efforts by two leading voices in earthquake risk assessment and
management illustrates the existence of this emerging consensus. In 1996, a consortium of nine
western U.S. universities defined the focus of the proposed Pacific Earthquake Engineering
Research Center (PEERC) to be the development of urban earthquake risk reduction
technologies. The Secretariat for the International Decade for Natural Disaster Reduction
(IDNDR) recently proposed the “Disaster Reduction for Sustainable Development” project,

which includes urban vulnerability reduction as one of the three thematic areas on which it
will concentrate (UN Secretariat for the IDNDR 1996). Second, by the year 2005, fifty
percent of the world's population will be gathered in urban areas, and by 2025, more than sixty
percent will be (United Nations. Dept. for Economic and Social Information and Policy
Analysis. Population Division 1995b), so by addressing only urban areas, most of the world's
population is still considered.
Third, earthquakes primarily affect people and the components of their built physical
environment, both of which are concentrated in urban areas. Furthermore, the unprecedented
size, complexity, and interconnectedness of today's megacities have created situations unlike any
that existed in the past. Smaller cities may expect to experience events similar to those that
occurred historically. There have been few large earthquakes that directly impacted a megacity,
however, so the potential implications of the new nature of cities on earthquake risk are unclear,
and pre-earthquake risk assessment takes on added importance for major urban
conglomerations. Fourth, the impact area of a single earthquake is of the same order as a
greater metropolitan area. Impact generally will not be contained within legal city limits; nor will
it extend over an entire country. Since many lifeline networks, economic, social, and political
functions are defined for or roughly uniform over greater metropolitan areas, that unit of area
provides the best common ground for assessing expected physical impact, emergency response
and recovery capability, and context of the impact.
It is generally understood that a greater metropolitan area is a large, continuous

urbanized area composed of many smaller cities or other administrative units whose economic
and social lives are interdependent. Nevertheless, no precise, universally applicable definition
exists. In the United States, the Bureau of the Census established the following definitions of a
Metropolitan Statistical Area (MSA) and a Consolidated Metropolitan Statistical Area
(CMSA):

“An MSA must include at least: (a) one city with 50,000 or more inhabitants,
or (b) a Census Bureau-defined urbanized area2 (of at least 50,000 inhabitants)
and a total metropolitan population of at least 100,000 (75,000 in New
England).” “An area that meets these requirements for recognition as an MSA
and also has a population of one million or more may be recognized as a
CMSA if: 1) separate component areas can be identified within the entire area
by meeting statistical criteria specified in the standards, and 2) local opinion
indicates there is support for the component areas” (U.S. Bureau of the Census
1995).
The Bureau of the Census also publishes lists of all MSA’s, CMSA’s, and the cities and
counties that are included in each.
The United Nations Population Fund says that a metropolitan area “includes the central
city, the suburbs and satellite communities whose economic and social life are tied to the city in
such a way that both the central city and its zone of influence form an indissoluble functional
whole” (Institut d’Estudis Metropolitans de Barcelona 1988). The determination of what areas
to include in a greater metropolitan area should be made for each urban center individually by
local authorities familiar with the area. To ensure consistency among indicators and among
cities, it is critical to establish at the outset exactly what areas are included in each greater
metropolitan area (city) in the analysis. For each city, the values for every indicator should refer
to the same area. It does not make sense to count the “number of hospitals” in Tokyo
prefecture, but determine the “number of housing units” from the entire five-prefecture Tokyo
metropolitan area. Similarly, comparing Boston as defined by its legal city limits, to the CMSA
of Los Angeles, which includes five counties would produce meaningless results. For the EDRI
sample analysis, the U.S. cities were defined as the appropriate CMSA, and the international
cities were defined using either the UN Population Fund guidelines, or definitions set forth by the
2
A census bureau-defined urbanized area “comprises one or more places (“central place”) and the adjacent
densely settled surrounding territory (“urban fringe”) that together have a minimum of 50,000 persons. The
urban fringe generally consists of contiguous territory having a density of at least 1,000 persons per square
mile. The urban fringe also includes outlying territory of such density if it was connected to the core of the
contiguous area by road and is within 1½ road miles of that core, or within 5 road miles of the core but
separated by water or other undevelopable territory. Other territory with a population density of fewer than
1,000 people per square mile may be included in the urban fringe if it eliminates an enclave or closes an
indentation in the boundary of the urbanized area” (U.S. Bureau of the Census 1993b).

local statistical agency. Appendix B describes which areas (e.g., counties, cities) are included in
each of the ten sample analysis cities.
2.5. Composite index form
Composite indexes have been used for a long time in a wide variety of disciplines to
measure complex, multi-dimensional concepts that cannot be observed or measured directly.
Their power lies in their ability to synthesize a vast amount of diverse information into a simple,
easily usable form. In the case of the EDRI, there are volumes available on the seismicity,
structural types, economic and political conditions of different cities. All of that material would
be helpful in understanding the relative earthquake disaster risk of those cities, but the amount of
information could be overwhelming. It is difficult to see the forest when lost among so many
trees. The EDRI offers a unique way to bring the many components of earthquake risk together
to reveal the big picture. Furthermore, the straightforwardness of the development, evaluation,
and interpretation of the EDRI makes the information easily accessible to the general public,
government agencies, insurance companies, and other potential users.
While the EDRI can provide a concise tool for measuring earthquake disaster risk that
currently is not available from any other source, clearly it cannot profess to provide all the insight
that the volumes of theory and data on seismicity, geology, structural engineering, economic
modeling, and other relevant topics would. For detailed understanding of each individual
component of earthquake disaster risk, the focused studies that address those topics are the
best sources. To understand the sum effect of all those components, the EDRI should prove
useful.
One drawback of the composite index form should be noted here. Since earthquake
disaster risk cannot be measured directly (in fact, the purpose of creating the EDRI is to provide
a way to do so), there is no satisfactory way to validate the EDRI. In the chapters that follow,
many aspects of the EDRI are validated to the extent possible. This includes, among other

things, demonstrating that the factors that are said to contribute to earthquake disaster risk do,
establishing that the indicators represent the concepts that they purport to represent, justifying
the proposed values of the weights, and verifying the reasonableness of the final results of the
sample analysis. Most often these efforts at validation fall short of scientific proof, and instead
rely on imperfect empirical evidence or even intuition. A sensitivity analysis was performed to
determine the robustness of the results in the face of the uncertainties in the EDRI development
and evaluation.
Nevertheless, the fact that scientific validation is impossible is not cause to discard the
idea of a composite index. The Consumer Price Index is just one reasonable representation of
the general level of prices. It is derived from the prices of a certain group of goods and the
quantity of each that were sold at those prices in a given year. While one could argue that a
better representation of the general level of prices would rely on a different set of goods, the fact
that the current derivation of the CPI has been accepted as a reasonable one, and that the index
has been tracked for many countries and many years, has endowed the CPI with a substantial
degree of meaning and usefulness. The EDRI has the potential to serve an analogous role in the
area of earthquake risk assessment.
The plethora of existing indexes are evidence of the usefulness of the composite index
form. The Consumer Price Index, Dow Jones Industrial Average, Economic Freedom Index
(Johnson and Sheehy 1996), Human Development Index (UNDP
1990-95), Physical Quality of Life Index (Morris 1979), and World Competitiveness Index
(IMD International and WEF 1992) are but a few examples of composite indexes successfully
measuring that which previously could not be measured.

Chapter Three
Factor Identification and Conceptual Framework
3. Introduction
The conceptual framework that describes the components of earthquake disaster risk
provides the foundation on which the Earthquake Disaster Risk Index rests. To determine
which indicators to include in the EDRI, the factors that they aim to represent must be well-
defined. Any errors or omissions in the framework will be perpetuated through the subsequent
steps of the project, resulting in an EDRI that does not represent earthquake disaster risk as
well as it could. This chapter presents the construction of the conceptual framework of
earthquake disaster risk. Section 3.1 describes the basic approach taken to complete this task,
and Section 3.2 discusses the challenges that arise in the process. Section 3.3 presents the
results of this first step from the sample analysis. It first reveals the conceptual framework that
was developed, then elaborates on each of the factors that comprise it. The discussion of each
factor includes a definition, a discussion of how it contributes to a city’s earthquake disaster
risk, and an explanation of the variables on which it depends.
3.1. Approach
The strategy for identifying the factors that contribute to earthquake disaster risk
involves starting with the most general concept—earthquake disaster risk, decomposing it into
the factors that together describe it, then decomposing each of those factors into sub-factors,
Chapter Three: Conceptual Framework 27

and so forth. For example, a thorough understanding of a city’s earthquake disaster risk
requires knowing something about the hazard to which the city is subjected. To be able to
assess a city’s hazard, one must know about the ground shaking hazard as well as the collateral
hazards, such as liquefaction, landslides, and tsunami. The process continues until all of the
factors and sub-factors have been identified. The development of the framework is an entirely
conceptual undertaking that should be performed without regard for issues of measurability or
data availability. The final conceptual framework should be as comprehensive as possible, even
if measurability and data constraints prevent some of its components from being included in the
final EDRI.
3.2. Challenges
The search for factors that contribute to earthquake disaster risk is complicated for a
few reasons. With earthquake disaster risk defined as broadly as it is, it becomes clear that the
contributing factors span an extremely wide range of disciplines. Furthermore, the factors are
not well-defined, distinct entities that exist a priori. The index developer must define them into
existence. Geological, economic, social, and political activities overlap and interact in the
complex web that forms the life of a city. The factors must be extracted from that web and
defined so that they are mutually exclusive and collectively exhaustive. Aspects of the city that
affect its earthquake disaster risk should not be double counted or omitted. For example, since
Economic External Context includes the economic effects of impact to a city that will be felt by
those outside the city, the Economic Vulnerability factor should refer only to the ease with which
activities within the city may be disrupted.
Factors should be defined so that they are neither too specific nor too general. They
should be disaggregated to the point that is most natural for their assessment. Factors only need
to be as specific as necessary to convey the concept of earthquake disaster risk clearly.
Making them more specific than required will succeed only in complicating the framework. If
the factors are defined to be too general, it may be difficult to hone in on the way in which they
28 Chapter Three: Conceptual Framework

affect earthquake disaster risk. For example, suppose the factor “Economic Wealth” was
proposed. It could be argued that risk both increases and decreases with Economic Wealth.
Cities with greater wealth generally have better enforcement of building codes, which would
indicate a lower risk; but they also produce more goods and services, so damage in wealthy
cities would create more indirect economic losses, thereby increasing the risk. A better
alternative would be to consider building code enforcement and the amount of production as
two separate factors, rather than considering the broader “Economic Wealth” factor.1
Whenever possible, a factor should be important to earthquake disaster risk in itself, not
because it affects something else that, in turn, is important to earthquake disaster risk.
“Enforcement of Building Codes” would be an example of the first case, a directly relevant
factor, while “Economic Wealth” could be an example of the second, a factor that is relevant
only indirectly.
However carefully the factors are defined, the innumerable interactions that exist among
the factors make it possible to argue that almost every factor both increases and decreases
earthquake disaster risk, at least indirectly. A large proportion of wood-frame structures, for
example, could be considered to increase the vulnerability to fire damage, and therefore to
increase the risk; or one could argue that wood-frame structures generally exhibit excellent
resistance to ground shaking damage, so a high proportion of them would indicate lower risk.
Following are other examples of interactions among factors:
• The components of the Exposure factor are related to each other, e.g., cities with
more people will have more physical infrastructure as well.
• The soft soil, liquefaction, and landslide components of Hazard are all related to a
city's soil conditions.
1
Note that because of data constraints, there is often less freedom in the selection of indicators for the EDRI
than in the definition of conceptual factors. There may be cases in which an indicator is used more than
once to represent more than one concept. For example, if it was impossible to find appropriate alternative
indicators for enforcement of building codes and amount of economic production, the per capita Gross
Domestic Product could be used to represent both. A different weight could be associated with each use.
This topic is discussed more thoroughly in Section 4.2.

• Economy Exposure, Economy Vulnerability, Economic External Context, and the
financial resources component of Emergency Response and Recovery are all related
to the structure and size of a city's economy, and therefore, they are related to each
other.
• Physical Infrastructure Vulnerability and Mobility and Access are related because
the former helps determine the amount of debris and the condition of the
transportation network following an earthquake, two conditions that, in turn, help
determine the latter.
• The Planning component of the Emergency Response and Recovery factor may be
related to the Hazard if cities with a history of earthquakes have been convinced by
their experience to plan for future events.
Given the complexity of the interactions among factors, the best that can be done is to
acknowledge this difficulty, define the factors in a way that avoids this problem as much as
possible, focus on the principal effect that each factor has on earthquake disaster risk, and
determine weights that reflect only that principal effect on risk. To continue the previous
example, if the proportion of structures that have a wood-frame is included as an indicator to
represent the city’s vulnerability to fire hazard, then the weight associated with that indicator
should be determined based only on the relative importance of the proportion of wood-frame
structures as it affects vulnerability to fire hazard, with no regard to what it indicates about the
vulnerability to ground shaking.
Due to the many interactions among factors, almost any aspect of a city's makeup could
be considered to affect the city's earthquake risk in some way. To be included in the final
framework then, factors must contribute significantly to the risk and must have values that
potentially vary from city to city, so that they may affect the relative rankings of different cities.
3.3. Sample analysis solution

Figure 3.1 illustrates the conceptual framework of earthquake disaster risk that was
developed in this work. It is hypothesized that the five main factors contribute to a city's
earthquake disaster risk: Hazard, Exposure, Vulnerability, External Context, and Emergency
Response and Recovery Capability. Each of these five main factors is disaggregated into the
more specific factors that comprise it. For simplicity, the framework does not portray
interactions among the factors.
EARTHQUAKE DISASTER RISK
Hazard Exposure Vulnerability External Emerg. Resp.&

Context Recovery Cap.
Ground Physical Physical Planning

Shaking Infrastructure Infrastructure Economy
Collateral Population Population Resources

Hazards Transportation
Economy Economy Mobility

Politics & Access
Social-Political Social-Political
System System Culture
Figure 3.1. Conceptual framework of earthquake disaster risk
3.3.1. Hazard factor
Hazard represents the geophysical phenomena that serve as initiating events of

earthquake disaster, the demand to which a city will be subjected. This factor describes, for
both ground shaking and collateral hazards (i.e., fire following, liquefaction, landslide, and
tsunami and seiche), the frequency of each possible severity level as it is distributed throughout
the city.
3.3.1.1. Ground Shaking

Ground Shaking is the most important component of Hazard because it is usually
directly responsible for the majority of damage suffered in an earthquake, and because most
other hazard types (i.e., liquefaction, landslide, fire) require a sufficient level of ground shaking
to trigger them. The characteristics of ground shaking that may be of interest include the
amplitude, frequency content, and duration. Amplitude is typically assessed in terms of
acceleration (e.g., peak ground acceleration, PGA, or spectral acceleration, Sa) or intensity
(e.g., Modified Mercalli Intensity, MMI, or Japan Meteorological Agency, JMA).
Given the occurrence of a particular earthquake, the amplitude of the resulting ground
shaking can be determined at every location throughout a city. In assessing the city’s overall
earthquake risk, however, it is important to know not only the amplitude of ground shaking that
will be caused by each conceivable earthquake, but also the probability that each of those
earthquakes will occur. By using Probabilistic Seismic Hazard Analysis (PSHA), these two
equally important quantities can be integrated into a single parameter, the expected value of the
magnitude of ground shaking for a specified future time period. Employing a short time period
shows how frequently ground shaking occurs, while using a longer time period conveys how
strong the ground shaking potentially can be. The probabilistically expected severity of ground
shaking at a given site depends on (1) the locations, recurrence relationships, and maximum
possible magnitudes of the faults and area sources that might produce the ground shaking; (2)
on the attenuation relationships that are associated with soil and topography and describe the
way in which ground motion changes in magnitude and frequency content as it travels from the
source to the site; and (3) on the ability of the soil at the site to amplify or reduce the ground
shaking. Topography and directivity of fault rupture have been identified as additional
determinants of the severity of ground shaking at some sites, but they have not been
incorporated into the PSHA yet.
3.3.1.2. Liquefaction

Liquefaction is a phenomenon in which soil “changes from a firm material into a viscous
semi-liquid material” (Gere and Shah 1984). “If a saturated sand is subjected to ground
vibrations, it tends to compact and decrease in volume; if drainage is unable to occur, the
tendency to decrease in volume results in an increase in pore water pressure, and if the pore
water pressure builds up to the point at which it is equal to the overburden pressure, the
effective stress becomes zero, the sand loses its strength completely, and it develops a liquefied
state” (Seed and Idriss 1983).
Liquefaction can cause structural damage to buildings and lifelines. When soil liquefies,
it loses its bearing strength, causing any buildings or other structures that rest on it to settle.
Structures may remain essentially intact, but tilted. In that case, the structure must be rebuilt, but
little loss of life results. Liquefaction can damage underground infrastructure, like pipelines and
septic tanks, by causing them to float up towards the surface of the ground. Roads and railroad
tracks can also be damaged when a portion of the road or track sits on soil that liquefies and
settles, while an adjacent portion does not. Liquefaction became widely recognized as a
significant collateral hazard after causing extensive damage in the Niigata, Japan and Alaskan
earthquakes of 1964. In the more recent Great Hanshin earthquake of 1995, the lateral
spreading of the ground associated with liquefaction, caused extensive damage.
Rather than measure degrees of liquefaction, it usually suffices to make a binary

assessment of whether liquefaction occurred or not, and refer instead to the area over which
liquefaction takes place. The liquefaction potential of an area (the probability that liquefaction
will occur) equals the convolution of the liquefaction opportunity (a function of the ground
shaking demand on the soil) and the liquefaction susceptibility (the probability that liquefaction
will occur given sufficient ground shaking). Liquefaction susceptibility is determined by the type
and relative density of the soil, and the water table depth. Liquefaction can occur only in
cohesionless, sandy soil that is saturated with water (RMS 1997).

3.3.1.3. Fire Following Earthquake
Fires that break out following earthquakes can be extremely destructive because there
may be many simultaneous ignitions, and because the fire suppression capability may be
impaired by damaged water supply systems, the unavailability of personnel, and restricted
mobility due to debris-filled, damaged roads. The severity of the fire following earthquake
hazard is often described in terms of the percentage of the city area that is expected to be
burned.
Three main components govern the fire hazard: the number and locations of potential
ignition sources, the ease with which fire will spread in the city, and the ease with which fire can
be suppressed. Potential post-earthquake ignition sources include gas line breaks, overturned
water heaters, electrical shorts, and flammable liquid spills. The fire spread depends on the
distribution of fuel (e.g., wooden structures, building contents) and the weather conditions (e.g.,
humidity, temperature, wind speed, and wind direction). Fire suppression depends on the
characteristics of the region’s fire suppression agencies—their management, fire-fighting
personnel, emergency water supply, and equipment (e.g., fire trucks, stations, and hydrants).
Conceptually, fire suppression could be considered either as part of the Hazard factor, as
discussed here, or as part of the Emergency Response and Recovery Capability factor, as
discussed below. To avoid double counting, it should be included in one but not both factors.
Since fire suppression is an activity that is carried out to reduce impact and not a characteristic
of the natural setting, it is considered to be a component of the Emergency Response and
Recovery Capability factor.
3.3.1.4. Landslide
Landslides can be triggered by the ground shaking associated with earthquakes.

Besides damaging buildings and other structures, landslides can increase the impact of an

earthquake by blocking and destabilizing roads, and by blocking rivers and lakes, and thus
inducing flooding. Like liquefaction, landslide hazard is discussed in terms of susceptibility,
opportunity, and potential. Landslide susceptibility depends on soil cohesion, water level, and
slope steepness. Landslide opportunity depends on ground shaking demand, precipitation, and
distance to the earthquake source.
3.3.1.5. Tsunami and Seiche
Tsunamis are “long period sea waves produced by uplifts or downdrops of the sea floor
during an earthquake. Once generated, a tsunami can travel thousands of miles, propagating at
speeds as high as five hundred miles per hour. Although virtually undetectable at sea, the
incident waves interact with shallow sea floor topography upon nearing a landmass, causing the
characteristic increase in wave height, or run-up (defined as the maximum elevation reached by
the wave about the initial water level)” (RMS and CUREe 1994). When a tsunami runs up on
land, everything in its path will be inundated with water, and often will be dragged out to sea by
the wave as it ebbs. Because they can occur quickly and without warning, and because they
will indiscriminately destroy everything in their paths, tsunamis can cause severe loss of property
and life. A seiche is a major water wave that occurs in a lake, bay, or other inland body of
water because of large uplifts or downdrops of land, e.g., in the 1959 Hebgen Lake, Montana
earthquake (RMS and CUREe 1994).
The severity of a tsunami, described in terms of its run-up height and the area that it
inundates, depends on three conditions: (1) the seismic source mechanism, i.e., the magnitude
and type of earthquake faulting that generates the tsunami; (2) the distance and seafloor
topography over which the tsunami travels as it propagates from the source to the site of
interest, and (3) the topography of the coast at the run-up site (RMS and CUREe 1994).

3.3.2. Exposure factor
Exposure describes the size of a city; a list of everything that is subject to the physical
demands imposed by the hazard. It includes the quantity and distribution of people and physical
objects, and the number and type of activities they support. Exposure is a necessary
component of risk. No matter how severe the hazard, without an exposed population and
infrastructure, there would be nothing to be damaged or disrupted, and therefore, there would
be no risk. The larger the exposure then, the greater the risk. Exposure can be addressed with
respect to the following components of a city: its physical infrastructure, population, economy,
and social-political system.
3.3.2.1. Physical Infrastructure Exposure
The physical infrastructure can be divided into several components, so that the exposure
and vulnerability of each can be assessed separately. Several schemes exist to categorize the
built physical environment into components with similar risk characteristics (i.e., vulnerability to
damage, value, occupancy patterns, functions, potential consequences of damage). RMS and
CUREe (1994) provides a good survey of the many available classification schemes. While all
the systems are somewhat similar, they vary depending on their intended uses, and the
techniques and resources that will be available to assess the exposure in each category. As an
example, the NIBS standardized loss estimation methodology (RMS and CUREe 1994, RMS
1997) refers to the basic components in Table 3.1.
In a general sense, the physical infrastructure exposure relates to the circumstances of a

city's development—when, where, how, how fast, and why structures have been built to serve
society's needs. Physical Infrastructure Exposure may be assessed in terms of the number, size
(e.g., square footage, height), geographical distribution, and monetary value of each
infrastructure component. The monetary value (value per square foot) depends on the cost of
labor and materials for construction and the

Table 3.1. Physical infrastructure components
Infrastructure component Sub-components
General building stock Residential, commercial, industrial, agriculture, religion/non-profit, government, education buildings
Buildings Essential facilities Hospitals, health care facilities, police and fire stations, emergency operations centers, communication
centers, planned shelters
High potential loss facilities Nuclear power plants, dams, military installations, industrial facilities
Highway systems Roads, bridges, tunnels
Railways Rail track, rail bridges, rail tunnels, rail stations, rail fuel facilities, rail dispatch facilities, rail
maintenance facilities
Light rail Light rail track, light rail bridges, light rail tunnels, light rail substations, light rail dispatch facilities,
light rail maintenance facilities
Transportation Bus Bus stations, bus fuel facilities, bus dispatch facilities, bus maintenance facilities
systems
Ports and harbors Port waterfront structures, port handling equipment, port warehouses, port fuel facilities
Ferry Ferry waterfront structures, ferry passenger terminals, ferry fuel facilities, ferry dispatch facilities, ferry
maintenance facilities
Airports Airport control towers, airport runways, airport terminal buildings, airport parking structures, airport
fuel facilities, airport maintenance and hanger facilities
Potable water supply Pipes, water treatment plants, wells, tanks, pumping plants
Waste water Pipes, waste water treatment plants, lift stations
Utilities Oil systems Pipes, refineries, pumping plants, tank farms
Natural gas systems Pipes, compressor stations
Electric power Substations, distribution circuits, power plants
Communications Central office
Source: RMS and CUREe (1994), and RMS (1997).
lifestyle of the buildings' users. The infrastructure will be more expensive in cities that give more
weight to its appearance, and the comfort and amenities it offers. A city’s physical infrastructure
exposure may be determined through a labor-intensive brute force count; by extrapolating from
censuses, tax assessors’ files, and other available databases; or by using indirect estimation
methods (Jones 1994).
3.3.2.2. Population Exposure
Population Exposure conveys the number and geographical distribution of the set of all
city residents, and of the socio-economic and other groups that exist within the city. It is only
necessary to consider separate groups of people if they exhibit different risk characteristics. It is
possible that all groups of people within a city could be equally vulnerable to earthquake losses
and equal recipients of emergency response and recovery efforts, regardless of their economic
class, ethnicity, religious affiliation, or other characteristics. Often that is not the case, however,
as the poor, ethnic minorities, the homeless, and other marginalized groups are frequently forced
to live in more hazardous areas and in more vulnerable structures, and reap a disproportionately
small share of the benefits of emergency response and recovery efforts. The elderly and the
very young can also be considered a group with distinct vulnerability characteristics, as they are
likely to be physically less resilient and less able to survive in a post-earthquake atmosphere of
hardship than others.
To assess the vulnerability of these groups separately, it is necessary to assess their

exposure separately as well. Population Exposure may also include an understanding of how
these quantities vary with the time of the day, the day of the week, and season of the year. This
factor is generally assessed by a straightforward count of the individuals in each group and in the
city as a whole, as reported in population censuses.

3.3.2.3. Economy Exposure
The economic circuit is the “continuous and timely flow of production, services, and
money between productive units (i.e., intermediate flows) and between these units and
households (i.e., final flows) within a framework of technical and economic interdependence”2
(Albala-Bertrand 1993). Economy Exposure describes the economic flow that takes place
within a city—the volume, type, origins, and destinations of goods, services, and money that are
transacted. Flow into and out of the city are considered in the Economic External Context
factor (see Section 3.3.4.1). The volume and types of economic flow within a city typically are
assessed by dividing an economy into its principal sectors (e.g., agriculture; mining; construction;
manufacturing; transportation; trade; finance, insurance, and real estate; service; and
government), and determining the total value of production in each sector and in the economy as
a whole. The sectors should be specific enough so that all the units and activity within each
exhibit similar economic vulnerability (see Section 3.3.3.3.). Input-output models describe the
interactions among sectors (see RMS 1997 for an overview). Those interactions are critical in
determining the extent and severity of economic disruption that will result from physical damage.
3.3.2.4. Social-Political System Exposure
Every city has an infrastructure of interconnected political and social institutions. Those
institutions carry out organized political, policy-making, bureaucratic, and administrative
activities related to government functions or social sectors. Government functions can be
executive, legislative, judicial, coercive, and informational. A social sector is a group of people
with some common cohesive characteristic (e.g., ethnicity, religion, social status, regional
identification) that defines the group and differentiates it from others (Albala-Bertrand 1993).
2
"Technical interdependence refers to the level of output that each of the industries of an economy should
produce to satisfy all input requirements as well as the final demands of that economy, so that no
bottlenecks will arise anywhere." "Economic interdependence, in turn, refers to the circular flow of income
and their leakages, i.e., the flow of payments and receipts between firms and households as well as public
finance and foreign payments" (Albala-Bertrand 1993).

The Social-Political System Exposure describes the number and types of those institutions, the
activities they engage in, and the complexity of the connections among them. The less formal
aspect of this factor may be described as a city’s social capital. Social capital is “the set of
human ties that a community’s members find in that place” (Kling 1992). Those ties may be
friendships, professional relationships, or an internal sense of stability or home. They provide
psychic income that is nontransferable (unlike job satisfaction) because it is attached to specific
people and is developed over time.
To clarify the difference between the two factors, Population Exposure refers to the
actual people who live in a city, while Social-Political System Exposure refers to the activities
and ties that exist among those people, and to the social and political institutions that support
those activities and ties. Social-Political System Exposure is directly proportional to the size of
the population.
3.3.3. Vulnerability factor
Vulnerability describes how easily and how severely a city's exposed entities can be
affected given a specified level of hazard. Vulnerability refers to the potential for the physical
infrastructure to be damaged or destroyed; members of the population to be injured, killed, or
left homeless, or to have their daily lives disrupted; and the economic and social-political
systems to be disrupted. Everything that is exposed must have an associated vulnerability, so
the same categories exist for the two factors.
3.3.3.1. Physical Infrastructure Vulnerability
The Physical Infrastructure Vulnerability describes the expected degree of direct

damage to the physical infrastructure, given a specified level of hazard demand. In general, the
severity of structural damage is assessed as a damage ratio, i.e., the repair cost divided by the
replacement cost, and structural vulnerability is portrayed using a damage curve, or fragility

curves. A damage curve (Fig. 3.2) depicts the expected severity of damage associated with
each level of hazard. Fragility curves (Fig. 3.3) come as a set, one curve for each damage state.
Each fragility curve shows how the probability that that damage state will occur increases with
the level of hazard.
1.00
0.75
Expected damage ratio
0.50
0.25
0.00
VI VII VIII IX X XI
Hazard level (MMI)
Figure 3.2. Schematic example of a damage curve. Based on RMS (1997).
1.00
Slight damage
Probability [damage state > specified
damage state | the level of hazard]
Moderate damage
0.75
Extensive damage
Complete damage
0.50
0.25
0.00
0 5 10 15 20 25
Hazard level (Spectral displacement in inches)

Figure 3.3. Schematic example of a set of fragility curves. Based on RMS (1997).
Two features make Physical Infrastructure Vulnerability a critical factor. First, it

determines the extent of physical damage a city will suffer, and physical damage indirectly
affects many other post-earthquake events. For example, the ability to carry out an effective
emergency response effort depends greatly on the degree of physical damage to the
transportation network, and the severity of business interruption losses can depend on the
degree of physical damage to all the lifeline networks. Second, the Physical Infrastructure
Vulnerability is one of the factors over which society has a relatively significant amount of
control. Some of the most effective mitigation strategies that have been implemented to date are
those aimed at reducing physical infrastructure vulnerability, e.g., the design and enforcement of
seismic building codes, and land use zoning.
In a general sense, the Physical Infrastructure Vulnerability depends on the

circumstances of the city's development—when, where, how, how fast, and why structures have
been built to serve society's needs, and how they have been maintained since their construction.
Although every city is unique, and the process of urban development is complex, some general
observations about the relationship between a city’s vulnerability and the history of its
development may be made. A city can develop an infrastructure with low vulnerability to
earthquake damage by chance (if the traditional construction styles happen to exhibit
earthquake-resistant characteristics), by deliberate effort, or by some combination of the two.
In the absence of good luck, a city must have the motive, means, and opportunity to create an
infrastructure with low vulnerability. That is, the city must have:
1. Motive
• Awareness of the problem of earthquake risk at the time the infrastructure is
built.
• Desire to address the problem, i.e., a perception that the risk is unacceptably
severe in the context of the other problems the city faces.

2. Means
• Knowledge of how to reduce the vulnerability. Organizations and individuals to
create, enact, and enforce effective seismic codes.
• Financial resources to pay for the code development, implementation, and
enforcement.
3. Opportunity
• Good timing. The need to construct the infrastructure must occur at a time
when the motive and means exist. The more recent the construction, the less
the deteriorating effects of aging.
• Ample physical space to accommodate the population growth so that the
infrastructure is not too dense, and there is room for lifeline redundancy.
• Ample time to construct the infrastructure in a quality-controlled way.
At a micro level, the vulnerability of the individual structures that make up a city can be
assessed by applying the principles of structural analysis. A city’s vulnerability could then be
considered the aggregation of the vulnerability of all the structures it contains. The ability of a
structure to resist earthquake forces depends on the following characteristics:
1. Structural system and construction material

The structure should be able to “undergo long-duration horizontal and vertical
shaking without excessive loss of stiffness or strength” (Rojahn 1994). For
example, in California, single-story wood-frame houses and ductile reinforced
concrete frames generally resist earthquake forces effectively. Unreinforced
masonry buildings and non-ductile concrete frames do not.
2. Configuration
The plan shape should be regular and symmetrical to minimize torsional response,
and the elevation should have no great discontinuities in stiffness, strength, or mass.
3. Structural continuity

Structural elements should be tied together well so that they remain as a unit during
shaking.
4. Changes in structural condition since construction
The effects of aging, maintenance (or lack thereof), damage in previous
earthquakes, and retrofitting can significantly diminish or improve a structure’s
original performance.
5. Proximity to other structures
A sufficient space should separate adjacent structures so that they do not pound
against one another during shaking.
6. Network redundancy
The more redundancy in lifeline infrastructure components the lesser the
consequences of physical damage to any one component of the lifeline network, and
therefore, the less the vulnerability.
The first approach addresses the city as a whole, and bases its assessment on the
processes that generally lead to lower vulnerability. It is an indirect, macro technique. Since
more is known about the general development of the city than about each structure it contains,
this first approach lends itself to relatively easy evaluation. Its main disadvantage is that shared
by all indirect, overly general measures. The occurrence of the processes described cannot fully
determine a city’s level of vulnerability. There surely will be cases in which other processes
interfere to create outcomes entirely different from those expected. The second approach
examines each structure individually as it appears, without regard for how it came to exhibit or
not exhibit the various earthquake resistant properties (by chance or by design). It is a direct,
micro technique. If all else is equal, direct approaches are always more desirable than indirect
ones, because it is more certain that they actually will represent the quantities that they intend to
represent. Often this directness comes at the price of an enormous increase in assessment
difficulty. It would be impossible to conduct a full structural analysis of every individual structure
in a city, and in its purest form, that is what the second approach demands.

The approach that loss estimation models use to assess the physical vulnerability of a
city combines aspects of each of the two methods. It finds a middle ground in resolution by
dividing the entire stock of infrastructure into components that exhibit similar levels of
earthquake resistance (see Section 3.3.2.1.), and assessing the vulnerability of each component
separately (i.e., creating damage or fragility curves for each). Although loss estimation models
aim to determine the vulnerability of areas smaller than an entire city, they could be used to
assess a city’s vulnerability by aggregating the results of each infrastructure component analysis,
weighing the vulnerability of each component by the proportion of the infrastructure that it
represents. Loss estimation models also lie somewhere between the extreme directness and
indirectness of the two approaches described above. They identify the typical characteristics of
each infrastructure component, and begin to apply the principles of structural analysis to
determine its vulnerability. When a relevant structural characteristic is unknown, the models
look to the processes that may have helped determine that characteristic. For example, if it is
not clear how well a particular type of building in a certain city will endure shaking, the models
instead would rely on information about the year in which seismic codes were adopted relative
to the average age of the building type, and about the degree to which those codes are enforced
in the city under consideration.
3.3.3.2. Population Vulnerability
Population Vulnerability describes the characteristics of individuals that make them more
or less likely to be injured, killed, displaced, or to have their daily lives disrupted as the result of
an earthquake. It also includes their ability to recover from any impact they do experience.
There can be considerable variation in the vulnerability of residents within a city, due, for
example, to their age, ethnic group, or economic status. An earthquake’s expected impact on a
city may be spread evenly over a city's entire population, or may be concentrated on a small
number of highly vulnerable people. All else being equal, a city exhibiting the latter situation
would have a higher earthquake disaster risk. This factor attempts to include the concept of
internal heterogeneity of the population. It contributes to a city’s earthquake disaster risk by

altering the distribution of risk so that certain highly vulnerable groups within a city experience an
especially great level of risk. Population Vulnerability can exist in three different forms: physical,
economic, and social.
Physical Population Vulnerability describes the physical characteristics of the city's

residents that make them more or less vulnerable. Groups that might be more vulnerable are
those who are young (less than five years old), old (sixty-five years old or older), sick, disabled,
or pregnant. These characteristics can be important due to their potential effect on a person's
ability to escape from a collapsing structure, recover from an injury, or survive in a post-
earthquake atmosphere of hardship. As noted in Blaikie et al. (1994), “studies of disaster
casualties have indicated that the young and the old are often most at risk. They are, for
example, less mobile (capable of evacuation), more dependent, have less resistance to disease,
and often command fewer resources.”
Economic Population Vulnerability highlights the fact that for residents who are
struggling financially before an earthquake, the poor or unemployed, it may be especially difficult
to recover from losses. They may be unable to afford the health care costs associated with
earthquake injuries. The financial loss they suffer will make up a greater percentage of their
income than it would for wealthier residents. The poor will have to rely more heavily on aid
because they do not have a buffer of personal savings or good credit.
Social Population Vulnerability accounts for the fact that individuals who are members
of certain marginalized groups often suffer a disproportionate share of the impact. Non-native
speakers, the homeless, certain ethnic groups, and other minorities often have limited access to
resources and political power. That limited access may translate into increased vulnerability if
those groups are relegated to more hazardous locations in a city, are left out of pre-earthquake
mitigation efforts, or receive a disproportionately small amount of emergency response and
recovery services (Blaikie et al. 1994).

3.3.3.3. Economy Vulnerability
The vulnerability of a city’s economy describes the expected severity and distribution of
the disruption in economic flows that will occur given a specified level of hazard. Economy
vulnerability relates to indirect losses, i.e., losses that arise in economic sectors not because of
direct physical damage they sustain, but because they are either forward-linked (rely on
obtaining supplies from) or backward-linked (rely on selling their output to) to sectors that do
experience direct physical damage, or because of a reduction in government, investment, or
export demands for goods and services (Cochrane 1996). While exposure and vulnerability of
the physical infrastructure largely determine the amount of direct economic losses, exposure and
vulnerability of the economy largely determine the extent of indirect economic losses. West
(1996) discusses the many types of indirect effects that may result from an earthquake.
The calculated indirect economic effects will depend on the perspective that is adopted
in conducting the analysis. First, the gains and losses that are included in the analysis will vary.
If a local government provides aid to a household, the amount constitutes a gain from the
perspective of the household. However, from a regional point-of-view, it is neither a gain or a
loss, but a redistribution of funds within the region. Second, the loss as a percentage of total
value will vary depending on the viewpoint of the analysis because each unit has a different total
wealth. A nation is much wealthier than a region, which is much wealthier than an individual
household.
Consider three possible perspectives on the indirect economic losses due to

earthquakes: (1) city (greater metropolitan area), (2) nation, and (3) individual households,
companies, sectors, or neighborhoods. Experience suggests that, from the standpoint of a
greater metropolitan area, the indirect losses following earthquakes are negligible (RMS 1997,
Munroe 1992). Earthquakes generally do not result in a significant, long-lasting effect on the
performance of the economy of a greater metropolitan area as a whole (i.e., its total output)
(Albala-Bertrand 1993). The outside assistance that enters a region following an earthquake

seems to offset the indirect losses that the region suffers. While there are several examples of a
regional or national economic depression following a natural disaster (e.g., Managua, Nicaragua
earthquake of 1972), Cochrane (1992) argues that if one compares the losses with and
without the earthquake, rather than before and after the earthquake, it can be seen that
“disasters tend to accelerate ongoing economic and social processes that were working prior to
the event.” They do not cause robust economies to collapse.
If economic effects are considered from a national perspective, then the negative
national implications of the federal assistance that comes into the affected city must be
considered. The cost of providing federal disaster relief is passed on either to another region of
the country or to a future generation by cutting lower priority programs, raising taxes, or
increasing the federal debt. RMS (1997) suggests that the negative effects of federal assistance
on other regions or generations roughly balance the positive effects for the recipient city. In that
case, the absolute value of indirect losses would be greater when viewed from a national rather
than a regional standpoint, because outside aid cannot be considered as a component that
offsets the indirect losses. Still, the total indirect losses would be a much smaller percentage of
the national economy than of the regional economy.
From the perspective of individual households, companies, sectors, and local

geographical areas, there will be winners and losers. Suppose a store is damaged and therefore
is unable to sell its product for six months, that store will suffer serious economic losses, but an
undamaged store down the street may experience economic gains by taking over business that
previously belonged to the damaged store. The vulnerability of a productive unit is determined
by the unit's profits, solvency, liquidity, stocks, client power, market access, business
connections, and political influence (Albala-Bertrand 1993). In general, size is a good indicator
of the vulnerability of a productive unit with larger units being less vulnerable.
Since the EDRI is focused on the greater metropolitan area as a unit of study, the
Economy Vulnerability factor should adopt the same perspective. It considers the potential for

disruption to the economic flows within the metropolitan area, including aid that comes into the
area as an mitigating factor. The Economic External Context factor considers any indirect
economic losses that are suffered outside the metropolitan area, including the negative effect of
offering federal assistance on other regions of the country. Even if the economic effects seem
negligible for the metropolitan area as a whole, however, an earthquake may cause devastating
indirect economic losses to individual households, companies, sectors, and neighborhoods.
Since the EDRI acknowledges not only the total economic losses in the metropolitan area, but
also the distribution of those losses within the area (see Fig. 2.1), a city in which certain
individual units (i.e., households, companies, sectors, or neighborhoods) suffer severe economic
losses would be considered to have a higher risk than one suffering the same amount of loss, but
evenly distributed among units.
Direct effects on labor and capital stock (i.e., inventories and fixed assets) can initiate
disruptions in local economic flows and impairment of productive processes. A disrupted labor
force and damaged production facilities or inventories can cause a loss of sectoral function (i.e.,
a reduction in the output of a sector). Loss of sectoral function can, in turn, lead to a shortage
of supplies for all forward-linked sectors, a reduction in demand for inputs from all backward-
linked sectors, and/or a reduction in government, investment, or export demands for goods and
services (Cochrane 1996). The vulnerability of a city’s economy depends on (1) the pattern of
direct physical damage in the city, (2) the internal and external capacity of its component
sectors, (3) the structure of the economy, and (4) the amount of outside aid that will be available
for emergency response, recovery, and reconstruction. The remainder of the section presents a
discussion of each of these four items in turn.
The Physical Infrastructure Vulnerability and Exposure factors determine the expected
severity and distribution of physical damage to inventories, facilities, and lifelines in the city.
That damage pattern, in turn, determines the amount of money necessary to repair the damage,
and the degree and duration of reduction in the output of each sector. West (1996) refers to
studies by Brady and Perkins (1991) and Cochrane (1995) in arguing that, for indirect losses,

“what is damaged is more important than how much is damaged.” If the damage is evenly
distributed among sectors, indirect effects will be less than if one or two sectors suffer a
disproportionate amount. According to Cochrane (1996), a uniform damage pattern (i.e., when
each sector suffers an equal damage ratio) causes the economy to shrink proportionately, so
that forward- and backward-linked sector losses disappear, and the economy remains
balanced. Damage to the city's lifeline infrastructure (e.g., transportation, communications) can
disrupt economic flows and affect the productivity of many sectors simultaneously.
Economic disruptions take place within the context of a dynamic economic system.
Indirect effects do not propagate through the economy unchecked. As soon as the physical
damage occurs, each sector will begin trying to restore the economy’s equilibrium by utilizing
available excess capacity. The amount of capacity in a sector describes the ability of that sector
to absorb the direct effects of an earthquake, thereby minimizing indirect losses to forward- and
backward-linked sectors. Capacity may be internal (i.e., originating within the city) or external
(i.e., originating outside the city). If direct damage to facilities and inventories causes a shortage
of supplies to forward-linked sectors, forms of capacity that might help reduce the indirect
losses to those forward-linked sectors include unemployed labor, idle inventory within the
region, and supplemental import sources. If direct damage causes a reduction in demand for
inputs from backward-linked sectors, forms of capacity that might help reduce the indirect
losses to those backward-linked sectors include the ability to accumulate inventory for future
sale (requiring that the goods be non-perishable, and that storage space exists), and new export
markets. The degree of capacity may depend on post-event household spending patterns, the
mobility of the workforce, the elasticity of supplies from the construction industry, and the
potential for product substitutions due to relative price changes (RMS 1997).
Given a certain level of disruption within each economic sector, the network of flows of
money, services, and goods among sectors will cause the effects of that disruption to propagate
through the economy. The economic structure, i.e., the nature of inter-industry flows and the
relative importance of the sectors, will determine the magnitude and distribution of indirect

effects within the economy. According to Jones (1992), some sectors of the economy are more
vulnerable to disruption than others, and if the more vulnerable sectors control a larger
proportion of the economic flows, the economy as a whole is more vulnerable. Specifically,
Jones (1992) argues that while “extractive industries are little dependent on linkages and
interaction at low levels of development,” “fabricative sectors are fairly robust and show
remarkable ability to recover. Distributive sectors are heavily dependent on lifelines and
infrastructure and are likely to be heavily impacted. Services are highly dependent on
transportation and communication and are perhaps the most vulnerable of all” (Jones 1992).
Since economies with a high per capita Gross Domestic Product (GDP) are dominated by the
service and distribution industries, they are more vulnerable to natural disasters than economies
with a low per capita GDP, in which the extractive and fabricative sectors have larger shares.
There is disagreement about how the complexity of a city’s economic network affects
the severity and distribution of losses it will suffer. A more extensive, interconnected network
might allow the losses to spread further, affecting more people and increasing the risk, as Jones
(1992) suggests, but it also might indicate more capacity for substitution of suppliers and buyers,
thereby reducing the vulnerability of the economy. As Cheng notes, “to ignore the benefits
provided by the economic linkages, including a greater capacity to deal with disasters, and to
view them primarily as the nodes of a network by which diseases are spread would be
theoretically wrong and very misleading.” (Cheng 1992).
Finally, considering economic vulnerability from the perspective of a city, the amount of
indirect losses depends on the percentage of spending on recovery and reconstruction that
come from outside aid and payments on insurance claims. Those outside sources can
significantly offset any indirect losses that are incurred.
3.3.3.4. Social-Political System Vulnerability

An earthquake disaster may disrupt individual institutions and organizations within a
city's social and political structure, and/or the connections among them. Institutions may
experience (1) functional deficiencies due to the damaged physical infrastructure and disrupted
work force, and (2) new demands that emerge as part of the response, recovery, and
reconstruction efforts (Albala-Bertrand 1993). The infrastructure that connects the many social
and political institutions may experience instability or sectoralization. This factor captures the
characteristics of a city's social-political infrastructure that determine the likelihood, severity, and
extent of disruption that will occur given a specified level of hazard. The Nicaragua earthquake
of 1972 provides a well-known example of the way in which the social-political situation can
affect the earthquake impact, and vice versa (Albala-Bertrand 1993). In general, the social-
political vulnerability depends on the type of political system, the strength and complexity of
linkages among local institutions and organizations, the tendency of groups to act centrifugally,
and the ability of in-built mechanisms (e.g., self-help, family-centered activities) to counteract
deficiencies in the infrastructure by relieving stress on formal institutions (Albala-Bertrand 1993).
3.3.4. External Context factor
In today's global community, major cities are increasingly interconnected. No city is an

island. Neither the factors that contribute to a city's risk, nor the consequences of an
earthquake disaster are confined within a city's borders. External Context is included to
describe how damage to a city affects those outside the city. It incorporates the reality that,
depending on a city's prominence with respect to economics, transportation, politics, and
culture, damage to certain cities may have more far-reaching effects than damage to others. A
city’s External Context rating indicates the degree to which economic losses, and disruption to
transportation networks, political processes, and social lives will extend beyond its borders.
The ripple effect may both (1) increase the total impact by creating additional economic losses

or disruption that would not occur if the city was completely isolated, and (2) redistribute any
impact that may otherwise have been confined within the city to regions outside the city.
Increasing the total impact would increase the overall risk, but spreading out losses that
otherwise would be concentrated within the city, may reduce the overall risk.
This factor may be divided into the external context that depicts the city’s relationship to
its country, and the external context that depicts the city’s relationship to the world. The two do
not necessarily convey the same information. A city may be important to its country, but not
very important to world affairs, relative to other cities worldwide. Santiago, Chile, for example,
is home to one-third of Chile’s population, and serves as the country’s political, economic, and
cultural center, but relative to Tokyo, New York, Paris, London, and other megacities, it does
not play a large role in global affairs. External context as it relates to both city-country
relationships and city-world relationships, consists of four main components: economics,
transportation, politics, and culture.
3.3.4.1. Economic External Context
Economic External Context describes the economic ripple effect of an earthquake

disaster. It depends on the number and strength of economic linkages between the city and
those outside city. This factor is related to, but not the same as total wealth. A city could be
extremely wealthy, but isolationist, so damage to it would not affect outsiders significantly. The
Economic External Context can be conceived as similar to a balance of payments account for a
city, i.e., a record of all transactions between a city's residents and the rest of the world. For a
more detailed understanding, consider the transactions included in a balance of payments:
merchandise trade, transportation service, tourist services, business, and professional services,
unilateral transfers (donations), direct investment, long-term portfolio investment, and short-term

capital flows. The need for and existence of alternatives for the particular types of imports and
exports concerned are also important.
3.3.4.2. Transportation External Context
Transportation External Context considers how many people travel to, from, and
through the city. Transportation might include the transport of goods as well as people, but the
movement of goods is accounted for already through the Economic External Context factor, so
only passengers are included in this component. All forms of transport should be considered:
road, railroad, air, and water. For the last three, the factor could focus on the rail terminals,
airports, and ports, assuming that all inter-city travelers would pass through the main facility.
For roads, there is no central facility that all travelers must use, so the amount of inter-city traffic
might be conceived of as the passenger-miles of trips that originate or end in the city.
3.3.4.3. Political External Context
Political External Context conveys the degree to which people outside a damaged city
are affected politically. As Albala-Bertrand (1993) notes, “disaster impacts on capital cities
(i.e., where both the central administration is located and a high proportion of the population
live, like Managua or Guatemala City) are institutionally more disruptive and pressing than
events in the countryside or secondary towns.” The factor depends on how many people are
affected and how significantly (i.e., how soon and in what way). Political functions are generally
concentrated in the capital of each province (or state) and country. A disruption in performance
of those political functions could affect people outside the city in two ways:
1. Domestic political functions

All the people governed by a particular capital could be affected by damage to the
capital city if government officials are disrupted in performance of their jobs. The
effects on constituents would vary between countries, depending on the form of the
government. In the U.S., for example, citizens could be affected, especially in the
long-term, through a slowing or temporary halt in program funding, tax collection,
and provision of public services. The extent of impact would depend on the
population of the country (or province) of which the city is capital, and the degree of
control and centralization of the government.
2. Foreign political functions

Other cities and countries could be affected by the disruption of their relations with
the impacted city's government. The extent of impact would depend on the degree
of political relations between the country of the impacted city and others, and how
much power the country has in those relations.
Note that the nominal political capital does not always retain all of a country’s political
power. Another city that was the country’s capital in previous years, or a city that serves as the
country’s economic and cultural center may hold some political power as well, either because
political functions are physically located there, or because it sets the political agenda. For
example, Ankara is currently the capital of Turkey, but because Istanbul was the capital of the
region until 1923 and still is the economic and cultural center of the country, it retains a degree
of the country’s political power as well.
3.3.4.4. Cultural External Context
A city’s Cultural External Context describes the degree to which people outside the city
rely on the city’s cultural resources. The cultural resources includes the stock of historic and
cultural assets (monuments, human artifacts, art) that “are important in providing identity and
continuity to a society’s culture,” and the social capital, i.e., intangible assets like human

relationship networks and individual’s sense of place (Kling 1992). The value of historic and
cultural assets lies in their ability “(1) to give a sense of place in history and an identity shared by
other people, times, and places, or (2) to provide examples of great aesthetic traditions.” They
are unique, irreplaceable, and enjoyed jointly and freely (Kling 1992). As discussed in Section
3.3.2.4., social capital is “the set of human ties that a community’s members find in that place”
(Kling 1992). Those ties may be friendships, professional relationships, or an internal sense of
stability or home. They provide psychic income that is nontransferable (unlike job satisfaction)
because it is attached to specific people and is developed over time.
Clearly every city has cultural significance, and it would be impossible to rank the
relative amount in different cities. This factor is included for completeness, to recognize that
cities are not only economic and political machines. They also have significance in other less
tangible, but equally important ways.
3.3.5. Emergency Response and Recovery Capability factor
Emergency Response and Recovery Capability describes how effectively and efficiently
a city can respond to and recover from short- and long-term impact through formal, organized
activities that are performed either after the earthquake, or before the earthquake, but with the
primary purpose of improving post-earthquake activities (e.g., planning). It is assumed that any
other actions taken before the earthquake (e.g., retrofitting) are significant only if they changed
the current snapshot of the city, in which case they are accounted for in other factors. Note also
that the Emergency Response and Recovery Capability factor does not include some less visible
response activities that happen after an earthquake, e.g., the in-built mechanisms that counteract
economic effects, like utilizing excess capacity (see Section 3.3.3.3.). Although those activities
help the city reduce and recover from impact, they occur without a deliberate, formal, organized
effort. They are so automatic and integrated into the economic, social, and political systems that
they are considered to be characteristics of those systems, and therefore, are included in the

various components of the Vulnerability factor. Following are twelve aspects of an effective
emergency response and recovery effort that are included in this factor. Appendix C describes
each in more detail.
1. Emergency management system

There must be an emergency management system to coordinate the response effort.
The system must manage problems associated with convergence, disputes over
access to the disaster site, and disputes over authority. It must also determine what
aid is needed, channel requests upward, and distribute assistance when it arrives.
2. Communications
The procedures and physical equipment must be ready for inter-organizational
communication (e.g., among government agencies, relief organizations, and utility
companies), inter-jurisdictional communication, and communication between the
public and officials. Each party should know who to talk to, what information to
convey and to receive, and what form of communication to use.
3. Financial arrangements
The financial arrangements should be in place to facilitate timely assistance to
households, businesses, and local governments for relief and reconstruction.
4. Legislation
Pre-disaster legislation should establish channels to request and receive assistance
efficiently so that emergency response and recovery efforts are not hindered by legal
delays.
5. Damage assessment

The procedures and equipment should be available to determine, in almost real time,
the type, severity, and location of damage and human losses, and to use that
information to identify and prioritize the emergency response and recovery tasks.
6. Search-and-rescue
The effort should be able to locate and rescue survivors trapped in collapsed
structures.
7. Secondary hazard control

The procedures and equipment should be available to control secondary hazards.
This may include fire prevention and suppression, removal of hazardous debris,
sanitation control, and security from looting.
8. Health care
Health care teams must be provide on-site first aid, transport the injured to health
care facilities as necessary, and treat patients at those facilities with the greatest
speed and highest quality possible.
9. Mass care
The effort must identify the type and quantity of basic needs required, determine
where they are needed, and provide them as necessary. Basic needs include
emergency food and water, water containers, blankets, cooking utensils, and
clothes.
10. Shelter

The effort should determine how many people need shelter and for how long, what
kind of shelter is appropriate for them, and how to provide it.
11. Clean-up
Undamaged items must be returned to their correct positions, damaged items should
be repaired, and destroyed items must be removed. Clean-up includes the task of
locating and burying the dead.
12. Restoration of services

Interrupted utility service must be restored, and damaged or blocked transportation
routes must be reopened.
Experience has demonstrated that certain aspects of a response and recovery effort are
important to its overall effectiveness. To estimate the effectiveness of a city's response and
recovery then, one can determine how well the effort achieves those important elements relative
to other cities. A city’s Emergency Response and Recovery Capability depend on:
1. Pre-earthquake organizational and operational planning

2. Resources available post-earthquake: Financial, equipment and facilities, and
trained manpower.
3. Mobility and access post-earthquake
Each of the three components should be measured with respect to the twelve main aspects of
the Emergency Response and Recovery Capability factor (see above). Following are
discussions of each of these three components.
3.3.5.1. Planning

The planning factor describes the number and quality of plans and procedures
developed before the earthquake to guide the twelve main aspects of emergency response and
recovery. Plans should be organizational, establishing the roles and responsibilities of each
involved party, and operational, creating the procedures that explain what should be done, how,
and when. A city’s historical experience with earthquakes (or lack thereof), as well as its
general cultural and economic setting may influence the degree of planning since a society must
have both the desire and the means to plan.
3.3.5.2. Resources Available Post-earthquake
Resources come in the form of money, equipment and facilities, and trained manpower.
The term available means that the city either has the resources on hand or can obtain them
from another source for use immediately following an event. The availability will depend on
what is available before the earthquake and the probability that it still will be available after the
earthquake, which in turn is a function of the level and geographical extent of physical damage.
The resources themselves may be damaged, or the ability to bring them to those in need may be
hindered because of damage to the transportation network. In identifying resources, it is helpful
to think of all the possible suppliers of the resources, and each of the twelve main response
tasks.
Money available for recovery and relief may come from local, state, or national
governments, insurance coverage for individual homeowners or businesses, residents' personal
savings, businesses, relief organizations like the Red Cross, or foreign sources (bilateral aid,
multilateral aid, or non-governmental organizations). Albala-Bertrand (1993) estimates that
domestic disaster assistance (especially local) is generally about twice as much foreign disaster
assistance, but the ratio may vary from region to region. Equipment and facilities necessary for
response includes equipment for emergency communications (e.g., emergency generators),
damage assessment (e.g., loss estimation computer models), search and rescue (e.g., the “jaws
of life”), secondary hazard control (e.g., fire trucks), health care (e.g., hospital beds), mass care

(e.g., food, water, blankets), temporary shelter (e.g., mobile homes), and clean-up (e.g., earth-
moving trucks). Possible trained manpower resources also can be identified through their role in
each of the twelve main aspects of response. For example, manpower resources could include
search-and-rescue teams, physicians, firefighters, public volunteers, and military personnel from
a nearby base.
3.3.5.3. Mobility and Access Post-earthquake
The ability to maneuver within and gain access to a city after an earthquake depend on
the condition of the post-earthquake transportation system, amount of debris, city layout, city
topography, remoteness, and weather conditions. The condition of the transportation system,
and the amount of debris are both determined by the severity of direct physical damage, and are
already incorporated into the EDRI through the Physical Infrastructure Vulnerability and
Exposure factors.

Chapter Four
Indicator Selection
4. Introduction
Once the factors have been identified, a set of simple, measurable, scalar indicators
must be selected to represent each in the Earthquake Disaster Risk Index. The meaningfulness
of the EDRI hinges on the ability of the indicators to represent the ideas of the conceptual
framework. Clearly there are no simple indicators that can convey the ideas of the framework
completely. Information necessarily will be lost in the move from complex, conceptual factors
to simple, measurable indicators. It is the unavoidable price of operationalization, but a price
that is willingly paid to enable the quantitative comparison that the EDRI aims to provide. The
process of indicator selection therefore, is an exercise in making tradeoffs between
comprehensiveness and simplicity. The challenge is to minimize the loss of information so that
together the indicators provide a meaningful measure of earthquake disaster risk.
This chapter begins with a list of the criteria that guide the selection of indicators.
Section 4.2 discusses the strategy adopted in the indicator selection process. Sections 4.3.1 to
4.3.5 describe the execution of the process for each of the five main factors. Each section
presents the indicators selected in the sample analysis to represent the associated factor,
explains why those indicators were chosen, discusses alternative indicators that were considered
and the reasons that they were not selected, and lists the aspects of the factor that the indicators
fail to represent.
Chapter Four: Indicator Selection 63

4.1 Criteria for indicator selection
The tasks of defining the factors and creating the conceptual framework are carried out
as if the framework is the ultimate goal, without concern for the issues of measurability and data
availability. These constraints enter the analysis as indicators are chosen to represent the
framework factors. Since there are no indicators that can be proven to be the “correct” ones, it
is helpful to list the criteria that an ideal indicator would satisfy, then judge the possible indicators
on the basis of their ability to meet them. Rossi and Gilmartin (1980) discuss many qualities
with which to assess and compare social indicators. Based on their ideas and the particular
needs of the EDRI, the criteria that should guide the indicator selection process are:
1. Validity
The indicators should represent the concepts they purport to represent. Unless a
convincing argument demonstrates that the proposed indicators accurately represent
the factors for which they are proxies, the EDRI will not be meaningful or useful.
2. Data availability and quality

The indicators should be measured by data that is reliable, available in a consistent
form for all the world's major cities, and relatively easy to collect. This criterion is
difficult to satisfy. The availability and quality of data vary significantly from city to
city, indicator to indicator, and year to year. Data for many indicators that might be
included in the EDRI have not been compiled into a consistent form from local
sources, or have not been collected at all. Most international comparative data is
available only for countries, not for metropolitan areas.
3. Quantitativeness and objectivity
64 Chapter Four: Indicator Selection

The EDRI aims to be transparent and replicable to earn greater credibility and to
provide greater insight into the composition of risk. By making the indicators that
comprise the EDRI quantitative and objective whenever possible, the EDRI can be
more easily evaluated and understood. If all else is equal then, indicators that are
evaluated with directly available quantitative data are preferable to those that rely on
subjective assessments.
4. Understandability
Indicators should be intuitively understandable, using familiar concepts and
quantities when possible.
5. Directness
An indicator is direct if it measures the concept of interest itself, indirect if it
measures “some other variable that is assumed (based on experience or theory) to
be closely related to the variable of interest.” “Direct and indirect indicators differ in
the extent to which a change in the indicator matches a corresponding change in the
aspect of society being considered. A direct indicator will tend to vary more
sensitively, changing its value to reflect changes in the value of the variable being
measured. An indirect indicator, on the other hand, can change without the
occurrence of corresponding change in the variable of concern (and vice versa).”
“Because of other possible influences, an indirect indicator will not be as valid and
accurate as a direct one. Thus, direct indicators are always preferable if they are
available” (Rossi and Gilmartin 1980). Unfortunately, data are often more easily
available for indirect indicators than for direct ones.
4.2. Strategy

The indicator selection process originates with the conceptually complete framework
factors. As the constraints of quantitativeness, objectivity, measurability, and data availability
are incorporated, the list of possible indicators gradually shrinks until the final set that will be
used in the EDRI is determined. Proceeding gradually from the framework factors to the
indicators that are the best possible under the true conditions of the current time offers two
benefits. First, the process clarifies the conceptual rationale that supports the indicator
selection. Second, it suggests ways in which the data availability and quality should be
improved, and ways in which an EDRI might be developed if more data gathering resources or
better access to data were available.
Indicators may be identified using a microscopic approach or a macroscopic approach.

For example, in identifying indicators to represent the physical infrastructure vulnerability, the
former would rely on assessing the vulnerability of each type of infrastructure component
separately and aggregating the results, while the latter would look to the general trends of the
city’s development and what they indicate about the vulnerability of the infrastructure for the city
as a whole. The microscopic approach, involves disaggregating a factor into its smallest
components, attempting to represent each with separate indicator, and combining the many
indicators to represent the entire factor. Compared to the macro approach, it generally uses
more direct indicators and focuses on precisely the specific concepts that the factor intends to
represent, but it requires more detailed analysis and data that is more difficult, if not impossible
to obtain. The macroscopic approach, which identifies indicators that represent the factor as a
whole, generally has less demanding data requirements, but often results in an indirect measure
that suffers from the associated disadvantages of indirect measurement (see Section 4.1,
Criterion 5).
The number of indicators that should be used to represent a factor is not predetermined,
nor is it necessarily the same for every factor. There are advantages and disadvantages
associated with using more or fewer indicators to represent a factor. Using more indicators

provides a more complete representation of the factor and makes the composite EDRI less
sensitive to any one indicator, but it requires more data and complicates the index. While the
Competitiveness Index includes over 330 indicators, the Human Development Index includes
only three. A balance must be found between completeness and simplicity so that a sufficient
number of indicators are included to represent the factor meaningfully, but no more.
Note that some indicators in the sample analysis are used more than once, to represent
more than one concept. For example, “per capita GDP” helps to represent three different
factors—economic exposure, physical infrastructure exposure, and financial resources available
for emergency response and recovery. In fact, an indicator may increase the earthquake
disaster risk in one use and decrease the risk in another. A higher “per capita GDP” indicates a
larger exposure, and therefore a higher risk; but it also indicates more financial resources for
recovery, and therefore a lower risk. Although multiple uses suggests that an indicator is
broader or more inclusive than would be ideal, it is not incorrect to use an indicator more than
once. Care must be taken, however, to ensure that the weight associated with the indicator
reflects only the relative importance of the concept it represents in the associated use. That is,
the weight associated with “per capita GDP” in the Exposure factor should reflect only the effect
that the “per capita GDP” has on the exposure, without taking into account its affect on the
recovery at all.
Figure 4.1 presents the indicators that were selected in the sample analysis. The
indicators are listed in the rightmost column, the five main factors they represent in the leftmost
column, and the factor components in the middle. The following five subsections (4.3.1 to
4.3.5) discuss the indicator selection process as it applies to the five main factors. For each of
these subsections, the discussion presents a list of the indicators selected to represent the
associated factor, and describes the rationale behind their selection, the assumptions on which
they rest, the procedure to calculate them, alternative indicators that were considered, and the

aspects of the factor that are omitted from the indicator set. The sensitivity of the results to the
indicator selection and definition is discussed in Chapter 7.
FIVE MAIN FACTOR COMPONENTS INDICATORS

FACTORS
XH1 : exp(MMI w/50-year return period)
Ground Shaking XH2 : exp(MMI w/500-year return period)
XH3 : Percent of urbanized area w/soft soil
Hazard XH4 : Percent of urbanized area with high
liquefaction susceptibility
Collateral Hazards XH5 : Percent of buildings that are wood
XH6 : Population density
XH7 : Tsunami potential indicator
XE1 : Population
Physical Infrastructure XE2 : Per capita GDP, constant 1990 US$
Exposure XE3 : Number of housing units
XE4 : Urbanized land area
Exposure
Population Exposure XE5 : Population
Economy Exposure XE6 : Per capita GDP, constant 1990 US$
XV1 : Seismic code indicator

XV2 : City wealth indicator
Physical Infrastructure
XV3 : City age indicator
Vulnerability
XV4 : Population density
Vulnerability
XV5 : City development speed indicator
Population Vulnerability XV6 : Percent of pop’n. aged 0-4 or 65+
Economic External Context XC1 : Economic context indicator

External Context
XC2: Political country context indicator
Political External Context XC3: Political world context indicator
Planning X R1: Planning indicator
X R2: Per capita GDP, constant 1990 US$

X R3: Ten-year average of annual real
Emergency growth in per capita GDP
Response and Resources X R4: Housing vacancy rate
Recovery Planning X R5: Num. hospitals per 100,000 people
X R6: Num. physicians per 100,000 people
X R7: Extreme weather indicator
Mobility and Access X R8: Population density
X R9: City layout indicator
Figure 4.1. Sample analysis indicators
4.3.1. Hazard factor

The seven indicators listed below are used to represent the Hazard factor. The first
three represent the ground shaking hazard, and the last four represent the collateral hazards of
liquefaction, fire following earthquake, and tsunami:
1. XH1: e(MMI with a fifty-year return period)

2. XH2: e(MMI with a fi ve-hundred-year return period)
3. XH3: Percentage of urbanized area with soft soil
4. XH4: Percentage of urbanized area with high liquefaction susceptibility
5. XH5: Percentage of buildings that are wood
6. XH6: Population density (people per square kilometer)
7. XH7: Tsunami potential indicator
Just as the Hazard factor was defined in Chapter 3 by considering ground shaking, then each of
the collateral hazards in turn, so will the discussion of the indicator selection be divided into the
same subsections—ground shaking, liquefaction, fire following earthquake, and tsunami. The
landslide hazard is not included in the sample analysis indicator set because no appropriate
indicators could be identified to represent it. One possible measurable landslide indicator might
be “percentage of urbanized area with high landslide potential (i.e., slopes steeper than a certain
grade and made of soil without adeqeuate cohesiveness).” Another might be derived from the
number and size of historical landslides. The data are not available for either of these
possibilities, however, so no landslide indicators are included.
4.3.1.1. Ground Shaking
As discussed in Section 3.3.1.1, the ground shaking hazard depends on (1) the
locations, recurrence relationships, and maximum possible magnitudes associated with the faults
and areas sources that might produce the ground shaking; (2) the attenuation relationships that
are associated with soil and topography and describe the way in which ground motion changes

in amplitude and frequency content as it travels from the source to the site; and (3) the ability of
the soil at the site to amplify or reduce the ground shaking. Not only do all of these
characteristics need to be included to create a complete representation of ground shaking
hazard, but their interactions must be acknowledged too.
Probabilistic Seismic Hazard Analysis (PSHA) is a technique that integrates these

components to produce a summary expression of the level of bedrock ground shaking hazard,
i.e., the probabilistically expected magnitude of ground shaking at a given site, in a specified
future time period. Using the output of the PSHA ensures that ground shaking is represented as
completely as possible. Since the typical PSHA output is a map showing contours of equal
hazard with a hazard curve of return period versus ground shaking for each (Fig. 4.2), one or
more scalar quantities must be derived from that output to serve as indicators in the EDRI. Two
points are selected from the hazard curve to capture the influence of both events with a high rate
of occurrence and small magnitude, and events with a low frequency and large magnitude.
According to the Poisson model of earthquake occurrence, these two points—the MMI with a
fifty-year return period, and the MMI with a five-hundred-year return period—correspond to
the levels of ground shaking with a ten percent chance of being exceeded in five and fifty years,
respectively.1
Because earthquake magnitudes are defined as logarithmic functions (e.g., Richter

magnitude = logA- logAo, where A is the maximum amplitude of a seismogram measured in
micrometers, and Ao is the zero amplitude), roughly speaking, the amplitude of ground shaking
varies as e(earthquake magnitude). The difference in ground shaking between earthquakes of
magnitudes eight and nine is much more severe than the difference between earthquakes of
magnitudes two and three. Therefore, e(MMI with a fifty-year return period)
1
According to the Poisson model, P = [probability of occurrence of ground shaking at least as severe as the
specified level, within t years] = 1 - e-(t/RP), where RP is the return period associated with the specified level of
ground shaking. Therefore, if P=0.1 and t=5 years, then RP≈50 years; and if P=0.1 and t=50 years, then
RP≈500 years.

and e(MMI with a five-hundred-year return period) (XH1 and XH2) are better indicators of the amplitude of
ground shaking than the MMI values themselves. To evaluate e(MMI with a fifty-year return period), a
single, average value is estimated for the entire city from available, published PSHA results;
likewise for e(MMI with a five-hundred-year return period).
Return period [years]
1000
500
100
50
10
1
MMI50 MMI 500
Ground shaking [MMI]
Figure 4.2. Schematic example of a hazard curve
By building on PSHA, the two bedrock ground shaking indicators provide a relatively
complete representation of a city’s bedrock ground shaking hazard. The cost of directly using
the output of PSHA studies to evaluate the indicators is that the results then depend on the
quality and consistency of those studies. If the MMI values are taken from analyses conducted
by different researchers, the methods and data may not be consistent across studies. It can also
be argued that PSHA is too complicated for use in the EDRI, which aims to be simple to
evaluate and interpret. However, the relative sophistication of XH1 and XH2 can be justified
because it is far easier to find the results of a PSHA for every major city, than to determine all
the pieces of the analysis (e.g., recurrence and attenuation relationships) from scratch, and the
representation of ground shaking hazard would be incomplete if some aspects of the PSHA
were omitted.
The indicator “percentage of urbanized area with soft soil” (XH3) represents the extent
to which soil conditions are likely to increase the expected bedrock ground shaking in the city.

While it would be useful to know the distribution of each soil type with respect to the
infrastructure and the expected bedrock ground shaking throughout a city, that information is
too unwieldy for direct inclusion in the EDRI. Instead, the “percentage of urbanized area with
soft soil” conveys only the information that is most critical in comparing the soil conditions of
different cities. “Soft soil” includes soils of type E or F (Table 4.1), as used in Hazus97, the
NIBS standardized earthquake loss estimation methodology (RMS 1997). It refers to those
soils that will amplify the ground shaking most severely. The indicator XH3 considers only the
“urbanized area,” i.e., the built-up area within the city’s boundaries, because soft soil will only
increase the city’s earthquake disaster risk if it supports a structure.
Table 4.1. Soil classifications (from the 1997 NEHRP Provisions)

Shear wave
Name General description velocity (m/s)
Minimum Maximum
A HARD ROCK
Eastern United States sites only. 1500
B ROCK 760 1500
VERY DENSE SOIL AND SOFT ROCK
C Undrained shear strength us ≥ 2000 psf (us ≥ 100 kPa) or N 360 760
≥ 50 blows/ft.
STIFF SOILS
D Stiff soil with undrained shear strength 1000 psf ≤ us ≤ 2000 180 360
psf (50 kPa ≤ us ≤ 100 kPa) or 15 ≤ N ≤ blows/ft.
SOFT SOILS 180
Profile with more than 10 ft. (3 m.) of soft clay defined as
E soil with plasticity index PI > 20, moisture content
w > 40% and undrained shear strength us < 1000 psf
(50 kPa) (N < 15 blows/ft.)
SOILS REQUIRING SITE SPECIFIC EVALUATIONS
1. Soils vulnerable to potential failure or collapse under
seismic loading, e.g., liquefiable soils, quick & highly
sensitive clays, collapsible weakly cemented soils.
2. Peats and/or highly organic clays (10 ft. (3 m.) or thicker
F layer).
3. Very high plasticity clays (25 ft. (8 m.) or thicker layer
with plasticity index > 75).
4. Very thick soft/medium stiff clays (120 ft. (36 m.) or

thicker layer).
Source: Reprinted from RMS (1997).
For a given city, the value of “percentage of urbanized area with soft soil” can be
estimated by superimposing the boundaries of the city’s urbanized area onto a map that depicts
the distribution of soft soil throughout the city, perhaps using Geographic Information Systems
(GIS) software. If a soil map portraying areas of soft soil (as it is defined in Table 4.1) is not
available, it may be necessary to determine the soft soil areas directly from a geological map of
the city, which should almost always be available.
The EDRI ground shaking hazard indicators do not convey any information about
topographic or directivity effects, or about the relative distributions of soft soil and bedrock
ground shaking throughout a city. Although MMI values were used in the sample analysis, it
would be better to use spectral acceleration or peak ground acceleration values if the PSHA
results were available for all cities.
4.3.1.2. Liquefaction
The liquefaction indicator (XH4) represents an approach similar to that of the soft soil
indicator. Considering the percentage of the city’s urbanized area that exhibits the
characteristics associated with the highest risk (i.e., soft soil or high liquefaction susceptibility)
conveys the most critical information in a single, simple value. “High liquefaction susceptibility”
refers to the soil deposits that would be classified as having either high or very high liquefaction
susceptibility in the NIBS standardized loss estimation methodology (according to Youd and
Perkins 1978; and Liao, Veneziano, and Whitman 1988). Table 4.2 indicates the Youd and
Perkins liquefaction susceptibility classifications for a variety of soil types. Table 4.3 defines
each liquefaction susceptibility category in terms of the associated probability of liquefaction
given a certain peak ground acceleration. As with the soft soil indicator, the liquefaction
indicator considers only the “urbanized area,” i.e., the built-up area within the city’s boundaries,

because liquefiable soil will only increase the city’s earthquake disaster risk if it supports a
structure or lifeline.
Table 4.2. Liquefaction susceptibility of sedimentary deposits
(from Youd and Perkins 1978)
Distribution of Likelihood that cohesionless sediments when saturated
cohesionless would be susceptible to liquefaction
Type of deposit (by age of deposit)
sediments in <500 yr. Holocene Pleistocene Pre-
deposits Modern <11 ka 11 ka-2 Ma Pleistocene
> 2Ma
(a) Continental deposits
River channel Locally variable Very high High Low Very low
Flood plain Locally variable High Moderate Low Very low
Alluvial fan and Widespread Moderate Low Low Very low
plain
Marine terraces Widespread --- Low Very low Very low
and plains
Delta & fan-delta Widespread High Moderate Low Very low
Lacustrine, playa Variable High Moderate Low Very low
Colluvium Variable High Moderate Low Very low
Talus Widespread Low Low Very low Very low
Dunes Widespread High Moderate Low Very low
Loess Variable High High High Unknown
Glacial till Variable Low Low Very low Very low
Tuff Rare Low Low Very low Very low
Tephra Widespread High High ? ?
Residual soils Rare Low Low Very low Very low
Sebka Locally variable High Moderate Low Very low
(b) Coastal zone
Delta Widespread Very high High Low Very low
Esturine Locally variable High Moderate Low Very low
Beach: High wave Widespread Moderate Low Very low Very low
energy
Beach: Low wave Widespread High Moderate Low Very low
energy
Lagoonal Locally variable High Moderate Low Very low
Fore shore Locally variable High Moderate Low Very low
(c) Artificial
Uncompacted fill Variable Very high --- --- ---
Compacted fill Variable Low --- --- ---

Table 4.3. Conditional probability relationship for liquefaction susceptibility categories

(from Liao, Veneziano, and Whitman 1988)
Susceptibility category P[Liquefaction|PGA=a]
Very high 9.09a-0.82
High 7.67a-0.92
Moderate 6.67a-1.00
Low 5.57a-1.18
Very low 4.16a-1.08
None 0.00
For a given city, the value of “percentage of urbanized area with high liquefaction
susceptibility” can be estimated by superimposing the boundaries of the city’s urbanized area
onto a map that depicts the distribution of high liquefaction susceptibility soil (as it is defined in
Tables 4.2 and 4.3) throughout the city, perhaps using GIS software. If no such liquefaction
susceptibility map is available, it may be necessary to estimate the value of XH4 directly from a
geological map of the city.
4.3.1.3. Fire Following Earthquake
Two indicators—“percentage of buildings that are wood” and “population density” (XH5
and XH6)—represent the fire hazard. The first conveys information about the amount of fuel
available to feed post-earthquake fires, and the second refers to the ease with which post-
earthquake fires can spread and be suppressed. Planners in Kobe, Japan used the same two
indicators to estimate the hazard of fire following earthquake before the Great Hanshin
earthquake of 1995 (Bauman 1997). Although the indicator XH5 refers to all buildings whose
primary construction material is wood, the value may be estimated from residential building
stock, for which data are generally available from housing censuses.

While fire spread is more directly related to infrastructure density than population
density, population data is more easily available than infrastructure data, so it is assumed that
infrastructure density and population density are proportional, and population density is used
instead. In the indicator XH6, population density is defined as the weighted average of the
densities of the counties (or smaller cities, neighborhoods, or other administrative units) that
make up the greater metropolitan area, where the weights are the populations of those counties:
XH6 = (Σ idipi)/ Σ ipi, (4.1)
where di is the density of county i, pi is the population of county i, and the summation is over all
counties i that comprise the greater metropolitan area. The weighted average provides a more
accurate representation of the city’s density than a simple ratio of total population divided by
total land area would, because some greater metropolitan areas are defined to include large,
sparsely populated surrounding areas that would make the densities of those cities appear
smaller than they are. Still, the population density calculation can be problematic because the
results will depend on how many counties (or administrative units) there are within the
metropolitan area, and on how their boundaries are defined. The sensitivity analysis compares
the two possible definitions of population density (Section 7.1).
The set of hazard indicators omit several important aspects of the fire hazard factor. No
information is included about the location and number of potential ignition sources, the
distribution of fuel, the existence of other types of fuel besides wood buildings, or the typical
weather conditions (e.g., precipitation, wind speed and direction) that may influence the fire
spread. Since fire suppression capability is accounted for within the Emergency Response and
Recovery Capability factor, no attempt is made to represent it with the Hazard indicators.
4.3.1.4. Tsunami and Seiche

The “tsunami potential indicator,” XH7, is defined as a three-level classification system to
rate a city’s susceptibility to tsunami damage. The indicator has three possible values (Table
4.4):
Table 4.4. “Tsunami potential indicator” definition

Tsunami Category description
potential XH7
0 None. No historical tsunamis. Physically impossible because city is not
adjacent to a body of water that contains an earthquake source capable of
generating a tsunami.
1 Low. Physically possible, but unlikely (i.e., tsunami has occurred less than
twenty times in historical record)
2 High. Physically possible, and not unlikely (i.e., tsunami has occurred
twenty or more times in historical record)
The value of XH7 for a given city can be assessed by examining the worldwide historical
database of tsunamis (NGDC 1997), and the city’s geographical location relative to possible
tsunami-generating sources. The indicator does not take into account a city’s coastal
topography, which can influence the run-up height and the extent of inundation should a tsunami
occur.
4.3.2. Exposure factor
The following six indicators represent the Exposure factor. The first four describe the
Physical Infrastructure Exposure, the fifth represents the Population Exposure, and the sixth
refers to the Economy Exposure. Note that two of the indicators are listed twice because they
are each used in two different capacities, as discussed below. For a given city, XE1 and XE5 will
always have the same value, as will XE2 and XE6. The indicators are separated for conceptual
clarity, so that, for example, the weight assigned to XE1 relates to the population as it conveys
information about the Physical Infrastructure Exposure, and the weight assigned to XE5 relates to
the population as it conveys information about the Population Exposure.

1. XE1: Population
2. XE2: Per capita Gross Domestic Product, in constant 1990 US dollars2
3. XE3: Number of housing units3
4. XE4: Urbanized land area (square kilometers)
5. XE5: Population
6. XE6: Per capita Gross Domestic Product, in constant 1990 US dollars
The Social-Political System Exposure is not represented explicitly by the set of Exposure
indicators, but it seems reasonable to assume that cities with relatively high Physical
Infrastructure, Population, and Economy Exposure, would have a relatively high Social-Political
System Exposure as well.
4.3.2.1. Physical Infrastructure Exposure
Jones (1994) proposes that “there are strong regularities in human behavior in
constructing elements to shelter people and their activities and to facilitate movement of people,
goods, and information, and that there will be strong patterns in the ways in which populations
occupy space and locate structures to accommodate their needs (Isard, 1956).” Based on
these assumptions, he and his colleagues have developed a methodology for making indirect
estimates of a city’s physical infrastructure—the amount (in terms of number, square footage,
and replacement value) and distribution (i.e., distance from the center of gravity of population of
the city) of each type of infrastructure component (e.g., residential buildings, commercial
buildings, roads). The model that he and his colleagues developed was calibrated and verified
with extremely labor-intensive, direct count inventories of all infrastructure components for a few
case study cities (Wichita, Kansas; Ithaca, New York; Kyoto, Japan). The studies that
produced this model and inventory data aimed to estimate the physical infrastructure in absolute
2
See Section 4.3.2.3. for a description of the difference between Gross Domestic Product and Gross National
Product.

terms, but for the purposes of the EDRI, which aspires to assess only the relative physical
infrastructure exposure of different cities, the most important findings are the following:
1. At the scale of a greater metropolitan area, the physical infrastructure exposure

depends primarily on the population, and the level of technological development and
income. “More developed and elaborately organized societies use time in more
specialized ways and have greater variety in structures to accommodate uses that
fulfill needs. Within a given culture at a level of technological development resulting
in a level of income, there should be little variation from place to place, and
differences should reflect particular geographical characteristics of places. Strong
regularities will exist, but the parameters [e.g., number of buildings per capita,
square footage per capita, and replacement cost per capita, for each building use
category] will differ with culture and over time with income” (Jones 1994).
2. The indirect, macroscopic approach to estimating a city’s physical infrastructure

exposure is reasonable, supported by theoretical assumptions and by empirical
evidence.
Based on these findings, an indirect, macroscopic approach was used to determine the
Physical Infrastructure Exposure indicators for the EDRI. Four simple indicators were identified
to represent the factor—“population” (XE1), “per capita Gross Domestic Product, in constant
1990 US dollars” (XE2), “number of housing units” (XE3), and “urbanized land area” (XE4).
Inclusion of the first two follow the arguments put forth by Jones et al., with “per capita GDP”
representing the city’s level of technological development and income. The third is added as a
direct measure to support the population. The fourth, “urbanized land area,” was included on
the hypothesis that it indicates the length roads and other lifelines in a city, because a larger area
of urbanization requires more extensive lifeline systems to service it.
3
The term “housing unit” is defined in Section 4.3.2.1.

Recognizing that many indicators of physical infrastructure exposure are well-correlated,
and that therefore a principal components analysis might be possible, an analysis was performed
to further verify the selection of the four Physical Infrastructure Exposure indicators (see Section
5.3.2 for more about principal components analysis). The analysis was conducted with two
separate databases of physical infrastructure information for cities worldwide. The first includes
data on ten indicators, for 160 cities4 in forty-five countries; the second includes data on six
indicators, for 76 cities5 in forty-six countries6 (Table 4.5). The complete data sets are
presented in Appendix D.
Table 4.5. Indicators in the two physical infrastructure exposure data sets
Data Set One Data Set Two
Statistics of World Large Cities Human Settlements Statistics
[160 cities in forty-five countries] [76 cities in forty-six countries]
1. Per capita real Gross Domestic Product 1. Per capita real Gross Domestic Product of
of country in which the city is located country in which the city is located (PPP$)
(PPP$)
2. Population 2. Population
3. Number of housing units7 3. Number of housing units8
4. Land area (sq. km.) 4. Total electricity consumption per day
(1000 kwh)
5. Total (domestic & industrial) 5. Total annual water consumption
consumption of electric light & power (1000 m3)
(1000 kwh)
6. Total (domestic and industrial) gas 6. Number of motor vehicles
consumption (1 million kcal)
7. Total (domestic and industrial) water
supply (1000 m3)
4
All cities with population over one million people are included in the database.
5
For each country, the capital city and up to three of the other largest cities were included in the database.
6
The data are not complete for all cities and indicators, so each regression analysis reported in Tables 4.6
and 4.7 includes only those cities with data for every indicator in that particular regression model. The
number of cities used in each regression is listed in the final column of Tables 4.6 and 4.7. All cities are used
in the principal components analysis described in Section 5.3.4.1.
7
The indicator is actually "number of dwellings," but for these purposes, it can be taken as equivalent to the
value of the "number of housing units" indicator.
8
The indicator is actually "number of occupied housing units," but for these purposes, it can taken as
equivalent to the value of the "number of housing units" indicator.

8. Length of roads (km)
9. Number of telephone subscriptions
10. Number of establishments
In both data sets, the per capita GDP data is taken from the Human Development Report
1995 (UNDP 1990-95). Data for the last nine indicators in Data Set One are from Statistics
of World Large Cities 1991 (TMG 1991) published by the Statistical Division of the Tokyo
Metropolitan Government; data for the last five indicators in Data Set Two are from the United
Nations’ Compendium of Human Settlements Statistics 1995 (UN Centre for Human
Settlements 1995). Note that Data Set Two does not contain data for the land area, so the
analysis is performed only for the other three indicators (XE1, XE2, XE3) in that case.
Both of the data sets contain a great deal of uncertainty, and the data for different
indicators and cities relate to roughly, but not exactly, the same year. Nevertheless, they are the
best available data on physical infrastructure exposure for cities, and they should suffice to
suggest if the four proposed indicators provide a reasonable representation of a city’s physical
infrastructure exposure. Furthermore, duplicating the analysis with the two, presumably
independent data sets, provides additional support for the results of each.
Using these two data sets, an analysis was conducted to determine if the four selected
indicators (XE1, XE2, XE3, XE4) do in fact provide a reasonable representation of a city’s
physical infrastructure exposure. The analysis first assumes that the complete set of indicators
(ten for Data Set One; six for Data Set Two) together represent a city’s physical infrastructure.
Intuitively, each indicator in Table 4.5 should relate to the physical infrastructure. For example,
to support a greater level of water, gas, or electricity consumption, there must be more
infrastructure to deliver each of those services. Second, the efficiency of the proposed
indicators (four for Data Set One; three for Data Set Two) in representing the complete sets of
indicators is evaluated by determining the percentage of the total variation of the complete set
that the proposed indicators explain. Let the proposed indicators be called the “retained”
indicators, and the other indicators (six for Data Set One; three for Data Set Two) be the

“discarded” indicators. To carry out this second step, each of the discarded indicators is
regressed on the retained indicators, in turn. The percentage of variation of the complete set
that is explained by the retained indicators equals:
(nr + Σ iR2i,r)/N, (4.2)
where nr is the number of retained indicators, R2i,r is the squared multiple correlation of the ith
discarded indicator with the r retained indicators, and N is the total number of discarded and
retained indicators in the complete set (Dunteman 1989).
Table 4.6. Results of principal components analysis for Data Set One
(Statistics of World Large Cities)
P-values of retained indicators
Discarded Popu- No. of Per cap. Land Inter- R2 Signif- No.
indicators lation housing GDP area cept icance F obs.
units
Elec. consum. 0.5341 0.0013 0.0044 0.9784 0.8479 0.497 3.43e-08 63
Gas consum. 0.9345 0.5274 0.5912 0.6830 0.0023 0.021 0.9052 52
Water supply 0.5900 0.2609 0.4450 0.6749 0.0012 0.028 0.7971 63
Roads length 0.0098 0.2663 4.01e-05 1.2e-14 7.57e-65 0.804 8.97e-18 57
No. phones 0.6545 0.7170 0.0551 0.5728 0.0568 0.104 0.1336 68
No. establish. 0.0056 6.2e-06 0.9726 0.0462 1.36e-05 0.695 7.28e-11 49
Sum of R2 values + nr = 6.149
Total amount of variation explained by four proposed indicators = 62%
Table 4.7. Results of principal components analysis for Data Set Two
(Human Settlements Statistics)
P-values of retained indicators
Discarded indicators Popu- No. hous- Per capita Inter- R2 Signif- No.
lation. ing units GDP cept icance F obs.
Elec. consumption 0.0012 4.94e-06 0.4989 0.0008 0.904 6.73e-06 15
Water consump. 0.3858 0.5586 0.9976 0.2524 0.140 0.7353 12
No. motor vehicles 0.0765 0.8757 0.0273 0.2513 0.593 0.0013 21
2
Sum of R values + nr = 4.637
Total amount of variation explained by three proposed indicators = 77%
Tables 4.6 and 4.7 display the results of the analyses conducted for the two data sets.
The four proposed indicators (population, number of housing units, per capita GDP, and land

area) explain 62% of the total variation of the complete set of indicators in Data Set One, and
the three proposed indicators (population, number of housing units, and per capita GDP)
explain 77% of the total variation of the complete set of indicators in Data Set Two. Since the
complete sets of indicators are not perfect representations of the physical infrastructure
exposure, it is not important that the proposed indicators represent all of the variation in the
complete set. In both cases, however, the proposed indicators explain a large enough
proportion of the total variation to suggest that they could reasonably represent the complete set
of indicators, and therefore, that they could reasonably represent a city’s physical infrastructure
exposure.
Finally, the P-values indicate which of the retained indicators help explain the variation in
each discarded indicator. For each retained indicator, the P-value, or observed significance
level, measures the extent to which the data disagrees with the null hypothesis that the β value
(weight) associated with that retained indicator should equal zero. The P-value is the probability
of observing results that are at least as contradictory of the null hypothesis as those computed
from the sample data (McClave and Dietrich 1988). The results in Tables 4.6 and 4.7 suggest
that in Data Set One, the retained indicator “population” helps predict “length of roads” and
“number of establishments”; “number of housing units” helps predict “electricity consumption”
and “number of establishments”; “per capita GDP” helps predict “electricity consumption,”
“length of roads,” and “number of phones”; and “land area” helps predict “length of roads” and
“number of establishments.” In Data Set Two, “population” helps predict “electricity
consumption” and “number of motor vehicles”; “number of housing units” helps predict
“electricity consumption”; and “per capita GDP” helps predict “number of motor vehicles.”
Each of the proposed indicators then is helpful in predicting at least one of the discarded
indicators. The results of the two data sets match in that in both cases, the number of housing
units relates to electricity consumption, and both population and per capita GDP relate to the
transportation network (number of motor vehicles or length of roads). They are inconsistent in
that per capita GDP helps predict electricity consumption in Data Set One, but not Data Set

Two; and population helps predict electricity consumption in Data Set Two, but not Data Set
One.
Despite the uncertainty associated with each data set in this analysis, the fact that they
both independently suggest that the proposed indicators may reasonably represent the physical
infrastructure, makes the analyses worthwhile. While the four proposed indicators will not be
perfect replacements for doing direct counts of the inventories of every city, the findings of the
Jones et al. studies and these principal component analyses suggest that the four proposed
indicators do convey roughly the same information as direct counts would, and much more
simply.
The “population” and “number of housing units” data for a city will be available from
censuses. The EDRI adopts the U.S. Bureau of the Census definition that a “housing unit is a
house, an apartment, a mobile home or trailer, a group of rooms or a single room occupied as a
separate living quarters or, if vacant, intended for occupancy as separate living quarters”9 (U.S.
Bureau of the Census 1993b). If it is unavailable directly, the “urbanized land area” can be
evaluated by estimating the percentage of the total area that is urbanized from a map showing
the urbanized area, and multiplying that percentage by the total land area, which is reported in
most country censuses and statistical yearbooks.
Finally, a couple of points regarding the indicator “per capita Gross Domestic Product,
in constant 1990 US dollars” deserve note. First, taking the USA as an example, Gross
Domestic Product (GDP) is defined as the “market value of the final goods and services
produced by resources located in the United States” (Johnson 1993). GDP data for a specified
country and year is generally reported in terms of the national currency, at the prices
corresponding to the specified year. To enable GDP comparison across countries, the values
9
“Separate living quarters are those in which the occupants live and eat separately from any other persons
in the building and which have direct access from outside the building or through a common hall. The
occupants may be a single family, one person living along, two or more families living together, or any other
group of related or unrelated persons who share living arrangements” (U.S. Bureau of the Census 1993b).

must all be converted to the same currency units using exchange rates or purchasing power
parity; and to enable comparisons across years, the values must be converted to constant
dollars using GDP deflators. So that comparisons of GDP among countries and over time
reflect only the change in volume of output (and not differences in currency or price inflation),
the GDP data used in the EDRI is measured in terms of constant 1990 US dollars.10
Second, GDP data is almost always calculated only for countries. To find the
equivalent of GDP for a city for use in the EDRI, it is assumed that GDP is distributed
throughout a country according to population distribution, i.e., that GDP per capita is constant
throughout a country and equal to the country GDP divided by the country population. Using
GDP and population data collected for regions within several countries, the veracity of this
assumption was tested. Figure 4.3 shows the GDP versus population for five countries—China,
France, India, Indonesia, and USA. As the figure demonstrates, the assumption seems to hold
fairly well, with the notable exception that in some cases the largest cities are responsible for a
disproportionate amount of the GDP. For those largest cities then, using the country per capita
GDP underestimates the city’s per capita GDP. Future refinements of the EDRI may take this
observation into account in evaluating the “per capita GDP” indicator.
10
Several sources exist that should be equally suitable for determining the per capita GDP. However,
because of subtle differences in the various sources of GDP data, it is best to calculate the per capita GDP
for each city in a consistent way, using data from a single source.

Chinese Gross Domestic Product vs. Population by Region [1993-94]
GDP 1993 [current 108 yuan] Guangdong

Jiangsu
3000
Liaoning
2000 Zhejiang Sichuan
Shanghai Henan
1000 Beijing
0
0 20 40 60 80 100 120
Region population 1994 [millions]
Figure 4.3a. Distribution of GDP in China. Source: People's Republic of China (1995).
French Gross Domestic Product vs. Population by Region [1989]

GDP 1989 [million francs]
Ile de France
1,600,000
1,200,000
Best-fit line without Ile de France outlier
800,000
400,000
0
0 2 4 6 8 10 12
Region population 1989 [millions]
Figure 4.3b. Distribution of GDP in France. Source: INSEE (1990).

Indian Gross Domestic Product vs. Population by State [1990]
700
[current Rs. billion]

600
GDP 1991-92
500
400
300
200
100
0
0 50 100 150
State population 1991-92 [millions]
Figure 4.3c. Distribution of GDP in India. Source: EIU (1995).

GDP 1990 [Billion rupees; 1983 prices]
Indonesian Gross Domestic Product vs. Population by Province [1990]

18,000
16,000
14,000 Jakarta
12,000
10,000 Riau
8,000 Aceh
6,000 Best-fit line without four labeled outliers
4,000 E. Kalimantan
2,000
0
0 5 10 15 20 25 30 35 40
Province population 1990 [millions]
Figure 4.3d. Distribution of GDP in Indonesia. Source: EIU (1996).

United States Gross Domestic Product vs. Population by State [1990]
GSP 1990 [current US$billion]
800
700
600
500
400
300
200
100
0
0 5 10 15 20 25 30
State population 1990 [millions]
Figure 4.3e. Distribution of GDP in the USA. Source: U.S. Bureau of the Census (1995).
The Physical Infrastructure Exposure indicators do not convey any information about
the within-city geographical distribution or volume of use of the various infrastructure
components.
4.3.2.2. Population Exposure
Population Exposure is the most straightforward factor to represent in the EDRI. The
ideal, direct measure of population exposure, i.e., “population” (XE5), is readily available in
censuses, so the indicator selection requires no simplification or justification. Using only
“population” to represent Population Exposure, however, does not convey any information
about the distribution of the population within the city, or the breakdown of different, particularly
vulnerable groups within the city. The percentage of the population that belongs to each group
is included instead in the Vulnerability section (see Section 4.3.3.2).
4.3.2.3. Economy Exposure

In Section 3.3.2.3., Economy Exposure was defined as the volume of the economic
flow of goods, services, and money that are transacted within a city. The indicator “per capita
Gross Domestic Product, in constant 1990 US dollars” (XE6) provides the best representation
of the volume of that economic flow. Another economic indicator that could be considered is
“per capita Gross National Product.” The difference between the two is explained as follows,
considering the USA as an example. GNP “relates to production by labor and other factor
supplied by residents of the United States. Thus, the income earned by American residents on
their investments in other countries is included in the gross national product.” GDP relates to
“production by factors physically located in the United States, no matter who owns them”
(Johnson 1993). Since the EDRI is concerned with the economic flows that could be disrupted
by direct physical damage, GDP seems the more appropriate quantity. Furthermore, the
Bureau of Economic Analysis has been changing the emphasis from GNP to GDP since 1991
(Johnson 1993). Of course, the discussion in Section 4.3.2.1 about calculating “per capita
Gross Domestic Product, in constant 1990 US dollars” holds for XE6 too.
4.3.3. Vulnerability factor
The Physical Infrastructure Vulnerability factor is represented by the first five indicators
listed below; Population Vulnerability by the last indicator.
1. XV1: Seismic code indicator

2. XV2: City wealth indicator
3. XV3: City age indicator
4. XV4: Population density (people per square kilometer)
5. XV5: City development speed indicator
6. XV6: Percentage of population aged 0-4 or 65+
Economy and Social-Political Vulnerability are not represented because indicators that reflect
those concepts could not be identified. If Jones’ assessments of the relative vulnerability of the

extractive, fabricative, distributive, and service sectors of the economy are assumed to be
correct (see Section 3.3.3.3), a possible method for representing Economy Vulnerability would
be to determine the percentage of each city’s GDP associated with each of those four sectors,
and conclude that those cities with a larger share of the economy in more vulnerable sectors
have a more vulnerable economy. Unfortunately, while GDP data are often reported by sector
share, those data are available only for countries, not metropolitan areas, and it would be
inappropriate to assume that the structure of a city’s economy would be the same as that of the
country in which it is located. In general, indirect economic losses are difficult to assess even
after an earthquake occurs (much less to estimate before an earthquake) because the indirect
effects are often shifted to other regions or generations, because the in-built mechanisms that
immediately go to work to reestablish equilibrium in the economy hide the indirect effects, and
because it is impossible to know how the economy would behave without the occurrence of the
disaster, and assessment of indirect effects requires a comparison of the economy with and
without (not before and after) the earthquake.
4.3.3.1. Physical Infrastructure Vulnerability
Section 3.3.3.1. discusses two methods by which the Physical Infrastructure

Vulnerability factor could be assessed—an indirect, macroscopic approach based on the
circumstances of a city’s development, and a direct, microscopic one based on structural
analysis principles. The Physical Infrastructure Vulnerability indicators in the sample analysis
were selected using the former method because of its simplicity. The characteristics of a city’s
development that generally influence the vulnerability of the infrastructure were identified, and an
indicator was derived to represent each. They are discussed in turn below. The characteristics
include knowledge of how to reduce vulnerability (XV1), financial resources to reduce
vulnerability (XV2), recentness of development to reduce effects of aging (XV3), ample space to
accommodate population growth so that the infrastructure is not too dense, and there is room
for lifeline redundancy (XV4), and ample time to construct the infrastructure in a quality-
controlled way (XV5). Although the infrastructure is the real quantity of interest, since population

data are more easily available, all five Physical Infrastructure Vulnerability indicators use
population data instead of physical infrastructure data, relying on the assumption that
infrastructure construction coincides with population growth.
The “seismic code indicator” (XV1) conveys an awareness of the problem of earthquake
risk at the time the infrastructure is built, a desire to address the problem (i.e., a perception that
the risk is unacceptably severe in the context of the other challenges the city faces), and the
existence of the knowledge of how to reduce the vulnerability. It demonstrates if a city has the
organizations and individuals to create, enact, and enforce effective seismic codes, and when the
codes were developed in relation to the development of the city.
The “seismic code indicator” can be calculated following a five-step process. First,
identify the benchmark years in which significant changes in the quality of the city’s seismic code
took place. Second, assess the sophistication of the seismic code during each inter-benchmark
period according to the Sophistication Rating Scale in Table 4.8. Third, determine the
percentage of the current city population that arrived in each inter-benchmark period. Fourth,
rate the quality of code enforcement in the city using the Enforcement Rating Scale in Table 4.9.
Fifth, evaluate the seismic code indicator using the following expression:
XV1 = e*[Σ k(sk)*(pk)], (4.4)
where e is the enforcement rating, sk is the sophistication rating for inter-benchmark period k, pk
is the percentage of the current population that arrived in inter-benchmark period k, and the
summation is over all inter-benchmark periods k. The summation is essentially a weighted
average of the sophistication of each inter-benchmark period, where the weights are the
percentages of the current infrastructure built (or population arrived) during each period.
Table 4.8. Seismic code sophistication rating scale

Sophistication Category description
level sk
0.1 No regulations.
0.3 Only requirement is to limit the base shear to an arbitrary constant

percentage of building weight. Not based on theory.
0.5 Includes a base shear equation with coefficients to account for some, but not
all of the following: soil type, building period, seismic zone, importance of
the structure, and ductility of the structural type.
0.8 Includes some, but not all of the basic components* of a seismic code.
0.9 Includes all of the basic components of a seismic code.
1.0 Refined version of a seismic code with all the basic components.
* Basic components include: (1) provisions for static and dynamic design methods; (2) an
equation to calculate the base shear; (3) coefficients in the base shear equation to account for
soil type, building period, seismic zone, importance of the structure, and ductility of the
structural type; (4) a method to distribute forces along the height of structure; (5) provisions
for addressing torsion; (6) provisions for addressing irregularities in plan or elevation; (7)
method to address inter-story and total drift; (8) special provisions for essential facilities; and
(9) provisions for detailing.
Table 4.9. Seismic code enforcement rating scale

Enforcement Category description
level e
0.2 Poor
0.4 Below average
0.6 Average
0.8 Above average
1.0 Excellent
The seismic code indicator evaluation includes a great deal of subjective assessment
because there is no purely quantitative, objective way available to include this essential
information about the seismic code development. Because the subjective enforcement rating
value will control the seismic code indicator values, it may seem unnecessary to require so much
detail in calculating the benchmark periods, and the values of sk and pk. Nevertheless, the
formulation is defined as it is for conceptual completeness, addressing the timing and quality of
the development of seismic codes, as well as the degree of their enforcement. The specific
values in the Sophistication and Enforcement Rating Scales were determined by calibrating the
final seismic code indicator values for the ten sample analysis cities so that the results appear
reasonable. Appendix E illustrates the calculation of the “seismic code indicator” for the sample
analysis.

The “city wealth indicator” (XV2) conveys the availability of resources in the city to
develop, implement, and enforce both a general building code and a seismic code. It relates to
a city’s general quality of design and construction. The “city wealth indicator” equals the
average gross domestic investment (GDI) per capita per year in constant 1987 US dollars.
GDI is one of the four main components of GDP (along with personal consumption
expenditures, net exports, and government purchases). Also known as gross fixed capital
formation (GFCF), GDI “measures the outlays for the addition of reproducible capital goods to
the fixed assets of private and public enterprises, private nonprofit institutions, and general
government (reduced by their net sales of used or scrapped capital goods) and the value of the
net increase or decrease in inventories” (World Bank 1980). Since there is no exact year in
which a city begins, and data certainly would not be available for so long ago anyway, a
historical time period had to be established over which the average would be taken. Using the
time period from the present back to the year in which the city had one-half of its current
population makes the number of years averaged different for each city, but makes the
percentage of the total existing infrastructure built with that GDI consistent among cities.
The “city age indicator” (XV3) represents the extent to which the effects of aging have
increased the vulnerability of the physical infrastructure. It is defined as a weighted average of
the age of the population:
XV3 = Σ i(pi)*(ai), (4.5)
where pi is the percentage of the current population that arrived in year i, ai is the age of the
population that arrived in year i (i.e., current year - year i), and the summation is over all the
years from the year in which the population was one-half of its current level to the current year.
Cities require ample physical space to accommodate population growth so that the
infrastructure is not too dense, and so that there is room for lifeline redundancy. When a city’s
development becomes too dense, more vulnerable areas (e.g., those with steep slopes or
artificial fill) are more likely to be occupied. The “population density” indicator (XV4) represents

the extent to which a city has had adequate space to develop. It is exactly the same quantity as
XH6, and is evaluated in the same way.
In many rapidly expanding cities in developing countries, the population growth is

occurring so quickly that a quality, low vulnerability infrastructure cannot be developed fast
enough to support the new residents. The “city development speed indicator” (XV5) conveys
whether or not the city has been developed with ample time to construct the infrastructure in a
quality-controlled way. It is defined as the number of years that have passed since the city
population was one-half of its current level.
Collectively, the five indicators (XV1 to XV5) offer an assessment of the Physical
Infrastructure Vulnerability factor. These assessments were validated using results obtained with
the loss estimation modeling approach to physical infrastructure vulnerability assessment. Risk
Management Solutions, Inc. developed aggregate damage curves for many regions throughout
the world (RMS 1996-97). Each curve is a weighted average of the damage curves for the
individual infrastructure components, with the weights equal to the proportion of the total
infrastructure in the region that is composed of the corresponding type of infrastructure
component. The curves are expressed as the mean damage ratio (MDR) versus MMI. To
compare the damage curves to the five-indicator EDRI assessments, two points were selected
from each curve—MDRs corresponding to MMI values of VII and XI (Fig. 4.4), the two
MDRs were scaled11, and the average of the scaled values were taken as the loss estimation
modeling assessment of physical infrastructure vulnerability. As shown in Fig. 4.5 and Table
4.10, the two methods—EDRI and loss estimation modeling—produce reasonably similar
results for a sample of seven cities. The Spearman rank correlation coefficient12 for the two
11
The MDR values were scaled in the same way as the EDRI indicators (see Section 5.2.7).
12
Spearman’s rank correlation coefficient (rs) measures the correlation between two sets of ranks. The
following formula is used to evaluate the Spearman coefficient when there are no ties in the ranks:
 6∑ (u i − vi ) 2 
rs = 1 −  i 2 
n(n − 1) 
 

methods is 0.75, suggesting reasonable, but not perfect correlation between the results of the
two methods for this small sample.
Example of a damage curve used in loss estimation modeling

Mean damage ratio [%]
100
50
0
VI VII VIII IX X XI XII
MMI
Figure 4.4. Example of a loss estimation model damage curve
4
3
EDRI
2
Loss
1 estimation
model
0
Boston Jakarta Mexico San Santiago St. Tokyo
City Francisco Louis
Figure 4.5. Comparison of the Physical Infrastructure Vulnerability factor assessment using
EDRI and loss estimation models
Table 4.10. Comparison of two methods for Physical Infrastructure Vulnerability factor
assessment
where u i and v i are the ranks of the ith observation in samples 1 and 2, respectively, and n is the number of
pairs of observations (number of observations in each sample).

Scaled values Ranks Squared
Loss differ-
City EDRI estimation EDRI Loss es- ences
timation
model
Boston 1.97 1.94 4 4 0
Jakarta 2.66 3.76 1 1 0
Mexico City 2.28 1.75 3 5 4
San Francisco 1.34 0.77 7 7 0
Santiago 2.44 2.24 2 3 1
St. Louis 1.94 2.48 5 2 9
Tokyo 1.37 1.06 6 6 0
Spearman rank correlation coefficient = 0.75
The five physical infrastructure vulnerability indicators (XV1 to XV5) together represent
the key characteristics of a city’s development that affect the vulnerability of its infrastructure.
Because each characteristic is explicit and separate, each component easily could be improved
as further study supports or contradicts the assumptions on which these sample analysis
indicators are based. The representation of Physical Infrastructure Vulnerability embodied by
these indicators has the advantage of relatively simple data requirements, compared to the
alternative loss estimation modeling approach that relies on assessing the exposure and
developing fragility curves for each component of the infrastructure individually.
The approach does suffer from a few disadvantages. The indicators are not as
objective as would be ideal. The indirect approach is based on the assumption that the
identified characteristics of a city’s development do in fact determine the vulnerability of the
city’s physical infrastructure. Ideally, the vulnerability indicators would depend on a city’s
hazard. The structural requirements should not necessarily be the same in different cities, but
they should be consistent with the city’s level of hazard. It is inappropriate for a city with low
hazard to have a seismic code as strict as that for a city with high hazard.
Finally, there are a few aspects of the Vulnerability factor that are not represented by
the sample analysis vulnerability indicators. Maintenance, previous damage, and retrofitting can

affect the vulnerability of the physical infrastructure, but because there are no good indicators
available to represent them, and they are probably relatively insignificant compared to the other
determinants, they are not included. Vulnerability to nonstructural and content damage are not
represented, because there is little data or theory to determine how these ideas could be
included, or if they would vary significantly from city to city. No indication of the redundancy of
the city’s lifeline networks is included either, because no consistent, city-level data is available
on the length of pipelines or other lifelines.
4.3.3.2. Population Vulnerability
Population vulnerability can exist in three forms: physical, economic, and social.
The “percentage of population aged 0-4 or 65+” indicator (XV6) is a direct measure of one type
of physical Population Vulnerability. It conveys the number of residents who may be less able
to escape from a collapsing structure, recover from an injury, or survive in a post-earthquake
atmosphere of hardship because they are very young or very old. The threshold ages of five-
years-old and sixty-five-years-old were chosen both because they are reasonable estimates of
the ages at which people become more vulnerable, and because census data is generally
reported for the age groups defined by those cutoffs.13 Theoretically, similar indicators could be
used to capture other types of population vulnerability, e.g., percentage of the population that is
sick, disabled, poor, homeless, or disenfranchised. Those indicators were not included in the
sample analysis because of the difficulty in defining and obtaining data for the other potentially
vulnerable groups. A value judgment is required to determine which, if any, minority groups are
more vulnerable, and the groups will not be the same from city to city.
4.3.4. External Context factor
13
The age thresholds appropriate for this indicator may vary regionally, because life expectancies are not
uniform across cultures and regions. Sixty-five may be too young for industrialized cities.

Three indicators convey the External Context factor in the sample analysis. The first
represents the Economic External Context; the last two represent the Political External Context.
They are:
1. XC1: Economic external context indicator

2. XC2: Political country external context indicator
3. XC3: Political world external context indicator
Transportation and Cultural External Context are not included explicitly because of the difficulty
in finding appropriate indicators. An accurate measure of transportation external context would
have to include all modes of transportation, because the relative predominance of different
modes varies considerably from city to city. One city may be a major port, but have little rail
traffic, while another might be a rail hub, but have no port. To compare the transportation
external context of these two cities, both rail and water would have to be assessed. Although
some transportation data exist, particularly for travel through airports and ports, it would be
difficult to address all modes consistently. Nevertheless, the information conveyed in the three
indicators listed above should be sufficient to convey the relative external context ratings of
different cities because in general, any major transportation hub or cultural center will exhibit a
high economic and/or political external context rating also. Appendix F shows the values for the
external context indicators (XC1, XC2, and XC3) for a set of 319 cities.
4.3.4.1. Economic External Context
The “economic external context indicator” (XC1) is calculated according to the following
equation:
Pop city Pop city
XC1 = (TFT) * (Pop country ) * ( ) *( ) (4.6)
Pop country Pop largest city

where TFT is the total foreign trade between the country in which the city is located and other
countries, Popcity, Popcountry, and Poplargest city are the population of the city of interest, of the
country in which the city is located, and of the largest city in that country, respectively. TFT is
the sum of the absolute values of a country’s current account credits and debits, capital account
credits and debits, and financial accounts assets and liabilities in US$millions (see IMF 1995 for
more information). The first two terms of the expression for XC1 convey the economic
importance of the country to the world—a function of the amount of economic exchange
between the country and others, and of the size of the country, measured in terms of population.
The last two terms convey the importance of the city to its country. Assuming a constant
economic importance per capita throughout the country, the first two terms must be multiplied
by the ratio of Popcity/Popcountry to determine the economic external context associated with the
city alone. Finally, since the economic importance per capita will probably be higher in the
largest cities in the country, particularly if there are only one or two principal cities (see Fig.
4.3), the final factor is included to increase the relative XC1 value for a country’s primary cities.
The reason for adopting this country-to-world, and city-to-country approach is that economic
exchange data is widely available for exchange between countries, but virtually nonexistent for
inter-city exchange.
4.3.4.2. Political External Context
The “political country external context indicator” (XC2) represents the political power of
the city in its country. Assuming that quantity depends largely on how many people are
governed by the city, XC2 is defined as:
Pop political region
X C2 = ( ) (4.7)
Pop country
where Poppolitical region and Popcountry are the population of the largest region of which the city is a
political capital, and the population of the country in which the city is located, respectively.
Poppolitical region is defined so that Popcity ≤ Poppolitical region ≤ Popcountry. Note that XC2 equals one
for all country capitals, because, without exploring the differences due to the type of

government, all country capitals are equally important to the people in their country. That is,
San Jose is as important to the people of Costa Rica as Washington D.C. is to the people of the
USA. If the city is not a capital of any state or country, XC2 = Popcity/Popcountry.
To distinguish among the country capitals, the indicator “political world external context
indicator” (XC3) is included, conveying the relative political power of cities on the world stage.
Assuming that wealthier country capitals will control more political power than poorer ones, XC3
is defined as:
XC3 = (per capita GDP)*(Poppolitical region), (4.8)
where per capita GDP is the “per capita Gross Domestic Product, in constant 1990 US
dollars,” the same quantity as indicators XE2 and XE6, and Poppolitical region is defined as above.
Equivalently, XC3 equals the GDP associated with the largest region of which the city is the
political capital.
The political external context indicators do not convey any information about the type of
government, and they assume that the nominal political capital controls all the political power.
As mentioned in Section 3.3.4.3, this is not always the case.
4.3.5. Emergency Response and Recovery Capability factor
The sample analysis uses the following nine indicators to capture a city’s Emergency
Response and Recovery Capability:
1. XR1: Planning indicator

2. XR2: Per capita Gross Domestic Product, in constant 1990 US dollars
3. XR3: Average annual real growth in per capita GDP during previous ten years
4. XR4: Housing vacancy rate
5. XR5: Number of hospitals per 100,000 residents
6. XR6: Number of physicians per 100,000 residents

7. XR7: Extreme weather indicator
8. XR8: Population density (people per square kilometer)
9. XR9: City layout indicator
The first one represents the planning component of Emergency Response and Recovery
Capability; the next five, the resources component; and the last three, the mobility and access.
4.3.5.1. Planning
“Years since the first emergency response plan was developed” or “number of revisions
to the emergency plan” are possible quantitative indicators of planning. Neither, however,
convey any of the crucial information about how comprehensive the plan is and how effectively
it will be used in the event of an earthquake. The “planning indicator” (XR1) was developed to
provide a more comprehensive representation of emergency response and recovery planning.
The three values of the “planning indicator” are defined in Table 4.11.
Table 4.11. Planning indicator definition

Planning level XR1 Category description
1 Minimal. Planning is inadequate AND inactive.
3 Good. Planning is inadequate OR inactive.
4 Excellent. Planning is adequate AND active.
In Table 4.11, “adequate” means that the city’s planning situation meets six or more of the
following criteria: (1) is organizational, i.e., establishes the roles and responsibilities of all
involved parties;, (2) is operational, i.e., includes procedures to explain what to do, how, and
when; (3) addresses all main aspects of emergency response and recovery, i.e., emergency
management system, communications, financial arrangements, legislation, damage assessment,
search-and-rescue, secondary hazard control, health care, mass care, shelter, clean-up,
restoration of services (see Section 3.3.5); (4) exists for the region as a whole, for local
governments, and/or for specific organizations and institutions;

(5) includes a revision completed within the last ten years; (6) addresses earthquake hazard
specifically; (7) addresses particular characteristics of local context; (8) is well-integrated (a)
through hierarchical levels, (b) across agencies, organizations, and departments, and (c) with
daily operations and on-going development goals; (9) is based on the city’s actual experience in
a major event in the last twenty years. “Active” means that at least one of the following is true:
(1) plan(s) is practiced regularly through training exercises; (2) plan(s) has been used in a real
event in the last twenty years; (3) all involved parties are familiar with the contents of the plan(s)
and are trained in its use.
The “planning indicator” can be evaluated most accurately through consultation with a
city’s local planning experts. Although its evaluation requires a subjective assessment and
allows only three possible values, the “planning indicator” offers a level of comprehensiveness
not possible with available completely objective indicators.
4.3.5.2. Resources Available Post-earthquake
The indicators “per capita Gross Domestic Product, in constant 1990 US dollars” (XR2)
and “average annual real growth in per capita GDP during previous ten years” (XR3) represent
the overall financial resources available for post-earthquake emergency response and recovery.
Rather than estimating the actual value of available resources from each possible source (e.g.,
local and national government, earthquake insurance, private savings) and summing them, XR2
and XR3 simply convey how well the city’s economy is doing in general, and assume that that
information reflects the city’s ability to finance an emergency response and recovery effort. As
discussed in Section 4.3.2.3, “per capita GDP” is the best available representation of a city’s
economic power. “Average annual growth rate” is added to convey the economy’s recent
performance. It indicates whether the economy is currently growing or shrinking. An economy
that has been shrinking in recent years will have become lean, and thus will be in a worse
position to finance response and recovery than an economy that has been growing in recent
years. “Per capita Gross Domestic Product, in constant 1990 US dollars” is calculated just as

in XE2 and XE6. To calculate “average annual real growth in per capita GDP during previous ten
years,” compute the annual change in per capita GDP for each of the previous ten years, and
take the average of those ten values.
To evaluate the equipment and facility resources available for emergency response and
recovery directly, it would be necessary to count the number of each type of equipment and
facility that could be useful. The sample analysis addresses only two types of equipment and
facility resources—housing and hospitals. The “housing vacancy rate” (XR4) conveys the
availability of the first; “number of hospitals per 100,000 residents” (XR5), the second. Several
experts (e.g., Comerio 1996; West 1996; and Eisner 1996) have indicated the importance of
the housing vacancy rate at the time of an earthquake in helping to determine the city’s
emergency response and recovery capability.
The values of XR4 and XR5 can be estimated from censuses and country statistical yearbook
data, but care should be taken to ensure that both public and private hospitals are included,
since they may be tabulated by different organizations and reported separately. Data
concerning other equipment and facilities may not be as easily or consistently available.
As with equipment and facilities, there are many types of trained manpower that could
be counted, but only one is included in the sample analysis—physicians. The possibility of
including the number of firefighters and police officers was explored as well, but the data were
too fragmented to be included (see discussion Section 6.1). Data on the “number of physicians
per 100,000 residents” (XR6) are generally reported together with the hospital data.
The resource indicators (XR2 to XR6) convey information about some of the resources
necessary for an emergency response and recovery effort, but neglect many other types of
resources. Financial resources from specific sources (e.g., insurance coverage for individual
homeowners or businesses, residents' personal savings, businesses, relief organizations like the
Red Cross, or foreign sources) are not included explicitly. No equipment resources (e.g.,
mobile homes for temporary housing, search-and-rescue equipment), no facility resources other

than hospitals and housing (e.g., police and fire stations, health centers), and no trained
manpower resources other than physicians (e.g., police officers, firefighters, nurses) are
represented directly. Another omission from the set of resource indicators is that they do not
address the availability of potential resources. Hospitals will be valuable resources if they are
functional following an earthquake, and thus are able to treat the injured. However, if a hospital
is damaged, it may not only forfeit its role as a resource, but instead become a liability. Finally,
resources do not necessarily have to be local, and the sample analysis resource indicators do
not indicate if a city will be able to acquire the necessary resources from outside sources.
4.3.5.3. Mobility and Access Post-earthquake
The degree of damage to the physical infrastructure will greatly influence the mobility
within a city and access to a city following an earthquake. The vulnerability indicators convey
this aspect of the Mobility and Access factor. Other determinants of post-earthquake mobility
and access are represented by the last three emergency response and recovery indicators (XR7
to XR9).
The “extreme weather indicator” (XR7) represents the potential for severe post-
earthquake weather (i.e., extreme hot or cold) to impede a response effort, and to increase the
hardship for victims who are left homeless. Originally, XR7 was going to be defined using a
direct measure of the likelihood of severe weather, the average number of days per year in
which the temperature either reached at least 90°F or dropped to 32°F or below.
Unfortunately, those data are not reported regularly for most cities outside of the USA.
However, the National Climatic Data Center does compile and report the monthly average of
daily minimum and maximum temperatures for all major cities (NCDC 1997). The “extreme
weather indicator” was defined so that it is a function of the available data, but is well-correlated
with the more direct measure that was originally proposed. The following expression defines the
“extreme weather indicator”:

Dec. Dec.
XR7 = ∑ (Tmax,m - Th) + ∑ (Tc - Tmin, m), (4.9)

m= Jan. m= Jan.
where Tmax,m and Tmin,m are the average daily maximum and minimum temperatures in month m,
respectively; and Th and Tc are the hot and cold threshold temperature constants, 85°F and
50°F, respectively. The quantities (Tmax,m - Th) and (Tc - Tmin, m) are displayed graphically in
Figure 4.6. Using data for twenty-six U.S. cities, the values of Th and Tc were determined to
achieve the highest possible correlation coefficient (ρ=0.892) between XR7 and the average
number of days per year in which the maximum temperature is at least 90°F or the minimum
temperature is no more than 32°F.
“Population density” (XR8) represents the idea that a densely populated city suffers from
congestion in the transportation network, congestion that could impede the ability of emergency
response workers to transport victims and resources through the city. The indicator XR8 is
evaluated exactly the same way that XH6 and XV4 are.
A natural feature, such as a bay or mountains may control the physical spread of a city,
resulting in an irregularly shaped city plan. Such an irregular shape may make it more difficult to
move within a city. For example, the effort of response workers in the greater San Francisco
metropolitan area, which surrounds a bay, would be hindered by the difficulty in crossing the
bay as they move from one neighborhood to another. The “city layout indicator” (XR9) conveys
whether or not a city’s emergency response and recovery effort might be made more difficult by
this circumstance. It is evaluated as a simple binary function: XR9 = 1 if a city exhibits such an
irregular layout, otherwise XR9 = 0. These three indicators do not represent possible effects of
the city topography or remoteness from other cities on post-earthquake mobility and access.

Monthly Average of Daily Maximum Temperature Tmax,Oct.-T h
95
Average Daily Maximum Temp.
85
[degrees Farenheit]
75 Boston
Jakarta
65 San Francisco
Santiago
55 St. Louis
Th
45
35
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month
Monthly Average of Daily Minimum Temperature

Average Daily Minimum Temp.
80
70
[degrees Farenheit]
60
Boston
50 Jakarta
Tc -Tmin,Aug. San Francisco
40 Santiago
St. Louis
30 Tc
20
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month
Figure 4.6. Graphical representation of “extreme weather indicator.”

Sources: Ruffner and Bair (1987) and NCDC (1997).

Chapter Five
Mathematical Combination
5. Introduction
Once all the indicators that will be included in the Earthquake Disaster Risk Index have
been selected, they must be combined mathematically to create the composite EDRI. Table 5.1
shows a schematic example of a table of indicator values for a sample of cities. The EDRI relies
on the assumption that comparing the set of values associated with City A to those associated
with City B should convey some information about the relative earthquake disaster risk of the
two cities. The task of the mathematical combination is to combine the set of values for each
city into a single number that captures this information. Determining the method of combination
constitutes the last step in defining the term earthquake disaster risk. Section 2.1 defined
earthquake disaster risk by identifying the factors that contribute to it. To understand the
concept more fully, the way the factors interact and their relative importance must be established
as well. This information is incorporated into the EDRI through the method by which the
indicators are mathematically combined.
To convey information about the determinants of earthquake disaster risk for each city,
separate composite indexes are calculated for the five main factors that contribute to the risk—
Hazard, Exposure, Vulnerability, External Context, and Emergency Response and Recovery
Capability. All the indicators that relate to Hazard are combined into the Hazard index, all the
indicators that relate to Exposure are combined into the Exposure index, and likewise for the
other three factors. The EDRI is a composite index of the five
Chapter Five: Mathematical Combination 107

Table 5.1. Schematic indicator-city table
Cities
Indicator A B ...j... Y Z
Ind. 1
City B values
City Z values
City A values
City Y values
Ind. 2
… xij
Ind. i
…
Ind. n
EDRI EDRIA EDRIB EDRI Y EDRI Z
main factor indexes. The following discussion of methods of combination relates to the
construction of the five main factor indexes, as well as the EDRI. Every reference to combining
indicators into the factor indexes applies equally to the combination of factors into the EDRI,
and vice versa.
This chapter explores the many options for combination that were considered,
explaining the advantages and disadvantages of each alternative, presenting the method that was
ultimately selected for the sample analysis, and demonstrating the reasoning behind the choice.
Four general approaches are considered: no combination, graphical methods, nonlinear
combination, and linear combination. Once it is established that the sample analysis EDRI will
be a linear combination, several alternative ways to perform each of the two main steps in
creating a linear combination—scaling and weighting—are explored. Finally, the chapter
summarizes the mathematical equations needed to evaluate the EDRI.
The method of combination chosen for the sample analysis represents one reasonable
approach. There are other possibilities that are equally correct mathematically, and there is no
way to prove that one method is better than another. Instead, the criteria by which the method
of combination will be judged must be explicitly established up front. Then the final approach
can be selected from the alternatives based on its ability to meet those criteria. The method of
mathematical combination should:
108 Chapter Five: Mathematical Combination

1. Be able to achieve the goals of the EDRI, i.e.,
(a) Convey the relative overall earthquake disaster risk (as defined in Section 2.1)
at a point in time.
(b) Convey the relative contribution1 of each component factor to a city’s overall
earthquake disaster risk.
(c) Convey the change in a city’s relative overall earthquake disaster risk over time.
(d) Be straightforward to develop, evaluate, and interpret.
2. Not require measuring earthquake disaster risk directly.
3. Not be sensitive to the values of just a few cities or indicators.
4. Be objective to the extent possible.
Ideally, the method of combination would not be sample-specific either, i.e., it would
not require a sample of data, and it would not produce a different index if a different sample of
data was used. Sample-specificity presents a disadvantage because it increases sensitivity to
data quality and sample city selection, restricts the generality of the final index, requires that the
entire analysis be repeated if a new city is added to the sample, and requires adjustments to
track the EDRI over time. In the case of the EDRI, however, these disadvantages are not
prohibitive. Ultimately, the EDRI should be evaluated for a single sample that will remain
roughly constant over time, i.e., the set of all the world’s largest cities. Since the combination
can be calculated easily using spreadsheet software, reevaluating the results is easy to do. A
principal goal of the EDRI is to measure the earthquake disaster risk of cities relative to other
cities, not necessarily to some hypothetical “best” or “worst” situation, so it is appropriate to
use a technique that relies on evaluating the risk with respect to other sample cities, i.e., a
sample-specific one.
5.1. General approach

Four possible general approaches to the index combination were considered:
(1) no combination, (2) graphical methods, (3) nonlinear combination, and (4) linear
combination. This section discusses the advantages and disadvantages of each, explaining why
the linear combination was ultimately selected for the sample analysis.
5.1.1. No combination
The possibility of not combining the indicators into a single composite index was
explored. It is tempting to avoid the technical difficulties associated with the mathematical
combination, and simply present the components of the concept being measured, allowing each
user to combine them as he deems appropriate. For all the reasons discussed in Section 2.5,
however, the usefulness of a single summary EDRI outweighs the desire to avoid the challenges
associated with combining indicators. As long as the assumptions and techniques that guide the
combination are explicit and clear, the user can interpret the combination based on his belief in
the appropriateness of the approach. Furthermore, since the indicators that comprise the EDRI
are presented as well as the EDRI, the user can always refer to the indicator values themselves,
and disregard the final EDRI if he wishes.
5.1.2. Graphical methods
Two graphical methods developed by Wroclaw could be used for comparing the
earthquake disaster risk of different cities (Hellwig 1972a, 1972b). In applying the first one to
this project, the “distance from the ideal” approach, each city would be represented by a point
Pi in n-space with coordinates (x1i', x2i', …, xni') equal to the scaled values of the n indicators
(Fig. 5.1). Point Po, the ideal, would have coordinates (x1o', x2o', …, xno') such that xso equals
the minimum value of the indicator s over all cities (or the maximum value if an increase in s
would result in a decrease in earthquake disaster risk). The value of the EDRI for city Pi would
1
Refers to the city-specific relative contribution (Section 5.1.4).

be computed as cio/co, where cio is the distance between Pi and Po, and co equals the mean plus
two standard deviations of the cio for all cities (i.e., c o = co + 2sc o ). Each city’s EDRI value
would represent the distance between the city under consideration and the ‘ideal” city, with
respect to earthquake disaster risk. The value of the EDRI would always be positive, and
almost always less than one.
x2
x2i Pi
cio
x2o Po
x1
x1o x1i
Figure 5.1. “Distance from ideal” graphical method of combination
The second graphical method would divide cities into groups that exhibit similar
indicator values (i.e., similar levels of earthquake disaster risk). It also involves locating each
city as a point Pi in n-space, and evaluating the distances between them (Fig. 5.2). The distance
between points Pi and Pj is denoted cij. The value of ci, the minimum cij for all j, measures the
resemblance between Pi and its most similar city, Pj. The cities are grouped so that each city Pi
is in a group with its most similar city Pj, and cities within the same group are never more than
the critical distance c(+) apart, where c(+) equals the mean plus two standard deviations of the ci
for all cities (i.e., c (+) = ci + 2s ci ).
Wroclaw’s techniques offer ways to rank cities according to their values for the set of
indicators identified in Chapter 4. The procedures are simple, objective, and intuitively based.
Nevertheless, neither presents an appropriate approach for the EDRI because neither conveys

any information about the relative contributions of the different factors to the final ranking, and
that is one of the main goals of the EDRI. By using the
x2
ci Pj
Pi
x1
Figure 5.2. “Similar groups” graphical method of combination
distances between points, each implicitly weighs all the indicators equally. Furthermore, results
from the “distance from the ideal” method are overly sensitive to the “ideal” indicator values,
and the “similar groups” method does not distinguish among cities within the broad groupings.
Depending on the sample, the second technique may result in too few or too many groups, and
it may not be obvious how to label the groups.
5.1.3. Nonlinear combination
Perhaps the way to capture the concept of earthquake disaster risk most realistically is
to combine the indicator values using a nonlinear function. A nonlinear function can represent
the interactions that exist among the factors, and between the factors and the weights. For
example, some argue that the relative importance (i.e., the weight) of the Emergency Response
and Recovery Capability factor depends on the values of the Hazard and Vulnerability factors.
The emergency response and recovery gains importance when there is more severe damage to
the physical infrastructure. Unfortunately, while these interactions seem intuitively reasonable,
there has been little systematic study either to prove that they exist, or to determine the exact
form of the interactions. The function describing the variation of the weight of Emergency

Response and Recovery Capability versus the Hazard and Vulnerability, for example, remains
unknown. With the current level of knowledge, the cost of introducing extra uncertainty and
complexity into the model outweighs any possible benefits. The “perceived” exactness of the
model would be inconsistent with the current understanding and quality of data. Finally,
depending on the form of a nonlinear model, the relative contributions of each factor may not be
clear.
5.1.4. Linear combination
If the EDRI is constructed as a linear combination, the EDRI value for city j would take
the form:
EDRIj = ∑i cij, (5.1)
where cij is the contribution of indicator i to the earthquake disaster risk of city j relative to other
cities in the sample. Let xij in Table 5.1 be the value of the indicator i for city j. The value of the
contribution cij depends on two comparisons: (1) how the value xij compares to the values of
the same indicator for other cities, and (2) how the importance of indicator i to earthquake
disaster risk compares to the importance of other indicators. In general, the contribution is
computed in two steps—scaling and weighting. Scaling performs the first comparison by
transforming each value, xij, into a corresponding scaled value, xij', that interprets the value
relative to other values of the same indicator. Scaling also converts the indicators into
compatible units of measurement. Weighting performs the second comparison by multiplying
the scaled values of each indicator by a constant, unitless coefficient whose magnitude
represents the importance of the indicator relative to other indicators. The form of the EDRI
becomes:
EDRI = ∑i wixi', (5.2)
where xi' is the scaled value of indicator i, and wi is the constant weight that represents the
relative importance of indicator i.

There are two types of relative contributions associated with the EDRI: definition-
related and city-specific. Weights describe the definition-related relative contributions of the
five main factors to the EDRI. They are the same for all cities, because they relate to the
concept of earthquake disaster risk, not the characteristics of a particular city. The assumptions
about the definition of earthquake disaster risk (i.e., the relative importance of each of the five
main factors to the concept) are isolated and made explicit in the weights. The city-specific
relative contributions are captured in the relative factor values (H, E, V, C, and R). These are
the relative contributions to which the second objective of the EDRI refers (Section 2.2). They
describe the characteristics of each city. The city-specific relative contribution of one factor is
the ratio of the value of that factor to the sum of the values of all factors. For example, the
relative contribution of Hazard to City j’s risk is:
Contribution of Hj = Hj/(Hj + Ej + Vj + Cj + Rj). (5.3)
The interpretation of an xij value depends on the indicator to which it refers. An xij
value of one means something different for the indicators “housing vacancy rate” and “per capita
Gross Domestic Product in constant 1990 U.S. dollars,” because the magnitudes characteristic
of these two indicators are different. The values for “housing vacancy rate” will range from zero
to one, while those for “per capita GDP” will be on the order of thousands. If the two values
were added directly, the “per capita GDP” would dominate the “housing vacancy rate” in every
case. Furthermore, a city-to-city difference of say, one-tenth (i.e., xiA- xiB = 0.1), takes on
different meanings for the two indicators because their typical dispersions are different. For
“housing vacancy rate” a difference of ten percent is very significant, while for “per capita GDP
in constant 1990 U.S. dollars,” a difference of ten cents is negligible. Scaling aims to make the
interpretation of the xij values and the city-to-city differences consistent among indicators by
transforming them so that all indicators take on values that are similar in magnitude and
dispersion. Magnitude is controlled by forcing all indicators to have either the same mean or the
same range of values, and dispersion is controlled by forcing all indicators to have either the
same standard deviation or the same range of values. Most scaling techniques establish one or
two benchmarks and compare the indicator value for each city to those benchmarks, and

thereby to the values for other cities. Scaling is strictly a mathematical operation. It does not
attempt to convey how each indicator relates to the overall earthquake disaster risk. Scaling
also converts all component indicators into compatible units of measurement. The “housing
vacancy rate” and “per capita GDP” values cannot be added directly, because the former is
measured in percentage points; the latter in monetary units. Only after scaling has transformed
all indicators so that have compatible magnitudes, dispersions, and units can the various
indicator values be added.
Once all the indicators have been scaled, their associated weights must be determined.
As shown in Figures 5.3a and 5.3b, scaling transforms an indicator’s unscaled value to the
associated scaled value, and weighting relates the scaled value to cij, the contribution it makes
to the relative earthquake disaster risk of city j. The weight of each indicator shows how
important it is to the overall earthquake disaster risk. Determining the weights is the final step in
defining earthquake disaster risk. Section 2.1 defined earthquake disaster risk by identifying
the factors that contribute to it. To complete the understanding of the concept of earthquake
disaster risk, the relative importance of each contributing factor must be established as well.
Weighting achieves that task. Much research has been done to demonstrate how each factor
plays a role in earthquake disaster risk. Unfortunately, the wealth of knowledge about
earthquake disasters is spread among many people, each of whom is an expert in a different
facet of the problem. The EDRI tries to capture the collective knowledge of all experts to
define earthquake disaster risk as it is conceived by the group as a whole. Weights are correct
only inasmuch as they achieve that goal.
The linear combination satisfies all of the four main goals of the EDRI. The final EDRI
value for each city offers a quantitative ranking of the relative overall risk. The ratio of each
factor to the sum of the values of all factors clearly and explicitly represents that factor’s city-
specific relative contribution to the overall earthquake disaster risk. The linear combination is
simple to evaluate and interpret, yet provides meaningful results. Almost all previously
developed composite indexes (e.g., Human Development Index, the Economic

Freedom Index, Competitiveness Index, Consumer Price Index) have relied on the
linear combination approach.
Scaled values [x']
xij'
xij Unscaled data values [x]
a. Scaling functions
Contribution to EDR [c]
slope = wi
cij
xij' Scaled values [x']
b. Weighting functions
cij
xij
Unscaled data values [x]

c. Contribution functions
Figure 5.3. Graphical comparison of scaling, weighting, and contribution functions
The principal disadvantage of expressing the EDRI as a linear combination is that it

implies that no interactions exist among the indicators, or between the indicators and the
weights. As discussed in Section 5.1.3, this assumption is undoubtedly incorrect. Since the
exact form of those interactions is unclear, however, it is preferable to define the EDRI as a
linear combination than to employ a highly uncertain nonlinear form. Furthermore, an EDRI that
is essentially a linear combination can incorporate some nonlinearities through the use of
indicators that are derived, nonlinear functions of the indicators for which data are gathered.
The derived indicators that are used in the sample analysis include the “economic external
context indicator,” “exp(MMI with a 50-year return period),” and “seismic code indicator,” whose formulas
are discussed in Chapter 4.
Another shortcoming of the linear combination becomes evident when examining

extreme values. Intuitively, if a city has absolutely no hazard (i.e., zero chance of experiencing
any ground shaking), its earthquake disaster risk should be zero as well. With the linear
combination, however, if the city had nonzero values for any of the other four main factors, its
risk would not be zero. Although this presents a hypothetical problem, in reality, no city could
have absolutely zero hazard (or exposure or vulnerability), so this situation is not realistic.
5.1.5. Sample analysis solution
The general approach adopted in the sample analysis is a linear combination that
includes several derived nonlinear indicators. For all the reasons discussed in Section 5.1.4, this
approach provides the best available technique for mathematically combining the indicators into
the composite EDRI. A summary of all the equations necessary to evaluate the EDRI follows:
EDRI = wHH + wEE + wVV + wCC + wRR (5.4a)

H = wH1x'H1 + wH2x'H2 + wH3x'H3 + wH4x'H4 + wH5x'H5 + wH6x'H6 + wH7x'H7 (5.4b)
E = wE1x'E1 + wE2x'E2 + wE3x'E3 + wE4x'E4 + wE5x'E5 + wE6x'E6 (5.4c)
V = wV1x'V1 + wV2x'V2 + wV3x'V3 + wV4x'V4 + wV5x'V5 + wV6x'V6 (5.4d)
C = wC1x'C1 + wC2x'C2 + wC3x'C3 (5.4e)
R = wR1x'R1 + wR2x'R2 + wR3x'R3 + wR4x'R4 + wR5x'R5 + wR6x'R6 + wR7x'R7 +
wR8x'R8 + wR9x'R9 (5.4f)
where H, E, V, C, and R are the values of the Hazard, Exposure, Vulnerability, External
Context, and Emergency Response and Recovery Capability indexes, respectively; x'i refer to
the scaled values of the indicators listed in Figure 4.1; and wi are the weights associated with
each indicator.
5.2. Scaling
Seven possible techniques for scaling the indicators were considered before the final
method, scaling with respect to the mean minus two standard deviations (see Sections 5.2.7 and
5.2.8), was selected for use in the sample analysis. Each of the following seven subsections
describes one option, its advantages and disadvantages. The alternatives are summarized in
Table 5.2 and Figure 5.4.
5.2.1. Contribution function
This first scaling option actually combines the tasks of scaling and weighting into a single
step. The technique involves determining a function, ci(xij), for each indicator i, that relates the
unscaled indicator value xij directly to cij, the contribution that value makes to the earthquake
disaster risk of city j relative to other cities in the sample
(Fig. 5.3c). Figure 5.5 presents a hypothetical example of one such function for the indicator
“population.” Comparing points A and B indicates that City A, with a population of xpop,A, has
a higher risk with respect to the population, than City B, with a population of xpop,B. The
indicator “population” would contribute (cpop,A - cpop,B) more to the EDRI value for City A than

for City B. Once a contribution function has been developed for each indicator listed in
Chapter 4, the EDRIj for city j is evaluated as the sum of the contribution of all the indicators,
i.e., EDRIj = ∑i ci(xij).

Table 5.2. Comparison of scaling techniques
Technique Advantages Disadvantages Function Range of scaled
values
Contributions • Can incorporate nonlinearities • Combines scaling and weighting user-specified user-specified
functions • Can’t determine city-specific relative
contributions
Min. and max. • Simple to calculate, interpret • Dependent on two values (x ij − minob i ) 0 to 1
observed • Objective • Need adjustment to track over time
• Sample-specific (maxob i − minob i )
Min. and max. • Easy to track over time • Dependent on two values (x ij − minpsi ) 0 to 1
possible • Simple to calculate, interpret • Difficult to determine min. and max. possible
• Not sample-specific • Subjective (maxps i − minps i )
Base city and year • Simple to calculate, interpret • Dependent on one value x ij --∞ to +∞
• Objective • Sample-specific
• Easy to track over time • Implicit weighting x i,base
• Dependent on arbitrary choice of base city
Consistent 0 to 10 • Easy to interpret • Subjective integers 0 to 10 0 to 10
scale • Not sample-specific • Low resolution
Mean • Objective • Results in negative scaled values x ij − x i -∞ to +∞
• Easy to calculate • Sample-specific
• Need adjustment to track over time si
Mean minus two • Objective • Sample-specific x ij − ( x i − 2s i ) -∞ to +∞
standard • Easy to calculate • Need adjustment to track over time
deviations* si
* Chosen for sample analysis.
1.0

1.0
Scaled [x']
0.5 0.5
0.0 0.0
Minobs Maxobs
Unscaled [x] Unscaled [x]
a. Contribution function b. Scaling: Min. and max. observed values
1.0
Scaled [x']
Scaled [x']
0.5
11.0
0.0
Minposs Unscaled [x] Maxposs Base city Unscaled [x]
c. Scaling: Min. and max. possible values d. Scaling: Base city
10
Scaled [x']
Scaled [x']
5 0
Mean
0
0 Unscaled [x] Unscaled [x]
e. Scaling: Zero to ten f. Scaling: Mean
Scaled [x']
Inversely proportional
inds.
Figure 5.4. Graphical comparison of
Directly
proportional
scaling techniques: (a) contribution
2
2.0
inds. function, (b) min. and max. observed,
(c) min. and max. possible, (d) base city,
(e) consistent zero to ten scale, (f) mean,
(g) mean minus two standard deviations.
Mean
Unscaled [x]
g. Scaling: Mean minus two stan. deviations

cpop,A A
cpop,B B
xpop,B xpop,A Population
Figure 5.5. Hypothetical contribution function for the indicator “population”
The principal advantage of this technique is that it does not require the contribution of
each factor be a linear function of the scaled values xij' (i.e., the weights do not have to be
constant). While interactions among different factors (as in the example in Section 5.1.3) still
cannot be addressed, this technique would allow the inclusion of any nonlinearities in the
relationship of a single factor to the overall earthquake disaster risk. This technique avoids the
task of assessing weight values explicitly.
Three major disadvantages prevent this approach from being the most attractive choice
for the EDRI. First, the contribution functions must be determined by a very difficult subjective
assessment. As discussed in Section 5.1.3, no real evidence exists to indicate what would be
the correct form of the relationships between each factor and the overall risk, much less
between an indicator that represents a factor and the overall risk. As with the nonlinear
combination, the added complexity may introduce more uncertainty than benefit. Second,
without separate weights, there is no clear way to separate the city-specific and definition-
related relative contributions. It is impossible to determine the city-specific relative contributions
of the various component factors to the overall risk, and that is one of the main objectives of the
EDRI (Chapter 5 introduction; Criterion 1b).
Third, the contribution functions essentially perform the tasks of both scaling (i.e.,
making magnitudes, dispersions, and units consistent across indicators) and weighting (i.e.,

conveying the relative importance of each indicator to the overall risk). Merging the scaling and
weighting into one step makes it difficult to know if the contribution function was determined
because of scaling concerns, weighting concerns, or a combination of both. It is preferable to
separate scaling and weighting. Since scaling is exclusively a mathematical exercise, all the
subjectivity about the indicators’ relative importance can be isolated in the weights. With the
more transparent approach of separating scaling and weighting, the user can easily decide if he
agrees with the subjective assessments included in the index combination, and alter them if not.
5.2.2. Minimum and maximum observed
Scaling may be achieved by normalizing with respect to the maximum and minimum
values of that indicator that are observed in the sample of cities under consideration.
Mathematically,
(xij − minobsi )
xij ' = (5.5)
(maxobs i − minobsi )
where xij is the unscaled value for indicator i and city j, xij' is the corresponding scaled value,
and maxobsi and minobsi are the maximum and minimum values (over all cities j) of indicator i
observed in the sample, respectively (Fig. 5.4b). The maximum and minimum observed values
serve as the benchmarks in this scaling technique. Each city is evaluated relative to the cities
that exhibit the highest and lowest risk with respect to each indicator. This scaling function
makes all indicators exhibit similar magnitudes and dispersions by forcing them all to take on
values between zero and one.
Because the maximum and minimum observed values will be different for different time
periods and different city samples, this method does not allow consistent comparison over time
or across samples. To overcome this difficulty and meet the criterion of enabling the EDRI to
be tracked over time, a base time period could be established, and the maximum and minimum
observed values for that time period would be used to evaluate the EDRI in every subsequent
time period.

This alternative, which Briguglio (1995) used in the development of an index of the
economic vulnerability of small island developing states, offers an objective, sample-specific,
easily calculated, and easily interpreted way to scale the indicator values, but it suffers a
significant disadvantage. The results of this scaling technique are completely dependent on the
maximum and minimum observed values, and therefore are dangerously sensitive to the quality
of data for those two values. Outlier values can skew the results seriously.
5.2.3. Minimum and maximum possible
The third approach, which is used in the Human Development Index (UNDP 1990-95)
and the Physical Quality of Life Index (Morris 1979), is similar to the previous one, except the
maximum and minimum values that are considered possible are used instead of the maximum
and minimum values observed in the sample of cities. Mathematically,
(xij − minposs i )
xij ' = (5.6)
(maxpossi − minpossi )
and maxpossi and minpossi are the maximum and minimum values (over all cities j) of indicator i
that are considered possible (Fig. 5.4c). As in the previous method, this approach makes all
indicators exhibit similar magnitudes and dispersions by forcing them all to take on values
between zero and one, and the maximum and minimum possible values serve as the
benchmarks.
Unlike the similar technique with the maximum and minimum observed values, this one
does not have a problem tracking the change in EDRI over time. It should be possible to assess
the maximum and minimum possible values so that they will be large and small enough,
respectively, to accommodate the indicator values as they change over time. This characteristic
points to one of the method’s main difficulties—how to assess the maximum and minimum
possible values. Should they be the maximum and minimum possible in today’s world? Over

the next ten years? Forever? The maximum and minimum theoretically possible, or the
maximum and minimum that can be practically expected? For example, theoretically, the
“housing vacancy rate” can range from zero to one, but practically, it should never exceed, say
0.2. The results are completely dependent on, and therefore, dangerously sensitive to these two
assessed values.
5.2.4. Base city and year
Another scaling possibility, used in the development of an index of country size (Jalan
1982), is to render every indicator unitless by normalizing it with respect to a single base city
and year, i.e.,
xij
xij ' = (5.7)
x i,base
and xi,base is the unscaled value for indicator i and the base city (Fig. 5.4d). In this case, there is
a single benchmark, and there is no difficulty tracking EDRI over time because the benchmark
does not vary with the time period. The scaled values can take on any nonnegative value. This
alternative is objective, sample-specific, and easy to use. It also produces perhaps the most
easily interpreted results because all cities are ranked relative to a single, real city, not a
hypothetical “riskiest” or “least risky” city.
While the base city scaling causes the magnitudes of the different indicators to gather
roughly around a value of one, it does not control magnitudes as restrictively as the other scaling
options do, and it does not make all the indicators have similar dispersion. In failing to control
dispersion, this technique introduces an implicit weighting into the mathematical combination.
Consider two indicators. One has values spread between one and ten; the other between one
and ten thousand. Because the dispersion characteristic of the first indicator is smaller than that
of the second, a city-to-city difference of, say one, is more significant for the first indicator than
the second. The base city scaling neglects this fact. Suppose the base city has a value of one

for each of the two indicators. After being scaled with respect to the base city, the range of
values for the first and second indicators will remain one to ten and one to ten thousand,
respectively. That is, for the other cities in the sample, the first indicator will have much smaller
values than the second, in effect reducing the weight of the first indicator relative to the second.
Section 7.3 presents an illustration of this implicit weighting.
There are three principal drawbacks of scaling with respect to a base city and year.
First, its inability to control dispersion introduces an undesirable implicit weighting effect, as
discussed. Second, the effect of this technique is very sensitive to the essentially arbitrary
choice of base city. If a base city is chosen that has a low value for indicator X1 (relative to
other cities) and a high value for indicator X2 (relative to other cities), the other cities will take
on scaled values that are high for X1, and low for X2. When the indicator values are added to
compute the composite index, in effect, X1 becomes more important to the other cities than X2.
On the other hand, if a base city is selected that has a relatively high value for indicator X1 and a
relatively low value for indicator X2, the reverse will be true. The indicator X1 becomes less
important to the other cities than X2. Third, the results are completely dependent on the data
availability and quality for that base city.
5.2.5. Consistent zero to ten scale
The Economic Freedom Index (Johnson and Sheehy 1996) scales the component
indicators by converting the natural measurement scale of each to a unitless scale from zero to
ten, with ten being associated with the most freedom (Fig. 5.4e). Continuous indicators are
scaled by discretizing the continuous scale into eleven parts, so that when the base year sample
of countries (for the EDRI, it would be cities) are assigned values, there are an equal number of
countries ranked with each value from zero to ten. Binary indicators are assigned possible
values of zero and ten. For indicators that are measured by some classification scheme with
more than two and less than infinite number of possible values, each possible state is assigned a

number from zero to ten, again so that the higher values correspond to more freedom. The
EDRI indicators could be scaled similarly.
This technique provides easily understandable scaled values, and achieves all the main
criteria of the mathematical combination listed in the introduction to Chapter 5, but it has two
disadvantages. First, it depends on the difficult, somewhat subjective task of determining the
scale for each indicator individually. Care would be necessary to ensure that the scale
conversions are determined solely on the basis of scaling concerns, without consideration of
weighting issues (i.e., without considering the relative importance of the indicators). Second,
converting all indicators to a zero to ten scale reduces the resolution of the results at the
beginning of the calculation, while the other more quantitative methods carry more numerical
precision through the calculations, rounding off the results at the conclusion of the analysis.
5.2.6. Mean
The scaling approach adopted in the World Competitiveness Index renders each
indicator unitless by scaling it with respect to the mean of a sample of cities, i.e.,
(x ij − x i )
xij ' = (5.8)
si
where xij and xij' are the unscaled and scaled values for indicator i and city j, respectively; and
x i and s i are the mean and standard deviation of a sample of cities for indicator i (Fig. 5.4f).
This objective, sample-specific technique achieves consistency of magnitude and dispersion by

making the mean of the scaled values equal to zero and the standard deviation equal to one, for
every indicator. It uses a single benchmark, the mean of the sample of cities. Scaled values
may range from negative to positive infinity. This method is as sensitive to the value of the mean,
as the methods discussed in Sections 5.2.2, 5.2.3, and 5.2.4 are to the maximum and minimum
observed values, the maximum and minimum possible values, and the base city value,
respectively. However, the benchmark used in this method is much more stable than those used

in the previously discussed approaches. The mean depends on the entire sample of data, not
just one or two values. This method directly evaluates each city relative to all others (via the
mean and standard deviation), while the others achieve the same comparison by evaluating each
city relative to the benchmarks, and by transitivity, relative to each other.
To allow consistent EDRI comparisons at different time periods, a base year should be
established, and the EDRI values for subsequent time periods should be evaluated using the
mean and standard deviation of the sample of cities in the base year. The most disturbing
disadvantage of scaling with respect to the mean, is that roughly half of the scaled values (and
perhaps some of the EDRI values) will be negative. While a negative value simply indicates that
the corresponding city falls below the mean for that indicator, it seems inappropriate to speak of
a city having a negative level of risk.
5.2.7. Mean minus two standard deviations
This final scaling option, the one selected for the sample analysis, is similar to the option
discussed in 5.2.6, but with a variation designed to relieve the undesirable characteristic of
producing negative scaled values. Mathematically,
x ij − (x i − 2si ) x ij − x i
x ij ' = = +2 (5.9)
si si
where xij' and xij are the unscaled and scaled values for indicator i and city j, respectively; and
x i and s i are the mean and standard deviation of a sample of cities for indicator i (Fig. 5.4g).
This scaling function simply equals the scaling function of 5.2.6 translated upward two units.
Instead of producing a set of scaled values with a mean of zero and a standard deviation of one,
this scaling formula delivers a set of scaled values with a mean of two and a standard deviation
of one. The result of this simple transformation is a scaling technique that shares all the benefits
of the option discussed in 5.2.6, but will rarely produce negative values, thus making the results

more easily understandable. A common rule of thumb says that there is a 95% chance that
values will lie within two standard deviations of the mean (McClave and Dietrich 1988).
5.2.8. Inversely related indicators
All the discussions of scaling alternatives above assumed that the indicator value is
directly proportional to a city’s earthquake disaster risk. The scaling formulas presented are all
increasing functions. They must be adjusted to treat indicators that are inversely related to
earthquake disaster risk (e.g., “housing vacancy rate”). The following function is a variation of
the scaling function in Section 5.2.7 that is suitable for use with inversely related indicators:
- x ij + (x i + 2s i )
x ij ' = (5.10)
si
It was defined so that it produces the same mean as Equation 5.9, and a slope that is equal in
magnitude, but negative (Fig. 5.4g). Indicators scaled with the inverse form of the scaling
function will still have a mean of two and standard deviation of one, so they remain consistent
with the directly scaled indicators.
5.3. Weighting
No amount of clever mathematical manipulation will uncover the “correct” weights for
the EDRI, because no single correct set of weights exists a priori. Rather, depending on the
weights that are defined, the concept measured by the EDRI will vary. For example, if the
weight of the Hazard factor is taken as one, and all other weights are zero, earthquake disaster
risk is, in effect, defined as hazard. Experience and study indicate that there are other factors
that are important in earthquake disaster risk, however, so the hazard weight should be less than
one, and the weights of other factors should be greater than zero. Nevertheless, although there
is no one set of “correct” weights, certainly some weight values will help achieve the goals of the
EDRI more effectively than others. The “best” weights for the EDRI are those that most
accurately represent the relative contributions of the corresponding indicators to the overall

earthquake disaster risk, as it is defined in Section 2.1. The best technique for determining the
weight values is the one that produces the best weight values while meeting the criteria listed in
the introduction to Chapter 5. Three alternatives were considered for determining the weights
to be used in the EDRI linear combination—regression, principal components analysis, and
subjective assessments. In the sample analysis, the weights are normalized so that their sum
equals one. This normalization does not affect the results, but allows the weights to be
interpreted as the percentage by which the associated indicator contributes to earthquake
disaster risk.
5.3.1. Regression
Multiple regression is a technique by which a model is derived to predict the values of a

dependent random variable Y using the values of a set of independent variables X1,…, Xn.
Multiple linear regression models take the form:
yi = βo + β1x1 +...+βn x n ,
~ (5.11)
where ~
yi is the predicted value of Y, and x1,…, xn are the values of the n independent
variables. The quantity y i −~yi , where yi is the observed value of Y, describes the error
between the observed and the predicted values of the variable Y (Fig. 5.6a). Using the least
squares method, regression computes the constants β 0, …, β n so that the model exhibits the
smallest possible total error. Minimizing the sum of the squared errors ensures that the model
describes the line of closest fit.
If one considers earthquake disaster risk to be the dependent variable, Y, the indicators
listed in Chapter 4 to form the set of independent variables X1,…, Xn, and the weights w1,…,
wn to be the constants β 1, …, β n, then multiple linear regression seems to be a promising
technique for computing the weights. Using regression to determine the weights would ensure
that the EDRI provides the “best” model for predicting earthquake disaster risk. Unfortunately,
regression requires a set of data that includes values for the dependent variable, Y. Since the

dependent variable, earthquake disaster risk, cannot be measured directly, it is impossible to
use regression directly to determine the weights of the EDRI.
Dependent variable [Y]
~
yi
error
yi
Independent variable [X]

a. Regression analysis
X2 First principal
component
(x1 ,x2 )
X1
b. Principal components analysis
Figure 5.6. Graphical comparison of regression and principal components analysis
5.3.2. Principal components analysis
Principal components analysis (PCA) offers another statistically-based technique for

determining the weights in a composite index. While PCA can be applied to achieve several
other goals, one of its uses is to derive weights for a composite index based on the first principal
component of a set of variables. PCA was used in this way by Ram (1982) in developing a

Physical Quality of Life Index and a Basic Needs Fulfillment Index, and by Downes (1988) in
developing a Country Size Index.
Principal components analysis “searches for a few uncorrelated linear combinations of

the original variables that capture most of the information of the original variables” (Dunteman
1989). The weights of the combinations are “mathematically determined to maximize the
variation of the linear composite, or equivalently, to maximize the sum of the squared
correlations of the principal component with the original variables” (Dunteman 1989). The
linear combinations (i.e., principal components) are ordered so that the first principal component
has the largest variation, i.e., explains the highest percentage of the total variation in the original
variables; the second principal component explains the second highest percentage, and so forth.
The first principal component thus offers the single linear combination of the original variables
that best summarizes the information captured in those variables, and the variation it explains
provides a measure of how well it does so. In creating the EDRI then, using the first principal
component weights would result in the EDRI that best captures the information in the indicators.
Note, however, that while the first principal component will be the best single linear
combination, it may not be a complete summary of the information in the original variables. The
variation of each principal component measures how much information it contains.
Like regression, PCA can determine a line of closest fit to the values of a set of
independent variables. However, PCA uses a different criterion to determine the closest fit,
namely that the procedure should maximize the sum of squared correlations of the linear
combination with the original variables. Graphically, instead of defining the line of closest fit as
the one that minimizes the sum of squared distances between the observed values and predicted
values in the direction of the dependent variable, principal components defines it as the line
that minimizes the sum of squared distances between the observed values and their
perpendicular projections onto the largest principal component (Dunteman 1989) (Fig. 5.6b).

The principal component method consists of four steps (see example in Section
5.3.4.1). First, the correlation matrix is determined for the set of indicators being combined.
Second, the eigenvalues and eigenvectors of the correlation matrix are calculated. Third, each
eigenvector is multiplied by the square root of the associated eigenvalue to obtain the factor
loadings. Each principal component is defined as the linear combination of the indicators, with
weights equal to their factor loadings. The first principal component is the one corresponding to
the largest eigenvalue; the second principal component, to the second largest eigenvalue, and so
forth. Fourth, the eigenvalue divided by the number of indicators being combined conveys the
percentage of the total variation in the indicators that is explained by the associated principal
component. Depending on how much of the total variation it explains, the first principal
component weights may be used in the composite index of the indicators.
Principal components offers an objective, statistically-based, interpretable procedure for

determining the weights. It can also provide a measure of how well the final linear combination
represents the indicator values. Nevertheless, PCA is not appropriate for general use in the
EDRI because it is based on the assumption that all the indicators being combined are well-
correlated, and that is not the case for the EDRI (e.g., expected ground shaking is not related to
population). PCA was used, however, to help determine the weights for the Physical
Infrastructure Exposure factor because the indicators representing that factor are well-
correlated (Section 5.3.4.1).
5.3.3. Subjective assessments
There are two possible approaches to using subjective assessments to determine the
values of the weights in the EDRI combination formula. The first method, which is used in the
sample analysis, consists of an assessor (or assessors) simply assigning values for the weights
based on his (their) best judgment. The second asks the assessor(s) to assign ranks to the
overall EDRI for a sample of cities. Assuming those assessed EDRI values are “correct,” they

are used as the dependent variable data, and a regression analysis is then performed to
determine the weights for each component factor.
Obtaining subjective assessments is easy to do, but difficult to do well. The many issues
associated with obtaining accurate subjective assessments (i.e., assessments that actually convey
exactly the knowledge they attempt to capture) include determining who should make the
assessments, how they should be elicited, and how they should be used. The assessments could
be made by a single expert, many experts, experts in specific subject areas, or experts in
specific geographical locations. The assessments may be obtained using questionnaires or
interviews, and they may be aggregated in various ways, such as the Delphi method. Chhibber,
Apostolakis, and Okrent (1992) summarize the many issues involved in obtaining expert
opinion.
There are three principal disadvantages of both subjective assessment approaches.

First, they lack a statistical or other explicitly replicable basis. Second, the assessor may not
have the knowledge that the index developer hopes to elicit. Third, even if the assessor does
have the knowledge, he may not be able to convey it accurately, because biases often corrupt
subjective assessments. Tversky and Kahneman (1974) describe many of the biases to which
subjective assessments are vulnerable. While the subjective assessment approaches are not
ideal for these reasons, they constitute the only feasible option for the EDRI, and offer a
reasonable way to capture the current state of understanding among earthquake risk experts
about the relative importance of the various component factors to earthquake disaster risk. The
first subjective assessment variation, assessing the weight directly, is preferable to the second
because more focused assessments are generally easier to make. Furthermore, by performing
the regression analysis on the assessments, the second method obscures the impact of those
assessments, and requires that the entire regression analysis be repeated if any changes in the
assessments are made.

5.3.4. Sample analysis solution
For the reasons discussed in the previous sections, direct subjective assessments of the
weights were employed to determine all weight values, except those used in the Physical
Infrastructure Exposure. The Physical Infrastructure Exposure weights were determined with
the help of the principal component method (Section 5.3.4.1). While all other weights ultimately
were established by the author alone, an attempt was made to assess whether or not well-
known earthquake risk experts would agree with the weight values assessed by the author.
That effort consisted of a questionnaire survey, the methodology, shortcomings, and findings of
which are presented in Section 5.3.4.2. The final weight values used in the sample analysis are
summarized in Section 5.3.4.3.
5.3.4.1. Physical Infrastructure Exposure weights
The indicators that describe the Physical Infrastructure Exposure—“population,”

“number of housing units,” “per capita GDP,” and “land area”—are well-correlated, and
therefore are amenable to the principal component method of determining weights. The
technique was performed both on all four indicators using the Statistics of World Large Cities
data set, and on the first three indicators using the Human Settlements Statistics data set.
There are not enough data points to include area in the Settlements data set analysis. Both data
sets are described in Section 4.3.2.1. The resulting factor loadings are shown in Tables 5.3 and
5.4.
Table 5.3. Results of principal components analysis to determine Physical Infrastructure

Exposure weights (Statistics of World Large Cities data set)
Indicator Principal Component Factor Loadings
P.C. 1 P.C. 2 P.C. 3 P.C. 4
Population 0.83 0.25 0.38 0.32
Number of housing units 0.89 -0.24 0.11 -0.37
Per capita GDP 0.30 -0.85 -0.38 0.20
Land area 0.44 0.58 -0.69 0.01

Eigenvalue 1.76 1.17 0.78 0.28
Percent total variation explained 44% 29% 20% 7%
Table 5.4. Results of principal components analysis to determine Physical Infrastructure

Exposure weights (Human Settlements Statistics data set)
Indicator Principal Component Factor Loadings
P.C. 1 P.C. 2 P.C. 3 P.C. 4
Population 0.95 -0.20 0.24 ---
Number of housing units 0.95 -0.18 -0.24 ---
Per capita GDP 0.39 0.92 0.01 ---
Land area --- --- --- ---
Eigenvalue 1.96 0.92 0.12 ---

Percent total variation explained 65% 31% 4% ---
The correlation matrices, eigenvalues, and eigenvectors calculated in each analysis are
presented in Appendix G. While the analysis suffers from imperfect data sets and would be
better if the first principal components explained more of the total variation, the results of the
first principal components suggest roughly that the “population” and “number of housing units”
indicators should have equal weights, that the “per capita GDP” and “land area” indicators
should have equal weights, and that the former should be a little more than twice the latter. As
shown in Figure 5.8, the weights of “population” and “number of housing units” were taken to
be 0.35, and the weights of “per capita GDP” and “land area” were assigned values of 0.15.
5.3.4.2. Questionnaire
A questionnaire was developed and distributed to a sample of experts in earthquake

disaster risk to elicit their subjective assessments of the relative importance of the five main
factors that contribute to a city’s earthquake disaster risk—Hazard, Exposure, Vulnerability,
External Context, and Emergency Response and Recovery Capability. Appendix H contains a
copy of the questionnaire. Although it was not a scientific survey, an attempt was made to
obtain a sample of respondents that represent a range of professional and geographical

backgrounds. The survey participants, all experts with a holistic knowledge of earthquake
disaster risk, either were identified through personal connections, or picked up the questionnaire
at a poster session at the Eleventh World Conference on Earthquake Engineering in Acapulco,
Mexico in June 1996. Eighteen questionnaires were returned out of about forty distributed.
Two were incomplete, and thus were not included in the summary of results. Responses came
from professors and practitioners in Australia (2), China, India, Japan (2), Mexico, Turkey, and
the United States (10). The respondents’ areas of expertise spanned a range of fields, including
structural and geotechnical engineering, building construction, seismology, risk management,
emergency response and recovery, sociology, regional planning, city planning, and public policy.
The questionnaire was designed to facilitate thoughtful assessments of weights

representing the relative contributions of factors to earthquake disaster risk. It aimed to lead the
respondent through a three-step thought process that began with relatively easy assessments
and progressed to the assessments of interest, the more difficult assessments of the five relative
factor weights (wH, wE, wV, wC, and wR). The first step (Questions
1 to 10; part a) asked simple, qualitative, binary comparisons (e.g., “Which factor contributes
more to earthquake disaster risk, Hazard or Exposure?”). The second step (Questions 1 to 10;
part b) asked the respondent to refine the first-step assessments by giving quantitative responses
to the same binary comparisons (e.g., How much more important is Hazard than Exposure?).
The third step (Question 11) asked to synthesize the results of the binary comparisons into a
five-way comparison of relative factor importance. Although the first ten questions provided no
information that was not also contained in Question 11 (as many respondents noted), they were
included to help lead the respondents to thoughtful answers.
Although the responses given in each of the three parts should provide the same
information about the relative importance of the factors, in at least four of the questionnaires
returned, the ratios of relative factor importance derived from Question 11 differed substantially
from those assessed in Questions 1 to 10. This inconsistency could indicate that the respondent
misinterpreted the questionnaire, or it simply could reflect the fact people are not naturally

consistent. In any case, inconsistency across the three parts should not necessarily invalidate the
Question 11 results. The primary purpose of Questions 1 to 10 was to force respondents to
think about comparisons of relative factor importance slowly and carefully before making the
final weight assessments. Whether or not the responses to those questions were consistent with
the response for Question 11, the process of answering them was probably sufficient to achieve
this goal.
Experts have not reached a consensus on the definitions of many key concepts in hazard
studies (e.g., risk, vulnerability, exposure). In light of lingering differences in taxonomy,
particularly across disciplines, the questionnaire included brief definitions to help ensure that all
respondents were thinking of the same concepts as they made their assessments. Nevertheless,
some comments received indicate that the concepts were not interpreted exactly as intended.
Unfortunately, there is no way to know for certain the extent to which the differences in
responses reflect different beliefs about the nature of earthquake disaster risk, and the extent to
which they simply reflect different understandings of the factor definitions.
As mentioned, the questionnaire exercise was not a strictly scientific survey, rather an
attempt to obtain a rough idea of the experts’ gut feelings about the relative contributions of the
five main factors to earthquake disaster risk. The results are summarized in Table 5.5 and
Figure 5.7. Keeping in mind the methodological shortcomings discussed above, the responses
lead to the following interesting observations.
First, some consensus exists among the respondents about what weight values are
appropriate. Standard deviations for the five factor weights are relatively small, between 0.05
and 0.09 (see Table 5.5). The respondents apparently consider all five factors to contribute
earthquake disaster risk. While it is a common public belief that hazard is by far the most
important determinant of earthquake disaster risk, the experts appear to believe otherwise.
Table 5.5. Five main factor weights obtained from questionnaire survey

External Emerg. Resp.
Respondent Hazard Exposure Vulnerability Context & Recovery
WH WE WV WC WR
1 0.350 0.200 0.200 0.125 0.125
2 0.200 0.171 0.286 0.171 0.171
3 0.222 0.333 0.167 0.111 0.167
4 0.385 0.231 0.154 0.115 0.115
5 0.206 0.294 0.235 0.176 0.088
6 0.385 0.192 0.192 0.115 0.115
7 0.294 0.176 0.265 0.118 0.147
8 0.233 0.233 0.233 0.070 0.233
9 0.200 0.267 0.333 0.067 0.133
10 0.167 0.167 0.417 0.042 0.208
11 0.391 0.130 0.391 0.043 0.043
12 0.156 0.281 0.156 0.094 0.313
13 0.242 0.242 0.303 0.030 0.182
14 0.278 0.194 0.250 0.083 0.194
15 0.190 0.238 0.238 0.143 0.190
16 0.105 0.211 0.263 0.211 0.211
Mean 0.250 0.223 0.255 0.107 0.165

Standard Deviation 0.088 0.053 0.078 0.051 0.064
Coeff. of Variation 0.352 0.238 0.306 0.477 0.388
Minimum Value 0.110 0.130 0.150 0.030 0.040
Maximum Value 0.390 0.330 0.420 0.210 0.310
Note: Weights have been normalized so the sum of the five weights equals one.
Weight values of five main factors

0.5
Weights (normalized to one)
0.4
0.3
0.250 0.255
0.2
0.223
0.165
0.107
0.1
0.0
Hazard Exposure Vulnerability External Emerg.Resp.
Context & Recovery
Maximum Mean + 1 st.dev. Mean Mean - 1 st.dev. Minimum
Figure 5.7. Summary of questionnaire survey results

Second, while the means of the Hazard, Exposure, and Vulnerability weights are all
about 0.25, the means for Emergency Response and Recovery Capability and External Context
weights are only about 0.16 and 0.11, respectively. The lower Response and Context weights
could reflect a genuine belief by the respondents that those two factors contribute less to
earthquake disaster risk than the other factors, or they could be due to the fact that Response
and Context are less familiar concepts in earthquake risk assessment than the other three
factors.
Third, the results provided guidelines for determining the weights that should be adopted
in the Earthquake Disaster Risk Index (wH = wE = wV = 0.25; wC = wR =0.125), and suggested
the range of weights that to be investigated in studying the sensitivity of the final EDRI values to
the weight values (see Section 7.4).
5.3.4.3. Sample analysis solution
Table 5.6 summarizes the final weight values employed in the sample analysis. The
values may be substituted directly into the equations in Section 5.1.5. Figure 5.8 illustrates how
the values were derived by first assessing weights for the factor components. For example, the
weight 0.298 for e(MMI w/50-year return period) is the product of the three factor component weights:
0.7 for being part of ground shaking versus collateral hazards, 0.85 for bedrock ground shaking
versus soft soil, and 0.5 for being part of short-, not long-term seismicity. Presenting the
weights explicitly allows the users to see the assumptions about the definition-related relative
contributions that were made in constructing the EDRI. If a user disagrees with any of these
assumptions made about the relative importance of different factors, the values of the
appropriate weights can be easily changed to reflect a different set of assumptions.

Table 5.6. Summary of sample analysis weights
Weight Weight
Factor Indicator
name value
exp(MMI w/50-year return period) WH1 0.298
exp(MMI w/500-year return period) WH2 0.298
Percentage urbanized area w/soft soil WH3 0.105
Hazard Percentage urbanized area w/high liquefaction suscept. WH4 0.100
Percentage of buildings that are wood WH5 0.050
Population density WH6 0.050
Tsunami potential indicator WH7 0.100
HAZARD WH 0.250
Population WE1 0.140
Per capita GDP in constant 1990 US$ WE2 0.060
Exposure Number of housing units WE3 0.140
Urbanized land area (sq.km.) WE4 0.060
Population WE5 0.400
Per capita GDP in constant 1990 US$ WE6 0.200
EXPOSURE WE 0.250
Seismic code indicator WV1 0.190
City wealth indicator WV2 0.285
Vulner- City age indicator WV3 0.190
ability Population density WV4 0.190
City development speed indicator WV5 0.095
Percentage of population aged 0-4 or 65+ WV6 0.050
VULNERABILITY WV 0.250
Economic external context indicator WC1 0.800
External Political country external context indicator WC2 0.100
Context Political world external context indicator WC3 0.100
EXTERNAL CONTEXT WC 0.125
Planning indicator WR1 0.333
Per capita GDP in constant 1990 US$ WR2 0.133
Avg. annual real growth in per cap. GDP in prev. 10 yrs. WR3 0.033
Emerg. Housing vacancy rate WR4 0.083
Resp. & Number of hospitals per 100,000 residents WR5 0.042
Recovery Number of physicians per 100,000 residents WR6 0.042
Extreme weather indicator WR7 0.111
Population density WR8 0.111
City layout indicator WR9 0.111
EMERGENCY REPONSE & RECOVERY WR 0.125

0.5
0.85 bedrock ground e(MMI w/50-yr. ret. per.)
0.7 shaking
Ground Shaking 0.5 e (MMI w/500-yr. ret. per.)
0.15
soft soil
Hazard
0.33
liquefaction 0.5
0.3 Collateral Hazards 0.33 % buildings wood
fire following eq. 0.5
0.33 population density
tsunami potential
0.4 Population Exposure
0.35
population
0.15
per capita GDP
Exposure 0.4 Physical 0.35
Infrastructure Exp. num. of housing units
0.15
urbanized land area
0.2 Economy Exposure
0.05
Population Vulner. 0.2 seismic code
Vulner- 0.3
ability city wealth
0.95 Physical 0.2 city age
Infrastructure Vul. 0.2
population density
0.1
city development speed
0.8 Economic Ext. Context
External
Context 0.5
0.2 Political Ext. Context political country ext. con.
0.5 political world ext. con.
0.8 per capita GDP
0.33 0.5
Planning financial resources
0.2 avg. annual per
cap. GDP growth
Emerg. 0.5
Response 0.33 Resources 0.25 housing vacancy rate
equipment & facilties
& Recov. 0.5 num.
Capability hospitals
0.25 trained manpower num. physicians
0.33
extreme weather
0.33Mobility and Access 0.33 population density
0.33 city layout indicator
Figure 5.8. Factor component weights
5.4. Mathematical combination summary

Chapter 5 had two purposes: (1) to determine the best method by which to combine the
indicators from Chapter 4 into the EDRI, and (2), in the process of completing the first goal, to
identify the alternatives available for this type of combination, and to compare them
systematically. To summarize the results of the first task, it was determined that the indicators
listed in Figure 4.1 should be combined in two steps. First, the value xij for each indicator i and
city j should be scaled using one of the following equations (reprinted from Sections 5.2.7 and
5.2.8):
x ij − (x i − 2s i )
x'ij =
si (5.9)
if the indicator is directly-related to earthquake disaster risk,
- x ij + (x i + 2s i )
x ij ' =
si (5.10)
if the indicator is inversely-related to earthquake disaster risk. Second, the following equations
are used to combine the scaled values into the five main factor indexes and the EDRI (reprinted
from Section 5.1.5):
EDRI = wHH + wEE + wVV + wCC + wRR (5.4a)
H = wH1x'H1 + wH2x'H2 + wH3x'H3 + wH4x'H4 + wH5x'H5 + wH6x'H6 + wH7x'H7 (5.4b)
E = wE1x'E1 + wE2x'E2 + wE3x'E3 + wE4x'E4 + wE5x'E5 + wE6x'E6 (5.4c)
V = wV1x'V1 + wV2x'V2 + wV3x'V3 + wV4x'V4 + wV5x'V5 + wV6x'V6 (5.4d)
C = wC1x'C1 + wC2x'C2 + wC3x'C3 (5.4e)
R = wR1x'R1 + wR2x'R2 + wR3x'R3 + wR4x'R4 + wR5x'R5 + wR6x'R6 + wR7x'R7 +
wR8x'R8 + wR9x'R9 (5.4f)
Context, and Emergency Response and Recovery Capability indexes, respectively; x'i refer to
the scaled values of the indicators listed in Figure 4.1; and wi are the weights listed in Table 5.6.

Although many other composite indexes are available to measure a wide range of
concepts, the publications in which they are described include little, if any, discussion of the
mathematics of their combinations. Many composite indexes use a linear combination
approach, but none explain why this method was chosen, or if other approaches were
considered. At least five different scaling techniques and three different weighting approaches
were represented among the many indexes that were examined in the literature survey, but
again, none of the publications provided a systematic comparison of the alternative techniques
available. It is not immediately obvious what, if any, implications are associated with choosing
one scaling option or weighting technique over another. This chapter aimed to fill this gap in the
literature by identifying and systematically comparing the various combination, scaling, and
weighting alternatives available. This comparison provides a sound analytical basis for
determining which methods are most appropriate for a given task, in particular for defining the
EDRI and its associated factor indexes. Sections 7.3 and 7.4 provide some empirical
comparison of the various scaling and weighting options.

Chapter Six
Data Gathering and Evaluation
6. Introduction
With the indicators selected and the method for their mathematical combination
established, the theoretical component of the Earthquake Disaster Risk Index development
process is essentially complete. This chapter describes the process of gathering data and
evaluating the five main factors and the EDRI for a sample of cities. The discussion of data
gathering identifies some potentially useful sources of data, common deficiencies in the available
data, and approaches to overcome those deficiencies (Section 6.1). The procedure for
evaluating the EDRI is outlined (Section 6.2), and the results of the data gathering and
evaluation step in the sample analysis are presented (Section 6.2). The indicator selection and
data gathering steps overlap considerably because an indicator cannot be included if the data to
measure it are not available. Some of the issues discussed in Section 6.1, therefore, could have
been included in Chapter 4. They are mentioned here instead, to keep all data-related issues
together.
6.1. Data sources
As mentioned in Section 4.1, indicators should depend on data that is reliable, available
in a consistent form for all the world’s major cities, and relatively easy to collect. Data are
reliable if the proportion of variance that is error variance is sufficiently small (Rossi and
Gilmartin 1980). It will be impossible, in general, to obtain statistically-based, quantitative
Chapter Six: Data Gathering and Evaluation 145

estimates of the error variance, but approximations can be surmised based on the quality of the
data sources used, and on whether the required data values are reported directly or are
estimated from related, available data. “In consistent form” indicates that the data for each
indicator should refer to the same quantity across cities, and data for each city should refer to
the same city boundaries across indicators. Data are considered relatively easy to collect if they
are publicly available, and any user can obtain them in a reasonable amount of time, with limited
or no financial expense. In the sample analysis, if the data could be gathered from the author’s
university, in a time frame of a couple of months1 for ten cities, they were considered to be
relatively easy to collect.
Sources of EDRI data come in three principal forms: compilations, focused sources, and
personal communications. Compilations published by the United Nations, World Bank,
International Monetary Fund, or other international organizations are a good starting point for
the data gathering expedition. A single compilation can provide data for all or several cities, and
presumably the task of ensuring consistency across indicators and cities has already been
undertaken before the data are published. Sources focused on one or two topics and/or cities
(e.g., journal articles, censuses, country-specific statistical yearbooks) can be used to fill in data
that have not been compiled already. These sources provide more detailed data, but gathering
information from many specific sources can be much more labor-intensive than using
compilations, and it is difficult to ensure consistency across sources. Finally, personal interviews
with local or topic experts can be particularly useful in providing subjective assessment
information about, for example, emergency response and recovery planning, or seismic code
enforcement. However, using personal interviews also may be more labor-intensive than using
published material, and the data may reflect the biases of the information provider.
1
In the sample analysis, data gathering was actually performed over the course of several months, because
it was undertaken at the same time as the indicator selection step. Simultaneously determining which
indicators to use and finding data for those indicators took more time than data gathering alone would if the
indicator list was already established.
146 Chapter Six: Data Gathering and Evaluation

Several problems may be encountered in gathering data for the EDRI. The data simply
may not exist, or they may exist, but not in the exact form in which they are needed. For
example, “length of gas pipelines” was considered as a potential Exposure indicator. After an
extensive library search and many conversations with local utilities personnel in Boston, St.
Louis, and San Francisco, however, it became clear that that information simply does not exist
in many cities. Accurate records were not kept at the time that many local pipelines were laid,
so that even the utility companies may not know the length of pipeline in a given city.
Following are three situations in which data may exist, but still be problematic. First,
data may be available, but not for the greater metropolitan areas defined in the EDRI. Most
compilations published by international organizations, like the United Nations or World Bank,
provide data by country. In the few compilations that present data by city (e.g., Statistics of
World Large Cities 1991), the definition of what areas are included in the city are unclear,
inconsistent across indicators, or different from those used in the EDRI, rendering the data
unusable. “Number of police officers,” under consideration as an Emergency Response and
Recovery Capability indicator, was eventually omitted from the EDRI because the data, while
they do exist, would be difficult to compile for different cities consistently. Police officers can be
local, state, or federal employees, and each jurisdiction compiles the data on the number of
police officers it employs separately. The quantity of interest to the EDRI is the total number of
police officers in a given city—local, state, and federal. Counting only one type could be
misleading, and compiling all three is difficult because it is not clear how many of the state and
federal employees work within the greater metropolitan area of interest. “Per capita gas
consumption,” considered as an Exposure indicator alternative to “length of gas pipelines,”
provides another example of an indicator for which data theoretically exist, but are not compiled
in a usable form. Some gas consumption data are available, but it is difficult or impossible to
find the consumption level for exactly the areas contained in the specified city of interest. The
data are available only disaggregated by company service area, state, or country, none of which
coincide with the boundaries of the greater metropolitan area.

Second, data may not be available in a consistent, standardized form. Different cities
may collect data using different units or using related, but different quantities. For example,
many censuses report a break down of the number of housing units into categories based on the
material used in the roof, but the set of possible materials varies from census to census. In the
Philippines census, the categories of roof materials are galvanized iron/aluminum;
tile/concrete/clay tile; half galvanized iron and half concrete; wood; cogon/nipa/anshaw;
makeshift/salvaged/improvised; and asbestos/others. In the Chile census, the categories are
zinc; concrete slab; slate; tiles; wood shingles; fonolita; straw and mud; and other. It is difficult
to reconcile these different classification systems.
Third, for some indicators, data are available, but not for all major cities worldwide, or
not for the year under consideration. For example, “average number of days per year that the
temperature reaches 90°F or more, or 32°F or less” was considered as a possible extreme
weather indicator for the Emergency Response and Recovery Capability factor, but data for it
are readily available only for cities in the USA.
The data deficiencies described above may be addressed in a few ways. First, it may
be possible to translate related data into a form that the EDRI can use through unit conversions.
Land area may be reported in square kilometers, square miles, or hectares. Ground shaking
may be presented in terms of MMI or PGA. Per capita Gross Domestic Product may be
available in Yen for Japan and dollars for the USA. In all these cases, standard conversion
formulas are available to make the data consistent for all cities. In some cases, more significant
manipulation is required to derive the desired data from the available data. For example, in the
sample analysis, since the Indonesian census reports the percentage of the population aged 0-4
or 65+ for urban Indonesia, but not for DKI Jakarta specifically, the latter quantity was
estimated to be equal to the former. The number of housing units in Lima was available only for
1981, so the 1990 value was estimated as:

 Population1990 
(Num. housing units)1990 = (Num. housing units) 1981 *   (6.1)
 Population1981  ,
assuming that the number of people per housing unit was the same for both years. For each
indicator-city value, the degree of manipulation and estimation required to transform the data as
reported into the data required by the EDRI values is reflected in the estimate of data
uncertainty used in the sensitivity analysis (Section 7.5).
If the desired indicator data cannot be estimated reliably from related, available data, it
may be necessary to select a different indicator to represent the concept of interest. The
“economic external context indicator” was defined so that it relies only on available data. While
the indicator originally hoped to measure the total economic exchange between the city and the
world, the only related data available was the total economic exchange between the country
and the world. Since a country’s economic exchange may be distributed unevenly among its
cities, it was necessary to develop the economic external context indicator used in the sample
analysis to capture the concept using only available data (see Section 4.3.4.1).
Finally, if data for an indicator cannot be found directly or estimated from related,
available data, and the indicator cannot be redefined to accommodate the data, it may be
necessary to omit an indicator entirely. Since the EDRI aims to be as comprehensive as
possible, omitting a concept should be considered a last resort. Nevertheless, if no reasonably
reliable and consistent data can be found to evaluate a given indicator, it is better to omit that
indicator than to incorporate an excessive amount of uncertainty into the model. In the sample
analysis, for example, the exposure indicators do not attempt to measure the percentage of the
population that lives in poverty because of the well-documented difficulties in obtaining
consistent, reliable assessments of that quantity.
6.2. Evaluation procedure

The five main factors and the EDRI for a set of cities can be evaluated in three steps.
First, an unscaled indicator-city table, like the sample analysis example in Table 6.2, should be
completed. For some indicators, the values can be taken directly from the data source and
entered into the table. For derived indicators, such as “economic external context indicator,”
“e(MMI with a fifty-year return period),” and “city development speed indicator,” the indicator values must
be calculated from the raw data before being entered into the unscaled indicator-city data table.
To reduce data uncertainty, it is best to evaluate cities as consistently as possible. For each
indicator, a single source and procedure should be used if possible. Second, the scaled
indicator-city table should be completed. The scaled table is similar to the unscaled version,
but with two variations. The value in each cell will be the scaled value xij' instead of the
unscaled value xij (see Sections 5.2.7 and 5.2.8), and extra rows are included for the values of
the five main factors and the EDRI. Third, as part of completing the scaled table, the values for
the five main factor indexes and the overall EDRI can be computed for each city, using the
appropriate linear combination equations and weights (see Section 5.1.5 and Table 5.6).
Constructing the charts and figures that will be used to display this information in graphical form
is discussed in Chapter 8.
Finally, when the main factor and EDRI values are evaluated in subsequent time
periods, the scaling should be performed using the mean and standard deviation of the base
year, to allow consistent comparison over time. Otherwise, a city’s EDRI may rise substantially
from one time period to the next, but if the values for every other city increase proportionally,
the final EDRI will appear to have remained constant.
In the sample analysis, data were gathered and the main factor and EDRI values were
computed for ten cities for the year 1990. These steps were performed on real cities to explore
the availability and quality of data for the indicator selection phase, to illustrate the data gathering

and evaluation process for real cities, and to obtain results that could be used to demonstrate
the subsequent sensitivity analysis, presentation, and interpretation steps of the process. The
following sample analysis cities were selected to represent a variety of values for each of the
main factors, and to explore a variety of data availability situations:
Boston, USA Mexico City, Mexico

Istanbul, Turkey San Francisco, USA
Jakarta, Indonesia Santiago, Chile
Lima, Peru St. Louis, USA
Manila, Philippines Tokyo, Japan
The sample represents developing and industrialized countries, four continents, populations from
two to thirty-five million people, and a range of levels of seismicity. As discussed in Section
2.4, it is critical to establish at the outset exactly what areas are included in each greater
metropolitan area (city) in the analysis. Appendix B lists the areas (e.g., counties, cities) that are
included in each of the ten sample analysis cities.
Table 6.1 summarizes for every indicator-city combination, the source(s) of the data
used in the sample analysis. Some values were reported by the source(s) directly, and some
were estimated from information provided in the sources. The number(s) in each cell
corresponds to the reference(s) listed in Figure 6.1.

Table 6.1. Sample analysis data sources
Mexico San
Factor Indicator Boston Istanbul Jakarta Lima Manila Santiago St. Louis Tokyo
City Fran.
xh1 exp(MMI w/50-year return period) 20 18 20 14 6, 7 9 20 20 20 20
xh2 exp(MMI w/500-year return period) 20 18 20 14 6, 7 9 20 20 20 20
xh3 Percentage urbanized area w/soft soil 20 18 20 15 8 10 40 20 20 20
Hazard xh4 Percentage urbanized area w/high liquefaction suscept. 20 18 20 15 8 10 20 20 20 20
xh5 Percentage of buildings that are wood 65 --- 64 13 5 19 65 32 65 64
xh6 Population density (people/sq.km.) 21 16, 17 20, 30 11 4 2 21 32 21 1
xh7 Tsunami potential indicator 33 33 33 33 33 33 33 33 33 33
xe1 Population (1000s) 21 17 30 11 4 2 21 32 21 1
xe2 Per capita GDP in constant 1990 US$ 22 22 22 22 22 22 22 22 22 22
Exposure xe3 Number of housing units (1000s) 21 17 30 11 5 2 21 32 21 1
xe4 Urbanized land area (sq.km.) 21 3, 18 20 11 4 2, 3 21 32 21 1
xe5 Population (1000s) 21 17 30 11 4 2 21 32 21 1
xe6 Per capita GDP in constant 1990 US$ 22 22 22 22 22 22 22 22 22 22
42, 45, 41, 45, 41, 45, 41, 45, 42, 45,
xv1 Seismic code indicator 45, 61 42, 45, 50 45, 56 43, 45, 46 44, 45, 61
59, 60 53-55 57, 58 51, 52 47-49
xv2 City wealth indicator 22 22 22 22 22 22 22 22 22 22
Vulner- xv3 City age indicator 24 17, 23 28, 29, 30 11,23 4 2, 23 24 25, 26, 27 24 1
ability xv4 Population density (people/sq.km.) 21 16, 17 20, 30 11 4 2 21 32 21 1
xv5 City development speed indicator 24 17, 23 28, 29, 30 11,23 4 2, 23 24 25, 26, 27 24 1
xv6 Percentage of population aged 0-4 or 65+ 21 16 30 11 4 2 21 36 21 1
External xc1 Economic external context indicator 37, 23, 21 17, 23, 37 23, 30, 37 11, 23, 37 4, 23, 37 2, 23, 37 21, 23, 37 23, 32, 37 21, 23, 37 1, 23, 37
Context xc2 Political country external context indicator 21 17 30 11 4 2 21 32 21 1
xc3 Political world external context indicator 21,22 17, 32 22, 30 11, 22 4, 22 2, 22 21, 22 22, 32 21, 22 1, 22
xr1 Planning indicator 62 34 34 34 34 34 63 34 34 34
xr2 Per capita GDP in constant 1990 US$ 22 22 22 22 22 22 22 22 22 22
xr3 Avg. annual real growth in per cap. GDP in prev. 10 yrs. 22 22 22 22 22 22 22 22 22 22
Emerg. xr4 Housing vacancy rate 21 --- --- 13 5 --- 21 32 21 1
Resp. & xr5 Number of hospitals per 100,000 residents 21 17 31 12 4 2 21 35 21 1
Recovery xr6 Number of physicians per 100,000 residents 21 17 31 12 4 2 21 35 21 1
xr7 Extreme weather indicator 38 39 39 39 39 39 38 39 38 39
xr8 Population density (people/sq.km.) 21 16, 17 20, 30 11 4 2 21 32 21 1
xr9 City layout indicator 34 34 34 34 34 34 34 34 34 34
Numbers refer to sources listed in Figure 6.1.
1. (Japan Statistics Bureau 1995) 34. (Author's assessment)
2. (Mexico INEGI 1994 and 1995) 35. (Chile OPN 1993)
3. (Institut d'Estudis de Barcelona 1988) 36. (Chile INE 1989)
4. (Philippines NEDA 1995) 37. (IMF 1995)
5. (Philippines NSO 1992) 38. (Ruffner and Bair 1987)
6. (Algermissen, Thenhaus, and Campbell 1996) 39. (NCDC 1997)
7. (Su 1978) 40. (US Geological Survey 1994)
8. (Rantucci 1994) 41. (Takizawa 1993)
9. (Wang et al. 1996) 42. (Watabe 1983)
10. (Romo, Jaime, and Resendiz 1988) 43. (Martin 1993)
11. (Peru INE 1993) 44. (Olshansky 1993)
12. (Webb and Baca 1993) 45. (Hirosawa et al. 1992)
13. (Peru INEI 1994) 46. (SEAOC 1975, 1990)
14. (Sharma and Candia-Gallegos 1992) 47. (UNCRD 1995)
15. (Alva-Hurtado 1994) 48. (RMS 1995)
16. (Turkey. Devlet Istatistik Enstitusu 1994) 49. (Otani 1995)
17. (Turkey. Devlet Istatistik Enstitusu 1992) 50. (Fraser 1983)
18. (Erdik 1994) 51. (Cruz 1989)
19. (Mexico INEGI 1990-92) 52. (Sarrazin 1992)
20. (Windeler 1995-97) 53. (Acuna 1992)
21. (U.S. Bureau of the Census 1930-94) 54. (Vega 1992)
22. (IMF 1996) 55. (Zegarra 1987)
23. (UN. Population Division 1995b) 56. (Nakata 1991)
24. (U.S. Bureau of the Census 1930-94) 57. (Rosenblueth 1987)
25. (Chile DGE 1931-35) 58. (Marsh 1993)
26. (Chile DEC 1963) 59. (Gulkan and Ergunay 1984)
27. (Chile INE 1958-90) 60. (Bayulke 1992)
28. (Indonesia BPS 1941-82) 61. (Morse 1989)
29. (UN. Population Division 1995a) 62. (Smith 1996)
30. (Indonesia BPS 1992) 63. (Eisner 1996)
31. (Indonesia NDIO 1993) 64. (Romer 1995-97)
32. (Chile INE 1992a, 1992b) 65. (Bouabid 1995-97)
33. (NGDC 1997)
Figure 6.1. Sources cited in Table 6.1.
Tables 6.2 and 6.3 are the unscaled and scaled sample analysis indicator-city tables,
respectively. In the unscaled table, a few values are in bold to indicate that they were set equal

to the mean of the corresponding indicator, because real data could not be obtained for that
indicator-city combination during the sample analysis. The scaled
Table 6.2. Unscaled indicator-city data
Mexico San
Factor Indicator D/ I* Boston Istanbul Jakarta Lima Manila Santiago St. Louis Tokyo
City Francisco
xh1 exp(MMI w/50-year return period) D 446 992 164 2981 1339 164 1212 602 493 1339
xh2 exp(MMI w/500-year return period) D 3641 4447 735 6634 2208 493 4024 1998 4915 3641
xh3 Percentage urbanized area w/soft soil D 35% 15% 7% 25% 5% 50% 17% 4% 0% 10%
Hazard xh4 Percentage urbanized area w/high liquefaction suscept. D 35% 7% 49% 25% 50% 5% 17% 4% 21% 20%
xh5 Percentage of buildings that are wood D 67% 42% 19% 4% 32% 2% 68% 59% 64% 60%
xh6 Population density (people/sq.km.) D 884.1 12161.1 12652.6 9862.8 19756.6 11999.6 1075.3 8130.9 745.2 8849.2
xh7 Tsunami potential indicator D 0 1 1 2 1 0 2 0 0 2
xe1 Population (1000s) D 5,686 5,328 8,228 6,397 7,928 16,621 6,253 4,311 2,493 35,467
xe2 Per capita GDP in constant 1990 US$ D 25,610 1,749 1,521 564 623 2,705 25,312 2,088 20,181 25,833
Exposure xe3 Number of housing units (1000s) D 2,271 1,194 1,548 1,200 1,483 3,377 2,457 1,056 1,027 14,046
xe4 Urbanized land area (sq.km.) D 1336.8 796.4 487.7 698.5 572.4 1614.9 2099.4 346.5 993.5 3525.9
xe5 Population (1000s) D 5,686 5,328 8,228 6,397 7,928 16,621 6,253 4,311 2,493 35,467
xe6 Per capita GDP in constant 1990 US$ D 25,610 1,749 1,521 564 623 2,705 25,312 2,088 20,181 25,833
xv1 Seismic code indicator I 14.18 40.82 8.76 9.83 10.78 39.81 52.88 19.70 10.08 52.79
xv2 City wealth indicator I 2780 318 100 135 130 489 2981 258 2858 4631
Vulner- xv3 City age indicator D 49.7 8.8 11.9 8.7 10.7 14.1 19.7 20.1 42.4 20.6
ability xv4 Population density (people/sq.km.) D 884.1 12161.1 12652.6 9862.8 19756.6 11999.6 1075.3 8130.9 745.2 8849.2
xv5 City development speed indicator I 106 18 22 18 20 23 37 32 89 36
xv6 Percentage of population aged 0-4 or 65+ D 19.80% 14.80% 13.70% 14.30% 14.90% 15.20% 18.10% 16.70% 20.40% 14.70%
External xc1 Economic external context indicator D 3.43E+12 3.28E+11 5.73E+11 8.21E+10 2.56E+11 2.49E+12 4.15E+12 1.13E+11 6.59E+11 4.20E+13
Context xc2 Political country external context indicator D 0.0241 0.1096 1.0000 1.0000 1.0000 1.0000 0.0250 1.0000 0.0100 1.0000
xc3 Political world external context indicator D 1.54E+11 1.08E+10 2.78E+11 1.22E+10 3.79E+10 2.29E+11 1.58E+11 2.75E+10 5.03E+10 3.19E+12
xr1 Planning indicator I 3 3 1 3 1 3 4 1 3 4
xr2 Per capita GDP in constant 1990 US$ I 25,610 1,749 1,521 564 623 2,705 25,312 2,088 20,181 25,833
xr3 Avg. annual real growth in per cap. GDP in prev. 10 yrs. I 2.13% 2.93% 10.17% -2.82% -0.71% -0.48% 0.29% 1.50% 2.20% 3.56%
Emerg. xr4 Housing vacancy rate I 7.00% 6.70% 6.70% 8.70% 3.20% 6.70% 5.20% 3.50% 8.30% 11.00%
Resp. & xr5 Number of hospitals per 100,000 residents I 1.88 1.60 1.99 3.19 2.17 0.85 1.45 1.83 1.98 5.57
Recovery xr6 Number of physicians per 100,000 residents I 332 180 36.5 172 32 187 299 68 253 164
xr7 Extreme weather indicator D 118.00 41.40 46.93 0.00 46.46 7.82 10.40 65.07 126.60 55.67
xr8 Population density (people/sq.km.) D 884.1 12161.1 12652.6 9862.8 19756.6 11999.6 1075.3 8130.9 745.2 8849.2
xr9 City layout indicator D 0 1 0 0 0 0 1 0 0 0
D = directly related to earthquake disaster risk. I = inversely related to earthquake disaster risk. If directly related to earthquake disaster risk,
use equation 5.9 to scale; if inversely related, use equation 5.10 to scale.
Boldfaced values were set equal to the sample mean because real data could not be obtained.
Table 6.3. Scaled indicator-city data
Mexico San
Factor Indicator Boston Istanbul Jakarta Lima Manila Santiago St. Louis Tokyo
City Francisco
xh1 exp(MMI w/50-year return period) 1.370 2.023 1.034 4.397 2.437 1.034 2.285 1.557 1.426 2.437
xh2 exp(MMI w/500-year return period) 2.191 2.612 0.677 3.751 1.445 0.551 2.391 1.335 2.855 2.191
xh3 Percentage urbanized area w/soft soil 3.156 1.887 1.367 2.522 1.252 4.108 2.014 1.189 0.935 1.570
Hazard xh4 Percentage urbanized area w/high liquefaction suscept. 2.701 1.029 3.515 2.104 3.597 0.909 1.626 0.850 1.865 1.805
xh5 Percentage of buildings that are wood 2.975 2.000 1.131 0.540 1.630 0.478 3.013 2.667 2.860 2.706
xh6 Population density 0.752 2.573 2.653 2.202 3.800 2.547 0.783 1.922 0.729 2.038
xh7 Tsunami potential indicator 0.972 2.114 2.114 3.256 2.114 0.972 3.256 0.972 0.972 3.256
HAZARD 1.945 2.120 1.405 3.362 2.129 1.242 2.281 1.397 1.835 2.285
xe1 Population 1.571 1.534 1.831 1.644 1.801 2.692 1.629 1.430 1.243 4.625
xe2 Per capita GDP in constant 1990 US$ 3.266 1.251 1.232 1.151 1.156 1.332 3.241 1.279 2.808 3.285
xe3 Number of housing units 1.825 1.553 1.642 1.555 1.626 2.104 1.872 1.518 1.511 4.794
Exposure xe4 Urbanized land area (sq.km.) 2.092 1.535 1.217 1.435 1.305 2.379 2.878 1.072 1.738 4.348
xe5 Population 1.571 1.534 1.831 1.644 1.801 2.692 1.629 1.430 1.243 4.625
xe6 Per capita GDP in constant 1990 US$ 3.266 1.251 1.232 1.151 1.156 1.332 3.241 1.279 2.808 3.285
EXPOSURE 2.078 1.463 1.612 1.491 1.579 2.237 2.157 1.382 1.717 4.284
xv1 Seismic code indicator 2.638 1.196 2.931 2.873 2.821 1.251 0.544 2.339 2.859 0.548
xv2 City wealth indicator 1.215 2.688 2.819 2.798 2.801 2.586 1.095 2.724 1.168 0.107
Vulner- xv3 City age indicator 4.043 1.166 1.384 1.159 1.297 1.538 1.930 1.963 3.525 1.995
ability xv4 Population density 0.752 2.573 2.653 2.202 3.800 2.547 0.783 1.922 0.729 2.038
xv5 City development speed indicator -0.104 2.706 2.578 2.706 2.642 2.546 2.099 2.259 0.439 2.131
xv6 Percentage of population aged 0-4 or 65+ 3.484 1.388 0.927 1.178 1.430 1.556 2.772 2.185 3.736 1.346
VULERABILITY 1.923 2.030 2.418 2.298 2.625 2.071 1.269 2.283 1.913 1.171
xc1 Economic external context indicator 1.847 1.608 1.627 1.589 1.602 1.774 1.902 1.591 1.633 4.827
External xc2 Political country external context indicator 0.803 0.976 2.774 2.774 2.774 2.774 0.805 2.774 0.775 2.774
Context xc3 Political world external context indicator 1.734 1.588 1.860 1.589 1.615 1.810 1.738 1.605 1.628 4.833
EXTERNAL CONTEXT 1.731 1.542 1.765 1.707 1.721 1.878 1.776 1.711 1.547 4.622
xr1 Planning indicator 1.659 1.659 3.363 1.659 3.363 1.659 0.807 3.363 1.659 0.807
xr2 Per capita GDP in constant 1990 US$ 0.734 2.749 2.768 2.849 2.844 2.668 0.759 2.721 1.192 0.715
xr3 Avg. annual real growth in per cap. GDP in prev. 10 yrs. 1.928 1.699 -0.374 3.344 2.740 2.675 2.454 2.108 1.908 1.518
Emerg. xr4 Housing vacancy rate 1.873 2.000 2.000 1.150 3.487 2.000 2.637 3.359 1.320 0.173
Resp. & xr5 Number of hospitals per 100,000 residents 2.283 2.502 2.198 1.283 2.063 3.073 2.611 2.321 2.205 -0.538
Recovery xr6 Number of physicians per 100,000 residents 0.465 1.926 3.306 2.003 3.349 1.859 0.783 3.003 1.225 2.080
xr7 Extreme weather indicator 3.538 1.757 1.886 0.795 1.875 0.977 1.037 2.308 3.738 2.089
xr8 Population density 0.752 2.573 2.653 2.202 3.800 2.547 0.783 1.922 0.729 2.038
xr9 City layout indicator 1.526 3.897 1.526 1.526 1.526 1.526 3.897 1.526 1.526 1.526
EMERGENCY RESPONSE & RECOVERY 1.632 2.242 2.548 1.780 2.908 1.931 1.448 2.695 1.694 1.122
EARTHQUAKE DISASTER RISK INDEX 1.907 1.876 1.898 2.223 2.162 1.864 1.830 1.816 1.772 2.653
indicator-city table includes the values of the five main factors and the EDRI. The factor and
EDRI values are calculated using the scaled data in the same table, the equations in Section
5.1.5, and the weights from Table 5.6. Tables 6.4 and 6.5 are auxiliary tables that show the
values of the components of each factor for each city (as in Figure 5.8), and the relative
contributions of each of the five main factors to the EDRI, respectively. The factor component
values are derived by combining only the indicators that relate to the specified factor component
(listed in the third column of Table 6.4), rescaling the weights so that the sum of the included
weights equals one. For example, ground shaking is calculated as: (6.2)
w H1 w H2 w H3
Ground shaking = x H1 + xH 2 + x
(w H1 + w H2 + w H3 ) (w H1 + w H2 + w H3 ) (w H1 + w H2 + w H3 ) H 3
where xH1, xH2, and xH3 are “e(MMI with a fifty-year return period),” “e(MMI with a five-hundred-year return period),” and
“percentage of urbanized area with soft soil,” respectively; and wH1, wH2, and wH3 are the
corresponding weights, presented in Table 5.6. For each of the five main factors, the city-
specific relative contribution to the EDRI equals the ratio of the factor value divided by the sum
of all five factor values (Section 5.1.4). For example, for Hazard,
H
Relative contribution of Hazard =
(H + E + V + C + R) (6.3)
Context, and Emergency Response and Recovery Capability indexes, respectively. Note that
the denominator is not equal to the EDRI value, because it does not include the factor weights.
157
Table 6.4. Factor component values
Indicators Mexico San
Factor Factor components Boston Istanbul Jakarta Lima Manila Santiago St. Louis Tokyo
included City Francisco
Hazard Ground shaking xh1-xh3 1.987 2.253 0.932 3.841 1.838 1.290 2.289 1.407 1.960 2.203
Short-term seismicity xh1 1.370 2.023 1.034 4.397 2.437 1.034 2.285 1.557 1.426 2.437
Long-term seismicity xh2 2.191 2.612 0.677 3.751 1.445 0.551 2.391 1.335 2.855 2.191
Soft soil xh3 3.156 1.887 1.367 2.522 1.252 4.108 2.014 1.189 0.935 1.570
Collateral hazards xh4-xh7 1.845 1.810 2.507 2.244 2.809 1.131 2.260 1.372 1.544 2.478
Liquefaction xh4 2.701 1.029 3.515 2.104 3.597 0.909 1.626 0.850 1.865 1.805
Tsunami xh5 0.972 2.114 2.114 3.256 2.114 0.972 3.256 0.972 0.972 3.256
Fire following xh6-xh7 1.863 2.287 1.892 1.371 2.715 1.513 1.898 2.295 1.794 2.372
HAZARD 1.945 2.120 1.405 3.362 2.129 1.242 2.281 1.397 1.835 2.285
Exposure Physical Infrastructure xe1-xe4 1.992 1.499 1.583 1.507 1.568 2.235 2.143 1.385 1.646 4.442
Population xe5 1.571 1.534 1.831 1.644 1.801 2.692 1.629 1.430 1.243 4.625
Economy xe6 3.266 1.251 1.232 1.151 1.156 1.332 3.241 1.279 2.808 3.285
EXPOSURE 2.078 1.463 1.612 1.491 1.579 2.237 2.157 1.382 1.717 4.284
Vulner- Physical Infrastructure xv1-xv5 1.840 2.064 2.497 2.357 2.688 2.098 1.190 2.288 1.817 1.161
ability Population xv6 3.484 1.388 0.927 1.178 1.430 1.556 2.772 2.185 3.736 1.346
VULERABILITY 1.923 2.030 2.418 2.298 2.625 2.071 1.269 2.283 1.913 1.171
External Economic xc1 1.847 1.608 1.627 1.589 1.602 1.774 1.902 1.591 1.633 4.827
Context Political xc2-xc3 1.269 1.282 2.317 2.181 2.194 2.292 1.272 2.189 1.201 3.803
EXTERNAL CONTEXT 1.731 1.542 1.765 1.707 1.721 1.878 1.776 1.711 1.547 4.622
Emerg. Planning xr1 1.659 1.659 3.363 1.659 3.363 1.659 0.807 3.363 1.659 0.807
Resp. & Resources xr2-xr6 1.298 2.323 2.258 2.173 2.960 2.451 1.632 2.804 1.426 0.674
Recovery Financial resources xr2-xr3 0.972 2.539 2.140 2.948 2.824 2.670 1.098 2.598 1.335 0.876
Equipment & facility resources xr4-xr5 2.771 3.001 2.798 1.622 3.700 3.382 3.499 3.787 2.350 -0.243
Trained manpower resources xr6 0.465 1.926 3.306 2.003 3.349 1.859 0.783 3.003 1.225 2.080
Mobility and access xr7-xr9 1.938 2.743 2.021 1.508 2.400 1.683 1.906 1.919 1.998 1.884
EMERG. RESP. & RECOV. 1.632 2.242 2.548 1.780 2.908 1.931 1.448 2.695 1.694 1.122
EDRI 1.907 1.876 1.898 2.223 2.162 1.864 1.830 1.816 1.772 2.653

Table 6.5. Relative contributions2 of each factor to the overall EDRI values
Hazard Exposure Vulnerability External Emerg. Resp.
Context & Recovery
Boston 21% 22% 21% 19% 18%

Istanbul 23% 16% 22% 16% 24%
Jakarta 14% 17% 25% 18% 26%
Lima 32% 14% 22% 16% 17%
Manila 19% 14% 24% 16% 27%
Mexico City 13% 24% 22% 20% 21%
San Francisco 26% 24% 14% 20% 16%
Santiago 15% 15% 24% 18% 28%
St. Louis 21% 20% 22% 18% 19%
Tokyo 17% 32% 9% 34% 8%
Although the EDRI data requirements pale in comparison to those associated with
engineering loss estimation methods, gathering data for each indicator and city still constitutes
the most time- and labor-intensive step in the EDRI development process. Since there had
been no previous international comparisons of urban earthquake risk like that that the EDRI
aims to achieve, only limited effort had been made previously to compile the type of data
needed for the EDRI. Even in fields that do have established mechanisms for gathering data,
information is rarely compiled in a systematic, consistent way for greater metropolitan areas.
Nevertheless, because these data restrictions were considered in selecting the indicators for the
EDRI, the data needs can be met to produce useful results with a reasonable amount of effort.
The findings presented here are examined in depth in the two chapters that follow. Chapter 7
explores the sensitivity of the results to the uncertainties involved in the analysis. Chapter 8
examines various ways of presenting the results to make them more easily accessible, and
interpreting them to understand their meaning and implications fully.
2
City-specific relative contributions (see Section 5.2.4 for explanation).

Chapter Seven
Sensitivity Analysis
7. Introduction
Although the EDRI development and evaluation was treated as a deterministic process
in the preceding chapters, the analysis actually involves many uncertainties. To determine the
robustness of the results, a sensitivity analysis was conducted to explore the potential effect of
each of these uncertainties on the final values of the five main factors and of the EDRI. Since
validation is not possible (see Section 2.5), the sensitivity analysis helps to determine if the
results are meaningful, or if they are rendered insignificant when the uncertainties considered.
Seven possible sources of uncertainty were investigated: indicator definitions, indicator selection,
scaling function, weights, data uncertainty, sample city selection, and city boundary definitions.
The procedure and results for each are described in Sections 7.1 to 7.7. The analysis aimed to
determine the conclusions that can be drawn about the relative city ranks in light of these
uncertainties, and to identify the specific indicators and cities that have the potential to affect the
values of the main factors and of the EDRI most significantly. Because the EDRI measures only
relative values, it is important to examine the sensitivity of relative, not absolute values. It does
not matter if changing the scaling technique increases the EDRI of every city by the same
amount, for example. Only changes in the relative city values are significant. Section 7.8
compares and summarizes the findings of the various components of the sensitivity analysis.
7.1. Indicator definitions
Chapter Seven: Sensitivity Analysis 159

In many cases, it is possible to define an EDRI indicator in more than one way, so that
each variation seems to measure exactly the same concept, but in a slightly different form. In
some cases, the choice among the variations seems arbitrary. For example, “number of
physicians per 100,000 residents” and “number of hospitals per 100,000 residents” could be
defined instead as “number of residents per physician” and “number of residents per hospital.”
It is not immediately obvious how much, or if, it would affect the final main factor and EDRI
values if one of the alternative definitions was used in place of the ones presented in Chapter 4.
To explore the effect of indicator definitions, alternative definitions were developed for three
sample analysis indicators, the analysis was rerun using each in turn, and the new results were
compared to the original results presented in Chapter 6. In addition to the alternative definitions
for “number of physicians per 100,000” and “number of hospitals per 100,000,” an alternative
definition of “population density” was taken to be the ratio of total population to total land area
(versus the weighted average definition provided in Section 4.3.1.3).
Figure 7.1 displays a comparison of the original results with those derived using the
alternative definitions. There is one figure for each of the five main factors and the EDRI, and
on each figure, there is one curve for each set of analysis results. (Note that the scales for each
figure are different.) The alternative definitions for “number of physicians per 100,000” and
“number of hospitals per 100,000” had a negligible effect on the EDRI results. The most that
the Emergency Response and Recovery Capability factor value changed for any city was 4.4%
for Lima and the most that the EDRI value changed for any city was 0.4% for Lima. The
alternative definition for “population density,” however, resulted in changes of 9.7% for the
Vulnerability and 4.8% for the EDRI in Jakarta. Sensitivity to the definitions of “population
density” is less disturbing than sensitivity to those for the “physician” and “hospital” indicators,
because the choice of “population density” definition is supported by reasoning. The chosen
“population density” definition represents the concept of interest more accurately than the
alternative one. In the case of the “physician” and “hospital” indicators, the choice was
160 Chapter Seven: Sensitivity Analysis

arbitrary. It weakens the power of the EDRI if its results are sensitive to arbitrary changes. In
this case, sensitivity to arbitrary changes in the indicator definitions does not appear significant.
3.5
Original
Alternative density def.
3.0
Relative Hazard factor values
Alternative hospitals def.
Alternative physicians def.

2.5
2.0
1.5
1.0
Boston Istanbul Jakarta Lima Manila Mexico San Santiago St. Louis Tokyo
City Francisco
Figure 7.1a. Sensitivity of Hazard factor to indicator definitions
4.5
Original
4.0 Alternative density def.
Relative Exposure factor values

3.5
3.0
2.5
2.0
1.5
1.0
City Francisco
Figure 7.1b. Sensitivity of Exposure factor to indicator definitions

2.8
Original
Relative Vulnerability factor values
2.4 Alternative physicians def.
2.2
2.0
1.8
1.6
1.4
1.2
1.0
City Francisco
Figure 7.1c. Sensitivity of Vulnerability factor to indicator definitions
5.0
Original
Relative External Context factor values

4.0
3.5
3.0
2.5
2.0
1.5
City Francisco
Figure 7.1d. Sensitivity of External Context factor to indicator definitions

3.5
Original
Relative Emerg. Resp. & Recov. factor values

3.0 Alternative hospitals def.
2.5
2.0
1.5
1.0
City Francisco
Figure 7.1e. Sensitivity of Emerg. Response and Recovery factor to indicator definitions
2.7
Original
2.5
Relative EDRI values

2.3
2.1
1.9
1.7
City Francisco
Figure 7.1f. Sensitivity of EDRI to indicator definitions
7.2. Indicator selection
The indicator selection phase of the six-step EDRI development process involved
decisions not only about how to define each indicator, but also about which indicators to include

and which to omit entirely. The next component of the sensitivity analysis explored the effect of
omitting an indicator from the EDRI. A computer program entitled sens.c (Appendix I) was
written to perform this and three other components of the sensitivity analysis (Sections 7.3 to
7.5). The program reads in the weights and the unscaled data values for each city and indicator,
and calculates the five main factor and EDRI values for each city. It then recalculates those
values for each of a series of trials. In each trial, the program makes a change in the input and
repeats the evaluation, keeping track of the original results and those associated with the
variation. Finally, the program computes some simple statistics, including the minimum and
maximum values, and the mean and standard deviation over all relevant trials, for the main
factors and the EDRI, for each city. A trial is considered relevant to a factor if it affects the
value of that factor for at least one city. The indicator selection sensitivity analysis included
thirty-six trials—one to omit each of the thirty-one indicators, and one to omit each of the five
main factors.
The results of the indicator selection sensitivity are displayed in Figure 7.2. There is one
figure for each of the five main factors and the EDRI. Five points are plotted for each city: the
minimum and maximum values over all relevant trials, the mean, and the mean ±standard
deviation for the set of all relevant trials. Omitting an indicator can affect the main factor and
EDRI values significantly. Still, some city rankings are clear, even considering the range of
values that could result from eliminating a single indicator. For example, the Exposure factor
value should be greatest for Tokyo, even if one of the indicators is omitted. Furthermore, while
it seems difficult to rank each city definitively if the possibility of changing the indicator set exists,
for each main factor and the EDRI, the sample of cities can, with some confidence, be
separated into groups with the similar values. For example, it seems that Santiago, Jakarta,
Mexico City have relatively moderate Hazard; Boston, San Francisco, Tokyo, Manila, and
Istanbul have relatively high Hazard; and Lima has a very high Hazard. Table 7.1 summarizes
the suggested categorical groupings that are possible for each factor, for the indicator selection,
weight, and data uncertainty components of the sensitivity analysis.

4.0
3.5
3.0
Relative Hazard factpr values
2.5 Min
Mean-sd
Mean
2.0
Mean+sd
Max
1.5
1.0
0.5
City Francisco
Figure 7.2a. Sensitivity of Hazard factor to indicator selection
5.0
4.5
4.0
3.5
Min
3.0 Mean-sd
Mean
Mean+sd
2.5
Max
2.0
1.5
1.0
City Francisco
Figure 7.2b. Sensitivity of Exposure factor to indicator selection

3.0
2.5
Min
2.0
Mean-sd
Mean
Mean+sd
1.5 Max
1.0
0.5
City Francisco
Figure 7.2c. Sensitivity of Vulnerability factor to indicator selection
5.0
4.5
4.0
3.5
Min
3.0 Mean-sd
Mean
Mean+sd
2.5
Max
2.0
1.5
1.0
City Francisco
Figure 7.2d. Sensitivity of External Context factor to indicator selection

3.3

2.8
2.3
Min
Mean-sd
Mean
1.8 Mean+sd
Max
1.3
0.8
City Francisco
Figure 7.2e. Sensitivity of Emergency Response & Recovery factor to indicator selection
3.3
3.1
2.9
2.7
2.5 Min
Mean-sd
2.3 Mean
Mean+sd
Max
2.1
1.9
1.7
1.5
City Francisco
Figure 7.2f. Sensitivity of EDRI to indicator selection

Table 7.1. Summary of categorical groupings from sensitivity analysis
Categories are labeled from A to E, with A for the lowest values and E for the highest.
Sensitivity to indicator selection

Influential Mexico San
indicators Boston Istanbul Jakarta Lima Manila City Francisco Santiago St. Louis Tokyo
Hazard --- C-D C-D B-C E D B D B C D
Exposure xe5, xe6 C-D B-C C B-C C D D B C E
Vulnerability xv2, xv3, xv4 B-C B-C C-D C-D C-D B-C A C B-C A
External Context xc1, xc2 B-C B C C C C B-C C B-C E
Emerg. Resp. & Recov. xr1, xr9 C D D-E C E C-D B-C D-E C B
EDRI xh, xe, xv C C C-D D C-D C-D C C C D-E
Sensitivity to weights
Hazard --- C D B E D B D B C D
Exposure xe5, xe6 D B C B-C C D D B C E
Vulnerability xv1, xv2, xv3 B-C B-C C C D C A C B A
External Context xc1, xc2 C B C C C C C C B E
Emerg. Resp. & Recov. xr1, xr2, xr7, xr9 C D D-E C E D C E C B
EDRI C C C D D C C C B E
Sensitivity to data uncertainty

Hazard xh1, xh2 C-D C-D B E C-D B D B C D
Exposure xe5, xe6 D B C B C D D B C E
Vulnerability xv1 C C E D E C A D C A
External Context xc1 C B C C C D C C B E
Emerg. Resp. & Recov. xr1 C C-E D-E C E C-D C D-E C B
EDRI xr1, xh2 C B-C C D D C B-C B-C B E
Eliminating some indicators has a more significant effect than others. For example,
omitting “population” or “per capita GDP” causes the largest changes in the Exposure factor
values. For the indicator selection, weight, and data uncertainty components of the sensitivity
analysis, Table 7.1 lists the indicators that are most “influential” to each factor, i.e., those that
lead to the minimum and maximum values over all relevant trials. The degree of sensitivity to the
indicator selection uncertainty is similar across factors and cities.
The sensitivity of the results to the indicator selection need not be considered a
deficiency in the EDRI. Changing the indicator set in effect alters the concept that the index
measures, so the relative values of different cities can be expected to change as well. In the
extreme, if all indicators except “population” are omitted, the resulting “EDRI” would measure
not earthquake disaster risk, but the number of people in the city. One would not expect the
city rankings to remain the same in that case.
7.3. Scaling
Section 5.2 introduced seven scaling techniques that were considered for use in the
EDRI, and explained why scaling with respect to the mean minus two standard deviations was
chosen for use in the sample analysis. The third component of the sensitivity analysis used the
computer program sens.c to investigate the effect of the choice of scaling technique on the final
main factor and EDRI values. The sample analysis was repeated four times, each time using
one of the following scaling techniques: mean minus two standard deviations, minimum and
maximum observed values, minimum and maximum possible values, and base city. Appendix J
lists the maximum and minimum possible values that were used for the third scaling option.
Clearly, the scaled values will not be the same for each scaling technique. The values
from the first will be centered around a mean of two; values from the second and third will
remain between zero and one; and values from the fourth technique will be more varied, with the
value for the base city, taken as San Francisco in the sample analysis, always equal to one.

Because the analysis aims to understand the relative values among cities, the ratio of the scaled
value of one city to that of another city might be of interest. However, even the ratios of the
scaled values of two cities will not be identical across scaling techniques. Table 7.2 illustrates
that for each scaling technique, the ratio of the City A scaled value of indicator i to the City B
scaled value of indicator i (x'iA/x'iB) measures a different quantity, and that therefore, the values
of those ratios should not be expected to be equal for each technique.
Table 7.2. Comparison of scaling techniques tested in sensitivity analysis

Scaling method Ratio of City A scaled value of indicator i to City B scaled value of i
Mean minus two x' iA (x iA − ( x i − 2s i )) si (x iA − (x i − 2s i ))
standard = * =
x' iB si (x iB − ( x i − 2s i )) (x iB − (x i − 2s i ))
deviations
Base city and x' iA x iA x i,basecity x iA
year = * =
x'iB x i,basecity x iB x iB
Minimum and x' iA (x iA − mini ) (max i − min i ) (x iA − min i )
= * =
maximum x'iB (max i − min i ) (x iB − mini ) (x iB − mini )
For a given indicator, if the above ratio (x'iA/x'iB) is greater than one for one scaling
technique, it should be greater than one for all four techniques, although its exact value may
vary. In general, except for the case of base city scaling, the same will hold true when
considering the values of the five main factor and EDRI composite indexes instead of the values
of a single indicator. The conclusion is not necessarily true when using the base city scaling
technique, because that option introduces an implicit weighting that alters the relative importance
of the indicators, and thereby alters the value of a composite index composed of those
indicators (see Section 5.2.4).
Figure 7.3 presents the results of the scaling sensitivity analysis. There is one figure for
each of the five main factors and the EDRI, and on each figure, there is one curve for each
scaling technique. Despite the analytical evidence that the results derived using each of the four
scaling techniques should not be identical, it turns out that the final

3.5
3.0
2.5
mean-2s
2.0 min/max
observed
min/max
possible
base city
1.5
1.0
0.5
0.0
City Francisco
Figure 7.3a. Sensitivity of Hazard factor to scaling function
4.5
4.0
3.5
3.0
mean-2s
2.5 min/max
observed
min/max
possible
2.0 base city
1.5
1.0
0.5
0.0
City Francisco
Figure 7.3b. Sensitivity of Exposure factor to scaling function

5.0
4.5
4.0
3.5
mean-2s
3.0
min/max
observed
2.5 min/max
possible
base city
2.0
1.5
1.0
0.5
0.0
City Francisco
Figure 7.3c. Sensitivity of Vulnerability factor to scaling function
16
14
12
10 mean-2s
min/max
observed
8 min/max
possible
base city
6
0
City Francisco
Figure 7.3d. Sensitivity of External Context factor to scaling function

4.0
3.5

3.0
2.5 mean-2s
min/max
observed
2.0 min/max
possible
base city
1.5
1.0
0.5
0.0
City Francisco
Figure 7.3e. Sensitivity of Emergency Response and Recovery factor to scaling function
4.0
3.5
3.0
mean-2s
2.5
min/max
observed
2.0 min/max
possible
base city
1.5
1.0
0.5
0.0
Boston Istanbul Jakarta Lima Manila Mexico City San Santiago St. Louis Tokyo
Francisco
Figure 7.3f. Sensitivity of EDRI to scaling function

relative values of the five main factors and the EDRI are similar in most cases, as evidenced by
the similar shapes of the four curves on a given plot. The base city technique provides a few
notable exceptions. For example, while the other three techniques conclude that the
Vulnerability in Tokyo is less than that in St. Louis, the base city method indicates that the
reverse is true. This conclusion is the result of the implicit weighting associated with using San
Francisco as the base city. Figure 7.4a shows how the range of values associated with each of
the unscaled vulnerability indicators are very different (on the order of thousands for “population
density”; tenths for “percentage of population aged 0-4 or 65+). When scaled with the mean
minus two standard deviations and the minimum and maximum observed techniques, the values
are forced to exhibit roughly similar magnitudes and dispersions across indicators (Figs. 7.4b
and 7.4c). The same is not true of the base city method (Fig. 7.4d). Because San Francisco’s
population density is relatively small compared to other cities, when using this technique with
San Francisco as the base city, the scaled “population density” values become very large and
spread out relative to the other indicator values. In effect, “population density” becomes more
important than the other indicators. Therefore, because Tokyo’s “population density” is larger
than St. Louis’s, and “population density” is given such a large implicit weight by the base city
scaling technique, Tokyo’s overall Vulnerability value becomes larger than St. Louis’s with that
scaling option. Similar examples of the implicit weighting effect of base city scaling can be seen
in the Emergency Response and Recovery Capability and EDRI values.

100000 5
Mean minus two st. dev.

scaled indicator values
Unscaled indicator values
10000 4
1000 3
100 2
10 1
1 0
0 Seismic City Pop’n. City % pop. -1

City Seismic City City Pop’n. City % pop.
code wealth age density dvlpt. aged 0-4
or 65+ code wealth age density dvlpt. aged 0-4
speed speed or 65+
a. Dispersion of unscaled vulnerability indicators b. Dispersion of vulnerability indicators scaled with
mean minus two standard deviations
Base city scaled indicator values

20
1
scaled indicator values
15
Min./max/ observed
10
0 0
-5
Seismic City City Pop’n. City % pop. Seismic City City Pop’n. City % pop.
code wealth age density dvlpt. aged 0-4 code wealth age density dvlpt. aged 0-4
speed or 65+ speed or 65+
c. Dispersion of vulnerability indicators scaled with min. d. Dispersion of vulnerability indicators scaled with
and max. observed values base city
Figure 7.4. Illustration of implicit weighting effect of base city scaling technique
7.4. Weights
The computer program sens.c was employed again to explore the effect of the weights
on the final main factor and EDRI values. The weight portion of the sensitivity analysis included
seventy-two trials, one to change each of the thirty-one indicators and five main factor weights
by +35%, and one to change each by -35%. The 35% value was taken from the fact that the
coefficients of variation of the questionnaire factor weight assessments were approximately 0.35
(see Section 5.3.4.2). For each factor and city, the program computed the minimum and
maximum values achieved over all seventy-two trials, and the mean and standard deviation of
the same seventy-two values.
Some variation in the five main factor and EDRI values results when any single weight is
altered by ±35% (Fig. 7.5). As with the indicator selection analysis, however, some city pair
comparisons remain clear, even considering the range of values, and for each factor, the sample
of cities can be divided into a few groups with similar ranking. Table 7.1 summarizes, for each
factor, the suggested categorical groupings and the indicator weights that lead to the minimum
and maximum values over all trials.
It should not be surprising that altering the values of the weights affects the final values of
the main factors and the EDRI. Just as the indicator set helps define the concept that is being
measured, so do the weight values. If the weights are changed, the concept being measured is
too, and the city rankings corresponding to the new concept should not necessarily equal those
associated with the original concept. In the extreme example given in Section 5.3, if the weight
of the Hazard factor is taken as one, and all other factor weights are assumed to be zero, the
resulting “EDRI” will measure a city’s earthquake hazard, not it’s earthquake disaster risk. Still,
since the weight values are difficult to assess, it was hoped that the final results would not vary
too drastically with minor changes in the assessments of the weights. The sensitivity analysis
suggests that that is indeed the case.

3.5
Relative Hazard factor values 3.0
2.5
Min
Mean-sd
Mean
2.0 Mean+sd
Max
1.5
1.0
City Francisco
Figure 7.5a. Sensitivity of Hazard factor to weights
4.5
4.0
3.5
3.0
Min
Mean-sd
2.5 Mean
Mean+sd
Max
2.0
1.5
1.0
City Francisco
Figure 7.5b. Sensitivity of Exposure factor to weights

2.8
2.6
2.4
2.2
2.0 Min
Mean-sd
1.8 Mean
Mean+sd
Max
1.6
1.4
1.2
1.0
City Francisco
Figure 7.5c. Sensitivity of Vulnerability factor to weights
5.0
4.5
4.0
3.5
Min
3.0 Mean-sd
Mean
Mean+sd
2.5
Max
2.0
1.5
1.0
City Francisco
Figure 7.5d. Sensitivity of External Context factor to weights

3.0
2.8

2.6
2.4
2.2
Min
2.0 Mean-sd
Mean
1.8 Mean+sd
Max
1.6
1.4
1.2
1.0
City Francisco
Figure 7.5e. Sensitivity of Emergency Response and Recovery factor to weights
2.9
2.7
2.5
Min
2.3 Mean-sd
Mean
Mean+sd
Max
2.1
1.9
1.7
City Francisco
Figure 7.5f. Sensitivity of EDRI to weights

7.5. Data uncertainty
Since changing the indicator set, the weight values, or the scaling technique necessarily
will and should change the final values of the main factors and the EDRI, sensitivity to those
three components cannot be considered deficiencies in the EDRI model. On the other hand,
sensitivity to data uncertainty theoretically is avoidable if the data are of sufficiently high quality.
If the final results are too sensitive to data uncertainty, it does weaken their ultimate usefulness.
The computer program sens.c was used once more to conduct the analysis of sensitivity to data
uncertainty. The analysis included 620 trials, one to change each of the indicator-city values by
+∆xij, and one to change each by -∆xij. The ∆xij values (Appendix K) were assessed for each
indicator-city combination separately based on the assumed quality of the associated data
source, and the degree of estimation and manipulation required to extract the required data from
the reported data.
As with the previously discussed sensitivity analysis results, varying the data to reflect
the uncertainty in their values results in some variation in the final main factor and EDRI values.
As shown in Figure 7.6, the variation is not uniform across factors or cities. Exposure and
External Context exhibit noticeably smaller variations than the other factors, because they rely
on data that are less uncertain than the other factors do (e.g., “population,” “per capita GDP”
are relatively more certain than “percentage of the urbanized area with high liquefaction
susceptibility”). The standard deviations generally appear small for the data uncertainty because
for a given city and factor, the relevant trials include any trial that changes the value for any city
and indicator related to that factor. For example, the trial in which Boston’s “population”
indicator is increased by ∆xpop,Boston is a relevant trial for Manila’s Exposure, because changing
the value for Boston changes the mean and standard deviation of “population,” which in turn
affects Manila’s “population” and Exposure values, but only slightly. In general, for a given city,
the minimum and maximum over all relevant trials are associated with changes to an indicator
value for that city. They are not unlikely to occur.

Although cities cannot be compared as precisely when data uncertainty is considered,
certain city comparisons are still possible, and the sample cities can be partitioned into groups
with similar values with some confidence. Table 7.1 summarizes, for each factor, the suggested
categorical groupings and the specific indicators that are responsible for the most significant
variation in the final results.
3.5
3.0
2.5
Min
Mean-sd
Mean
2.0 Mean+sd
Max
1.5
1.0
City Francisco
Figure 7.6a. Sensitivity of Hazard factor to data uncertainty

4.5
4.0
3.5
3.0 Min
Mean-sd
Mean
2.5
Mean+sd
Max
2.0
1.5
1.0
City Francisco
Figure 7.6b. Sensitivity of Exposure factor to data uncertainty
2.8
2.6
2.4
2.2
2.0 Min
Mean-sd
1.8 Mean
Mean+sd
Max
1.6
1.4
1.2
1.0
City Francisco
Figure 7.6c. Sensitivity of Vulnerability factor to data uncertainty

5.0
4.5
4.0
3.5 Min
Mean-sd
Mean
3.0
Mean+sd
Max
2.5
2.0
1.5
City Francisco
Figure 7.6d. Sensitivity of External Context factor to data uncertainty
3.5
3.0
2.5
Min
Mean-sd
Mean
2.0 Mean+sd
Max
1.5
1.0
City Francisco
Figure 7.6e. Sensitivity of Emergency Response and Recovery factor to data uncertainty

2.8
2.6
2.4
Min
2.2 Mean-sd
Mean
Mean+sd
Max
2.0
1.8
1.6
City Francisco
Figure 7.6f. Sensitivity of EDRI to data uncertainty
7.6. Sample city selection
Sensitivity of the final main factor and EDRI values to the set of cities used in the sample
was conducted similarly to the indicator definition sample analysis. Three alternative sets of
sample cities were considered, the EDRI evaluation was repeated using each, and the results
were compared with those obtained for the original ten-city sample. The alternative sample city
sets are (1) the original ten cities, not including Istanbul; (2) the original ten cities, not including
Tokyo; and (3) the original ten cities, not including Tokyo or Istanbul.
The results of the original and the three alternative city samples are displayed in Figure
7.7. Removing Istanbul from the sample of cities did not affect the results significantly. For a
given city, the main factor and EDRI values changed by less than 5%. Removing Tokyo from
the sample of cities had a much more pronounced effect on the results, particularly for Exposure
and External Context. For example, when Tokyo was removed, the value of the Mexico City

Exposure jumped from 2.237 to 3.502, and the ratio of Mexico City Exposure to Lima
Exposure, from 1.500 to 2.193. As a result of the changes in Exposure and External Context,
the overall EDRI values changed significantly as well. Removing both Tokyo and Istanbul
produced results similar to those associated with removing only Tokyo.
3.5
Original
Without Istanbul
3.0
Without Tokyo
Without Tokyo or Istanbul
2.5
2.0
1.5
1.0
City Francisco
Figure 7.7a. Sensitivity of Hazard factor to city sample
4.5
Original
4.0 Without Istanbul
Without Tokyo
3.5
3.0
2.5
2.0
1.5
1.0
City Francisco
Figure 7.7b. Sensitivity of Exposure factor to city sample

2.8
Original
2.6 Without Istanbul
Without Tokyo
2.4 Without Tokyo or Istanbul
2.2
2.0
1.8
1.6
1.4
1.2
1.0
City Francisco
Figure 7.7c. Sensitivity of Vulnerability factor to city sample
5.0
Original
4.5
Without Istanbul
4.0 Without Tokyo
3.5 Without Tokyo or Istanbul
3.0
2.5
2.0
1.5
1.0
City Francisco
Figure 7.7d. Sensitivity of External Context factor to city sample

3.5
Original
Without Istanbul
Relative Emerg. Resp. & Recov. values

3.0 Without Tokyo
2.5
2.0
1.5
1.0
City Francisco
Figure 7.7e. Sensitivity of Emergency Response and Recovery factor to city sample
2.7
Original
Without Istanbul
2.5
Without Tokyo
2.3
2.1
1.9
1.7
1.5
City Francisco
Figure 7.7f. Sensitivity of EDRI to city sample
While, at first glance, this limited investigation seems to suggest that the final main factor
and EDRI results are quite sensitive to the city sample, that conclusion may not hold in general.
When Istanbul is removed from the sample, the change in the results is negligible. The profound
effect of removing Tokyo can be attributed to two extenuating circumstances. First, the original

sample has only ten cities, so any effect is more pronounced when changing the sample from ten
to nine cities, than it would be if it was changed from 301 to 300. Second, Tokyo is a city
unique among all cities worldwide, in that it is so large and wealthy that it breaks the curves, so
to speak, on Exposure and External Context. Ultimately, when the number of cities in the
sample is on the order of two or three hundred, varying the sample by adding or removing a few
cities should not have a significant effect on the results, especially if Tokyo is not among those
cities that are added or removed.
7.7. City boundary definitions
The way in which each city is defined can affect the final EDRI values. Each city may
be defined as only the area within its legal city limits, as an expansive greater metropolitan area
that includes the downtown and the surrounding suburbs, or as any variation in between. In the
sample analysis, each city is defined as a greater metropolitan area that includes several
surrounding counties, prefectures, or similar administrative units (see Appendix B). Even
without a detailed analysis, it is clear that if a city is defined by its legal city limits or some other
standard instead, the results will change significantly. For example, the population of Tokyo
prefecture is 11.9 million, versus 35.5 million in the five-prefecture greater metropolitan area.
Similarly, if San Francisco is defined as the ten-county greater metropolitan area, the population
density is 1075 people per sq. km.; if only the area contained within its legal city is used, the
population density becomes 5947 people per sq. km. Clearly, the final main factor and EDRI
values are highly sensitive to the city boundary definitions.
7.8. Conclusions
Each aspect of the sensitivity analysis explore the effects of a single variation in the
EDRI input or model, e.g., one indicator-city unscaled data value, one weight, the scaling
technique, or the sample of cities. These effects are not mutually exclusive, however. What
would happen if the population of San Francisco was actually delta more than the assumed

value, and the urbanized land area of Manila was actually delta less than the assumed value?
While the sensitivity analysis would be more comprehensive if it investigated the effects of each
combination of these variations, the large number of possible combinations makes such an
analysis prohibitively complex. Instead, let it suffice to compute the sensitivity to each
component individually, and note that in many cases, the effects of multiple changes will add
linearly, since the EDRI is a linear combination, and that it is possible that even the rankings that
held under all conditions tested in the above analyses (e.g., in all 620 data uncertainty trials,
Lima exhibited the highest Hazard value) may change if multiple uncertainties are considered
simultaneously.
Spearman’s rank correlation coefficient (rs) measures the correlation between two sets
of ranks. The following formula is used to evaluate the Spearman coefficient when there are no
ties in the ranks:
 6∑ (u i − vi ) 2 
rs = 1 −  i 2 
n(n − 1)  (7.1)
 
where ui and vi are the ranks of the ith observation in samples 1 and 2, respectively, and n is the
number of pairs of observations (number of observations in each sample). In the case of the
EDRI sensitivity analysis, each observation corresponds to one city in the sample (n = 10), one
of the samples is always the original results from Chapter 6, and the other sample is from the
results computed after the input data are varied.
Table 7.3 compares the results of five components of the sensitivity analysis—indicator
definitions, indicator selection, scaling function, weight values, and data uncertainty. Sample city
selection sensitivity is not included in the table, because each trial has a different number of
cities, so the Spearman coefficient cannot be calculated. The Spearman coefficient values are
all very high, indicating that the rankings do not change dramatically in the face of the various
uncertainties, especially those associated with the weights and data. As expected, scaling
produces the lowest Spearman

Table 7.3. Comparison of average Spearman rank correlation coefficient (rs) values
Components of sensitivity analysis
Indicator Indicator Scaling Weights Data
definition selection uncertainty
Hazard 0.996 0.990 0.926 0.998 0.980
Exposure 1.000 0.994 0.953 1.000 1.000
Vulnerability 0.960 0.982 0.965 0.998 0.996
Ext. Context 1.000 0.984 0.823 0.994 1.000
Emer. Resp. & Recov. 0.992 0.996 0.953 1.000 0.996
EDRI 0.968 0.927 0.911 0.989 0.981
Num. of trials 3 36 3 72 620
coefficient values, because of the base city scaling option. Another way to compare the various
components of the sensitivity analysis is simply by comparing Figures 7.1 to 7.3, and 7.5 to 7.7.
As mentioned before, sensitivity to indicator selection, scaling function, weight values, and city
boundary definitions do not indicate deficiencies in the EDRI model. Sensitivity to indicator
definitions, data uncertainty, and the sample city selection, however, should be minimized if
possible.
The results of the sensitivity analysis can be summarized in the following two
conclusions. First, the many uncertainties involved in developing and evaluating the EDRI lead
to a range of possible values for the main factor and EDRI of each city. In some cases, the
ranges of values for two cities overlap, making it unclear which city has a higher value for the
associated factor. Nevertheless, even considering the range of possible values, it is clear that
some cities have higher factor or EDRI values than others. Furthermore, the sample of cities
can be divided into groups with similar values (e.g., low, medium, and high Hazard). While
establishing broad categorical groups is less precise than declaring than City A has a Hazard
that is 1.35 times as large as that of City B, it is still extremely valuable information that is not
currently available from any other source, at least not in such a systematically developed way.
Even if all that can be confidently extracted from the results is that five cities have a low EDRI
and the other five have a high EDRI, it may be that each of the five high EDRI cities have a high

risk for a different reason, and the disaggregated results will show that. The EDRI results are
meaningful, even when the uncertainties in the analysis are considered.
The second conclusion of the sensitivity analysis is that any improvement in the quality of
data or the assessment of the weights will improve the ability of the EDRI to provide definitive
conclusions about the relative values of the five main factors and the overall EDRI. It is worth
some time and effort to make the input data as reliable as possible. Even with the limited
resources available for the sample analysis presented here, however, the EDRI results provide
valuable insights into the relative earthquake disaster risk of different cities.

Chapter Eight
Presentation and Interpretation of Results
8. Introduction
The tables and figures presented in Chapters 6 and 7 tell a story. They contain
information that may not be immediately obvious at first glance into the sea of numbers. This
chapter is about extracting that information and translating it into a language that its potential
users can understand. The story describes the big picture of urban earthquake disaster risk—
how serious it is, how widespread, how it compares among cities worldwide, and what
contributes to it. There are also some smaller storylines that can be uncovered by examining
groups of cities with similar characteristics (e.g., industrialized versus developing countries), or
comparing pairs of factors or pairs of cities to determine specific information. Following a
general discussion of the goals of the presentation phase of the EDRI development, this chapter
introduces the results of the presentation and interpretation step for the sample analysis. The
many graphical forms used to present the EDRI results are displayed and interpreted, until the
entire story has been told.
8.1. Goals of the presentation
One of the primary potential uses of the EDRI is to raise awareness of the issues
involved in urban earthquake disaster risk among decision-makers and interested members of
the public, and to disseminate the current state of knowledge of earthquake risk. The
presentation of the results, therefore, is critical to ensuring that the information contained in the
EDRI is easily understood by its potential users.
Raising awareness requires capturing the attention of an audience that may not have a
well-developed interest in the topic yet. The presentation, therefore, must be engaging. It
should be creative, relying on whatever combination of figures, graphs, tables, and prose
summaries convey the EDRI’s information so that it is as clear, and easy to understand and
interpret as possible, keeping in mind that the audience may range from experts within the
earthquake community to the uninformed public. The presentation must convey the information
contained in the EDRI accurately. There are many uncertainties in the EDRI analysis, so it
would be irresponsible to imply a higher degree of precision than the analysis warrants.
Nonetheless, as the sensitivity analysis demonstrates, the EDRI does contain a lot of valuable
information, and that should be communicated.
As a summary of urban earthquake risk worldwide, the EDRI should serve as a starting
point for earthquake risk information. It cannot possibly convey all the relevant information that
is currently known, but it can provide summary and comparative information about the many
aspects of our natural, built, and social environments that contribute to earthquake risk, and the
wide range of geographical areas that are affected by the problem. That synopsis can help raise
awareness of the nature and magnitude of urban earthquake risk, and spark interest in the
challenge of mitigating that risk. The EDRI presentation should raise questions, and direct the
user to other sources that can provide more detailed information. Specifically, the presentation
of the EDRI analysis results should convey the relative overall earthquake disaster risk of
different cities, relative values of each of the five main contributing factors across cities, and
relative contributions of the main factors within each city.
Consider three possible forms of presentation for the EDRI results: categorical
groupings, the numerical form used in Chapter 6, and a rescaled numerical rating system. Using
categorical groupings has the advantage of being easy to understand, particularly if the
Chapter Eight: Results 193

categories are given labels with intuitive understanding, such as Low, Medium, and High
earthquake disaster risk. The disadvantage of the approach is that the low resolution makes it
impossible to distinguish between cities at the low and high ends of a category. The second
option, using the scaled EDRI values from Table 6.3 directly, has higher resolution, but is not
user-friendly. Expressing a city’s EDRI value as a number to three decimal places makes the
values appear more precise than they are. For most users, a simple integer rating scale from
one to one hundred would be easier to understand than an open-ended scale in which cities
have values between 1.772 and 2.653. For this reason, the final main factor and EDRI values
are rescaled for the presentation using the following equation:
 (Fscaled as in Chap. 6 − minvalue) 

Frescaled for presentation =  99 * +1
 (maxvalue − minvalue) 
(8.1)
where minvalue and maxvalue are taken as zero and five in the sample analysis, respectively,
and F can be any of the five main factors or the EDRI. It would be possible to use different
values for minvalue and maxvalue, but zero and five seem natural choices given the sample
analysis factor and EDRI values, as explained below. The resulting values of the five main
factors and the EDRI will be expressed as integers between one and one hundred.1 Given the
Section 5.2.3 discussion of the deficiencies of the scaling with respect to minimum and maximum
possible values, it may seem inappropriate to perform this final rescaling. Nevertheless, the
circumstances associated with the task make this scaling technique appropriate in this case.
These rescaled final main factor and EDRI values are not going to be combined into any other
composite index, so the choice of minimum and maximum possible values will have no
consequence, except to determine how clustered or dispersed the values appear in Figures 8.1
and 8.2. This final scaling puts artificial bounds on the range of possible EDRI values, when the
EDRI is meant to measure risk on an open-ended scale. Nevertheless, from the user’s point-
of-view, there would appear to be bounds anyway, because even if the values were not
rescaled, a range still would be chosen for the y-axis of Figures 8.1 and 8.2. That range would
probably be zero to five. By using zero and five for the minvalue and maxvalue then, the final
1
The scaling does not include the value zero in its range, because while the risk in some cities may be
relatively very small, there is always some risk.
194 Chapter Eight: Results

rescaling simply serves to change the values of the y-axis from zero to five, to one to one
hundred. The relative heights of the sample city columns in the graph appear the same with or
without the final rescaling. Finally, this technique achieves the goal of expressing the final main
factor and EDRI values as simple integers that are more agreeable to the user than the
alternative fractional values.
8.2.1. Overall EDRI and five main factor values
The first components of the presentation of the sample analysis results are the bar charts
that represent the overall EDRI, the five main factors, and their components. Those tables and
their accompanying table of values are displayed in Figures 8.1 and 8.2, and Table 8.1. As
discussed in the previous section, to improve the presentation and interpretation, the relative
values of the EDRI, the five main factors, and their components have been rescaled so that the
values range from one to one hundred. Using Figures 8.1 and 8.2, the results are briefly
discussed below, factor by factor.
According to Figure 8.2a, both short- and long-term seismicity are highest for Lima.
Short-term seismicity is also relatively high in Istanbul, Manila, San Francisco, and Tokyo.
Long-term seismicity, however, is highest in Boston, Istanbul, San Francisco, St. Louis, and
Tokyo. Combining the short- and long-term seismicity with the soft soil indicator results in
overall ground shaking component values that are dominated by Lima, roughly similar for
Boston, Istanbul, Manila, San Francisco, St. Louis, and Tokyo, and somewhat less for Jakarta,
Mexico City, and Santiago. The relative values of the collateral hazard component are highest
for Jakarta, Manila, and Tokyo, driven by the liquefaction, liquefaction and fire, and tsunami
hazards, respectively. Collateral hazard is lowest for Mexico City and Santiago.

EDRI = 0.25*(Hazard) + 0.25*(Exposure) + 0.25*(Vulner.) + 0.125*(Ext. Context) + 0.125*(Emer. Resp.)
100
90
80
70
Relative values
60
50
40
30
20
10
0
Hazard Exposure Vulnerability External Emer. Resp. EDRI
Context & Recov.
Boston Istanbul Jakarta Lima Manila Mexico City San Francisco Santiago St. Louis Tokyo
Figure 8.1. Sample analysis results: Relative values of EDRI and five main factors
HAZARD = Ground shaking + Collateral
100 Ground shaking = 0.425*short-term seis. + 0.425*long-term seis. + 0.15*soil
90 Collateral = 0.33*liq. + 0.33*tsunami + 0.33*fire
80
70
Relative values
60
50
40
30
20
10
0
short-term long-term soft soil Ground liquefaction tsunami fire Collateral HAZARD
seismicity seismicity shaking following hazards
Figure 8.2a. Sample analysis results: Relative values of Hazard components

EXPOSURE = 0.4*Physical Infra. + 0.4*Population + 0.2*Economy
100
90
80
70
Relative values
60
50
40
30
20
10
0
Physical Infrastructure Population Economy EXPOSURE
Figure 8.2b. Sample analysis results: Relative values of Exposure components

100
VULNERABILITY = 0.95*Physical Infra. + 0.05*Population
90
80
70
Relative values
60
50
40
30
20
10
0
Physical Infrastructure Population VULERABILITY
Figure 8.2c. Sample analysis results: Relative values of Vulnerability components

100
EXTERNAL CONTEXT = 0.8*Economic + 0.2*Political
90
80
70
Relative values
60
50
40
30
20
10
0
Economic Political EXTERNAL CONTEXT
Figure 8.2d. Sample analysis results: Relative values of External Context components
100
EMERG. RESP. & RECOV. = 0.33*Planning + 0.33*Resources + 0.33*Mob. & access
90
Resources = 0.5*financial + 0.25*equip. + 0.25*manpower
80
70
Relative values
60
50
40
30
20
10
0
Planning financial equipment trained Resources Mobility EMERG.
resources & facility manpower and access RESP. &
resources resources RECOV.
Figure 8.2e. Sample analysis results: Relative values of Emergency Response. and Recovery components
Table 8.1. Sample analysis results for Figures 8.1. and 8.2
Indicators Mexico San
Factor included Factor components Boston Istanbul Jakarta Lima Manila City Francisco Santiago St. Louis Tokyo
Hazard xh1 short-term seismicity 28 41 21 88 49 21 46 32 29 49
xh2 long-term seismicity 44 53 14 75 30 12 48 27 58 44
xh3 soft soil 63 38 28 51 26 82 41 25 20 32
xh1-xh3 Ground shaking 40 46 19 77 37 27 46 29 40 45
xh4 liquefaction 54 21 71 43 72 19 33 18 38 37
xh5 tsunami 20 43 43 65 43 20 65 20 20 65
xh6-xh7 fire following 38 46 38 28 55 31 39 46 37 48
xh4-xh7 Collateral hazards 38 37 51 45 57 23 46 28 32 50
HAZARD 40 43 29 68 43 26 46 29 37 46
Exposure xe1-xe4 Physical Infrastructure 40 31 32 31 32 45 43 28 34 89
xe5 Population 32 31 37 34 37 54 33 29 26 93
xe6 Economy 66 26 25 24 24 27 65 26 57 66
EXPOSURE 42 30 33 31 32 45 44 28 35 86
Vulner- xv1-xv5 Physical Infrastructure 37 42 50 48 54 43 25 46 37 24
ability xv6 Population 70 28 19 24 29 32 56 44 75 28
VULERABILITY 39 41 49 46 53 42 26 46 39 24
External xc1 Economic 38 33 33 32 33 36 39 33 33 97
Context xc2-xc3 Political 26 26 47 44 44 46 26 44 25 76
EXTERNAL CONTEXT 35 32 36 35 35 38 36 35 32 93
Emerg. xr1 Planning 34 34 68 34 68 34 17 68 34 17
Resp. & xr2-xr3 financial resources 20 51 43 59 57 54 23 52 27 18
Recovery xr4-xr5 equipment & facility resources 56 60 56 33 74 68 70 76 48 -4
xr6 trained manpower resources 10 39 66 41 67 38 16 60 25 42
xr2-xr6 Resources 27 47 46 44 60 50 33 57 29 14
xr7-xr9 Mobility and access 39 55 41 31 49 34 39 39 41 38
EMERG. RESP. & RECOV. 33 45 51 36 59 39 30 54 35 23
EDRI 39 38 39 45 44 38 37 37 36 54
Tokyo dominates the physical infrastructure and population components, and therefore
the overall values of the Exposure factor (Fig. 8.2b). Boston, Mexico City and, San Francisco
have the next highest physical infrastructure exposure. Mexico City holds a commanding
second place for population exposure. Economy exposure has two tiers, the three U.S. cities
and Tokyo having substantially higher values than the other cities.
Physical infrastructure vulnerability is highest in Jakarta, Lima, Manila, and Santiago;

lowest in San Francisco and Tokyo (Fig. 8.2c). Boston, San Francisco, and St. Louis post the
highest values for the much less important population component. Overall vulnerability rankings
are lowest for San Francisco and Tokyo, and highest for Jakarta, Lima, Manila, and Santiago.
Figure 8.2d shows that Tokyo dominates both major components of External Context,
economic and political. Boston, Mexico City, and San Francisco have the next highest
economic external context values. Jakarta, Lima, Manila, Mexico City, and Santiago, all
country capitals, constitute the next highest tier of political external context values.
Planning values are concentrated at three levels (Fig. 8.2e). Jakarta, Manila, and
Santiago have the highest values for planning (i.e., associated with the worst planning and
highest risk); San Francisco and Tokyo have the lowest values; and the remaining cities are in
the middle. Tokyo is set apart with the lowest value for resources (i.e., it has the most
resources available), with the U.S. cities being the three runners-up. Manila and Santiago have
the highest resources values (i.e., they have the fewest resources). Most cities have similar
mobility and access values, except Istanbul and Manila, which are somewhat higher, and Lima
and Mexico City, which are slightly lower.
Combining the five main factors results in overall EDRI values that are similar for most
cities (36 to 39). Only Tokyo is significantly set apart with a relative EDRI value (54) that is
much higher than the other cities, because of its high exposure and external context values.

Lima and Manila exhibit the next highest overall earthquake disaster risk, with values of 45 and
44, respectively.
8.2.2. Relative contributions of different factors
A pie chart for each city in the sample analysis displays the city-specific relative
contributions of the five main factors to the overall EDRI of that city (Fig. 8.3). In interpreting
the information in Figure 8.3, remember that the five main factor weights are not included in the
computation of relative contribution percentages (see Section 6.3).
Two observations are immediately clear on comparison of the ten figures. First, in
every city, all five factors contribute significantly to the risk. Only in two cases does a factor
have a relative contribution of less than 10%. Second, the relative contributions are not the
same from city to city. The factor contributions are most uneven in Tokyo, where Exposure and
External Context contribute 32% and 34%, respectively, and Vulnerability and Emergency
Response and Recovery Capability contribute 9% and 8%, respectively. Boston and St. Louis,
however, show a very even distribution across factors. For those two cities, all five factors
contribute 18% to 22%.
Perhaps more interesting than a comparison of the evenness of the distribution across
factors, is a comparison of which factors contribute the most, relative to other factors, for each
city. For Lima and San Francisco, Hazard contributes the most, with Exposure a close second
for San Francisco. Vulnerability and Emergency Response and Recovery Capability produce
the highest relative contribution for Jakarta (V: 25%; R: 26%) , Manila (V: 24%; R: 27%), and
Santiago (V: 24%; R: 28%). In Mexico City, Hazard (13%) contributes significantly less than
the other factors; and Exposure contributes the most (24%). Tokyo and Istanbul exhibit
opposite distributions. In Tokyo, Exposure and External Context contribute the most of all
factors, while in Istanbul they contribute the least, 16% each.

18% 21% 23% 23%
19%
16%
21% 16%
14%
21% 26% 22%
a. Boston b. Istanbul
17%
17% 19%
18% 27%
31% 25%
c. Jakarta
16% 14%
Hazard 16%
14%
22% Exposure 24%
d. Lima Vulnerability e. Manila

13% 16%
21% External Context 26%
Emer. Resp. & Recov.
24%
20% 20%
15% 24%
28% 14%
22%
f. Mexico City g. San Francisco
15%
8% 17%
19% 21%
18% 24%
h. Santiago
18% 34%
20% 32%
22% 9%
j. Tokyo
i. St. Louis
Figure 8.3. Sample analysis results: Relative contributions of five main factors to EDRI

8.2.3. Earthquake disaster risk map
The maps most often associated with earthquake risk are seismicity maps, isoseismal
maps showing the distribution of damage due to an actual earthquake, and ground shaking and
loss estimation maps with contours displaying expected levels of ground shaking and loss for a
region. Seismicity maps are the only ones generally plotted for the entire world. Plotting
earthquake disaster risk on a world map, and comparing it to a world seismicity map would
illustrate vividly that urban earthquake disaster risk exists all over the world, not only along the
plate boundaries. Figure 8.4 presents the sample analysis version of an Earthquake Disaster
Risk Map. There is a column at the location of each city in the sample analysis. The total height
of the column is proportional to the relative overall EDRI value (shown in brackets), and the
relative heights of the layers describe the relative contributions of the five main factors to that
city’s earthquake disaster risk.
8.2.4. City summaries
Since some potential EDRI users may have a particular interest in one or more specific
cities, it is helpful to reiterate the most pertinent information for each city in city summary pages.
The idea is modeled after the Economic Freedom Index presentation of a two-page summary of
results for each country (Johnson and Sheehy 1996). Figure 8.5 shows an example of a city
EDRI summary for Tokyo.

Istanbul
Boston [39] [38]
Tokyo [54]
San St. Louis [36]
Francisco
[37] Manila [44]
Mexico City
[38]
Lima
[45] Jakarta [39]
Santiago
[37]
Hazard Exposure Vulnerability External Context Emerg.Resp.&Recov.

Note: Number in barckets equals overall EDRI value. Layer heights are proportional to relative contributions of five main factors.
Figure 8.4. Earthquake disaster risk map

Tokyo, Japan
EDRI City Summary
Located at the intersection of four major tectonic plates, in one of the most
seismically active countries in the world, Tokyo exhibits a relatively high
short-term and long-term seismicity. A serious threat of fire following
earthquake, and the possibility of a tsunami, place Tokyo third out of the ten-
city sample in collateral hazards factor component, giving it an overall hazard
value second only to Lima.
Saitama
Kanagawa Tokyo Chiba
Shizuoka
Japan
Enlarged five-prefecture region
Tokyo dominates population and physical infrastructure exposure, sharing the

lead in economy exposure with the U.S. cities, and putting it in a
commanding position as the city with the largest overall exposure. It also
dominates all components of external context, thanks to its role as a leading
financial and political center for Japan and the world.
Emerg.
A wealthy city with a long Resp. &
Recov. 8% Hazard
history of seismic codes and
17%
emergency planning for
External
earthquakes, Tokyo has the Context
lowest physical infrastructure 34%
and overall vulnerability, and
the best emergency response
and recovery capability rating
Exposure
in the ten-city sample.
Vulnerability 32%
9%
Figure 8.5. Example of a city summary for Tokyo, Japan

Tokyo, Japan
EDRI City Summary
Indicator Tokyo Average St. Dev.
xh1 exp(MMI w/50-year return period) 1339 973 838
xh2 exp(MMI w/500-year return period) 3641 3274 1919
xh3 Percentage urbanized area w/soft soil 10% 17% 16%
xh4 Percent. urb'd. area w/high liquefaction suscept. 20% 23% 17%
xh5 Percentage of buildings that are wood 60% 42% 26%
xh6 Population density (people per sq. km.) 8849 8910 5955
xh7 Tsunami potential indicator 2.0 0.9 0.9
xe1 Population (1000s) 35,467 9,248 9,479
xe2 Per capita GDP in constant 1990 US$ 25,833 10,955 11,287
xe3 Number of housing units (1000s) 14,046 2,767 3,820
xe4 Urbanized land area (sq.km.) 3525.9 1157.1 967.8
xe5 Population (1000s) 35,467 9,248 9,479
xe6 Per capita GDP in constant 1990 US$ 25,833 10,955 11,287
xv1 Seismic code indicator 52.79 25.96 18.48
xv2 City wealth indicator 4631 1533 1600
xv3 City age indicator 20.6 20.3 13.6
xv4 Population density 8849.2 8910.3 5955.4
xv5 City development speed indicator 36.0 39.4 29.8
xv6 Percent. of population aged 0-4 or 65+ 14.70% 16.08% 2.34%
xc1 Economic external context indicator 4.20E+13 4.96E+12 1.24E+13
xc2 Political country external context ind. 1.0000 0.6517 0.4839
xc3 Political world external context indicator 3.19E+12 3.81E+11 9.37E+11
xr1 Planning indicator 4.00 2.60 1.17
xr2 Per capita GDP in constant 1990 US$ 25,833 10,955 11,287
xr3 Avg. ann. real per cap. GDP growth, prev. 10 yrs. 3.56% 2.24% 3.53%
xr4 Housing vacancy rate 11.00% 7.14% 2.68%
xr5 Num. hospitals per 100,000 residents 5.57 2.11 1.33
xr6 Num. physicians per 100,000 residents 164 167.4 100.1
xr7 Extreme weather indicator 55.67 50.85 40.95
xr8 Population density 8849.2 8910.3 5955.4
xr9 City layout indicator 0.00 0.18 0.40
Selected references
Hadfield, Peter. 1992. Sixty seconds that will change the world: The coming Tokyo
earthquake. Boston: C.E. Tuttle Co.
Japan. Statistics Bureau. 1995. Japan Statistical Yearbook. Tokyo: Sorifu.
Otani, Shunsuke. 1995. A brief history of Japanese seismic design requirements.
Concrete International 17, no. 12: 46-53.
Risk Management Solutions, Inc. 1995. What if the 1923 Earthquake Strikes
Again?: A Five-Prefecture Tokyo Region Scenario. Menlo Park, Calif.: RMS.
UNCRD. 1995. Comprehensive Study of the Great Hanshin Earthquake. Res.
Report Series, no. 12. Nagoya, Japan: UNCRD.
Figure 8.5 (cont’d). Example of a city summary for Tokyo, Japan

8.2.5. Group comparisons
The EDRI results may be compared for groups of cities with similar characteristics.
While the insight gained through these comparisons may be more powerful in the context of a
larger set of sample cities, the idea can be illustrated using the ten cities in the sample analysis.
The ten cities are divided into those located in high development countries and those in medium
development countries, as defined by the Human Development Report 1995 (UNDP 1990-
95). Boston, Mexico City, San Francisco, Santiago, St. Louis, and Tokyo comprise the first
group; the remaining four cities comprise the second. Figure 8.6 displays the average five main
factor and EDRI values for each group. It seems intuitively understandable that the Vulnerability
and Emergency Response and Recovery Capability would be higher in the medium
development cities, while the External Context would be higher in the high development cities.
Figure 8.6 also suggests that differences in the factor values largely compensate for one another,
because the overall EDRI values for the two groups are similar.
100
High human dvlpt. countries:
USA, Mexico, Chile, Japan
Medium human dvlpt. countries:
Relative values
Turkey, Indonesia, Peru, Philippines
50
0
Hazard Expos. Vulner. Context ER&R EDRI
Figure 8.6. Comparison of EDRI for cities with high and

medium human development
8.2.6. Hazard versus earthquake disaster risk

A comparison between the Hazard factor values and the EDRI values provides another
interesting smaller story that can be extracted from the EDRI results. Figures 8.7 and 8.8 offer
a couple examples of the several ways in which the comparison can be illustrated. Figure 8.7
plots EDRI versus the Hazard factor, and divides the grid into four quadrants: Low Hazard-
Low EDRI, Low Hazard-High EDRI, High Hazard-High EDRI, High Hazard-Low EDRI. The
quadrant boundaries correspond to the mean values for the sample of cities. While the figure
suggests that a high EDRI implies a high Hazard, the converse is not true. A high Hazard value
does not imply a high EDRI. Furthermore, some cities exhibit similar EDRI values, but very
different Hazard values (e.g., Santiago and San Francisco), while others have a similar Hazard,
but different EDRI (e.g., Tokyo and San Francisco). Figure 8.8 compares the factor values for
San Francisco and Tokyo. The columns illustrate clearly that the two cities have the same value
for the Hazard factor, but because of differences in the remaining four factors, particularly
Exposure and External Context, the final EDRI value is significantly higher in Tokyo than in San
Francisco. Both figures demonstrate that the two concepts—hazard and earthquake disaster
risk—are related, but not equivalent.
80
Low Haz. High Haz.
High EDRI Tokyo High EDRI
40
Santiago San Francisco
Low Haz. High Haz.

Low EDRI Low EDRI
0
0 40 80
Figure 8.7. Comparison of EDRI and Hazard factor

Relative factor values
Emerg. Resp.
& Recov.
Ext. Context
Vulnerability
Exposure
Hazard
San Francisco Tokyo
Figure 8.8. Equal Hazard, unequal EDRI
8.3. Reminders
Throughout the interpretation of the EDRI results, the limitations of the EDRI
development should be kept in mind. The difficulties of the indicator selection process led to the
omission of some of the concepts of the conceptual framework. The EDRI’s linear form means
that some nonlinearities and dependencies are not conveyed. The sensitivity analysis indicates
that there is some uncertainty in the values of the factor and EDRI values for all cities. Finally,
the EDRI goals were stated explicitly in Section 2.2. The index aims to measure earthquake
disaster risk as it was defined in Section 2.1. It includes all impact parameters, not only financial
losses and not only deaths. It conveys the risk only as it is understood relative to other cities,
not on an absolute scale. No information is provided about the distribution of risk within each
city. While the EDRI can provide some potentially very useful information, these characteristics
should be noted repeatedly to ensure that it is not misinterpreted or misused.

Chapter Nine
Future Work and Conclusions
9. Introduction
This final chapter looks to the future through a discussion of some natural extensions of
the EDRI project described in the previous eight chapters. It explores refinements that might
improve the EDRI, and suggests possible variations of the index that might be useful to specific
users. The chapter concludes with some final summary thoughts on the contributions of this
study.
9.1. Future work
Future work related to the Earthquake Disaster Risk Index could proceed in one of two
directions: refining the index, or developing variations of it. As with any research project, time
and resource constraints imposed limitations on the scope of the EDRI development and
evaluation. As a result, each of the six steps in the EDRI development procedure could be
repeated and refined to produce a similar, but improved EDRI. The conceptual framework
could be expanded to address the interactions among factors more fully. The indicator selection
process could be improved by exploring more indicator options, and buttressing those choices
with more analytical or empirical evidence that the indicators represent that factors for which
they are proxies. In keeping with the possible inclusion of factor interactions in the conceptual
framework, the mathematical combination could be altered to model nonlinearities and
dependencies among the indicators. The determination of the weights also provides a significant
opportunity for refining the EDRI. A more comprehensive, scientific questionnaire survey might
be conducted to provide more support for the assessment of weights. A similar survey might be
used to verify the final index rankings of the five main factors and the overall EDRI.
In the data gathering and evaluation step, the more time and effort spent reducing the
uncertainty in the data, the better the final EDRI results will be. Additional effort might go
toward more personal communications with subject and local area experts, or continuing library
research. Another obvious goal of future work is to evaluate more cities, gradually expanding
the ten-city sample until the EDRI of all major cities in the world have been assessed. Many
options exist to improve the presentation of the EDRI results. When a larger set of cities has
been evaluated, a glossy booklet, like the Human Development Report or the Economic
Freedom Index report, could be published. A colorful, laminated, poster-size Earthquake
Disaster Risk map might be created, with paragraphs of explanation, tables, and other auxiliary
information filling in the space around the edge of the map. Finally, a software package could
be developed that takes raw data for each city and indicator as input, performs the evaluation
and sensitivity analysis for the EDRI, and produces the final tables and figures as output. The
package might include some databases that would be useful for all cities, such as the historical
tsunami database, or population and GDP data for all countries. It also could offer flexibility in,
for example, the choice of weight values, the deltas used in the sensitivity analysis, and form of
the final output. Such a software tool could make it easy for any potential user to evaluate the
EDRI for any set of cities.
The development of variations on the EDRI constitutes the second vein of possible
future work. As mentioned in Section 2.2, one reason that the objectives of the EDRI were
stated explicitly at the beginning of the development procedure was to guide the decisions
encountered during the process. If the objectives are altered to reflect the needs of a specific
user, some aspects of the development and evaluation will have to be changed as well.
Different users might be interested in measuring a slightly different concept than the earthquake
disaster risk defined in Section 2.1. They may have different resources available for gathering
data, or they might be focused on a different unit of study or a specific subset of cities. In each
Chapter Nine: Future Work and Conclusions 215

case, the basic development and evaluation procedure of the EDRI described in this study
should be able to accommodate the user’s specific needs. It only may be necessary to modify
the specific factors that are included in the conceptual framework, change the set of indicators,
use a different sample for the evaluation, and/or present the results differently.
The following are some of the possible variations of the EDRI that might be desirable to
specific users. A multi-hazard disaster risk index would be similar to the EDRI, but would
address the risk due to hurricanes, tornadoes, floods, and/or technological hazards, as well as
earthquakes. The Hazard and Vulnerability factors of the conceptual framework for such an
index would have to be altered, and therefore the indicators chosen to represent those factors
would change as well. A multi-hazard index would be even more comprehensive, following the
emerging social science paradigm’s shift away from a hazard-based approach. It would be
especially useful to government and international aid organizations that are less concerned about
the agent of damage, loss, and disruption, than the magnitude and distribution of that impact.
A national government or national organization might be interested in a country-specific

EDRI, i.e., one that is evaluated only for cities within the specified country. Some modifications
will be necessary to tailor the EDRI to the specific circumstances of the country of interest. For
example, the country’s data availability situation may be different than that for a set of
international cities, and the External Context will take on a different role for a country-specific
EDRI. Similarly, while some aspects of earthquake disaster risk make more sense for a greater
metropolitan area than an entire country, an EDRI that uses a country as the unit of study might
be useful for some international organizations, and it would gain the advantage of improved data
availability compared to that associated with cities or regions.
Another variation of the EDRI might focus on one or more specific impact parameters,
rather than all possible impact parameters. For example, an insurance company, reinsurance
company, or bank might be interested only in the economic aspects of earthquake disaster risk.
216 Chapter Nine: Future Work and Conclusions

A revised EDRI that measures, not earthquake disaster risk as it is defined in Section 2.1, but
an economic earthquake risk, might be of more interest to those users.
9.2. Conclusions
The research presented in this dissertation establishes a methodology for the

development of the many possible variations of the EDRI that are discussed in the previous
section. It lays out the development procedure so that future researchers can repeat the
process, altering the details to fit their specific goals and resources. It includes both an outline of
the steps involved, and many specifics about the particular issues that arise in the course the
development of the EDRI. The previous eight chapters explore the current state of knowledge
of earthquakes and their effects on the life of a greater metropolitan area, the challenges and
options for the selection of indicators to represent a city’s earthquake disaster risk, the
mathematics of composite index construction, the data availability situation for cities worldwide,
the procedure for performing a sensitivity analysis, and the issues and possibilities for
presentation of the index results. With this methodology established, future researchers can
create variations of the EDRI with relative ease.
In addition to the methodology, this project offers two substantive products. The first is
the EDRI itself, an easy-to-understand summary of urban earthquake disaster risk. The index
can raise awareness of the nature and magnitude of urban earthquake disaster risk, and spark
interest in the challenge of mitigating that risk. It is the user interface for the second product, the
conceptual framework of earthquake disaster risk on which the index is based. The framework
is the behind-the-scenes support that helps to interpret the findings of the EDRI, and can
provide a launching point for further investigation of the concepts that the EDRI represents.
Although the development proceeded in the other order, a user would see the EDRI results first,
then to pursue his investigation of the topic, would learn more about the conceptual framework
and its component factors. Both the conceptual framework and the final index are significant.
The former provides meaning to the index; the latter summarizes and delivers the information to
Chapter Nine: Future Work and Conclusions 217

the user. This project has demonstrated that it is possible to create a meaningful Earthquake
Disaster Risk Index that allows direct comparison of the overall earthquake disaster risk of
cities worldwide, and describes the relative contributions of various factors to that overall risk.
218 Chapter Nine: Future Work and Conclusions

Appendix A
Selected Previously Proposed Definitions of Disaster
Quantitative, operational definitions

• A disaster is an event which causes ten deaths, and/or affects one hundred people, and/or
leads to an appeal for international assistance (International Federation of Red Cross and
Red Crescent Societies, and the Centre for Research on the Epidemiology of Disasters
1994).
• A disaster is an event for which at least one of the following is true(U.S. OFDA 1990):
i. The "Chief U.S. Diplomatic Mission in an affected country determined that a disaster
existed which warranted U.S. government response."
ii. It causes more than five deaths.
iii. The sum of the number of deaths and the number of people injured is greater than
twenty-four.
iv. More than one thousand people are affected.
v. It causes damage greater than $1 million.
Qualitative, comprehensive definitions

• Natural disasters are one type of social crisis. Disaster "refers to a condition in which the
traditional structure, due to the impact of a precipitating geophysical event, is destroyed
and/or no longer collectively defined as an appropriate guide for social behavior in the
altered system. Within a community a disaster represents a crisis of relatively high intensity."
The event "creates demands upon the community system that cannot be met by its
traditional, institutional structure—including its set of community-emergency and emergency-
relevant organizations. In such a setting, significant alterations in community structure can be
observed to occur as the system develops an emergent, disaster structure to facilitate its
response to the event and increase its capability to fulfill the created demands" (Wenger
1978).
• Disasters as "social disruption and changes brought about by the physical agent and its
impact" (Quarantelli 1978).
• A “sudden and disruptive impact in which disaster demands exceed the capabilities of
routine social structures in the community” (Kreps 1978).
• Natural disasters create needs in a community often unmet by established organizations
(Forrest 1978).
• Disaster is a crisis which exceeds the capacity of society to manage it (Anderson 1992).
• Disaster is an "event associated with the impact of a natural hazard, which leads to
increased mortality, illness and/or injury, and destroys or disrupts livelihoods, exceptional
and requiring external assistance for recovery" (Cannon 1994).
• A disaster is “an event, concentrated in time and space, in which a society (or a community)
undergoes severe danger and incurs such losses to its members and physical appurtenances
that the social structure is disrupted and the fulfillment of all or some of the essential
functions of the society is prevented” (Fritz 1961).
• “A major incident is a serious disruption to life, arising with little or no warning, causing or
threatening death or serious injury to, or rendering homeless, such numbers of persons in
excess of those which can be dealt with by the public services operating under normal
procedures, and which calls for the special mobilisaton and organization of those services.”
(U.K. Humberside County Council).
Appendix A: Disaster Definitions 221

Appendix B
Sample City Definitions
Boston, USA
Includes the following ten counties (according to the U.S. Census-defined Boston
Consolidated Metropolitan Statistical Area):
Bristol, Massachusetts Suffolk, Massachusetts
Essex, Massachusetts Worcester, Massachusetts
Middlesex, Massachusetts Hillsborough, New Hampshire
Norfolk, Massachusetts Rockingham, New Hampshire
Plymouth, Massachusetts Stafford, New Hampshire
Istanbul, Turkey
Includes the following fifteen district municipalities (Institut d'Estudis Metropolitans de
Barcelona 1988):
Adalar Gaziosmanpasa
Bakirkoy Kadikoy
Besiktas Kartal
Beykoz Sariyer
Beyoglu Sisli
Eminonu Uskudar
Eyup Zeytinbinburnu
Fatih
Jakarta, Indonesia
Defined as DKI Jakarta, which is equivalent to the following:
Jakarta Selatan
Jakarta Timur
Jakarta Pusat
Jakarta Barat
Jakarta Utara
222 Appendix A: Disaster Definitions

Lima, Peru
Defined as the sum of Lima and Callao Provinces, which is equivalent to the sum of the
following forty-nine districts; forty-three in Lima, six in Callao (Institut d'Estudis
Metropolitans de Barcelona 1988):
Lima Province:
Lima Pachacamac
Ancon Pucusana
Ate Puente piedra
Barranco Punta Hermosa
Brena Punta Negra
Carabayllo Rimac
Chaclacayo San Bartolo
Chorrillos San Borja
Cieneguilla San Isidro
Comas San Juan de Lurigancho
El Agustino San Juan de Miraflores
Independencia San Luis
Jesus Maria San Martin de Porres
La Molina San Miguel
La Victoria Santa Anita
Lince Santa Maria del Mar
Los Olivos Santa Rosa
Lurigancho Santiago de Surco
Lurin Surquillo
Magdalena del mar Villa El Salvador
Magdalena la vieja Villa Maria del Triunfo
Miraflores
Constitutional Province of Callao:

Callao
Bellavista
Carmen de la Legua Reynoso
La Perla
La Punta
Ventanilla
Appendix B: City Definitions 223

Manila, Philippines
Defined as the National Capital Region (NCR), which is equivalent to the sum of the
following four cities and thirteen municipalities (Institut d'Estudis Metropolitans de
Barcelona 1988):
Manila City Muntinlupa
Caloocan City Navotas
Pasay City Paranaque
Quezon City Pasig
Las Pinas Pateros
Makati San Juan del Monte
Malabon Taguig
Mandaluyong Valenzuela
Marikina
Mexico City, Mexico

Includes the entire Federal District, plus the following fifty-three districts from the State
of Mexico (Institut d'Estudis Metropolitans de Barcelona 1988):
Acolman Melchor Ocampo
Amecameca Naucalpan
Atenco Nextlapan
Atizapan de Zaragoza Nezahualcoyotl
Atlautla Nicolas Romero
Axapusco Nopaltepec
Ayapango Otumba
Coacalco Ozumba
Cocotitlan Papalotla
Coyotepec La Paz
Cuautitlan Izcalli Sn. Martin de las Piramides
Cuautitlan de Romerio Rubio Tecamac
Chalco Temamatla
Chiautla Temascalapa
Chicoloapan Tenango del Aire
Chiconcuac Teotihuacan
Chimalhuancan Teoloyucan
Ecatepec Tepetlaoxtoc
Ecatzingo Tepetlixpa
Huehuetoca Tepotzotlan
Huixquilucan Texcoco
Isidro Fabela Tezoyuca
Ixtapaluca Tlalmanalco
Jaltenco Tlalnepantla
Jilotzingo Tultepec
Juchitepec Tultitlan
224 Appendix B: City Definitions

Zumpango
Appendix B: City Definitions 225

San Francisco, USA
Includes the following ten counties (according to the U.S. Census-defined San
Francisco Consolidated Metropolitan Statistical Area):
Alameda Santa Clara
San Mateo Santa Cruz
Contra Costa Sonoma
Marin Napa
San Francisco Solano
Santiago, Chile
Defined as Provincia Santiago, which is equivalent to the sum of the following thirty-two
comunas (Institut d'Estudis Metropolitans de Barcelona 1988):
Santiago Penalolen Estacion Central
Independencia La Florida Cerrillos
Conchali San Joaquin Maipu
Huechuraba La Granja Quinta Normal
Recoleta La Pintana Lo Prado
Providencia San Ramon Pudahuel
Vitacura San Miguel Cerro Navia
Lo Barnechea La Cisterna Renca
Las Condes El Bosque Quilicura
Nunoa Pedro Aguirre
La Reina Cerda
Macul Lo Espejo
St. Louis, USA

Includes the following eleven counties and one city (according to the U.S. Census-
defined St. Louis Consolidated Metropolitan Statistical Area):
Clinton, Illinois Jefferson, Missouri
Jersey, Illinois Lincoln, Missouri
Madison, Illinois St. Charles, Missouri
Monroe, Illinois St. Louis County, Missouri
St. Clair, Illinois Warren, Missouri
Franklin, Missouri St. Louis City, Missouri
Tokyo, Japan
Includes the following five prefectures (RMS 1995):
Saitama
Chiba
Tokyo
Kanagawa
Shizuoka
226 Appendix B: City Definitions

Appendix C
Emergency Response and Recovery Tasks
The following questions elucidate what is involved in each of the twelve tasks associated
with post-earthquake emergency response and recovery. Sources: California Seismic Safety
Commission (1990); Tierney (1985); UNDHA (1994); UNDRO (1979); UNDRO (1984);
and UNDRO (1986).
Emergency management system

1. Does the city have a general emergency response plan?
2. Who developed the city plan?
3. How old is the city plan?
4. Has the city plan ever been executed? When? Have there been exercises to practice
execution of the plan?
5. What is the quality of the city plan?
a. Is the plan comprehensive, yet as simple as possible?
b. Does the plan include an earthquake-specific section?
c. Is the plan both organizational and operational? Does it define and organize roles and
responsibilities well? Is it usable?
d. Is the plan well-integrated?
i. through all hierarchical levels, i.e., community, city, county, nation, region (esp. for
island nations), and international? Does it include direction for inter-jurisdictional
coordination (e.g., mutual aid agreements)?
Appendix C: Emerg. Response and Recovery Tasks 227

ii. across agencies, organizations, and departments (e.g., government agencies, Red
Cross, utilities) through task forces, emergency councils, or other arrangements?
Does it include direction for interorganizational coordination?
iii. with daily operations and on-going development goals?
e. Does the plan include detailed information about who is responsible for which specific
subtasks, and how to prioritize and perform those subtasks?
f. Does the plan clearly define a set of standard priorities?
g. Does the plan define a control structure that:
i. has an established physical location for headquarters?
ii. allows for mobile management operations?
iii. has procedures for delegation of responsibility and authority?
iv. has liaison capability to other jurisdictions, organizations, and the public?
v. includes local officials in all major decisions (not just national or international
officials)?
vi. allows close monitoring of actions?
vii. offers a good balance between being centralized (no duplication) and being
dispersed (can cover many tasks simultaneously, has redundancy)?
viii. relies on written, not verbal communication when possible?
ix. provides a clear chain of command for each operation?
x. relies on a minimum number of coordination meetings, but includes representatives
from all sectors in those meetings?
h. Does the plan incorporate the capabilities of existing organizations, the military,
volunteers, and the resources of affected people?
i. Does the plan rely on existing organizations, rather than creating new ones, whenever
possible?
j. Does the plan call for a quick, but comprehensive impact assessment to focus (confirm
or alter) anticipated needs and priorities?
k. Does the plan establish procedures for giving and receiving information?
228 Appendix C: Emerg. Response and Recovery Tasks

6. Are all involved personnel aware of the capabilities and limitations of the city plan, and well-
trained in its use?
7. Which, if any, specific organizations and agencies (e.g., hospitals, fire departments, law
enforcement) have their own specialized emergency response plans?
8. Repeat items 2-6 for plans related to specific organizations and agencies.
Communications
1. Are there established interorganizational communications links (e.g., between government
agencies, relief organizations, utilities)? Are there established relationships and equipment
hookups among the organizations?
2. Are there established interjurisdictional (local, national, international) communication links
(relationships and equipment)?
3. Are there arrangements for officials to gather information from the public (for damage
assessment) and disseminate information to the public (“big picture” impact assessment and
instructions)?
a. Is there a partnership between the local government and the media to guide information
transfer between officials and the public?
b. Is there a prepared format for conveying information to the public that spells out the
consequences of the threat and the probability of occurrence, and that is specific,
urgent, repeated, and presented by someone with public confidence?
c. Is there a public information/press relations officer to talk to media representatives and
other groups (e.g., all the people in a certain area, all teachers)?
4. Do all involved organizations know who to talk to, what information to convey, what
information to request, and what form of communication to use?
5. Will damage to normal equipment be minimized so that it remains functional after an
earthquake?
6. Are there redundant and alternate emergency systems of communication (two-way radios,
cellular phones, fax and computer lines, dispatch and 911 systems, microwave systems) to

replace normal systems that are not functioning? Does the design of the emergency
communications systems consider:
a. the distances over which communications will be needed?
b. the nature of terrain?
c. the type of communication product desired (e.g., voice, hard copy)?
d. the degree of reliability needed?
e. the acceptable duration of delay?
f. the volume of communication (quantity and lengths of messages)?
g. if two-way communication is needed?
h. if equipment will be reliable in adverse operational conditions?
i. if additional maintenance facilities or operator training are needed?
j. the degree of compatibility needed between communication lines?
k. the expected number of trained personnel available to operate?
l. the costs?
Financial arrangements
1. Have pre-disaster arrangements been made to provide timely financial assistance to
households, businesses, local governments, and communities for relief and reconstruction, in
both the short- and long-term?
a. Are there any programs to assist households?
b. Are there any programs to assist businesses?
c. Are there any subsidy programs?
d. Are there any assistance programs from government agencies, private sources, or
voluntary associations?
e. Do households, businesses, or governments have earthquake insurance?
i. What is the market penetration for each sector (residential, commercial,
governmental)?
ii. What are the terms of the coverage?
iii. Is the coverage reliable?

2. Does a State of Emergency declaration trigger emergency financial arrangements?
3. How difficult will it be, in terms of money, time, and political will, to develop financial
arrangements of the type discussed in item 1, after the earthquake?
4. Is there an Emergency Reserve Fund, Emergency Funding Committee, or equivalent body
that is empowered to vote on funding if the normal governing body is not in session when
the disaster occurs?
5. Are there provisions to provide streamlined delivery of assistance? Have arrangements
been made:
a. for post-disaster waivers of customs duties for relief supplies, and of landing fees for
crafts carrying them?
b. for post-disaster price controls?
c. to provide compensation to people giving personal services or people whose property
is used by emergency relief officials?
Legislation
1. Will relief workers have the necessary legal power to execute relief measures without delay?
2. If legal holdups are expected to delay the provision of relief, how costly (in terms of time
and money) are they expected to be?
3. Has pre-disaster legislation established channels to request and receive international
assistance efficiently? Does it define who gives what relief, under what circumstances, and
how to file relief requests? Does the legislation include arrangements for duty exemption,
granting landing and overflying rights, lifting border-crossing restrictions, establishing the
chain of command when international teams are involved, exchange of information, and
cost-sharing?
Damage assessment
1. Are there established procedures to determine the type, degree, location, and geographic
distribution (centralized or dispersed) of casualties and damage immediately after an

earthquake (especially impact on medical services, water supply, liquid waste disposal,
shelter, food, and transport)?
a. Is it clear who is responsible for what portion of the information gathering, and how the
information can be interpreted to determine short- and long-term needs?
b. Is there a procedure to obtain information fromthe public (telephone reports)?
c. Are there arrangements for using aerial photography to collect information?
2. Are there resources (e.g., people, authority, money) that will be available to perform the
damage assessment?
3. Is there a pre-prepared city profile available that includes: hazard and/or risk maps,
information on the structural vulnerability, economic situation, number and location of people
who may be especially vulnerable (e.g., minorities, elderly)?
Search and rescue

1. Are there established procedures to locate and rescue any survivors trapped in collapsed
structures?
2. Are there local resources that will be available to help with the search and rescue?
a. Is there skilled manpower (search-and-rescue teams, fire and law enforcement
officers)?
b. Are there dog teams, sound equipment, or debris-moving equipment available?
3. Are there established procedures and the necessary communications links to summon
international search-and-rescue teams?
Secondary hazard control

1. Are there established procedures to prevent and suppress fires following an earthquake?
2. Is there general public awareness about how to prevent post-earthquake fires?
3. Have the gas companies implemented any fire prevention efforts?
4. Are there nearby resources (e.g., skilled fire fighters, fire-fighting equipment) to suppress
fires?

5. Are there established procedures to remove and dispose of hazardous debris following an
earthquake?
6. Are there nearby resources (e.g., manpower, debris-moving equipment) to remove the
hazardous debris?
7. Will there be a special need for sanitation control? If so, are there established procedures
and available resources to ensure adequate standards of sanitation?
Health care
1. Are there established procedures to asses the type and quantity of health care needs?
These needs may be based on: general information about the extent and area of damage,
population affected, and functional damage to public services (e.g., transportation and utility
networks), plus detailed inventory of functional health facilities and their capabilities (e.g.,
status of facilities, availability of medical staff and medical supplies).
2. Are there plans describing how to provide on-site first aid to injured, to transport injured to
health care facilities, and to treat patients at the health care facilities as effectively and
efficiently as possible?
3. Are there established communications links among medical providers?
4. Will there be enough skilled personnel (e.g., search-and-rescue teams, emergency medical
technicians) available to provide first aid?
5. Will there be enough ambulances, and open transportation routes to transport the injured to
health care facilities?
6. To what extent will health care facilities remain functional? Functional includes having
enough equipment, staff, medical supplies, medicine, working lights, water, utilities, power
for machinery, record-keeping, working elevators or alternate transport; and not being
flooded by pipeline breaks.
7. Are there provisions to minimize structural and non-structural damage to hospitals and other
health care facilities?
8. Are there provisions to minimize damage to utilities serving hospitals, or provide backup
service through emergency generators, emergency water supply systems, etc.?

Mass care
1. Are there established procedures to assess the type and quantity of basic needs that are
required (e.g., emergency food and water, water containers, blankets, cooking utensils,
clothes)?
2. Are there established procedures and resources to provide, in a timely manner and with as
little waste as possible, basic needs to those who do not have them and cannot receive them
from family or friends?
a. Are there established procedures to request, receive, and distribute needed relief
supplies from private sources, stockpiles, local, national, and/or international
governments?
b. Will the equipment necessary for receipt and distribution of supplies (trucks, manpower,
temporary distribution headquarters) be available?
c. Are there tanker trucks available to deliver water, and plans to use water from pools
and reservoirs?
Shelter
1. Are there established procedures and resources to assess how many people need shelter
and for how long, what kind of shelter is appropriate for them, and how to provide it?
2. In estimating shelter needs and determining the most appropriate and accessible emergency
shelter types, are the following considered:
a. number of housing units destroyed or not functional?
b. capability of community “coping mechanisms” to provide emergency shelter with family
and friends?
c. feasibility and likelihood of survivors making their own shelter with salvaged materials?
d. weather conditions?

e. accessibility of impacted areas?
f. risks of secondary disaster?
g. manpower at site capable of helping erect emergency shelter?
h. language, religious, and cultural needs?
i. desire for family cohesiveness?
j. desire to minimize distance from shelters to work, schools, stores, and community?
3. Are there established procedures and resources to provide short-term shelter: establish
sites, shelters, and facilities to support shelters (e.g., bathrooms)?
a. Are there local facilities, schools, churches, community centers that may be used for
short-term shelter?
b. Are there established procedures and necessary communications links to request
additional short-term shelter from outside sources (e.g., mobile homes, tents, campers)?
4. Are there established procedures and resources to provide long-term shelter: help with
reconstruction, or provide long-term (more than two weeks) temporary shelter (establish
sites, shelters, and facilities to support shelters)?
a. Are there local sites and facilities that may be used for long-term shelter?
b. Are there established procedures and necessary communications links to request
additional long-term temporary shelter from outside sources (e.g., mobile homes, tents,
campers)?
Clean-up
1. Are there established procedures to reposition undamaged items that are moved during an
earthquake?
2. Are there established procedures to repair damaged items?
3. Are there established procedures to remove destroyed items?
4. Is there equipment and manpower to clear debris created by damaged and collapsed
structures, and to unblock transportation routes?
5. Are there established procedures to find and bury the dead?

Restoration of services
1. Are there established procedures and resources to restore utility services?

Appendix D
Physical Infrastructure Exposure Data Sets
D.1. Physical Infrastructure Exposure data set from Compendium of Human Settlements
Statistics 1995 (UN Centre for Human Settlements 1995).
Total
Real per Num. of electricity Total annual Num. of
capita GDP occupied consumption water Num. of data
[PPP$], housing per day consumption motor points
Country City 1993 Population units [1000 kwh] [1000 cu.m.] vehicles (out of 6)
1 Angola Luanda 674 923,842 2
2 Burkina Faso Ouagadougou 780 459,826 2
3 Ethiopia Addis Ababa 420 1,423,111 259,555 3
4 Ghana Accra 2,000 957,157 2
5 Kenya Nairobi 1,400 1,324,570 2
6 Mali Bamako 530 658,275 105,394 3
7 South Africa Cape Town 3,127 2,614,051 495,000 3,817,791 259,959 522,488 6
8 Pretoria 3,127 442,664 127,680 5,177,830 98,919 597,671 6
9 Tunisia Tunis 4,950 815,795 168,464 3
10 Canada Montreal 20,950 3,127,240 1,239,909 3
11 Ottawa 20,950 920,855 266,738 3
12 Toronto 20,950 3,893,045 1,316,623 3
13 Cuba Havana 3,000 2,119,059 500,198 2,708,132 438,887 5
14 Guatemala Guatemala City 3,400 754,243 140,917 3
15 Honduras Tegucigalpa 2,100 539,590 102,573 421,721 30,274 54,043 6
16 Mexico Mexico City 7,010 15,047,685 3,120,673 3
17 Guadalajara 7,010 2,870,417 2
18 Monterrey 7,010 2,590,545 2
19 USA Chicago 24,680 8,065,633 70,852,787 3
20 Los Angeles 24,680 14,531,529 21,742,536 3
21 New York 24,680 18,087,251 35,474,230 3
22 Washington DC 24,680 3,923,574 9,848,445 3
23 Bolivia La Paz 2,510 669,400 156,500 3
24 Chile Santiago 8,900 3,902,315 2
25 Colombia Bogota 5,790 4,154,404 847,523 785,952 58,209 5
26 Cali 5,790 1,429,026 253,794 311,136 21,828 5
27 Medellin 5,790 2,121,174 275,208 538,112 32,675 5
28 Ecuador Guayaquil 4,400 1,508,444 319,900 3
29 Quito 4,400 1,100,847 262,709 3
30 Peru Lima 3,320 6,011,036 1,099,242 324,047 206,942 5
31 Venezuela Caracas 8,360 1,822,465 2
32 India Bombay 1,240 8,243,405 9,276,780 612,291 4
33 Calcutta 1,240 3,305,006 5,616,100 537,441 4
34 Delhi 1,240 5,729,283 2,064,000 3
35 Madras 1,240 3,841,396 2,623,085 110,827 686,556 5
Appendix D: Physical Infra. Exposure Data 1

Total
Real per Num. of electricity Total annual Num. of
capita GDP occupied consumption water Num. of data
[PPP$], housing per day consumption motor points
Country City 1993 Population units [1000 kwh] [1000 cu.m.] vehicles (out of 6)
36 Indonesia Bandung 3,270 1,810,917 1,181,419 24,552 246,831 5
37 Jakarta 3,270 9,406,477 1,740,214 5,917,812 111,503 1,435,731 6
38 Iraq Baghdad 3,413 3,842,000 419,876 421,837 4
39 Japan Osaka 20,660 2,623,801 963,380 726,834 4
40 Sapporo 20,660 1,671,742 572,850 663,575 4
41 Tokyo 20,660 8,163,573 3,112,590 2,545,027 4
42 Yokohama 20,660 3,220,331 1,001,800 1,058,719 4
43 Korea, Repub. of Incheon 9,710 1,816,827 4,243,166 177,486 179,952 5
44 Pusan 9,710 3,795,181 7,341,267 326,252 346,846 5
45 Seoul 9,710 10,618,395 16,027,052 138,775 4
46 Taegu 9,710 2,227,715 5,794,947 218,513 293,896 5
47 Myanmar Yangon 650 2,513,023 2
48 Pakistan Karachi 2,160 5,208,132 811,850 3
49 Lahore 2,160 2,988,486 447,509 3
50 Singapore Singapore 19,350 2,705,115 14,194,400 374,222 542,352 5
51 Thailand Bangkok 6,350 6,370,000 902,871 1,516,763 646,830 1,644,018 6
52 Turkey Ankara 4,210 3,306,327 299,771 3
53 Istanbul 4,210 5,842,985 605,933 3
54 Izmir 4,210 2,317,829 232,178 3
55 Viet Nam Ho Chi Minh 1,040 2,795,140 2
56 Austria Vienna 19,115 1,531,346 717,608 5,704,724 129,912 654,671 6
57 Bulgaria Sofia 4,320 1,201,719 5,203,000 193,269 4
58 France Paris 19,140 2,135,197 2
59 Germany Berlin (West) 18,840 2,130,525 1,043,719 3
60 Hamburg 18,840 1,626,220 779,605 3
61 Munich 18,840 1,206,683 626,425 3
62 Hungary Budapest 6,059 2,016,774 775,523 6,124,189 32,182 540,076 6
63 Italy Rome 18,160 2,693,383 988,543 3
64 Netherlands Amsterdam 17,340 702,444 363,740 255,155 4
65 Norway Oslo 20,370 461,190 244,434 3,242,000 213,459 5
66 Poland Warsaw 4,702 1,655,272 581,454 191,329 4
67 Romania Bucharest 3,727 2,067,545 761,156 3,225,200 367,677 345,372 6
68 Russian Federation Moscow 4,760 8,677,177 2,879,323 34,647,000 644,736 5
69 St. Petersburg 4,760 4,435,167 1,457,732 19,206,000 323,173 5
70 Spain Barcelona 13,660 1,643,541 2
71 Madrid 13,660 3,010,499 2
72 Sweden Stockholm 17,900 674,680 259,235 3
73 Ukraine Kiev 3,250 2,559,312 833,000 411,026 4
74 Australia Brisbane 18,530 1,149,401 2
75 Melbourne 18,530 2,832,893 2
76 Sydney 18,530 3,364,858 2
Number of data points (out of 76) 76 76 41 29 19 35

Mean 9,489 3,433,526 806,922 10,440,123 235,140 571,780
Standard deviation 7,963 3,397,374 750,421 14,822,308 165,130 563,708
Sources:
[1] GDP data: UNDP. Human Development Report 1996. New York: UN.
[2] All other data: UN Centre for Human Settlements (Habitat). Compendium of Human Settlements
Statistics 1995. 5th issue. New York: UN.
Notes:
Source [1] data is for 1993.
Source [2] data is for as close to 1990 as possible, but may be for any year 1980-92.
A few entries have been adjusted as noted:
Bamako, Tunis, Guat. city, Moscow, and St. Pete. entries have been adjusted assuming 1.1 households
per housing unit.
Bangkok water consumption reduced by 10% because data included Nonthaburi and Samut Prakan.
Warsaw water consumption increased by 20% because data included only household consumption.
Appendix D: Physical Infra. Exposure Data 2

D.2. Physical Infrastructure Exposure data set from Statistics of World Large Cities 1991 (TMG 1991).
Real per Num. of

capita Electricity Gas Length of telephone Num. of Num. of
GDP Num. of Land area consumption consumption Water supply roads subscrip- establish- data points
Country City [PPP$] Population dwellings [sq.km.] [1000 kwh] [1 mill. kcal] [1000 cu.m.] [km] tions ments (out of 10)
1 Egypt Alexandria 3,540 2,893,000 899 52,000 4
2 Egypt Cairo 3,540 6,399,000 214 6,288,600 1,047,457 4,750 757,098 4,191 8
3 Ethiopia Addis Ababa 330 1,686,300 263,751 217 409,778 38,012 1,120 84,471 46,484 9
4 South Africa Johannesburg 3,799 896,000 193,300 450 5,797,969 2,062,940 124,852 2,692 763,565 9
5 USA Chicago 23,760 3,021,912 1,154,800 228 69,149,000 7,158,799 1,426,460 5,917 5,084,344 116,931 10
6 USA Detroit 23,760 1,117,000 550,000 362 889,571 4,502 830,000 35,653 8
7 USA Houston 23,760 1,729,720 764,224 1,431,000 3,893,560 478,300 15,393 7
8 USA Los Angeles 23,760 3,070,710 1,189,475 1,207 19,100,000 3,134,675 722,020 3,904,905 8
9 USA New York 23,760 7,218,534 2,949,300 782 29,390,000 6,114,145 2,028,991 10,303 2,957,000 9
10 USA Philadelphia 23,760 1,646,156 698,868 129 11,841,285 19,154,609 1,980 2,565 5,100,000 23,619 10
11 Canada Montreal 20,520 1,752,585 443,560 494 14,240,129 2,300,000 2,194 3,030,000 38,045 9
12 Cuba Havana 3,412 2,096,054 555,036 727 2,769,663 657,496 375,117 489 236,277 16,231 10
13 Mexico Guadalajara 7,300 2,047,455 188 6,307 517,800 5
14 Mexico Mexico City 7,300 10,263,275 1,747,102 1,483 18,499,143 36,483 1,290,274 265,160 8
15 Argentina Buenos Aires 8,860 3,078,011 1,216,325 200 5,177,700 2,992,614 763,951 2,534 1,205,400 9
16 Brazil Belem 5,240 1,418,533 1,138,920 707,519 240,293 5
17 Brazil Belo Horizonte 5,240 2,331,528 182,808 335 2,900,000 263,416 2,981 285,000 8
18 Brazil Brasilia 5,240 1,567,709 5,771 1,729,917 4
19 Brazil Curitaba 5,240 1,537,324 432 1,966,209 40,119 66,023 572,163 7
20 Brazil Fortaleza 5,240 1,588,709 255,096 336 1,193,837 30,268 91,329 7
21 Brazil Nova Iguacu 5,240 1,460,000 328,993 764 1,174,796 312 32,032 7
22 Brazil Porto Alegre 5,240 1,254,890 239,337 469 2,021,689 176,078 180,115 14,811 8
23 Brazil Recife 5,240 1,293,400 209 1,590,849 172,651 2,905 211,184 7
24 Brazil Rio de Janiero 5,240 5,603,388 1,301,104 910 11,765,185 328,896 555,737 1,295,320 8
25 Brazil Salvador 5,240 1,927,667 299,028 324 1,645,429 403,029 6
26 Brazil Sao Paolo 5,240 11,097,954 1,509 19,569,589 130,151 779,119 724 4,652,226 24,842 9
27 Chile Santiago 8,410 5,342,913 2,026 710,563 4
28 Colombia Bogota 5,480 4,672,324 1,587 5,770,632 273,090 1,313 943,076 7
29 Colombia Cali 5,480 1,642,200 298,645 564 2,274,237 132,822 209,628 7
30 Colombia Medellin 5,480 1,643,339 333,341 395 5,818,068 173,851 147 445,130 8
31 Ecuador Guayaquil 4,350 1,572,615 1,098 1,489,644 62,871 5
32 Peru Lima 3,300 6,053,900 1,074,000 3,850 7,900 270,343 202,225 7
33 Uruguay Montevideo 6,070 1,309,200 420,000 542 1,979,808 87,900 223,034 35,320 8
34 Venezuela Maracaibo 8,520 1,248,853 543 206,965 80,952,979 8,954 118,783 7
35 Bangladesh Chittagong 1,230 1,391,877 65 349,879 14 18,051 51,924 7
Real per Num. of
36 Bangladesh Dacca 1,230 12,380,000 518,612 7,470 1,083,370 12,253 109,846 53,461 51,110 9
37 China Anshan 1,950 1,375,500 622 4,428,000 11,303 339 29,213 7
38 China Botou 1,950 1,180,500 2,153 2,537,000 21,759 393 29,691 7
39 China Changchun 1,950 2,069,100 1,116 1,865,000 16,041 664 64,407 7
40 China Changsha 1,950 1,301,200 367 902,000 24,588 384 42,947 7
41 China Chengdu 1,950 2,776,200 1,382 1,870,000 39,063 582 65,210 7
42 China Chongqing 1,950 1,471,100 1,534 6,211,000 151,700 31,651 1,190 34,345 40,920 9
43 China Dalian 1,950 2,368,500 2,415 3,390,000 18,631 680 65,589 7
44 China Fushun 1,950 1,334,700 675 4,210,000 22,658 349 32,494 7
45 China Fuzhou 1,950 1,269,100 1,043 1,273,000 27,238 412 40,602 7
46 China Guangzhou 1,950 3,543,900 1,444 4,040,460 56,260,061 823,070 945 337,913 8
47 China Guiyang 1,950 1,488,100 2,436 2,848,000 13,403 152 10,635 7
48 China Hangzhou 1,950 1,328,400 430 2,017,000 33,644 498 67,676 7
49 China Harbin 1,950 2,798,200 1,637 2,623,000 26,715 1,627 90,038 7
50 China Huainan 1,950 1,173,200 1,091 1,472,000 108,685 309 12,080 7
51 China Jilin 1,950 1,253,400 1,213 4,173,000 9,671 448 29,195 7
52 China Jinan 1,950 2,215,000 1,190,133 1,943 4,228,850 316,396 173,910 934 40,007 18,771 10
53 China Kaohsiung 1,950 1,374,231 295,384 154 363,722 619 484,512 68,935 8
54 China Kunming 1,950 1,505,100 2,081 1,862,000 10,425 255 33,729 7
55 China Lanzhou 1,950 1,482,200 1,632 4,942,000 36,829 455 53,875 7
56 China Liupanshui 1,950 1,798,900 6,272 739,000 597 49 8,950 7
57 China Nanchang 1,950 1,326,200 617 1,836,720 25,974 265 43,455 7
58 China Nanjiang 1,950 2,469,000 947 4,016,000 94,554 819 93,375 7
59 China Peking 1,950 6,920,500 2,738 10,638,000 88,181 2,784 469,378 7
60 China Pinxiang 1,950 1,376,900 2,765 630,000 2,520 29 7,576 7
61 China Qingdo 1,950 2,036,300 1,103 3,682,130 13,730 10,349 583 3,809 4,671 9
62 China Qiqihar 1,950 1,365,000 4,365 2,352,000 11,667 611 26,784 7
63 China Shanghai 1,950 7,777,900 749 20,328,000 4,530 218,698 1,340 233,400 8
64 China Shenyang 1,950 4,502,200 3,495 4,585,000 54,519 976 6
65 China Shijianzhuang 1,950 1,296,900 307 2,602,000 32,515 338 54,393 7
66 China Tai'an 1,950 1,402,700 2,089 113,000 3,451 644 11,517 7
67 China Taipei 1,950 2,702,678 660,024 272 7,711,093 276,481,211 483,888 1,387 1,186,350 204,815 10
68 China Taiyan 1,950 1,907,300 1,460 5,039,020 129,000 133,840 538 8,714 8
69 China Tangshan 1,950 1,488,800 1,090 344,878 8,791 518 39,908 3,450 8
70 China Tianjin 1,950 5,697,700 4,276 9,125,000 65,810 2,901 188,213 7
Real per Num. of
71 China Wuhan 1,950 3,706,700 1,607 5,461,000 81,297 1,141 107,554 7
72 China Xi'an 1,950 2,708,000 1,100 2,772,000 32,547 1,016 110,741 20,521 8
73 China Zaozhuang 1,950 1,700,800 3,065 2,971,000 9,240 497 10,288 7
74 China Zhengzhou 1,950 1,662,300 248,491 1,010 3,390,610 75 24,037 340 173,232 2,295 10
75 China Zibo 1,950 2,433,200 202,308 2,961 3,895,240 177,516 123,610 2,508 20,383 5,682 10
76 India Ahmedabad 1,230 2,854,988 190 1,675,274 130,921 1,214 138,114 7
77 India Bangalore 1,230 3,326,723 223,227 151 1,294 115,625 6
78 India Bombay 1,230 8,578,000 438 1,526,786 593,904 5
79 India Delhi 1,230 8,250,500 1,483 5,600,000 168,000 630,102 18,073 514,280 8
80 India Hyderabad 1,230 2,093,488 169 571,034 61,675 5
81 India Madras 1,230 3,785,600 64,922 170 1,541,829 636,600 112,634 2,581 113,422 2,081 10
82 Indonesia Bandung 2,950 1,404,360 41,776 66,645 3,451 21,500 28,118 7
83 Indonesia Jakarta 2,950 7,003,267 1,114,575 663 5,917,812 122,376 140,810 3,583 325,101 565,532 10
84 Indonesia Semarang 2,950 1,096,271 373 267,582 8,932 1,550,934 6
85 Indonesia Surabaja 2,950 2,431,547 290 550 76,058 5
86 Iran Esfahan 5,420 986,753 202 934,307 1,372,271 54,768 162 67,833 8
87 Iran Mashhad 5,420 2,086,600 249,162 156 1,548,005 2,485 60,610 87,949 8
88 Iran Teheran 5,420 6,964,700 1,154,746 567 7,043,749 346,000 312,000 3,654 280,165 9
89 Iraq Baghdad 3,413 2,350,000 863 1,423,700 158,320 106,467 6
90 Japan Fukuoka 20,520 1,237,062 419,740 336 6,092,243 1,603,718 130,526 3,571 750,823 75,613 10
91 Japan Hiroshima 20,520 1,085,705 361,560 740 5,846,860 1,216,028 139,229 3,511 538,817 57,737 10
92 Japan Kawasaki 20,520 1,173,603 388,360 143 8,817,191 3,593,029 159,352 2,358 550,968 44,973 10
93 Japan Kitakyushu 20,520 1,026,455 343,220 782 6,387,417 1,444,340 116,475 3,764 509,625 61,073 10
94 Japan Kobe 20,520 1,477,410 482,440 544 6,804,813 3,401,909 185,520 5,006 686,246 82,770 10
95 Japan Kyoto 20,520 1,461,103 516,320 610 6,669,560 4,370,520 209,733 3,051 761,240 105,383 10
96 Japan Nagoya 20,520 2,154,793 716,420 326 13,056,422 5,511,279 283,757 6,062 1,254,081 153,129 10
97 Japan Osaka 20,520 2,623,801 963,380 220 19,169,124 10,190,148 487,405 3,873 1,940,459 276,229 10
98 Japan Sapporo 20,520 1,671,742 572,850 1,121 5,060,824 1,221,793 157,570 4,811 817,200 78,768 10
99 Japan Tokyo 20,520 8,163,573 3,112,590 618 45,741,359 25,390,476 1,144,719 11,377 5,810,773 665,863 10
100 Japan Yokohama 20,520 3,220,331 1,001,800 432 12,450,594 7,283,980 404,471 8,522 1,418,200 117,945 10
101 Korea, Repub. of Inchon 9,250 1,616,017 274,245 208 4,680,412 362,618 131,608 896 437,688 59,365 10
102 Korea, Repub. of Pusan 9,250 3,857,312 427,308 526 6,531,786 251,611 290,284 1,750 1,108,124 155,100 10
103 Korea, Repub. of Seoul 9,250 10,627,790 1,463,081 605 16,027,052 502,213 1,005,588 7,376 3,929,984 483,986 10
104 Korea, Repub. of Taegu 9,250 2,288,441 300,295 456 3,773,640 14,598,500 193,491,527 111 625,783 97,856 10
105 Pakistan Lahore 2,890 3,808,661 328 493 4
Real per Num. of
GDP Num. of Land area consumption consumption [1 Water supply roads subscrip- establish- data points
Country City [PPP$] Population dwellings [sq.km.] [1000 kwh] mill. kcal] [1000 cu.m.] [km] tions ments (out of 10)
106 Philippines Manila 2,550 1,835,290 150,642 38 1,661,769 7,300 59,900 766 119,800 30,320 10
107 Philippines Quezon City 2,550 1,504,361 210,810 154 1,142 5
108 Singapore Singapore 18,330 2,647,100 623 11,734,800 668,700 296,934 2,810 1,220,000 8
109 Syria Aleppo 4,960 1,191,151 6,026 3
110 Syria Damascus 4,960 1,219,448 3,021 3
111 Thailand Bangkok 5,950 5,546,937 1,565 15,828,600 866,673 2,785 1,011,498 7
112 Turkey Ankara 5,230 2,632,906 628,071 788 799,998 51,776 92,738 3,455 395,000 9
113 Turkey Istanbul 5,230 5,560,908 1,512 348,286 4
114 Turkey Izmir 5,230 2,023,100 1,500 130,354 4
115 Viet Nam Ho Chi Minh City 1,010 3,934,000 2,029 380,743 4
116 Austria Vienna 18,710 1,545,663 847,087 415 8,238,300 215,000 2,753 1,294,806 66,127 9
117 Bulgaria Sofia 4,250 1,206,908 1,310 258,666 252,000 5
118 Czechoslovakia Prague 7,190 1,214,885 516,127 496 3,590,972 11,613,800 158,693 2,566 815,417 9
119 France Paris 19,510 2,152,333 1,282,000 105 8,837,000 8,060 207,597 1,310,000 262,820 9
120 German Demo. Rep. East Berlin 21,120 1,260,921 603,658 403 623,670 5
121 Germany, Fed'l Rep. West Berlin 21,120 2,130,525 1,074,821 480 10,327,000 1,014,500 183,900 2,937 1,228,325 87,217 10
122 Germany, Fed'l Rep. Hamburg 21,120 1,626,220 787,258 755 11,974,362 12,585,910 120,978 3,889 1,072,500 77,735 10
123 Germany, Fed'l Rep. Munich 21,120 1,268,366 655,077 311 5,901,803 21,714,305 132,595 2,263 995,279 71,070 10
124 Hungary Budapest 6,580 2,104,700 823,000 525 6,048,276 20,624,656 322,966 3,134 809,463 9
125 Italy Milan 18,090 1,449,403 703,379 182 4,979,227 739,026 252,236 1,396 1,139,645 107,908 10
126 Italy Naples 18,090 1,204,149 310,997 117 3,005,971 1,083,645,726 148,825,250 701 508,942 49,760 10
127 Italy Rome 18,090 2,817,227 1,337,247 1,508 6,728,653 658,348 388,800 4,500 1,142,485 124,235 10
128 Italy Torino 18,090 1,002,813 445,932 130 3,279,577 595,600 184,775 1,393 737,965 72,591 10
129 Poland Warsaw 4,830 1,671,376 586,956 485 392,170 3,697 6
130 Romania Bucharest 2,840 2,318,889 942,799 1,820 4,550,237 2,849,200 476,518 1,874 581,000 9
131 Spain Barcelona 13,400 1,714,355 699,031 99 4,476,575 2,141,864 145,978 1,237 851,483 82,837 10
132 Spain Madrid 13,400 3,102,846 1,181,414 606 7,098,254 233,258 1,890,010 116,686 8
133 United Kingdom Birmingham 17,160 992,500 395,680 264 5,272,512 9,684,000 116,942 2,120 24,491,000 9
134 United Kingdom London, Greater 17,160 6,770,435 1,579 3,340,300 166,218 5
135 Yugoslavia Belgrade 1,216,000 517,988 360 4,639,474 162,615 1,233 524,502 4,226 8
136 Australia Melbourne 18,220 3,043,500 1,086,614 6,129 22,786 5
137 Australia Sydney 18,220 3,530,953 1,225,257 12,407 20,191,430 48,951 471,203 20,307 1,628,506 190,215 10
138 USSR Alma-Ata 4,270 1,121,395 302,499 180 2,770,000 820,400 276,872 1,235 236,757 9
139 USSR Baku 2,550 1,794,874 794 6,067 2,971,000 517,359 2,180 216,935 8
140 USSR Chelyabinsk 6,140 1,143,000 486 13,314,000 5,546,200 238,860 166,524 7
Real per Num. of
141 USSR Dnepropetrovsk 5,010 1,179,000 5,331,100 225,586 162,343 5
142 USSR Donetsk 5,010 1,148,936 541 1,725,900 204,669 164,780 6
143 USSR Erevan 2,420 1,199,000 90 2,320,600 485,701 231,162 6
144 USSR Gorky 6,140 1,438,000 334 3,832,800 288,725 228,744 6
145 USSR Kazan 6,140 1,094,000 285 3,387,800 211,964 102,243 6
146 USSR Kharkov 5,010 1,611,000 303 3,440,100 305,630 275,926 6
147 USSR Kiev 5,010 2,572,212 777 8,493,000 6,230,800 514,829 623,501 7
148 USSR Kuibyshev 6,140 1,257,268 357,926 459 8,180,000 3,454,400 298,066 188,443 8
149 USSR Leningrad 6,140 4,990,749 570 19,248,000 10,826,900 939,335 3,069 1,593,130 8
150 USSR Minsk 6,440 1,607,077 159 6,336,000 3,883,600 230,970 364,934 7
151 USSR Moscow 6,140 8,875,579 1,059 34,914,000 24,920,300 2,125,791 3,542,610 7
152 USSR Novosibirsk 6,140 1,436,000 447 5,551,000 1,081,400 329,598 214,665 7
153 USSR Odessa 5,010 1,115,000 2,666,000 1,365,500 199,419 151,702 6
154 USSR Omsk 6,140 1,148,000 436 8,906,000 125,303 118,795 6
155 USSR Perm 6,140 1,091,000 5,735,000 4,852,500 167,466 93,362 6
156 USSR Rostov-na-Donu 6,140 1,020,000 372 5,844,000 1,681,100 209,489 120,115 7
157 USSR Sverdlovsk 6,140 1,404,016 405,200 1,025 3,789,900 189,535 1,325 208,487 8
158 USSR Tashkent 2,650 2,060,206 250 4,757,000 3,488,500 717,658 310,765 7
159 USSR Tblisi 2,300 1,246,936 349 3,319,000 2,000,700 406,205 217,825 7
160 USSR Ufa 6,140 1,083,000 8,976,000 4,958,300 183,822 101,600 6
Number of data points (out of 160 ) 159 160 74 151 132 84 138 110 146 61
Mean 7,437 2,647,654 700,010 1,098 6,667,131 20,665,946 3,330,065 9,239 837,873 128,681
Standard deviation 7,066 2,228,199 546,849 1,534 8,704,193 121,334,069 21,721,001 67,303 2,240,394 226,165
Sources:
[2] All other data: Tokyo Metropolitan Government. Bureau of General Affairs. Statistical Division. 1992. Statistics of World Large Cities 1991.
Notes:
Electricity consumption = Total (domestic + industrial) consumption of electric light and power
Gas consumption = Total (domestic + industrial) gas consumption
Water supply = Total (domestic + industrial) water supply
Source [1] data is for 1992.
Source [2] data is for as close to 1990 as possible, but may be for any year in the 1980s.
Appendix E
Sample Analysis Evaluation of “seismic code indicator”
Seismic code benchmark years

Benchmark yr. Boston Istanbul Jakarta Lima Manila Mex. City San Fran. Santiago St. Louis Tokyo
1 1975 1975 1983 1977 1981 1987 1988 1966 1989 1981
2 1968 1970 1968 1972 1976 1976 1949 1971
3 1949 1959 1957 1956 1935 1950
4 1940 1942 1947 1924
5 1933 1920
Sophistication rating of seismic codes between benchmark years [sk]

Inter-benchmark
period k
post-b.y. 1 0.8 0.9 0.9 0.9 0.9 1.0 1.0 0.9 0.9 1.0
b.y.2 to b.y.1 0.1 0.8 0.8 0.5 0.8 0.9 0.9 0.5 0.1 0.9
b.y.3 to b.y.2 0.5 0.1 0.1 0.5 0.8 0.8 0.3 0.8
b.y.4 to b.y.3 0.3 0.1 0.3 0.5 0.1 0.3
b.y.5 to b.y.4 0.1 0.1 0.3 0.1
pre-b.y.5 0.1
Percentage of 1990 city population that arrived between benchmark years [pk]
Inter-benchmark
period k
post-b.y. 1 6.0 44.7 13.7 38.7 22.7 3.6 3.0 31.8 0.1 8.4
b.y. 2 thru b.y.1 94.0 16.1 32.6 20.4 22.9 21.5 15.9 27.7 99.9 13.1
b.y. 3 thru b.y.2 23.6 53.8 40.9 24.2 45.9 27.8 13.6 34.7
b.y. 4 thru b.y.3 8.0 30.1 18.9 13.3 26.9 15.2
b.y. 5 thru b.y.4 7.6 10.1 13.4 2.6
pre-b.y.5 26.6 26.0
Enforcement rating [e]

Below
Qualitative Excellent Average Poor Poor Poor Average Excellent Average Excellent Excellent
Numerical 1.0 0.6 0.2 0.2 0.2 0.6 1.0 0.4 1.0 1.0
Code factor rating [XV1]

Unscaled 14.18 40.82 8.76 9.83 10.78 39.81 52.88 19.70 10.08 52.79
Scaled 2.64 1.20 2.93 2.87 2.82 1.25 0.54 2.34 2.86 0.55
* Higher value = better code, lower risk
Sophistication Enforce-
level s k Category description ment level e Category description
0.1 No regulations. 1.0 Excellent
0.3 No theory. Only base shear as percent of weight. 0.8 Above Average
0.5 Rudimentary base shear equation. 0.6 Average
0.8 Some of the basic components of a seismic code. 0.4 Below Average
0.9 All of the basic components of a seismic code. 0.2 Poor
1.0 Refinement of code with all basic components.
240 Appendix E: "Seismic code indicator"

XV1 = e*[Σ k(sk)*(pk)]
Appendix E: "Seismic code indicator" 241

Appendix F
External Context Indicator Values for 319-city Data Set
[A] [B] [C] [D] [E]

Scaled XC2 Scaled XC3
Scaled XC1 Political Political Total External
City Economic City country City world City political City Context
external external external external factor
context context context context C
=0.5*[B] + 0.5*[C] =0.8*[A] + 0.2*[D]
1 Tokyo 13.925 Abidjian 3.656 Washington D.C. 16.016 Washington D.C. 9.836 Tokyo 12.498
2 New York 12.982 Accra 3.656 Tokyo 9.920 Tokyo 6.788 New York 10.799
Los Angeles/ Los Angeles/
3 Riv./ SB 8.728 Addis Ababa 3.656 Berlin 5.967 Berlin 4.812 Riv./ SB 7.373
4 Paris 5.549 Algiers 3.656 Paris 5.047 Paris 4.351 Paris 5.309
5 London 5.191 Alma-Ata 3.656 Rome 4.768 Rome 4.212 London 4.970
6 Essen 5.022 Amman 3.656 London 4.514 London 4.085 Essen 4.377
7 Osaka 3.923 Amsterdam 3.656 Sacramento 3.469 Ottawa 3.459 Washington D.C. 3.865
8 Chicago 3.795 Ankara 3.656 Ottawa 3.261 Teheran 3.443 Osaka 3.521
9 Milan 3.030 Athens 3.656 Teheran 3.230 Madrid 3.382 Chicago 3.387
San Francisco/
10 San Jose 2.913 Baghdad 3.656 Madrid 3.108 Beijing 3.151 Berlin 3.088
Mexico City/ Mexico City/
11 Naucalpan 2.801 Baku 3.656 New York 2.688 Amsterdam 3.109 Naucalpan 2.844
12 Frankfurt 2.748 Bangkok 3.656 Dusseldorf 2.687 Delhi 3.097 Milan 2.769
13 Moscow 2.667 Beijing 3.656 Beijing 2.646 Seoul 3.038 Rome 2.687
14 Berlin 2.657 Belgrade 3.656 Amsterdam 2.561 Pyongyang 3.038 San Jose 2.670
15 Philadelphia 2.567 Berlin 3.656 Delhi 2.537 Stockholm 3.022 Seoul 2.621
16 Seoul 2.517 Bogota 3.656 Riv./ SB 2.493 Naucalpan 3.015 Frankfurt 2.511
17 Shanghai 2.488 Brasilia 3.656 Seoul 2.419 Brussels 2.982 Madrid 2.499
18 Cologne 2.445 Brazzaville 3.656 Pyongyang 2.419 Brasilia 2.933 Beijing 2.428
19 Naples 2.394 Brussels 3.656 Columbus 2.391 Helsinki 2.898 Philadelphia 2.388
20 Detroit 2.385 Bucharest 3.656 Stockholm 2.389 Copenhagen 2.893 Toronto 2.351
21 Toronto 2.384 Budapest 3.656 Naucalpan 2.373 Ankara 2.887 Dusseldorf 2.350
22 Dusseldorf 2.376 Buenos Aires 3.656 Osaka 2.363 Buenos Aires 2.873 Shanghai 2.301
Cairo/ Shubra El-
23 Washington D.C. 2.372 Khemia 3.656 Munich 2.358 Cape Town 2.854 Cologne 2.286
24 Sao Paulo 2.341 Cape Town 3.656 Brussels 2.308 Pretoria 2.854 Naples 2.273
25 Hamburg 2.309 Caracas 3.656 Toronto 2.300 Riyadh 2.850 Brussels 2.245
26 Rome 2.306 Conakry 3.656 Stuttgart 2.269 Jakarta 2.847 Sao Paulo 2.240
27 Stuttgart 2.286 Copenhagen 3.656 Brasilia 2.210 Bangkok 2.826 Detroit 2.239
28 Madrid 2.278 Dakar 3.656 Nagoya 2.206 Warsaw 2.823 Amsterdam 2.228
29 Beijing 2.247 Damascus 3.656 Hanover 2.161 Athens 2.801 Teheran 2.222
30 Dallas 2.242 Dar es Salaam 3.656 Sapporo 2.161 Baghdad 2.798 Jakarta 2.214
Miami/ Ft.
31 Lauderdale 2.228 Delhi 3.656 Chicago 2.160 Lisbon 2.797 Stuttgart 2.213
32 Hong Kong 2.190 Dhaka 3.656 Atlanta 2.142 Dublin 2.777 Khemia 2.190
33 Montreal 2.189 Dublin 3.656 Helsinki 2.139 Khemia 2.777 Delhi 2.173
34 Houston 2.161 Guatemala City 3.656 Copenhagen 2.130 Caracas 2.773 Hamburg 2.167
35 Munich 2.157 Hanoi 3.656 Ankara 2.118 Kuala Lumpur 2.770 Stockholm 2.167
36 Bombay 2.127 Harare 3.656 Boston 2.116 Metro Manila 2.764 Pyongyang 2.164
37 Boston 2.127 Havana 3.656 Essen 2.109 Singapore 2.764 Bangkok 2.159
38 Birmingham 2.127 Helsinki 3.656 Buenos Aires 2.089 Bogota 2.761 Ottawa 2.152
39 Tianjin 2.122 Hong Kong 3.656 Indianapolis 2.089 Budapest 2.758 Buenos Aires 2.134
40 Manchester 2.120 Jakarta 3.656 San Jose 2.063 Brazzaville 2.754 Singapore 2.127
41 St. Petersburg 2.065 Kabul 3.656 Naples 2.061 Lagos 2.752 Munich 2.123
42 Brussels 2.061 Kampala 3.656 Cape Town 2.051 Santiago 2.750 Dallas 2.122
43 Jakarta 2.056 Khartoum 3.656 Pretoria 2.051 Rabat 2.749 Lauderdale 2.110
44 Calcutta 2.051 Kiev 3.656 Riyadh 2.043 Yangon 2.747 Montreal 2.100
Cairo/ Shubra El-
45 Khemia 2.044 Kinshasa 3.656 Jakarta 2.038 Bucharest 2.747 Riyadh 2.082
Appendix F: External Context Data Set 8

[A] [B] [C] [D] [E]
46 San Diego 2.035 Kuala Lumpur 3.656 Sydney 2.038 Khartoum 2.746 Copenhagen 2.072
47 Barcelona 2.029 Kuwait City 3.656 Philadelphia 2.014 Damascus 2.746 Metro Manila 2.071
48 Rio de Janeiro 2.018 La Paz 3.656 Bangkok 1.996 Dhaka 2.741 Lagos 2.061
49 Amsterdam 2.008 Lagos 3.656 Warsaw 1.990 Kuwait City 2.739 Houston 2.055
50 Rotterdam 2.006 Lima 3.656 Milan 1.988 Yaounde 2.732 Boston 2.048
51 Sydney 1.999 Lisbon 3.656 Detroit 1.984 Tunis 2.731 Cape Town 2.046
52 Atlanta 1.997 London 3.656 Phoenix 1.982 Lima 2.731 Athens 2.044
53 Bangkok 1.992 Luanda 3.656 Melbourne 1.971 Abidjian 2.729 Helsinki 2.041
54 Minneapolis 1.981 Lusaka 3.656 Hiroshima 1.964 Quito 2.727 Bombay 2.037
55 Mannheim 1.973 Madrid 3.656 Florence 1.963 Addis Ababa 2.726 Birmingham 2.032
56 Phoenix 1.970 Managua 3.656 Denver 1.961 Nairobi 2.725 Lisbon 2.031
57 Singapore 1.967 Maputo 3.656 Dallas 1.957 Guatemala City 2.723 Ankara 2.031
Miami/ Ft.
58 Nagoya 1.960 Metro Manila 3.656 Lauderdale 1.954 Montevideo 2.722 Caracas 2.031
Mexico City/
59 St. Louis 1.957 Naucalpan 3.656 Oklahoma City 1.952 Dakar 2.722 Warsaw 2.027
60 Istanbul 1.956 Minsk 3.656 Kyoto 1.949 Accra 2.722 Manchester 2.026
61 Stockholm 1.953 Montevideo 3.656 Barcelona 1.948 Harare 2.722 Brasilia 2.025
62 Buenos Aires 1.949 Moscow 3.656 Athens 1.946 Santo Domingo 2.722 Santiago 2.023
63 Baltimore 1.948 Muqdisho 3.656 Houston 1.939 La Paz 2.721 Bogota 2.018
64 Pyongyang 1.945 Nairobi 3.656 Baghdad 1.939 San Jose 2.721 Sydney 2.016
65 Melbourne 1.943 Ottawa 3.656 Lisbon 1.938 Panama City 2.721 Kuala Lumpur 2.015
66 Delhi 1.942 Panama City 3.656 Montreal 1.938 Amman 2.720 Dublin 2.012
67 Leeds 1.926 Paris 3.656 Kitakyushu 1.934 Kinshasa 2.719 Pretoria 2.011
68 Seattle 1.926 Port-au-Prince 3.656 Lille 1.929 Kampala 2.718 Lima 2.007
69 Vienna 1.925 Prague 3.656 Cologne 1.924 Lusaka 2.718 Tianjin 2.006
70 Tampa 1.920 Pretoria 3.656 Sendai 1.918 Dar es Salaam 2.718 Budapest 2.004
71 Kiev 1.918 Pyongyang 3.656 Mashhad 1.915 Port-au-Prince 2.718 Dhaka 1.999
72 Teheran 1.917 Quito 3.656 Tabriz 1.912 Maputo 2.716 Bucharest 1.996
73 Cleveland 1.914 Rabat 3.656 Edmonton 1.910 Managua 2.715 Baghdad 1.993
74 Pittsburgh 1.914 Riga 3.656 San Diego 1.908 Sacramento 2.521 Kuwait City 1.993
75 Hanover 1.913 Riyadh 3.656 Birmingham 1.903 Dusseldorf 2.247 Damascus 1.990
76 Kitakyushu 1.908 Rome 3.656 Manchester 1.901 Lahore 2.224 Tunis 1.988
77 Bielefeld 1.908 San Jose 3.656 Dublin 1.897 Toronto 2.215 Rabat 1.987
Cairo/ Shubra El-
78 Turin 1.906 San Juan 3.656 Khemia 1.897 Sydney 2.081 Abidjian 1.987
79 Metro Manila 1.897 Santiago 3.656 Turin 1.897 New York 2.066 Brazzaville 1.986
80 Zurich 1.897 Santo Domingo 3.656 Lucknow 1.896 Munich 1.990 Yangon 1.985
Los Angeles/
81 Vancouver 1.893 Seoul 3.656 Minneapolis 1.892 Riv./ SB 1.953 Kinshasa 1.985
82 Denver 1.893 Singapore 3.656 Brisbane 1.892 Melbourne 1.942 Khartoum 1.984
83 Riyadh 1.890 Sofia 3.656 Caracas 1.890 Stuttgart 1.920 Panama City 1.982
84 Nuremberg 1.890 Stockholm 3.656 St. Louis 1.884 Osaka 1.913 Santo Domingo 1.982
85 Lagos 1.888 Tashkhent 3.656 Kuala Lumpur 1.884 Columbus 1.893 Nairobi 1.982
86 Pusan 1.885 Tbilisi 3.656 Zurich 1.883 Auckland 1.856 Yaounde 1.981
87 Katowice 1.878 Teheran 3.656 Baltimore 1.881 Hanover 1.836 Quito 1.981
88 Shenyang 1.876 Tokyo 3.656 Marseilles 1.875 Sao Paulo 1.832 Montevideo 1.981
89 Aachen 1.872 Tripoli 3.656 Seattle 1.873 Tel Aviv-Yafa 1.813 Addis Ababa 1.981
90 Norfolk 1.870 Tunis 3.656 Metro Manila 1.872 Nagoya 1.812 Sacramento 1.980
91 Copenhagen 1.867 Vienna 3.656 Salt Lake City 1.871 Karachi 1.811 Dakar 1.980
92 Karachi 1.867 Warsaw 3.656 Tampa 1.871 Guayaquil 1.809 Amman 1.980
93 Kansas City 1.863 Washington D.C. 3.656 Singapore 1.871 Zurich 1.796 Accra 1.980
94 Lyons 1.861 Yangon 3.656 Genoa 1.869 Essen 1.795 Guatemala City 1.980
95 Karlsruhe 1.859 Yaounde 3.656 Cleveland 1.869 Surabaja 1.793 Harare 1.980
96 Milwaukee 1.858 Yerevan 3.656 Pittsburgh 1.869 Naples 1.790 San Jose 1.979
97 Marseilles 1.857 Lahore 2.619 Sao Paulo 1.866 Lucknow 1.785 Lusaka 1.979
98 Cincinnati 1.856 Toronto 2.130 Bogota 1.865 Aleppo 1.783 Dar es Salaam 1.979
99 Madras 1.855 Sydney 2.124 Lyons 1.862 Sapporo 1.783 La Paz 1.979
100 Athens 1.854 Melbourne 1.914 Hamburg 1.861 Brisbane 1.778 Barcelona 1.978
101 Saarland 1.854 Auckland 1.909 Shiraz 1.861 Bandung 1.777 Maputo 1.978
102 Portland 1.852 Guayaquil 1.840 Budapest 1.860 Barcelona 1.775 Kampala 1.978
103 Wuhan 1.849 Karachi 1.826 Esfahan 1.859 Semarang 1.759 Port-au-Prince 1.978
104 Kyoto 1.849 Tel Aviv-Yafa 1.823 Chengdu 1.857 Chicago 1.758 Managua 1.977
105 Bremen 1.849 Dusseldorf 1.807 Taegu 1.857 Atlanta 1.748 Calcutta 1.972
106 San Antonio 1.848 Sao Paulo 1.798 Vancouver 1.856 Montreal 1.746 San Diego 1.951
107 Caracas 1.846 Aleppo 1.781 Mannheim 1.852 Boston 1.732 Rio de Janeiro 1.948
108 Sacramento 1.845 Surabaja 1.760 Brazzaville 1.851 Taegu 1.728 Atlanta 1.947
109 Cape Town 1.844 Banghazi 1.754 Norfolk 1.849 Milan 1.725 Melbourne 1.943
110 Genoa 1.844 Bandung 1.732 Lagos 1.848 Mashhad 1.718 Rotterdam 1.933
111 Guangzhou 1.842 Zurich 1.709 Kansas City 1.846 Indianapolis 1.717 Nagoya 1.930
112 Santiago 1.841 Semarang 1.699 Santiago 1.844 Tabriz 1.714 Phoenix 1.907
113 Kharkov 1.840 Lucknow 1.673 Milwaukee 1.843 Edmonton 1.710 Minneapolis 1.906

[A] [B] [C] [D] [E]
114 Lisbon 1.840 Brisbane 1.664 Leeds 1.843 Maracaibo 1.707 Istanbul 1.901
115 New Orleans 1.839 Medellin 1.625 Bombay 1.842 Medellin 1.706 Hanover 1.898
116 Sapporo 1.838 Maracaibo 1.624 Cincinnati 1.842 Florence 1.703 Mannheim 1.897
San Francisco/
117 Bogota 1.833 Munich 1.622 Rabat 1.842 San Jose 1.701 Lahore 1.895
118 Lille 1.832 Peshawar 1.603 Portland 1.840 Peshawar 1.694 St. Louis 1.885
119 Buffalo 1.832 Barcelona 1.602 Patna 1.840 Belo Horizonte 1.694 Baltimore 1.877
120 Florence 1.831 Taegu 1.599 Zhengzhou 1.840 Chengdu 1.688 Zurich 1.877
121 Columbus 1.831 Sacramento 1.574 Adelaide 1.839 Istanbul 1.681 Leeds 1.860
122 Hyderabad 1.830 Khulna 1.573 Perth 1.839 Khulna 1.676 Seattle 1.859
123 Tel Aviv-Yafa 1.830 Stuttgart 1.571 Yangon 1.839 Bombay 1.674 Karachi 1.856
124 Jeddah 1.829 Casablanca 1.566 San Antonio 1.838 Casablanca 1.673 Tampa 1.854
125 Chongqing 1.829 Belo Horizonte 1.564 Jinan 1.837 Philadelphia 1.673 Kitakyushu 1.854
126 Indianapolis 1.829 Montreal 1.555 Bucharest 1.837 Adelaide 1.670 Turin 1.853
127 Tashkhent 1.829 Istanbul 1.551 Bielefeld 1.836 Patna 1.669 Cleveland 1.849
128 Bangalore 1.828 Cali 1.543 Khartoum 1.836 Perth 1.669 Pittsburgh 1.849
129 Warsaw 1.828 Katowice 1.540 Damascus 1.835 Rio de Janeiro 1.668 Columbus 1.844
130 Belgrade 1.828 Donetsk 1.538 Calcutta 1.834 Katowice 1.668 Denver 1.843
131 Helsinki 1.827 Ibadan 1.530 Hyderabad 1.833 Lille 1.667 Vancouver 1.843
Santiago de los
132 Orlando 1.827 Caballos 1.523 New Orleans 1.832 Cali 1.663 Bielefeld 1.843
133 Kuala Lumpur 1.826 Mashhad 1.522 Nuremberg 1.831 Calcutta 1.657 Katowice 1.836
134 Algiers 1.826 Rio de Janeiro 1.521 Bhopal 1.831 Zhengzhou 1.656 Pusan 1.834
Providence-
135 Lima 1.826 Chengdu 1.520 Warwick 1.830 Hiroshima 1.656 Nuremberg 1.828
136 Chengdu 1.825 Naples 1.519 Saarland 1.830 Detroit 1.655 Sapporo 1.827
137 Ottawa 1.825 Tabriz 1.517 Lahore 1.830 Ibadan 1.655 Tel Aviv-Yafa 1.826
138 Harbin 1.825 Hanover 1.511 Rotterdam 1.828 Hyderabad 1.654 Shenyang 1.818
139 Taipei 1.825 Edmonton 1.509 Buffalo 1.827 Phoenix 1.654 Auckland 1.815
140 Taegu 1.824 Bombay 1.506 Madras 1.827 Birmingham 1.654 Aachen 1.813
Providence-
141 Warwick 1.823 Thessaloniki 1.504 Dhaka 1.826 Bhopal 1.652 Madras 1.812
142 Guadalajara 1.823 Adelaide 1.501 Aachen 1.826 Manchester 1.651 Brisbane 1.812
143 Xian 1.822 Perth 1.499 Surabaja 1.826 Taejon 1.651 Norfolk 1.812
144 Memphis 1.822 Patna 1.499 Nanjing 1.825 Jinan 1.651 Lyons 1.811
145 Johannesburg 1.821 Douala 1.497 Orlando 1.824 Cologne 1.651 Marseilles 1.810
Santiago de los
146 Brisbane 1.820 Cordoba 1.496 Taejon 1.824 Caballos 1.648 Kyoto 1.808
147 Alexandria 1.820 Salvador 1.483 Belo Horizonte 1.823 Cordoba 1.648 Indianapolis 1.806
148 Dublin 1.820 Essen 1.481 Bandung 1.822 Shiraz 1.648 Surabaja 1.806
149 West Palm Beach 1.820 Calcutta 1.480 Kuwait City 1.822 Thessaloniki 1.646 Florence 1.806
150 Belo Horizonte 1.820 Taejon 1.478 Guangzhou 1.822 Kyoto 1.646 Kansas City 1.806
151 Salt Lake City 1.819 Valencia 1.477 Karlsruhe 1.821 Salvador 1.646 Taegu 1.805
152 Edmonton 1.819 Hyderabad 1.476 Memphis 1.820 Esfahan 1.645 Bandung 1.805
153 Oklahoma City 1.819 Zhengzhou 1.472 Shijiazhuang 1.820 Turin 1.643 Wuhan 1.801
154 Tripoli 1.818 Bhopal 1.472 Changsha 1.820 Rotterdam 1.643 Karlsruhe 1.801
155 Dnepropetrovsk 1.818 Jeddah 1.471 West Palm Beach 1.819 Madras 1.643 Milwaukee 1.801
156 Nanjing 1.818 Jinan 1.465 Semarang 1.819 Denver 1.642 Cincinnati 1.800
157 Ankara 1.817 Osaka 1.463 Hefei 1.817 Vancouver 1.640 Guayaquil 1.800
158 Louisville 1.817 Milan 1.462 Louisville 1.816 Dallas 1.640 Genoa 1.799
Miami/ Ft.
159 Casablanca 1.817 Dnepropetrovsk 1.460 Pusan 1.815 Lauderdale 1.638 Lille 1.799
160 Prague 1.817 Madras 1.459 Rio de Janeiro 1.815 Oklahoma City 1.637 Saarland 1.799
161 Monterrey 1.816 Rotterdam 1.458 Wuhan 1.815 Kitakyushu 1.637 Guangzhou 1.798
162 Dalian 1.816 Porto Alegre 1.450 Bangalore 1.814 Douala 1.637 Chengdu 1.798
163 Ahmedabad 1.816 Pusan 1.447 Guadalajara 1.811 Jeddah 1.632 Edmonton 1.797
164 Budapest 1.815 Curitiba 1.447 Jaipur 1.811 Pusan 1.631 Portland 1.796
165 Donetsk 1.815 Maracay 1.445 Istanbul 1.811 Houston 1.629 Aleppo 1.795
166 Novosibirsk 1.815 New York 1.444 Frankfurt 1.811 Valencia 1.629 Hyderabad 1.795
167 Nizhni Novgorod 1.815 Guadalajara 1.443 Bremen 1.810 Nanjing 1.628 Lucknow 1.795
168 Odessa 1.814 Florence 1.442 Salvador 1.808 Guadalajara 1.627 Belo Horizonte 1.795
169 Jinan 1.814 Alexandria 1.442 Yaounde 1.807 Porto Alegre 1.626 San Antonio 1.792
170 Dhaka 1.814 Shiraz 1.434 Tunis 1.806 Sendai 1.626 Bremen 1.791
171 Porto Alegre 1.813 Kharkov 1.434 Nanning 1.806 Curitiba 1.624 Jeddah 1.790
172 Perth 1.813 Nanjing 1.432 Hangzhou 1.805 Guangzhou 1.624 Semarang 1.789
173 Inchon 1.813 Esfahan 1.431 Lima 1.805 Marseilles 1.620 Casablanca 1.788
174 Lahore 1.813 Medan 1.426 Shenyang 1.803 Shijiazhuang 1.620 Bangalore 1.786
175 Ekaterinburg 1.812 Guangzhou 1.426 Tel Aviv-Yafa 1.803 Changsha 1.619 Maracaibo 1.785
176 Bandung 1.812 Vancouver 1.424 Nanchang 1.803 Genoa 1.618 New Orleans 1.785
177 Recife 1.811 Bangalore 1.420 Abidjian 1.803 Bangalore 1.617 Mashhad 1.784
178 Changchun 1.810 Shijiazhuang 1.420 Porto Alegre 1.802 Hefei 1.614 Perth 1.784
Puebla de
179 Taiyuan 1.810 Changsha 1.419 Zaragoza 1.802 Maracay 1.613 Guadalajara 1.784
180 Adelaide 1.809 Nagoya 1.418 Kunming 1.802 Alexandria 1.611 Oklahoma City 1.783
181 Surabaja 1.809 Recife 1.416 Auckland 1.802 San Diego 1.611 Adelaide 1.782

[A] [B] [C] [D] [E]
182 Samara 1.809 Lvov 1.414 Curitiba 1.802 Wuhan 1.611 Jinan 1.781
183 Bucharest 1.809 Odessa 1.412 Trivandrum 1.801 Jaipur 1.610 Tabriz 1.781
Los Angeles/
184 Omsk 1.807 Riv./ SB 1.412 Cordoba 1.800 Lyons 1.609 Nanjing 1.780
185 Zaporozhye 1.807 Hefei 1.411 Harbin 1.800 Medan 1.607 Alexandria 1.779
186 Chelyabinsk 1.806 Jaipur 1.410 Xian 1.799 Recife 1.606 Medellin 1.778
Puebla de Puebla de
187 Salvador 1.806 Zaragoza 1.407 Guiyang 1.799 Zaragoza 1.605 Buffalo 1.778
Ho Chi Minh
188 Kuwait City 1.806 City 1.407 Quito 1.797 Minneapolis 1.602 Harbin 1.776
189 Hiroshima 1.806 Wuhan 1.406 Fuzhou 1.796 Hamburg 1.601 Porto Alegre 1.776
190 Pune 1.805 Lille 1.405 Recife 1.796 Leeds 1.597 Hiroshima 1.776
191 Kazan 1.805 Sapporo 1.405 Karachi 1.796 St. Louis 1.597 Johannesburg 1.775
192 Ufa 1.805 Birmingham 1.404 Katowice 1.796 Fortaleza 1.596 Orlando 1.774
193 Perm 1.805 Manchester 1.402 Taiyuan 1.795 Baltimore 1.595 Salvador 1.774
194 Maracaibo 1.805 Johannesburg 1.401 Addis Ababa 1.795 Mannheim 1.595 Zhengzhou 1.774
195 Auckland 1.805 Barranquilla 1.400 Monterrey 1.795 Nanning 1.594 Peshawar 1.774
196 East Rand 1.804 Fortaleza 1.399 Nairobi 1.794 Johannesburg 1.593 Xian 1.774
197 Fortaleza 1.804 Izmir 1.397 Jeddah 1.793 Izmir 1.593 Salt Lake City 1.774
Providence-
198 Lvov 1.804 Columbus 1.395 Fortaleza 1.793 Hangzhou 1.592 Warwick 1.772
199 Izmir 1.804 Mendoza 1.393 Changchun 1.792 Seattle 1.591 Taipei 1.772
200 Zhengzhou 1.804 Turin 1.390 Inchon 1.792 Trivandrum 1.591 Chongqing 1.770
201 Rostov-on-Don 1.803 St. Petersburg 1.390 Guatemala City 1.791 Salt Lake City 1.590 Recife 1.770
202 Sendai 1.803 Ujung Pandang 1.389 Lanzhou 1.791 Tampa 1.590 Monterrey 1.770
203 Durban 1.803 Nanning 1.382 Hohhot 1.790 Mendoza 1.590 Patna 1.770
204 Kunming 1.803 Zaporozhye 1.381 Maracaibo 1.789 Barranquilla 1.589 Taejon 1.770
205 Volgograd 1.803 Gdansk 1.381 Montevideo 1.789 Cleveland 1.588 Cali 1.770
206 Medan 1.803 Trivandrum 1.380 Thessaloniki 1.789 Pittsburgh 1.588 Memphis 1.769
207 Kanpur 1.802 Hangzhou 1.379 Taipei 1.788 Shenyang 1.588 Khulna 1.769
208 Sofia 1.802 Monterrey 1.379 Izmir 1.788 Nanchang 1.587 West Palm Beach 1.768
209 Tunis 1.802 Cologne 1.377 Dakar 1.788 Monterrey 1.587 Sendai 1.768
210 Abidjian 1.802 Shenyang 1.373 Accra 1.788 Ujung Pandang 1.587 Esfahan 1.767
211 Tangshan 1.802 Nanchang 1.372 Medan 1.788 Kunming 1.586 Ibadan 1.767
212 Guiyang 1.801 Goiania 1.371 Harare 1.788 Harbin 1.583 Thessaloniki 1.766
213 Krasnoyarsk 1.801 Rosario 1.371 Santo Domingo 1.788 Bielefeld 1.582 Bhopal 1.766
214 Lanzhou 1.801 Kunming 1.371 Goiania 1.788 Gdansk 1.581 Curitiba 1.766
215 Curitiba 1.801 Ekaterinburg 1.370 Mendoza 1.786 Xian 1.580 Inchon 1.766
216 Saratov 1.801 Adana 1.370 La Paz 1.786 Guiyang 1.580 Shiraz 1.765
217 Kinshasa 1.801 Cracow 1.370 Peshawar 1.786 Goiania 1.579 Valencia 1.765
218 Pretoria 1.801 Lodz 1.368 San Jose 1.786 Nuremberg 1.578 Louisville 1.765
219 Mashhad 1.801 Genoa 1.368 Medellin 1.786 Adana 1.577 Cordoba 1.765
Santiago de los
220 Voronezh 1.801 Palembang 1.366 Panama City 1.786 Rosario 1.577 Caballos 1.764
221 Damascus 1.801 Marseilles 1.365 Belem 1.786 Saarland 1.577 Medan 1.763
222 Kaohsiung 1.801 Harbin 1.365 Johannesburg 1.786 Norfolk 1.577 Douala 1.763
223 Kwangju 1.800 East Rand 1.363 Aleppo 1.786 Inchon 1.576 Shijiazhuang 1.763
Ho Chi Minh
224 City 1.800 Belem 1.362 Kwangju 1.786 Fuzhou 1.576 Changsha 1.763
225 Anshan 1.800 Rostov-on-Don 1.362 Urumqi 1.785 Kansas City 1.575 Taiyuan 1.763
226 Qiqihar 1.800 Xian 1.362 Adana 1.784 Cracow 1.575 Fortaleza 1.763
227 Lodz 1.800 Guiyang 1.361 Ujung Pandang 1.784 Taiyuan 1.574 Changchun 1.762
228 Fushun 1.800 Inchon 1.361 Rosario 1.784 Belem 1.574 Izmir 1.762
229 Nanchang 1.800 Durban 1.359 Amman 1.783 Aachen 1.574 Ahmedabad 1.760
230 Hangzhou 1.800 Nizhni Novgorod 1.359 Shanghai 1.783 Lodz 1.574 Dalian 1.760
231 Qingdao 1.800 Chicago 1.356 Cali 1.783 Palembang 1.574 Maracay 1.760
232 Valencia 1.799 Lyons 1.356 Valencia 1.782 Milwaukee 1.573 Kunming 1.760
233 Taejon 1.799 Ufa 1.356 Bursa 1.782 Cincinnati 1.573 Jaipur 1.760
234 Urumqi 1.799 Fuzhou 1.356 Kinshasa 1.782 East Rand 1.572 Hefei 1.760
235 Changsha 1.799 Bursa 1.354 Palembang 1.781 Portland 1.572 Hangzhou 1.758
Puebla de
236 Fuzhou 1.799 Chittagong 1.354 East Rand 1.781 Karlsruhe 1.570 Zaragoza 1.758
237 Shijiazhuang 1.799 Atlanta 1.353 Gdansk 1.781 San Antonio 1.570 East Rand 1.758
238 Jilin 1.799 Taiyuan 1.353 Casablanca 1.781 Durban 1.570 Nanning 1.757
239 Aleppo 1.798 Leeds 1.352 Durban 1.781 Bursa 1.568 Nanchang 1.757
240 Luanda 1.798 Chelyabinsk 1.352 Alexandria 1.781 Changchun 1.568 Guiyang 1.757
241 Nagpur 1.798 Kazan 1.351 Kampala 1.781 New Orleans 1.567 Durban 1.756
Providence-
242 Brasilia 1.798 Boston 1.349 Maracay 1.781 Warwick 1.566 Lodz 1.755
243 Baotou 1.798 Hiroshima 1.347 Ibadan 1.781 Lanzhou 1.565 Fuzhou 1.754
244 Nanning 1.798 Perm 1.347 Cracow 1.780 Buffalo 1.564 Gdansk 1.754
245 Lucknow 1.798 Indianapolis 1.345 Tianjin 1.780 Chittagong 1.564 Lanzhou 1.754
246 West Rand 1.798 Changchun 1.343 Lodz 1.780 Hohhot 1.563 Ujung Pandang 1.753
247 Gdansk 1.798 Kyoto 1.343 Lusaka 1.780 Orlando 1.562 Trivandrum 1.753
248 Luoyang 1.798 Samara 1.342 Srinagar 1.780 Frankfurt 1.562 Barranquilla 1.753
249 Minsk 1.798 Hamburg 1.342 Dar es Salaam 1.780 Kwangju 1.562 Kwangju 1.753

[A] [B] [C] [D] [E]
250 Esfahan 1.798 Krasnoyarsk 1.342 Khulna 1.779 Bremen 1.562 Cracow 1.753
San Francisco/ Leon de los
251 Alma-Ata 1.798 San Jose 1.340 Aldamas 1.779 Taipei 1.561 Mendoza 1.752
252 Panama City 1.797 West Rand 1.340 Port-au-Prince 1.779 Memphis 1.560 Palembang 1.752
253 Tabriz 1.797 Lanzhou 1.340 Ciudad Jarez 1.779 West Rand 1.559 Belem 1.752
254 Rabat 1.797 Kitakyushu 1.339 West Rand 1.779 West Palm Beach 1.559 Adana 1.752
255 Handan 1.797 Kwangju 1.338 Guayaquil 1.779 Louisville 1.558 Pune 1.752
256 Cracow 1.797 Voronezh 1.338 Tijuana 1.778 Urumqi 1.555 Goiania 1.751
257 Santo Domingo 1.797 Novosibirsk 1.338 Manaus 1.778 Shanghai 1.551 Rosario 1.750
258 Jaipur 1.797 Hohhot 1.337 Barranquilla 1.777 Davao 1.550 Urumqi 1.750
259 Guayaquil 1.797 Mannheim 1.337 Campinas 1.777 Manaus 1.550 West Rand 1.750
260 Surat 1.797 Saratov 1.337 Maputo 1.776 Faisalabad 1.548 Bursa 1.749
Leon de los
261 Datong 1.797 Volgograd 1.334 Santos 1.776 Aldamas 1.548 Hohhot 1.749
262 Semarang 1.797 Sendai 1.334 Douala 1.776 Ciudad Jarez 1.547 Kanpur 1.749
263 Campinas 1.797 Taipei 1.334 Ahmedabad 1.776 Srinagar 1.546 Chittagong 1.749
264 Maracay 1.797 Philadelphia 1.332 Chongqing 1.775 Tijuana 1.546 Tangshan 1.748
265 Medellin 1.797 Bielefeld 1.328 Pune 1.775 Tianjin 1.546 Kaohsiung 1.747
Puebla de
266 Zaragoza 1.796 Detroit 1.327 Dalian 1.775 Gujranwala 1.545 Anshan 1.747
267 Palembang 1.796 Phoenix 1.327 Kanpur 1.775 Campinas 1.545 Qiqihar 1.747
268 Cali 1.796 Davao 1.325 Nagpur 1.774 Rawalpindi 1.543 Fushun 1.747
Santiago de los
269 Belem 1.796 Omsk 1.325 Caballos 1.774 Multan 1.543 Qingdao 1.747
270 Nairobi 1.796 Nuremberg 1.325 Davao 1.774 Santos 1.543 Campinas 1.746
271 Thessaloniki 1.796 Urumqi 1.325 Chittagong 1.774 Hyderabad 1.542 Jilin 1.746
Leon de los
272 Hefei 1.796 Saarland 1.324 Bogor 1.774 Bogor 1.539 Aldamas 1.746
273 Daqing 1.796 Denver 1.323 Surat 1.774 Ahmedabad 1.539 Nagpur 1.746
274 Wuxi 1.796 Dallas 1.323 Faisalabad 1.774 Tanjung Karang 1.538 Faisalabad 1.746
Miami/ Ft.
275 Xuzhou 1.796 Lauderdale 1.322 Tangshan 1.774 Pune 1.537 Manaus 1.746
276 Benxi 1.795 Faisalabad 1.322 Tanjung Karang 1.774 Chongqing 1.537 Baotou 1.745
277 Yichun 1.795 Aachen 1.322 Managua 1.774 Kanpur 1.536 Luoyang 1.745
278 Hohhot 1.795 Oklahoma City 1.322 Kaohsiung 1.774 Dalian 1.536 Ciudad Jarez 1.745
279 Amman 1.795 Manaus 1.321 Anshan 1.774 Nagpur 1.536 Davao 1.745
280 Adana 1.795 Houston 1.320 Qiqihar 1.774 Surat 1.535 Surat 1.745
281 Jixi 1.795 Shanghai 1.320 Fushun 1.774 Tangshan 1.535 Tijuana 1.745
282 Montevideo 1.795 Karlsruhe 1.319 Qingdao 1.774 Kaohsiung 1.535 Santos 1.745
Leon de los
283 Banghazi 1.795 Aldamas 1.317 Gujranwala 1.774 Coimbatore 1.535 Handan 1.745
284 Ujung Pandang 1.795 Gujranwala 1.316 Coimbatore 1.774 Kochi 1.535 Datong 1.744
285 Jinzhou 1.795 Ciudad Jarez 1.315 Kochi 1.774 Anshan 1.535 Srinagar 1.744
Leon de los
286 Aldamas 1.795 San Diego 1.315 Jilin 1.774 Qiqihar 1.535 Gujranwala 1.744
287 Santos 1.795 Frankfurt 1.313 Vadodara 1.774 Vadodara 1.535 Bogor 1.744
288 Coimbatore 1.795 Campinas 1.313 Indore 1.774 Indore 1.535 Rawalpindi 1.744
289 Faisalabad 1.795 Tijuana 1.313 Madurai 1.774 Madurai 1.535 Daqing 1.743
290 Shantou 1.795 Bremen 1.313 Baotou 1.774 Fushun 1.535 Multan 1.743
291 Shiraz 1.795 Rawalpindi 1.313 Ulhasnagar 1.774 Qingdao 1.535 Wuxi 1.743
292 Dakar 1.795 Srinagar 1.313 Visakhapatnam 1.774 Ulhasnagar 1.534 Xuzhou 1.743
293 Kochi 1.795 Multan 1.312 Luoyang 1.774 Visakhapatnam 1.534 Benxi 1.743
294 Ibadan 1.795 Minneapolis 1.312 Varanasi 1.774 Varanasi 1.534 Yichun 1.743
295 Bogor 1.795 Tianjin 1.311 Rawalpindi 1.774 Jilin 1.534 Tanjung Karang 1.743
296 Patna 1.795 St. Louis 1.311 Ludhiana 1.774 Ludhiana 1.534 Hyderabad 1.743
297 Vadodara 1.795 Hyderabad 1.310 Handan 1.774 Agra 1.534 Jixi 1.743
298 Indore 1.795 Baltimore 1.310 Multan 1.774 Baotou 1.534 Coimbatore 1.743
299 Madurai 1.795 Santos 1.309 Agra 1.774 Luoyang 1.534 Jinzhou 1.743
300 Douala 1.795 Seattle 1.309 Datong 1.774 Jabalpur 1.534 Kochi 1.743
301 Quito 1.795 Salt Lake City 1.308 Jabalpur 1.774 Handan 1.534 Vadodara 1.743
302 Bursa 1.795 Tampa 1.308 Hyderabad 1.774 Datong 1.534 Indore 1.743
303 Chittagong 1.795 Cleveland 1.308 Allahabad 1.774 Allahabad 1.534 Madurai 1.743
304 Ciudad Jarez 1.795 Pittsburgh 1.308 Jamshedpur 1.774 Jamshedpur 1.534 Shantou 1.743
305 Suzhou 1.795 Norfolk 1.305 Meerut 1.774 Meerut 1.534 Ulhasnagar 1.743

306 Bhopal 1.795 Kansas City 1.304 Vijayawada 1.774 Vijayawada 1.534 Visakhapatnam 1.742
307 Ulhasnagar 1.795 Bogor 1.304 Daqing 1.774 Dhanbad 1.534 Varanasi 1.742
308 Visakhapatnam 1.794 Milwaukee 1.304 Dhanbad 1.774 Kozhikode 1.534 Suzhou 1.742
309 Manaus 1.794 Cincinnati 1.304 Wuxi 1.774 Daqing 1.534 Ludhiana 1.742
310 Varanasi 1.794 Portland 1.303 Xuzhou 1.774 Wuxi 1.534 Fuxin 1.742
311 Fuxin 1.794 San Antonio 1.303 Kozhikode 1.774 Xuzhou 1.534 Liuzhou 1.742
312 Liuzhou 1.794 Tanjung Karang 1.303 Benxi 1.774 Benxi 1.534 Agra 1.742
313 Ludhiana 1.794 New Orleans 1.302 Yichun 1.774 Yichun 1.534 Jabalpur 1.742
Providence-
314 Tijuana 1.794 Warwick 1.302 Jixi 1.774 Jixi 1.534 Allahabad 1.742
315 Yangon 1.794 Buffalo 1.301 Jinzhou 1.774 Jinzhou 1.534 Jamshedpur 1.742
316 Addis Ababa 1.794 Ahmedabad 1.301 Shantou 1.774 Shantou 1.534 Meerut 1.742
317 Goiania 1.794 Orlando 1.301 Suzhou 1.774 Suzhou 1.534 Vijayawada 1.742
318 Lusaka 1.794 Memphis 1.300 Fuxin 1.774 Fuxin 1.534 Dhanbad 1.742
319 Dar es Salaam 1.794 West Palm Beach 1.300 Liuzhou 1.774 Liuzhou 1.534 Kozhikode 1.742
Notes:
All data are for 1990.
The capital of Nigeria moved from Lagos to Abuja in 1991.
Sources:
Population data: United Nations. 1995. World urbanization prospects: 1994 revision. New York: UN.
Total foreign trade data: International Monetary Fund. 1995. Balance of payments yearbook 1995. IMF.
GDP data: International Monetary Fund. 1996. International financial statistics yearbook 1996. IMF.

Appendix G
Principal Components Analysis of Physical Infrastructure
Exposure Data Sets
G.1. Physical Infrastructure Exposure data set from Compendium of Human Settlements
Statistics 1995 (UN Centre for Human Settlements 1995).
Correlation matrix [R]
Population 1.000 0.880 0.189
Num. housing units 0.880 1.000 0.206
Per capita GDP (PPP$) 0.189 0.206 1.000
Eigenvectors [v] Principal component

3 2 1
Population 0.705 -0.208 0.678
Num. housing units -0.709 -0.187 0.680
Eigenvalues [e] 0.120 0.919 1.961

Percentage variation explained 4.0% 30.6% 65.4%
Factor loadings [v*sqrt(e)]

Population 0.244 -0.199 0.949
Num. housing units -0.245 -0.179 0.953
Factor loadings--Normalized so the sum over all indicators is one.

Population 0.41
Num. housing units 0.42
Per capita GDP (PPP$) 0.17
Sources:
[2] All other data: UN Centre for Human Settlements (Habitat). Compendium of Human Settlements
Statistics 1995. 5th issue. New York: UN.
Appendix G: Principal Components Analysis 14

G.2. Physical Infrastructure Exposure data set from Statistics of World Large Cities 1991
(TMG 1991).
Correlation matrix [R]
Population 1.000 -0.053 0.603 0.247
Per capita GDP (PPP$) -0.053 1.000 0.347 -0.093
Num. of housing units 0.603 0.347 1.000 0.175
Land area (sq.km.) 0.247 -0.093 0.175 1.000
Eigenvectors [v] Principal component

4 3 2 1
Population 0.608 0.431 0.235 0.624
Per capita GDP (PPP$) 0.382 -0.434 -0.785 0.223
Num. of housing units -0.696 0.127 -0.218 0.672
Land area (sq.km.) 0.010 -0.781 0.530 0.330
Eigenvalues [e] 0.280 0.784 1.175 1.761

Percentage variation explained 7.0% 19.6% 29.4% 44.0%
Factor loadings [v*sqrt(e)]

Population 0.322 0.381 0.254 0.828
Per capita GDP (PPP$) 0.202 -0.384 -0.851 0.296
Num. of housing units -0.369 0.113 -0.236 0.892
Land area (sq.km.) 0.005 -0.691 0.575 0.438
Factor loadings--Normalized so the sum over all indicators is one.

Population 0.34
Per capita GDP (PPP$) 0.12
Num. of housing units 0.36
Land area (sq.km.) 0.18
Sources:
[2] All other data: Tokyo Metropolitan Government. Bureau of General Affairs. Statistical Division. 1992.
Statistics of World Large Cities 1991.
Appendix G: Principal Components Analysis 15

Appendix H
Questionnaire
June 23, 1996
Dear Earthquake Professional,
As part of my doctoral thesis I am developing a multidisciplinary Earthquake Disaster Risk

Index (EDRI) that will:
1. allow straightforward comparison of the relative overall earthquake disaster risk of

different cities throughout the world, and
2. describe the relative contributions of various factors to that overall risk.
The following questionnaire is intended to capture the opinions of experts concerning the relative
importance of the five main factors comprising a city's earthquake disaster risk—Hazard,
Exposure, Vulnerability, Response Capability, and External Context.
Please take a few minutes to complete this questionnaire. Return it now, or take it with you and
send your response to the mailing address, e-mail address, or fax number listed below. If you
include your return address in Question 14, I will forward a copy of the results to you when they
are available.
Thank you very much for your time and interest.
Sincerely,
Rachel Davidson
Appendix H: Questionnaire 247

DEFINITIONS
In completing this survey, please apply the following definitions of earthquake disaster and of the
five main factors of which it is comprised:
Earthquake disaster is a function of not only the physical impact of an earthquake (as
determined by the hazard, exposure, and vulnerability of the infrastructure), but also the response
capability of the affected city, and the perceived relevance of the physical impact to the city and
to world affairs. The social, economic, political, and cultural context of the damage determines
whether an earthquake will create a disaster situation and how extensively its effect will be felt.
Hazard refers to the likelihood of each of the possible levels of severity of ground shaking, and of
collateral hazards like fire, landslide, liquefaction, and tsunami.
Exposure conveys the number and geographical distribution of the city's population, buildings, and
lifelines.
Vulnerability describes how easily the components of the infrastructure can be damaged and the
population can be harmed in the absence of any response activities. The vulnerability depends on
the (1) structural types, (2) quality of design and construction, (3) changes in structural condition
since construction, including the effects of aging, maintenance, damage in previous earthquakes,
and retrofitting, (4) expected nonstructural and content damage, and (5) extent of redundancy in
the utility and transportation networks.
Response Capability defines how effectively and efficiently a city can recover from short- and
long-term impact. Response relates to (1) the level of pre-earthquake operational and
organizational planning, (2) the financial, manpower, equipment, and facility resources available,
and (3) the ability to gain access to the city and maneuver within it following an event.
External Context describes the extent to which damage to a city affects those who live outside
the city. It is included in recognition of the fact that, depending on cities' prominence with respect
to economics, politics, transportation, and culture, damage to certain cities may create a more
significant ripple effect than damage to others.
QUESTIONS
In answering Questions 1 to 10, it may be helpful to use the following scenario to conceptualize
the proposed comparisons.
Consider two hypothetical cities, City A and City B. Suppose that you have assessed the
two cities with respect to each of the five factors defined above (i.e., Hazard, Exposure, ...,
External Context). Assume that the factors are defined so that a higher factor value always
corresponds to higher overall risk. You have determined that the two cities are identical in
every way, except that:
a. Factor 1 is greater in City A than in City B, and
b. Factor 2 is greater, by the same margin, in City B than in City A.

Would you consider City A to have an overall earthquake disaster risk that is higher than,
lower than, or roughly equivalent to that of City B?
250 Appendix H: Questionnaire

In your opinion, which of the two factors in each What ratio of Factor 1 to Factor 2 describes the
box is more important in determining a city's relative importance of the two factors?
earthquake disaster risk? Circle the appropriate ______________________________________
letter. _
For example, if Exposure is twice as important as
Hazard, write Hazard:Exposure::1: 2 .
______________________________________
_
1a. A. Hazard
B. Exposure
C. About the same 1b.
Hazard:Exposure::1:_____
2a. A. Exposure
B. Vulnerability
Exposure: Vulnerability::1:_____
3a. A. Response Capability
B. External Context
Response Cap.: External Context::1:_____
4a. A. Exposure
B. External Context
Exposure: External Context::1:_____
5a. A. Hazard
B. External Context
Hazard: External Context::1:_____
6a. A. Vulnerability
B. Response Capability
Vulnerability: Response Cap.::1:_____
7a. A. Hazard
B. Vulnerability
Hazard: Vulnerability::1:_____
8a. A. Exposure
Exposure: Response Capability::1:_____
9a. A. Hazard
Hazard: Response Capability::1:_____
10a. A. Vulnerability
B. External Context
Vulnerability: External Context::1:_____

11. Assign weighting factors, wi, such that
a. wi describes the relative importance of factor i to the overall earthquake disaster
risk of a city.
b. Each weight wi has a value from 1 to 10 (i.e., 1≤wi≤10 for all wi).
c. The factor(s) that are MOST important have a weight wi=10, and all other
factors are weighted relative to the most important one(s). The sum of the weights
is unimportant.
FACTOR i WEIGHT wi
Hazard ____________
Exposure ____________
Vulnerability ____________
Response Capability ____________
External Context ____________
12. List any factors that, in your opinion, are important in determining a city's earthquake
disaster risk, but were not considered in this survey.
13. What is your field of expertise?__________________________________________

In which country do you live?___________________________________________
14. (Optional) If you are interested in receiving a copy of the results of this survey, please write
the address or fax number to which they can be sent:
_____________________________________________________________________
_____________________________________________________________________
_____________________________________________________________________
_____________________________________________________________________

Minimum Maximum
Factor Indicator possible possible
value value
xh1 exp(MMI w/50-year return period) 20 4024
xh2 exp(MMI w/500-year return period) 148 8103
xh3 Percentage urbanized area w/soft soil 0% 75%
Hazard xh4 Percentage urbanized area w/high liquefaction suscept. 0% 75%
xh5 Percentage of buildings that are wood 0% 90%
xh6 Population density (people/sq. km.) 300 25000
xh7 Tsunami potential indicator 0 2
xe1 Population (1000s) 2,000 40,000
xe2 Per capita GDP in constant 1990 US$ 200 30,000
Exposure xe3 Number of housing units (1000s) 1,000 18,000
xe4 Urbanized land area (sq.km.) 100 2000
xe5 Population (1000s) 2,000 40,000
xe6 Per capita GDP in constant 1990 US$ 200 30,000
xv1 Seismic code indicator 2 60
xv2 City wealth indicator 50 5000
Vulner- xv3 City age indicator 5 55
ability xv4 Population density (people/sq. km.) 300 25,000
xv5 City development speed indicator 10 120
xv6 Percentage of population aged 0-4 or 65+ 10.00% 25.00%
External xc1 Economic external context indicator 100 5.50E+13
Context xc2 Political country external context indicator 0.0005 1
xc3 Political world external context indicator 2.00E+08 6.00E+12
xr1 Planning indicator 1 4
xr2 Per capita GDP in constant 1990 US$ 200 30,000
xr3 Avg. annual real growth in per cap. GDP in prev. 10 yrs. -15% 15%
Emerg. xr4 Housing vacancy rate 0% 20%
Resp. & xr5 Number of hospitals per 100,000 residents 0.5 10.0
Recovery xr6 Number of physicians per 100,000 residents 10 500
xr7 Extreme weather indicator 0 300
xr8 Population density (people/sq. km.) 300 25,000
xr9 City layout indicator 0 1

Appendix I
Code for sens.c Sensitivity Analysis Computer Program
/***********************************************************************
* Rachel Davidson February 19, 1997
*
* Program to perform a sensitivity analysis on the EDRI results for a sample of cities
*
* Input results for EDRI analysis from file "SensIn".
* Output senstivity results into files "SensOut1," "SensOut2," and "SensOut3"
*
* Compile with: gcc sens.c -lm -o sens -lrecipes_c
*
***********************************************************************/
#include <stdio.h>
#include <math.h>
#include <nrutil.h>
#define m 10 /* number of cities in sample */
#define nmax 10 /* max of nh, .., nr, 5 */

#define SWAP(a,b) itemp=(a)=(b); (b)=itemp; /* for ranking function */
#define M 11 /* for ranking function */
#define NSTACK 50 /* for ranking function */
#define basecity 2 /* num. of city used as base city in scaling w.r.t. base city */
/***********************************************************************
* Function prototypes
***********************************************************************/
void scale(
int option,
int inv,
double X[],
double Xprime[],
double minposs,
double maxposs
);
void normwgts(
int numwgts,
Appendix I: "sens.c" Code 16

double wgts[]
);
Appendix I: "sens.c" Code 17

void lincombo( );
int numterms,
double wgts[], double calcrs(
double X[][m], unsigned long U[],
double combo[] unsigned long V[]
); );
void rank( /*********************************************************

unsigned long n, ******
double arr[], * Main program
unsigned long ranks[] **********************************************************
); *****/
void updatestats1( main()

double min1[][m], {
double max1[][m], FILE *fpin; /* file pointer to read input file */
double mean1[][m], FILE *fpout1; /* file pointer to write to output file 1 */
double sd1[][m], FILE *fpout2; /* file pointer to write to output file 2 */
double sum1[][m], FILE *fpout3; /* file pointer to write to output file 3 */
double sumsq1[][m],
double F[][m+1], int n[6]; /* num. of indicators for each main factor and EDRI */
double Forig[][m+1], double x[5][nmax][m]; /* matrix of unscaled data x values */
int t, double xorig[5][nmax][m]; /* original matrix of unscaled data x */
int tskip[6], double xprime[6][nmax][m]; /* matrix of scaled x values */
int min1trial[][m], double w[6][nmax]; /* weights */
int max1trial[][m] double worig[6][nmax]; /* original weights */
); double dw; /* percent change for weights */
int inv[5][nmax]; /* inverse-direct indicator */
void printstats1( double minposs[5][nmax]; /* min. possible value for scaling */
double min1[][m], double maxposs[5][nmax]; /* max. possible value for scaling */
double max1[][m], int dupe[5][nmax]; /* duplicate indicator id */
double mean1[][m], int dup;
double sd1[][m], int option; /* to identify which scaling option to use */
int min1trial[][m], double d[5][nmax][m]; /* deltas for data uncertainty */
int max1trial[][m], double Forig[6][m+1]; /* original main factor and EDRI values */
FILE *fpout3 /* ranks of orig. main factor & EDRI values */
unsigned long Forigranks[6][m+1]; }
int i, j, k, p, q, r, s; /* looping indices */
int t; /* number of trials */ fprintf(fpout1, "\nInput data.\n\nFactor Num. inds.");
int tskip[6]; /* num. trials to skip in updating stats1 */
double F[6][m+1]; /* calculated main factor & EDRI values */ for(i=0; i<6; i++) {
/* ranks of main factor & EDRI values */ fscanf(fpin, "%d", &n[i]);
unsigned long Franks[6][m+1]; fprintf(fpout1,"\n %d %d", i, n[i]);
double sum1[6][m]; /* sum for main factor & EDRI values */ }
double sumsq1[6][m]; /* sumsq for main factor & EDRI values */
double min1[6][m]; /* min for main factor & EDRI values */ fprintf(fpout1,"\n\nFactor Indicator weights");
double max1[6][m]; /* max for main factor & EDRI values */ for(i=0; i<6; i++) {
double mean1[6][m]; /* mean for main factor & EDRI values */ for(j=0; j<n[i]; j++) {
double sd1[6][m]; /* standard dev. for main factor & EDRI values */ fscanf(fpin, "%lf", &w[i][j]);
int min1trial[6][m]; /* num. of the trial that gives min1 */ worig[i][j] = w[i][j];
int max1trial[6][m]; /* num. of the trial that gives max1 */ }
double rs2[6]; /* Spearman rank correlation coefficient */ fprintf(fpout1,"\n %d ", i);
for(j=0; j<n[i]; j++) {
/****************************************************** fprintf(fpout1," %6.4lf", w[i][j]);
***** }
* Open input and output files }
* Read input from files "SensIn", and print input data to SensOut1
fscanf(fpin, "%lf", &dw);
****************************************************** fprintf(fpout1,"\n\ndw = %5.3lf", dw);
*****/
fprintf(fpout1,"\n\nFactor Indicators (1=Direct. 0=Inverse)");
if((fpin = fopen("SensIn", "r") ) == NULL) { for(i=0; i<5; i++) {
printf("Couldn't open input file SensIn."); for(j=0; j<n[i]; j++) {
} fscanf(fpin, "%d", &inv[i][j]);
if((fpout1 = fopen("SensOut1", "wr")) == NULL) { }
printf("Couldn't open output file SensOut1."); fprintf(fpout1,"\n %d ", i);
} for(j=0; j<n[i]; j++) {
if((fpout2 = fopen("SensOut2", "wr")) == NULL) { fprintf(fpout1," %d", inv[i][j]);
printf("Couldn't open output file SensOut2."); }
} }
if((fpout3 = fopen("SensOut3", "wr")) == NULL) {
printf("Couldn't open output file SensOut3."); fprintf(fpout1,"\n\nFactor Minimum possible values for scaling");
for(i=0; i<5; i++) { for(k=0; k<m; k++) {
for(j=0; j<n[i]; j++) { fscanf(fpin, "%lf", &x[i][j][k]);
fscanf(fpin, "%lf", &minposs[i][j]); xorig[i][j][k] = x[i][j][k];
} fprintf(fpout1,"%9.2e", x[i][j][k]);
fprintf(fpout1,"\n %d ", i); }
for(j=0; j<n[i]; j++) { }
fprintf(fpout1,"%9.2e", minposs[i][j]); }
}
} for(i=0; i<5; i++) {
fprintf(fpout1,"\n\nFactor %d delta values", i);
fprintf(fpout1,"\n\nFactor Maximum possible values for scaling"); for(j=0; j<n[i]; j++) {
for(i=0; i<5; i++) { fprintf(fpout1,"\n");
for(j=0; j<n[i]; j++) { for(k=0; k<m; k++) {
fscanf(fpin, "%lf", &maxposs[i][j]); fscanf(fpin, "%lf", &d[i][j][k]);
} fprintf(fpout1,"%9.2e", d[i][j][k]);
fprintf(fpout1,"\n %d ", i); }
for(j=0; j<n[i]; j++) { }
fprintf(fpout1,"%9.2e", maxposs[i][j]); }
} fprintf(fpout1,"\n\n\nOutput data\n");
}
fprintf(fpout1,"\n\nFactor Indicators (Duplicate id. 0=no dups.)"); /******************************************************

for(i=0; i<5; i++) { *****
for(j=0; j<n[i]; j++) { * Evaluate original five main factor and EDRI results and print in
fscanf(fpin, "%d", &dupe[i][j]); * SensOut2:
} * Scale all x[i][j][k] values
fprintf(fpout1,"\n %d ", i); * Evaluate five main factor and EDRI values
for(j=0; j<n[i]; j++) { * Calculate city ranks with respect to each of five main factors
fprintf(fpout1," %d", dupe[i][j]); * and EDRI
} ******************************************************
} *****/
for(i=0; i<5; i++) { option = 0; /* scaled all unscaled data x values */

fprintf(fpout1,"\n\nFactor %d raw data values", i); for(i=0; i<5; i++) {
for(j=0; j<n[i]; j++) { for(j=0; j<n[i]; j++) {
fprintf(fpout1,"\n"); scale(option, inv[i][j], x[i][j], xprime[i][j], minposs[i][j],
maxposs[i][j]); sumsq1[i][k] = 0;
} min1[i][k] = Forig[i][k];
} max1[i][k] = Forig[i][k];
min1trial[i][k] = 0;
fprintf(fpout2, "\nOriginal main factor and EDRI values\n\nCity:\n"); max1trial[i][k] = 0;
for(k=0; k<m; k++) { }
fprintf(fpout2, " %d", k); tskip[i] = 0;
} }
fprintf(fpout2, "\nFactor:"); t = 0;
for(i=0; i<6; i++) { /* calculate five main factor and EDRI values */
lincombo(n[i], w[i], xprime[i], Forig[i]);
rank(m, Forig[i], Forigranks[i]); /* calculate ranks */
if(i!=5) { /******************************************************
for(k=0; k<m; k++) { *****
xprime[5][i][k] = Forig[i][k]; * Perform sensitivity analysis to data uncertainty
} ******************************************************
} *****/
fprintf(fpout2, "\n%d", i);
for(k=0; k<m; k++) { fprintf(fpout1,"\nSensitivity to data uncertainty.\n");
fprintf(fpout2, " %6.3lf", Forig[i][k]); fprintf(fpout2,"\nSensitivity to data uncertainty.\n");
} for(s=0; s<2; s++) { /* loop through +delta and -delta */
} for(p=0; p<5; p++) { /* loop through five main factors */
fprintf(fpout2, "\n"); for(q=0; q<n[p]; q++) { /* loop thru all inds. in each factor */
for(r=0; r<m; r++) { /* loop through all cities */
t++; /* increment num. trials */
/****************************************************** if (s==0) { /* change x by +delta */
***** x[p][q][r] = x[p][q][r] + d[p][q][r];
* Initialize statistics set 1 and number of trials t /* check if indicator is used more than once */
if (dupe[p][q] != 0) {
****************************************************** dup = dupe[p][q];
*****/ for(i=0; i<5; i++) {
for(j=0; j<n[i]; j++) {
for(i=0; i<6; i++) { if (dupe[i][j] == dup) {
for(k=0; k<m; k++) { x[i][j][r] = x[i][j][r] + d[i][j][r];
sum1[i][k] = 0; }
} }
} updatestats1(min1, max1, mean1, sd1, sum1, sumsq1, F,
} Forig, t, tskip, min1trial, max1trial);
} else if (s==1) { /* change x by -delta */ /* print results to “SensOut2” */
x[p][q][r] = x[p][q][r] - d[p][q][r]; fprintf(fpout2, "\nT%d\t", t);
/* check if indicator is used more than once */ for(k=0; k<m; k++) {
if (dupe[p][q] != 0) { fprintf(fpout2, "%d ", k);
dup = dupe[p][q]; }
for(i=0; i<5; i++) { fprintf(fpout2, "rs2");
for(j=0; j<n[i]; j++) {
if (dupe[i][j] == dup) { for(i=0; i<6; i++) {
x[i][j][r] = x[i][j][r] - d[i][j][r]; fprintf(fpout2, "\n%3d ", i);
} for(k=0; k<m; k++) {
} fprintf(fpout2, "%6.3lf", F[i][k]);
} }
} fprintf(fpout2, "%6.3lf", rs2[i]);
} /* scale x values. */ }
/*calculate factor & EDRI values & ranks */ fprintf(fpout2, "\n");
option = 0; /* reset unscaled data to their original values, */
scale(option, inv[p][q], x[p][q], xprime[p][q], /* and rescale them for next sensitivity analysis */
minposs[p][q], maxposs[p][q]); for(i=0; i<5; i++) {
for(i=0; i<6; i++) { for(j=0; j<n[i]; j++) {
lincombo(n[i], w[i], xprime[i], F[i]); for(k=0; k<m; k++) {
rank(m, F[i], Franks[i]); x[i][j][k] = xorig[i][j][k];
if (i!=5) { }
for(k=0; k<m; k++) { }
xprime[5][i][k] = F[i][k]; }
} }
} /* calculate rs */ }
rs2[i] = calcrs(Franks[i], Forigranks[i]); }
} }
option = 0;
if (s==0) { for(i=0; i<5; i++) {
fprintf(fpout1, "\nTrial %4d. Value x[%d][%d][%d]" for(j=0; j<n[i]; j++) { "is changed by +%6.3lf.", t, p, q, r, d[p][q][r]);
} else if (s==1) { scale(option, inv[i][j], x[i][j], xprime[i][j], minposs[i][j],
fprintf(fpout1, "\nTrial %4d. Value x[%d][%d][%d]" maxposs[i][j]);"is changed by -%6.3lf.", t, p, q, r, d[p][q][r]);
} /******************************************************
} *****
* Perform sensitivity analysis to weights
/****************************************************** ******************************************************
***** *****/
* Print statistics set #1 results of sensitivity analysis:
* min, max, mean, s.d. over all trials for each factor-city fprintf(fpout1,"\n\nSensitivity to weights.\n");
* For sensitivity to data uncertainty fprintf(fpout2,"\n\nSensitivity to weights.\n");
for(s=0; s<2; s++) { /* loop through +dw and -dw */
****************************************************** or(p=0; p<6; p++) { /* loop through all factors and EDRI */
*****/ for(q=0; q<n[p]; q++) { /* loop thru all inds. for each factor */
t++;
fprintf(fpout3,"\n\nSensitivity to data uncertainty.\n");
printstats1(min1, max1, mean1, sd1, min1trial, max1trial, fpout3); if (s==0) { /* change w by +dw percent */
w[p][q] = w[p][q]*(1+dw);
/* renormalize weights so their sum equals one */
normwgts(n[p], w[p]);
/****************************************************** } else if (s==1) { /* change w by -dw percent */
***** w[p][q] = w[p][q]*(1-dw);
* Initialize statistics set 1 and number of trials t /* renormalize weights so their sum equals one */
****************************************************** normwgts(n[p], w[p]);
*****/ }
/* calculate factor and EDRI values, and ranks */
for(i=0; i<6; i++) { for(i=0; i<6; i++) {
for(k=0; k<m; k++) { lincombo(n[i], w[i], xprime[i], F[i]);
sum1[i][k] = 0; rank(m, F[i], Franks[i]);
sumsq1[i][k] = 0; if(i!=5) {
min1[i][k] = Forig[i][k]; for(k=0; k<m; k++) {
max1[i][k] = Forig[i][k]; xprime[5][i][k] = F[i][k];
min1trial[i][k] = 0; }
max1trial[i][k] = 0; } /* calculate rs */
} rs2[i] = calcrs(Franks[i], Forigranks[i]);
tskip[i] = 0; }
} /* calculate statistics & print results in SensOut2 */
t = 0; if (s==0) {
fprintf(fpout1, "\nTrial %4d. Value w[%d][%d] is"
" changed by +%6.3lf percent.", t, p, q, dw); * For sensitivity to weights
} else if (s==1) { ******************************************************
fprintf(fpout1, "\nTrial %4d. Value w[%d][%d] is" *****/
" changed by -%6.3lf percent.", t, p, q, dw);
} fprintf(fpout3,"\n\nSensitivity to weights\n");
updatestats1(min1, max1, mean1, sd1, sum1, sumsq1, F, printstats1(min1, max1, mean1, sd1, min1trial, max1trial, fpout3);
Forig, t, tskip, min1trial, max1trial);
fprintf(fpout2, "\nT%d\t", t); /******************************************************

for(k=0; k<m; k++) { *****
fprintf(fpout2, "%d ", k); * Initialize statistics set 1 and number of trials t
} ******************************************************
fprintf(fpout2, "rs2"); *****/
for(i=0; i<6; i++) { for(i=0; i<6; i++) {

fprintf(fpout2, "\n%3d ", i); for(k=0; k<m; k++) {
for(k=0; k<m; k++) { sum1[i][k] = 0;
fprintf(fpout2, "%6.3lf", F[i][k]); sumsq1[i][k] = 0;
} min1[i][k] = Forig[i][k];
fprintf(fpout2, "%6.3lf", rs2[i]); max1[i][k] = Forig[i][k];
} min1trial[i][k] = 0;
fprintf(fpout2, "\n"); max1trial[i][k] = 0;
/* reset weights to their original values */ }
for(i=0; i<6; i++) { tskip[i] = 0;
for(j=0; j<n[i]; j++) { }
w[i][j] = worig[i][j]; t = 0;
}
}
} /******************************************************
} *****
} * Perform sensitivity analysis to indicator selection
******************************************************
/****************************************************** *****/
*****
* Print statistics set #1 results of sensitivity analysis: fprintf(fpout1,"\n\nSensitivity to indicator selection.\n");
* min, max, mean, s.d. over all trials for each factor-city fprintf(fpout2,"\n\nSensitivity to indicator selection.\n");
fprintf(fpout2, "%6.3lf", rs2[i]);
for(p=0; p<6; p++) { /* loop through all factors and EDRI */ }
for(q=0; q<n[p]; q++) { /* loop through all indicators in each factor */ fprintf(fpout2, "\n");
t++; /* increment num. of trials */
/* remove indicator by setting its weight to zero */ for(i=0; i<6; i++) { /* reset weights to their original values */
w[p][q] = 0; for(j=0; j<n[i]; j++) {
normwgts(n[p], w[p]); /* renormalize weights */ w[i][j] = worig[i][j];
/* calculate factor & EDRI values, & ranks */ }
for(i=0; i<6; i++) { }
lincombo(n[i], w[i], xprime[i], F[i]); }
rank(m, F[i], Franks[i]); }
if(i!=5) {
for(k=0; k<m; k++) { /******************************************************
xprime[5][i][k] = F[i][k]; *****
} * Print statistics set #1 results of sensitivity analysis:
} * min, max, mean, s.d. over all trials for each factor-city
rs2[i] = calcrs(Franks[i], Forigranks[i]); /* calculate rs */ * For sensitivity to indicator selection
} ******************************************************
/* calculate statistics and print results in SensOut2 */ *****/
fprintf(fpout1, "\nTrial %4d. Value w[%d][%d] is set to "
"zero, i.e., indicator[%d][%d] is removed.", t, p, q, p, q); fprintf(fpout3,"\n\nSensitivity to indicator selection\n");
printstats1(min1, max1, mean1, sd1, min1trial, max1trial, fpout3);
updatestats1(min1, max1, mean1, sd1, sum1, sumsq1, F, Forig,
t, tskip, min1trial, max1trial); /******************************************************
*****
fprintf(fpout2, "\nT%d\t", t); * Initialize statistics set 1 and number of trials t
for(k=0; k<m; k++) { ******************************************************
fprintf(fpout2, "%d ", k); *****/
}
fprintf(fpout2, "rs2"); for(i=0; i<6; i++) {
for(k=0; k<m; k++) {
for(i=0; i<6; i++) { sum1[i][k] = 0;
fprintf(fpout2, "\n%3d ", i); sumsq1[i][k] = 0;
for(k=0; k<m; k++) { min1[i][k] = Forig[i][k];
fprintf(fpout2, "%6.3lf", F[i][k]); max1[i][k] = Forig[i][k];
} min1trial[i][k] = 0;
max1trial[i][k] = 0; } else if (s==1) {
} fprintf(fpout1, "\nTrial %d. When scaling w.r.t. the "
tskip[i] = 0; "min. and max. observed values.", t);
} } else if (s==2) {
t = 0; fprintf(fpout1, "\nTrial %d. When scaling w.r.t. the "
"min. and max. possible values.", t);
/****************************************************** } else if (s==3) {
***** fprintf(fpout1, "\nTrial %d. When scaling w.r.t. a "
* Perform sensitivity analysis to scaling function "base city.", t);
****************************************************** }
*****/ updatestats1(min1, max1, mean1, sd1, sum1, sumsq1, F, Forig, t,
tskip, min1trial, max1trial);
fprintf(fpout1,"\n\nSensitivity to scaling function.\n");
fprintf(fpout2,"\n\nSensitivity to scaling function.\n"); fprintf(fpout2, "\nT%d\t", t);
for(s=0; s<4; s++) { /* loop through four scaling options */ for(k=0; k<m; k++) {
option = s; fprintf(fpout2, "%d ", k);
t++; }
for(i=0; i<5; i++) { /* scale all indicators */ fprintf(fpout2, "rs2");
for(j=0; j<n[i]; j++) {
scale(option, inv[i][j], x[i][j], xprime[i][j], minposs[i][j], for(i=0; i<6; i++) {
maxposs[i][j]); fprintf(fpout2, "\n%3d ", i);
} for(k=0; k<m; k++) {
} fprintf(fpout2, "%6.3lf", F[i][k]);
for(i=0; i<6; i++) { /* calculate factor & EDRI values, & ranks */ }
lincombo(n[i], w[i], xprime[i], F[i]); fprintf(fpout2, "%6.3lf", rs2[i]);
rank(m, F[i], Franks[i]); }
if(i!=5) { fprintf(fpout2, "\n");
for(k=0; k<m; k++) { }
xprime[5][i][k] = F[i][k]; option = 0;
}
} /******************************************************
rs2[i] = calcrs(Franks[i], Forigranks[i]); /* calculate rs */ *****
} * Print statistics set #1 results of sensitivity analysis:
if (s==0) { /* calculate statistics and print results in SensOut2 */ * min, max, mean, s.d. over all trials for each factor-city
fprintf(fpout1, "\nTrial %d. When scaling w.r.t. the " * For sensitivity to scaling function
"mean minus two standard deviations.", t);
sumx = sumx + X[j];
****************************************************** sumsqx = sumsqx + (X[j]*X[j]);
*****/ }
fprintf(fpout3,"\n\nSensitivity to scaling function.\n"); meanx = sumx/((double) m);

printstats1(min1, max1, mean1, sd1, min1trial, max1trial, fpout3); sdx = sqrt( (((double) m)*sumsqx - sumx*sumx) /
} (((double) m)*((double) (m-1))) );
/********************************************************* for(j=0; j<m; j++) {

****** if (inv==0) {
* scale(): Function to scale a vector of values X[i] into a vector Xprime[j] = ( X[j] - (meanx - 2.0*sdx) )/sdx;
* Xprime[i]. } else if (inv==1) {
* Four scaling options are possible, and the scaling may be for Xprime[j] = (-X[j] + (meanx + 2.0*sdx) )/sdx;
* an indicator that }
* is directly or inversely related to earthquake disaster risk. } /* scaling with respect to the min. and max. observed */
********************************************************** } else if (option==1) {
****/ minobs = X[0];
maxobs = X[0];
void scale( for(j=0; j<m; j++) {
int option, /* indicates which scaling option to use */ if (X[j] < minobs) {
int inv, /* indicates if ind. is directly or inversely related to EDRI */ minobs = X[j];
double X[], /* input vector of unscaled values */ } else if (maxobs < X[j]) {
double Xprime[], /* output vector of scaled values */ maxobs = X[j];
double minposs, /* min. and max. possible values for option 2 */ }
double maxposs }
) for(j=0; j<m; j++) {
{ if (inv==0) {
int j; Xprime[j] = (X[j]-minobs)/(maxobs-minobs);
double sumx, sumsqx; } else if (inv==1) {
double meanx, sdx; Xprime[j] = (maxobs-X[j])/(maxobs-minobs);
double minobs, maxobs; }
/* scaling with respect to the mean minus two */ } /* scaling with respect to the min. and max. possible */
if (option==0) { } else if (option==2) {
sumx = 0; for(j=0; j<m; j++) {
sumsqx = 0; if (inv==0) {
for(j=0; j<m; j++) { Xprime[j] = (X[j]-minposs)/(maxposs-minposs);
} else if (inv==1) { }
Xprime[j] = (maxposs-X[j])/(maxposs-minposs);
} /*********************************************************
} ******
} else if (option==3) { * lincombo():
/* scaling
Function
with respect
to calc.tolinear
a basecombos.
city */ of a matrix of values.
for(j=0; j<m; j++) { * Input number of terms in linear combination (numterms),
if (inv==0) { * vector of weights (wgts[]), matrix of values (X[][]).
Xprime[j] = X[j]/X[basecity]; * Output vector of linear combination values (combo[]), one
} else if (inv==1) { * per city.
Xprime[j] = 2.0 - (X[j]/X[basecity]); **********************************************************
} *****/
}
} void lincombo(
} int numterms, /* number of terms to be combined */
double wgts[], /* weight values */
/********************************************************* double X[][m], /* matrix of values to be combined */
****** double combo[] /* vector of linear combination values */
* normwgts(): Function to normalize weights so that their sum equals )
* one. {
********************************************************** int j, p;
*****/
for(j=0; j<m; j++) {
void normwgts( combo[j] = 0;
int numwgts, for(p=0; p<numterms; p++) {
double wgts[] combo[j] = combo[j] + (wgts[p]*X[p][j]);
) }
{ }
int i; }
double sumwgts = 0;
/*********************************************************
for(i=0; i<numwgts; i++) { ******
sumwgts = sumwgts + wgts[i]; * rank(): Function to take a vector of n values as input arr[], order them from
} * largest to smallest, and output the ranks corresponding to each
for(i=0; i<numwgts; i++) { * value in the output vector ranks[]. From Numerical recipes in C.
wgts[i] = wgts[i]/sumwgts; **********************************************************
} *****/
SWAP(indx[k], indx[l+1]);
void rank( if (arr[indx[l]] > arr[indx[ir]]) {
unsigned long n, /* number of values in arr[] */ SWAP(indx[l], indx[ir])
double arr[], /* input vector of values */ }
/* output vector of ranks corresponding to values in arr[] */ if (arr[indx[l+1]] > arr[indx[ir]]) {
unsigned long ranks[] SWAP(indx[l+1], indx[ir])
) }
{ if (arr[indx[l]] > arr[indx[l+1]]) {
unsigned long indx[n]; SWAP(indx[l], indx[l+1])
unsigned long i, indxt, ir=n, itemp, j, k, l=1; }
int jstack=0, *istack; i = l+1;
double a; j = ir;
int p; indxt = indx[l+1];
a = arr[indxt];
for(p=m-1; p>=0; p--) { for(;;) {
arr[p+1] = arr[p]; do i++; while (arr[indx[i]] < a);
} do j--; while (arr[indx[j]] > a);
if (j< i) break;
istack = ivector(1, NSTACK); SWAP(indx[i], indx[j])
for(j=1; j<=n; j++) indx[j] = j; }
for(;;) { indx[l+1] = indx[j];
if (ir-1 < M) { indx[j] = indxt;
for(j=l+1; j<=ir; j++) { jstack += 2;
indxt = indx[j]; if (jstack > NSTACK) nrerror("NSTACK too small in rank()");
a = arr[indxt]; if (ir-i+1 >= j-1) {
for(i=j-1; i>=l; i--) { istack[jstack] = ir;
if (arr[indx[i]] <= a) break; istack[jstack-1] = i;
indx[i+1] = indx[i]; ir = j-1;
} } else {
indx[i+1] = indxt; istack[jstack] = j-1;
} istack[jstack-1] = l;
if (jstack == 0) break; l = i;
ir = istack[jstack--]; }
l = istack[jstack--]; }
} else { }
k = (l+ir) >> 1; free_ivector(istack, 1, NSTACK);
for(i=0; i<6; i++) {
for(j=1; j<=n; j++) ranks[indx[j]] = (n-j+1); skip = 0; /* skip update because the change did not affect the */
for(p=0; p<m; p++) { for(k=0; k<m; k++) { /* value of this factor for any city */
arr[p] = arr[p+1]; if (F[i][k] != Forig[i][k]) {
} skip++;
} }
}
/********************************************************* if (skip == 0) {
****** tskip[i]++;
* updatestats1(): Function to update statistics set 1—min., max., mean, } else if (skip != 0) {
and for(k=0; k<m; k++) {
* standard deviation of each factor-city value over all if(F[i][k] < min1[i][k]) {
* trials. One value per trial. min1[i][k] = F[i][k];
* Input current values of stats, F, num. trials (t) for updating min1trial[i][k] = t;
********************************************************** }
*****/ if(F[i][k] > max1[i][k]) {
max1[i][k] = F[i][k];
void updatestats1( max1trial[i][k] = t;
double min1[][m], }
double max1[][m], trel[i] = t-tskip[i];
double mean1[][m], sum1[i][k] = sum1[i][k] + F[i][k];
double sd1[][m], sumsq1[i][k] = sumsq1[i][k] + (F[i][k]*F[i][k]);
double sum1[][m], mean1[i][k] = sum1[i][k]/((double) trel[i]); /* update m
double sumsq1[][m], sd1[i][k] = (sumsq1[i][k] - ( ((double) trel[i])*
double F[][m+1], (mean1[i][k]*mean1[i][k]) )) / ((double) (trel[i]-1)); /* update sd1 */
double Forig[][m+1], }
int t, }
int tskip[6], }
int min1trial[][m], }
int max1trial[][m]
) /*********************************************************
{ ******
int i, k; * printstats1(): Function to print statistics set 1 to output file “SensOut2”
int trel[6]; /* num. trials relevant to each factor */ **********************************************************
int skip; /* skip>0 means don't skip update; skip=0 means do */ *****/
void printstats1( }
double min1[][m], fprintf(fpout3,"\n %d sd1 ", i);
double max1[][m], for(k=0; k<m; k++) {
double mean1[][m], fprintf(fpout3, "%10.2e", sd1[i][k]);
double sd1[][m], }
int min1trial[][m], fprintf(fpout3,"\n %d Min1trial", i);
int max1trial[][m], for(k=0; k<m; k++) {
FILE *fpout3 fprintf(fpout3, "%6d", min1trial[i][k]);
) }
{ fprintf(fpout3,"\n %d Max1trial", i);
int i, k; for(k=0; k<m; k++) {
fprintf(fpout3, "%6d", max1trial[i][k]);
fprintf(fpout3, "\nFactor/City "); }
for(k=0; k<m; k++) { }
fprintf(fpout3, "%d ", k); }
}
for(i=0; i<6; i++) {
fprintf(fpout3,"\n %d Min1 ", i);
for(k=0; k<m; k++) { /*********************************************************
fprintf(fpout3, "%6.3lf", min1[i][k]); ******
} * calcrs(): Function to calculate Spearman's rank correlation coefficient rs
fprintf(fpout3,"\n %d Mean1-sd1", i); * for two vectors of ranks U[] and V[].
for(k=0; k<m; k++) { **********************************************************
fprintf(fpout3, "%6.3lf", (mean1[i][k]-sd1[i][k])); *****/
}
fprintf(fpout3,"\n %d Mean1 ", i); double calcrs(
for(k=0; k<m; k++) { unsigned long U[],
fprintf(fpout3, "%6.3lf", mean1[i][k]); unsigned long V[]
} )
fprintf(fpout3,"\n %d Mean1+sd1", i); {
for(k=0; k<m; k++) { int j;
fprintf(fpout3, "%6.3lf", (mean1[i][k]+sd1[i][k])); unsigned long diffsq[m];
} unsigned long sumdiffsq = 0;
fprintf(fpout3,"\n %d Max1 ", i); double rs;
for(k=0; k<m; k++) {
fprintf(fpout3, "%6.3lf", max1[i][k]); for(j=0; j<m; j++) {
diffsq[j] = ( (U[j]-V[j])*(U[j]-V[j]) );
sumdiffsq = sumdiffsq + diffsq[j];
}
rs = 1.0 - ( (6.0*sumdiffsq)/(m*( (m*m)-1 )) );
return rs;
}
Appendix J
Minimum and Maximum Possible Values for Scaling
Sensitivity Analysis
Appendix J: Minimum and Maximum Possible Values 255

Appendix K
Values of ∆xij for Data Uncertainty Sensitivity Analysis

Delta values for data uncertainty sensitivity analysis
San
Factor Ind. Boston Istanbul Jakarta Lima Manila Mexico City Santiago St. Louis Tokyo
Francisco
xh1 150 200 100 900 500 100 350 200 150 140
xh2 1100 1000 250 1500 900 500 1200 600 1500 370
xh3 5.0% 7.5% 5.0% 10.0% 10.0% 7.5% 5.0% 5.0% 5.0% 5.0%
Hazard xh4 8.0% 10.0% 8.0% 10.0% 10.0% 10.0% 8.0% 8.0% 8.0% 8.0%
xh5 5.0% 20.0% 10.0% 5.0% 7.5% 5.0% 5.0% 5.0% 5.0% 5.0%
xh6 27 365 506 296 593 360 32 244 22 265
xh7 0 0 0 0 0 0 0 0 0 0
xe1 28,500 53,000 82,000 64,000 79,000 166,000 31,000 43,000 12,500 177,500
xe2 768 52 46 17 19 81 759 63 605 775
Exposure xe3 45,417 35,833 61,915 35,989 29,657 101,299 49,144 21,124 20,543 280,920
xe4 14 56 34 49 40 113 22 24 10 141
xe5 28,500 53,000 82,000 64,000 79,000 166,000 31,000 43,000 12,500 177,500
xe6 768 52 46 17 19 81 759 63 605 775
xv1 3.5 10.2 2.2 2.5 2.7 10.0 13.2 4.9 2.5 13.2
xv2 111 13 4 5 5 20 119 10 114 185
Vulner- xv3 2.5 1.0 1.0 1.0 1.0 1.0 1.0 1.0 2.1 1.0
ability xv4 27 365 506 296 593 360 32 244 22 265
xv5 5.3 1.0 1.0 1.0 1.0 1.2 1.9 1.6 4.5 1.8
xv6 0.10% 0.10% 0.30% 0.10% 0.10% 0.10% 0.10% 0.10% 0.10% 0.10%
External xc1 1.03E+11 9.83E+09 1.72E+10 2.46E+09 7.68E+09 7.47E+10 1.24E+11 3.38E+09 1.98E+10 1.26E+12
Context xc2 0.0002 0.0011 0.0100 0.0100 0.0100 0.0100 0.0003 0.0100 0.0001 0.0100
xc3 4.62E+09 3.23E+08 8.34E+09 3.65E+08 1.14E+09 6.86E+09 4.75E+09 8.24E+08 1.51E+09 9.57E+10
xr1 1 1 1 1 1 1 0 1 1 0
xr2 768 52 46 17 19 81 759 63 605 775
xr3 0.02% 0.02% 0.02% 0.02% 0.02% 0.02% 0.02% 0.02% 0.02% 0.02%
Emerg. xr4 1% 5% 5% 3% 5% 5% 1% 1% 1% 1%
Resp. & xr5 0.28 0.24 0.30 0.48 0.33 0.08 0.22 0.27 0.30 0.56
Recovery xr6 23.24 18 3.65 17.2 3.2 13.09 20.93 6.8 17.71 11.48
xr7 17.70 6.21 1.88 1.20 6.97 1.17 1.56 9.76 18.99 8.35
xr8 27 365 506 296 593 360 32 244 22 265
xr9 0 0 0 0 0 0 0 0 0 0
Note: See Figure 4.1 for indicator definitions corresponding to the variables xi.
References
Acuna, Jack Lopez. 1992. Reedition history of seismic design code in Peru. Bulletin of the
International Institute of Seismology and Earthquake Engineering 26: 395-403.
Albala-Bertrand, J. M. 1993. The political economy of large natural disasters with special
reference to developing countries. Oxford: Clarendon Press.
Algermissen, S. T., Paul C. Thenhaus, and Kenneth Campbell. 1996. Global earthquake hazard
and risk assessment. In Proceedings of the eleventh world conference on
earthquake engineering in Acapulco, Mexico, June 23-28, 1996, by the
International Association of Earthquake Engineering, paper no. 1577. [CD-ROM]. n.p.:
Elsevier.
Alva-Hurtado, J. 1994. Seismic safety of the Lima Metropolitan Area. In Issues in urban
earthquake risk, ed. Brian Tucker, Mustafa Erdik, and Christina Hwang. Series E:
Applied Sciences, vol. 271. Dordrecht: Kluwer Academic Publishers.
Anderson, Mary 1992. Metropolitan areas and disaster vulnerability: A consideration for
developing countries. In Managing natural disasters and the environment, ed. Alcira
Kreimer and Mohan Munasinghe, World Bank Discussion Papers, no. 168.
Washington, D.C.: World Bank.
Applied Technology Council. 1985. Earthquake damage evaluation data for California.
ATC-13. Redwood City, California: ATC.
Applied Technology Council. 1988. Rapid visual screening of buildings for potential
seismic hazards: A handbook. ATC-21. Redwood City, California: ATC, developed
under FEMA contract.
Australian Bureau of Statistics. 1996. Year book Australia 1996. No. 78. Canberra: Australian
Bureau of Statistics.
Bauman, Catherine, Planning Department, City of San Francisco. 1997. Personal
communication, February, Stanford, California.
Bayulke, Nejat. 1992. Reedition history of seismic design code of Turkey. Bulletin of the
Berg, Glen V. 1982. Seismic design codes and procedures. Engineering Monographs on
Earthquake Criteria, Structural Design, and Strong Motion Records, ed. Mihran S.
Agbabian, vol. 6. Berkeley, California: Earthquake Engineering Research Institute.
Blaikie, Piers, Terry Cannon, Ian Davis, and Ben Wisner. 1994. At risk: Natural hazards,
people’s vulnerability, and disasters. London: Routledge.
Bouabid, Jawhar, Risk Management Solutions, Inc. 1995-97. Interviews by author. Menlo
Park, California.
Brady, R. J., and J. B. Perkins. 1991. Macroeconomic effects of the Loma Prieta
earthquake. Oakland, California: Association of Bay Area Governments.
Briguglio, Lino. 1995. Small island developing states and their economic vulnerabilities. In
Global conference on the sustainable development of small island developing
states in Bridgetown, Barbados, April 25-May 6, 1994. United Nations.
Burton, Ian, Robin W. Kates, and Gilbert F. White. 1993. The environment as hazard. 2nd
ed. revised. New York: Guilford.
California. Emergency Planning and Response Committee. 1990. Earthquake emergency

preparedness and response: A report to the California Seismic Safety
Commission. Sacramento, California: The Commission, no. SSC 90-08.
Cannon, Terry. 1994. Vulnerability analysis and the explanation of natural disasters. In
Disasters, development and environment, ed. Ann Varley. London: Wiley.
Caves, Richard E., and Ronald W. Jones. 1985. World trade and payments: An
introduction. 4th ed. Boston: Little, Brown.
Cheng, Leonard K. 1992. Presentation. In The economic consequences of a catastrophic

earthquake: Proceedings of a forum, August 1-2, 1990, by the National Research
Council. Committee on Earthquake Engineering. Division of Natural Hazard Mitigation.
Commission on Engineering and Technical Systems, 145-49. Washington, D.C.:
National Academy Press.
References 255
Chibber, Sumeet, George Apostolakis, and Okrent, David. 1992. A taxonomy of issues
related to the use of expert judgments in probabilistic safety studies. Reliability
Engineering and System Safety 38: 27-45.
Chile. Direccion de Estadistica y Censos. 1963. Entidades de poblacion de Chile 1960.

Santiago: Direccion de Estadistica y Censos.
Chile. Direccion General de Estadistica. Comision Central del Censo. 1931-35. Resultados del
X censo de la poblacion efectuado el 27 de Noviembre de 1930 y estadisticas
comparativas con censos anteriores. Santiago, Chile: Imp. Universo.
Chile. Instituto Nacional de Estadisticas. 1958-1990. Demografia, 1958-1990 editions.

Santiago, Chile: Instituto Nacional de Estadisticas.
Chile. Instituto Nacional de Estadisticas. 1989. Demodato '89. Santiago, Chile: El Instituto.
Chile. Instituto Nacional de Estadisticas. 1992a. Censo de poblacion y vivienda: Chile 1992:
Resultados generales. Santiago, Chile: El Instituto.
Chile. Instituto Nacional de Estadisticas. 1992b. XVI Censo de poblacion y V de vivienda, 22

de abril de 1992. Resultados preliminares, regiones, provincias y comunas.
Santiago, Chile: El Instituto.
Chile. Oficina de Planificacion Nacional. 1993. Compendio de estadisticas regionales.

Santiago, Chile: OPN.
Coburn, A., R. Spence, and A. Pomonis. 1992. Factors determining human casualty levels in
earthquakes: Mortality prediction in building collapse. In Proceedings of the tenth
world conference on earthquake engineering in Madrid, Spain,
July 19-24, 1992, 5989-94. Rotterdam, Netherlands: Balkema.
Cochrane, Harold C. 1992. Presentation. In The economic consequences of a catastrophic

Cochrane, Harold C. 1995. Indirect economic losses. Chapter 16 in Development of a

standardized loss estimation methodology, vol. II. Menlo Park, California: Risk
Management Solutions, Inc.
Cochrane, Harold C. 1996. Is Northridge a model for future events?: What are the economic
consequences with and without insurance and federal assistance? In The Northridge
256 References
earthquake. Analyzing economic impacts and recovery from urban earthquakes:
Issues for policy makers. Proceedings of a conference in Pasadena, California,
October 10-11, 1996, by the Earthquake Engineering Research Institute and Federal
Emergency Management Agency.
Comerio, Mary. 1996. Housing losses. In The Northridge earthquake. Analyzing economic
impacts and recovery from urban earthquakes: Issues for policy makers.
Proceedings of a conference in Pasadena, California, October 10-11, 1996, by the
Earthquake Engineering Research Institute and Federal Emergency Management
Agency.
Cornell, C.A. 1968. Engineering seismic risk analysis. Bulletin of the Seismological Society of
America 58, no. 5: 1583-1606.
Cret, L., and Katayama, T. 1992. On-line damage estimation for lifeline systems during
earthquakes. In Proceedings of the tenth world conference on earthquake
engineering in Madrid, Spain, July 19-4, 1992, 5501-10. Rotterdam, Netherlands:
Balkema.
Cruz, Ernesto. 1989. Present state of seismic design and analysis regulations in Chile.
Earthquake Spectra 5, no. 4: 661-74.
Downes, A.S. 1988. On the statistical measurement of smallness: A principal component

measure of country size. In Social and Economic Studies 37, no. 3.
Dunteman, G.H. 1989. Principal components analysis. Newbury Park: Sage Publications.
Earthquake Engineering Research Institute. 1996. Northridge Earthquake Reconnaissance

Report, Supplement C, vol. 2. Earthquake Spectra.
Economist Intelligence Unit. 1995. Country profile: India Nepal 1994-95. London: EIU.
Economist Intelligence Unit. 1996. Country profile: Indonesia 1995-96. London: EIU.
Eisner, Richard K., regional administrator, OES Coastal Region, Governor's Office of
Emergency Services. 1996. Interview by author, March 20, Oakland, California.
Erdik, Mustafa. 1994. Developing a comprehensive earthquake disaster masterplan for

Istanbul. In Issues in urban earthquake risk, ed. Brian Tucker, Mustafa Erdik, and
Christina Hwang. Series E: Applied Sciences, vol. 271. Dordrecht: Kluwer Academic
Publishers.
References 257
Felderer, Bernhard, and Stefan Homburg. 1992. Macroeconomics and new
macroeconomics. 2nd ed. Berlin: Springer-Verlag.
Forrest, Thomas R. 1978. Group emergence in disasters. In Disasters: Theory and research,
ed. E. L. Quarantelli, Sage Studies in International Sociology 13. Beverly Hills,
California: Sage Publications.
Fraser, I. A. N. 1983. A new draft code for seismic design of buildings in Indonesia. New
Zealand Society for Earthquake Engineering Bulletin 16, no. 3: 247-54.
Fritz, C.E. 1961. Disaster. In Contemporary social problems, ed. by Merton and Nisbet.
New York: Harcourt. Quoted in United Nations. Office of United Nations Disaster
Relief Co-ordinator (UNDRO). Preparedness aspects. Disaster Prevention and
Mitigation: A Compendium of Current Knowledge, vol. 11. New York: United
Nations, 1984.
Gere, James M., and Haresh C. Shah. 1984. Terra non firma: Understanding and
preparing for earthquakes. Stanford, California: Stanford Alumni Association.
Green, Norman B. 1987. Earthquake resistant building design and construction. 3rd ed.
New York: Elsevier.
Gulkan, Polat, and Oktay Ergunay. 1984. Mitigation of earthquake induced losses in Turkey:
The legislative aspect. In International symposium on earthquake relief in less
industrialized areas, Zurich, 28-30 March 1984. Rotterdam: A. A. Balkema.
Heikkala, Susan G., ed. 1982. Urban scale vulnerability: Proceedings of the U.S.-Italy
colloquium on urban design and earthquake hazard mitigation, in Rome, Italy, 12-
16 October, 1981, by the University of Washington and University of Rome.
Hellwig, Zdzislaw. 1972a. On the problem of weighting in international comparisons. In

Towards a system of human resources indicators for less developed countries, ed.
Zygmunt. Gostkowski, 91-111. Ossolineum, Poland: Polish Academy of Sciences
Press.
Hellwig, Zdzislaw. 1972b. Procedure of evaluating high-level manpower data and typology of
countries by means of the taxonomic method. In Towards a system of human
resources indicators for less developed countries, ed. Zygmunt Gostkowski, 115-
34. Ossolineum, Poland: Polish Academy of Sciences Press.
Hewitt, Kenneth, ed. 1983. Interpretations of calamity from the viewpoint of human
ecology. Winchester, Massachusetts: Allen & Unwin.
258 References
Hirosawa, M., S. Sugano, H. Takizawa, T. Kubo, and M. Yoshimura. 1992. List of summaries
on earthquake design criteria of earthquake-prone countries in the world. In
Earthquake disaster reduction handbook 1992. Japan. Ministry of Construction.
Building Research Institute, 103-79. Japan: Building Research Institute.
IMD International and World Economic Forum. 1992. The world competitiveness report
1992. 12th ed. Cologny/Geneva, Switzerland: IMD International and World Economic
Forum.
Indonesia. Biro Pusat Statistik. Statistical pocket book of Indonesia, 1941, 1962, 1970-71,
1982 editions. Jakarta, Indonesia: Biro Pusat Statistik.
Indonesia. Biro Pusat Statistik. 1992. Population of Indonesia: Results of the 1990
population census. Jakarta: Biro Pusat Statistik.
Indonesia. National Development Information Office. 1993. Indonesia source book. Jakarta,
Indonesia: NDIO.
Institut d'Estudis Metropolitans de Barcelona. 1988. Cities: Statistical, administrative, and

graphical information on the major urban areas of the world, sponsored by United
Nations Population Fund, Ministerio de Obras Publicas y Urbanismo, Espana,
Corporacio Metropolitana de Barcelona, with the collaboration of Ministerio de
Asuntos Exteriores, Espana. Barcelona, Spain: Institut d'Estudis Metropolitans de
Barcelona.
Institut National de la Statistique et des Etudes Economiques. 1990. Statistiques et

indicateurs des regiones francaises. Paris: INSEE.
International Federation of Red Cross and Red Crescent Societies, and the Centre for
Research on the Epidemiology of Disasters. 1994. World Disasters Report. Issued by
the. Dordrecht: Martinus Nijhoff.
International Monetary Fund. Balance of Payments Division. 1995. Balance of payments

statistics yearbook. Washington, D.C.: The Fund.
International Monetary Fund. 1996. International financial statistics yearbook 1996.

Washington, D.C.: IMF.
Isard, Walter. 1956. Some empirical regularities of the space-economy. Chapter 3 of Location
and space economy. New York: Technology Press of Massachusetts Institute of
Technology and John Wiley and Sons, Inc. Quoted in Barclay G. Jones. Development
of a methodology for making indirect estimates of the built physical environment. In
References 259
Research accomplishments, 1986-1994, 107-20. Buffalo, NY: National Center for
Earthquake Engineering Research, 1994.
Jalan, B. 1982. Classification of economies by size. In Problems and policies in small

countries. London: Groom Helm.
Japan. Statistics Bureau. Bureau of Management and Coordination Agency. 1995. Japan
statistical yearbook. Tokyo: Sorifu, Tokeikyoku.
Johnson, Bryan T., and Thomas P. Sheehy. 1996. 1996 Index of economic freedom.
Washington, D.C.: The Heritage Foundation.
Johnson, David B. (under the general editorship of Alan S. Binder). 1993. Finding & using
economic information: A guide to sources and interpretation. Mountain View,
California: Mayfield Pub. Co.
Jones, Barclay G. 1992. Population growth, urbanization and disaster risk and vulnerability in
metropolitan areas: A conceptual framework. In Environmental management and
urban vulnerability, ed. Alcira Kreimer and Mohan Munasinghe, World Bank
discussion papers, no. 168, 51-76. Washington, D.C.: World Bank.
Jones, Barclay G. 1994. Development of a methodology for making indirect estimates of the
built physical environment. In Research accomplishments, 1986-1994, 107-120.
Buffalo, NY: National Center for Earthquake Engineering Research.
Jones, Barclay, professor, Cornell University. 1996. Phone interview by author, December.
Jones, Barclay G., and Stephanie E. Chang. 1993. Indirect methods for estimating the built
physical environment for risk and damage assessment and relief, recovery, and
reconstruction planning. In Proceedings of the 1993 national earthquake
conference—Earthquake hazard reduction in the Central and Eastern United
States: A time for examination and action, in Memphis, Tennessee, May 3-5,
1993, by the Central United States Earthquake Consortium, 593-602. Memphis:
CUSEC.
Jones, Barclay G., and Stephanie E. Chang. 1994. A comparison of indirect and direct
estimates of the built physical environment in the Memphis region. In Proceedings of
the fifth United States national conference on earthquake engineering in Chicago,
Illinois, July 10-14, 1994, by the Earthquake Engineering Research Institute, vol. III,
975-984. Berkeley, California: EERI.
Jones, Barclay G., and Blane D. Lewis. 1990. Estimating size distributions of building areas for
natural hazards assessment. Earthquake Spectra 6, no. 3 (August): 497-505.
260 References
Jones, Barclay G., Donald M. Manson, Charles M. Hotchkiss, Michael J. Savonis, and
Kimberley A. Johnson. 1987. Estimating building stocks and their characteristics.
Ithaca, N.Y.: Cornell University, Program in Urban and Regional Studies.
King, Stephanie A., and Anne S. Kiremidjian. 1994. Regional seismic hazard and risk
analysis through Geographic Information Systems. The John A. Blume Earthquake
Engineering Center, report no. 111. Stanford, California: Blume Center.
Kling, Robert W. 1992. Presentation. In The economic consequences of a catastrophic

Commission on Engineering and Technical Systems, 82-97. Washington, D.C.: National
Academy Press.
Kreimer, Alcira, and Mohan Munasinghe, eds. 1991. Managing natural disasters and the
environment. Washington, DC: World Bank.
Kreimer, Alcira, and Mohan Munasinghe, eds. 1992. Environmental management and
urban vulnerability. World Bank discussion papers, no. 168. Washington, D.C.:
World Bank.
Kreps, G. 1978. The organization of disaster response: Some Fundamental theoretical issues. In
Disasters: Theory and research, ed. E. L. Quarantelli, Sage Studies in International
Sociology 13. Beverly Hills, California: Sage Publications.
Liao, S. S., D. Veneziano, and R. V. Whitman. 1988. Regression models for evaluating
liquefaction probability. Journal of Geotechnical Engineering 114, no. 4 (April).
Quoted in Risk Management Solutions, Inc. 1997. Hazus97: Earthquake loss
estimation methodology. Technical Manual. Prepared for National Institute of Building
Sciences. Menlo Park, California.
Marsh, James W. 1993. Earthquakes: Steel structures performance and design code
developments. Engineering Journal, American Institute of Steel Construction (second
quarter): 56-65.
Martin, H. W. 1993. Recent changes to seismic codes and standards: Are they coordinated or
random events? In Proceedings of the 1993 national earthquake conference—
Earthquake hazard reduction in the Central and Eastern United States: A time
for examination and action, in Memphis, Tennessee, May 3-5, 1993, by the Central
United States Earthquake Consortium, 367-376. Memphis: CUSEC.
References 261
McClave, James T., and Frank H. Dietrich, II. 1988. Statistics. 4th ed. San Francisco: Dellen
Pub.
Mexico. Instituto Nacional de Estadistica, Geografia e Informatica. 1990-92. Mexico XI censo

general de poblacion y vivienda 1990. Resultados definitivos 1990-1992.
Aguascalientes, Mexico: INEGI.
Mexico. Instituto Nacional de Estadistica, Geografia e Informatica. 1994. Anuario estadistico

del Distrito Federal. Mexico, D.F.: INEGI.
Mexico. Instituto Nacional de Estadistica, Geografia e Informatica. 1995. Anuario estadistico

del Estado de Mexico. Mexico, D.F.: INEGI.
Mexico. Instituto Nacional de Estadistica, Geografia e Informatica. 1990-92. Mexico XI censo

general de poblacion y vivienda 1990. Resultados definitivos 1990-1992.
Aguascalientes, Mexico: INEGI.
Morris, M.D. 1979. Measuring the condition of the world's poor: The Physical Quality of
Life Index. Pergamon Policy Studies, 42. New York: Pergamon Press.
Morse, Dan. 1989. Seismic codes: Preparing for the unknown. Civil Engineering
(November): 70-73.
Munroe, Tapan. 1992. Presentation. In The economic consequences of a catastrophic

Nakata, Shinsuke. 1991. Outline of damage on the earthquake in Philippine of July 16, 1990.
Bulletin of the International Institute of Seismology and Earthquake Engineering
25: 73-93.
Nasser, Aladdin. 1996. Interview by author, Stanford, California.
National Climatic Data Center. Climate Visualization (CLIMVIS) weather database; availablr
from http://www.ncdc.noaa.gov/ncdc.html; Internet; accessed 1996-97.
National Geophysical Data Center. Solid Earth Geophysics Division. Tsunami database.
Boulder, Colorado http://www.ngdc.noaa.gov/seg/hazard/hazards.html; Internet;
accessed 1996-97.
262 References
National Research Council. Committee on Earthquake Engineering. Division of Natural Hazard
Mitigation. Commission on Engineering and Technical Systems. 1992. The economic
consequences of a catastrophic earthquake: Proceedings of a forum, August 1-2,
1990. Washington, D.C.: National Academy Press.
Nishenko, Stuart, seismologist, Mitigation Directorate, Federal Emergency Management

Agency. 1997. Interview by author, April, Stanford, California.
Olshansky, Robert B. 1993. Selling seismic building codes in the central United States. In
Proceedings of the 1993 national earthquake conference—Earthquake hazard
reduction in the Central and Eastern United States: A time for examination and
action, in Memphis, Tennessee, May 3-5, 1993, by the Central United States
Earthquake Consortium, 649-658. Memphis: CUSEC.
Otani, Shunsuke. 1995. A brief history of Japanese seismic design requirements. Concrete
International 17, no. 12: 46-53.
Paz, Mario. 1994. International handbook of earthquake engineering: Codes, programs,

and examples. New York: Chapman & Hall.
People’s Republic of China. State Statistical Bureau of China. 1995. China statistical
yearbook. Beijing.
Peru. Instituto Nacional de Estadistica. Direccion de Estudios Socio-Economicos. Direccion

General de Indicadores Economicos y Sociales. 1993. Compendio estadistico. Lima
y Callao 1992-93. Lima, Peru: El Instituto.
Peru. Instituto Nacional de Estadistica e Informatica. Direccion Nacional de Censos y

Encuestas. 1994. Peru censos nacionales 1993, IX de poblacion, IV de vivienda.
Resultados definitivos, tomo III. Lima, Peru: INEI.
Philippines. National Economic and Development Authority. 1995. Philippine statistical

yearbook. Manila: NEDA.
Philippines. National Statistics Office. 1992. 1990 Census of population and housing:
Housing statistics: Philippines. National Capital Region, report no. 4. Manila: NSO.
Quarantelli, E.L., ed. 1978. Disasters: Theory and Research. Sage Studies in International
Quarantelli, E.L. 1987. What should we study? Questions and suggestions for researchers
about the concept of disasters. International Journal of Mass Emergencies and
Disasters 5, 1 (March): 7-32.
References 263
Ram, R. 1982. Composite indices of physical quality of life, basic needs fulfillment, and income.
Journal of Development Economics 11: 227-247.
Rantucci, Giovanni. 1994. Geological disasters in the Philippines: the July 1990
earthquake and the June 1991 eruption of Mount Pinatubo: Description, effects
and lessons learned. Quaderni di vita italiana, no. 5. Rome: Presidenza del Consiglio
Dei Ministri, Dipartimento Per L'Informazione e L'Editoria.
Renshaw, Vernon, Edward A. Trott, Jr., and Howard L. Friedenberg. 1988. Gross state
product by industry, 1963-86. Survey of Current Business, by the U.S. Department of
Commerce. Bureau of Economic Analysis. 68, no. 5 (May): 30-46.
Risk Management Solutions, Inc. 1995. What if the 1923 earthquake strikes again?: A five-
prefecture Tokyo region scenario. Topical Issues Series. Menlo Park, California:
RMS.
Risk Management Solutions, Inc. 1996-97. Damage curves data. Unpublished.
Risk Management Solutions, Inc. 1997. Hazus97: Earthquake loss estimation methodology.
Technical Manual. Prepared for National Institute of Building Sciences. Menlo Park,
California.
Risk Management Solution, Inc. and California Universities for Research in Earthquake
Engineering. 1994. Assessment of the state-of-the-art earthquake loss estimation
methodologies. Prepared for National Institute of Building Sciences. Earthquake
Hazards Reduction Series 70. Washington, DC: NIBS.
Rojahn, Christopher. 1994. Estimation of earthquake damage to buildings and other structures
in large urban areas. In Issues in urban earthquake risk, ed. Brian Tucker, Mustafa
Erdik, and Christina Hwang. Series E: Applied Sciences, vol. 271. Dordrecht: Kluwer
Academic Publishers.
Romer, Mark, Risk Management Solutions, Inc. 1995-97. Interviews by author, Menlo Park,
California.
Romo, M. P., A. Jaime, and D. Resendiz. 1988. General soil conditions and clay properties in
the Valley of Mexico. Earthquake Spectra 4, no. 4 (November): 731-52.
Rosenblueth, Emilio. 1987. Emergency regulations and the new building code. In The Mexico
earthquakes—1985: Factors involved and lessons learned, eds. M. Cassaro and E.
Romero, 440-54. New York: American Society of Civil Engineers.
264 References
Rossi, R. and K. Gilmartin. 1980. The handbook of social indicators: Sources,
characteristics, and analysis. New York: Garland STPM.
Ruffner, J. and F. Bair, ed. 1987. Weather of U.S. cities. 3rd ed. Detroit: Gale Research
Company.
Sarrazin, Mauricio J. 1992. History of Chilean seismic regulations. Bulletin of the

Scawthorn, Charles. 1986. Fire losses from earthquakes: Los Angeles region. In Future
earthquake losses in the Los Angeles, California region: A one-day seminar, in
Santa Monica, California, December 10, 1986, by the Committee on Seismic Risk of
the Earthquake Engineering Research Institute. El Cerrito, California: EERI.
Seed, H. Bolton, and Idriss, I. M. 1983. Ground motions and soil liquefaction during
earthquakes. Engineering Monographs on Earthquake Criteria, Structural Design, and
Strong Motion Records, ed. Mihran S. Agbabian, vol. 5. Oakland, California:
Earthquake Engineering Research Institute.
Shah, H., Boyle, R., and Dong, W. 1991. Geographic Information Systems and Artificial
Intelligence: An application for seismic zonation. In Proceedings of the fourth
international conference on seismic zonation, in Stanford, California,
August 25-29, 1991, by the Earthquake Engineering Research Institute, 487-517.
Oakland, California: EERI.
Sharma, Sunil, and Mario Candia-Gallegos. 1992. Seismic hazard analysis of Peru.
Engineering Geology 32 (1992): 73-79.
Smith, John C., Massachusetts Emergency Management Agency. 1996. Interview by author,
April, Framingham, Massachusetts.
Structural Engineers Association of California. Seismology Committee. 1975. Recommended

lateral force requirements and commentary. 4th ed. San Francisco: SEAOC.
Structural Engineers Association of California. Seismology Committee. 1990. Recommended

lateral force requirements and commentary. 5th ed. San Francisco: SEAOC.
Su, S. S. 1978. Earthquake risk analysis for Metro Manila. In Proceedings of the U.S.-
Southeast Asia symposium on engineering for natural hazards protection, in
Manila, Philippines, 1977, edited by A. Ang. Manila, 33-43. Urbana, Illinois:
Department of Civil Engineering, University of Illinois and Urbana-Champaign.
References 265
Takizawa, Haruo. 1993. An overview of earthquake-resistant design standards in Latin-
american countries. Bulletin of the International Institute of Seismology and
Earthquake Engineering 27: 301-29.
Tierney, Kathleen J. 1985. Report on the Coalinga earthquake of May 2, 1983.

Sacramento, California: California Seismic Safety Commission, no. SSC 85-01.
Tokyo Metropolitan Government. Statistical Division. Bureau of General Affairs. Statistics of

world large cities 1991. Tokyo: TMG.
Trott, Jr., Edward A. Ann E. Dunbar, and Howard L. Friedenberg. 1991. Gross state product
by industry, 1977-89. Survey of Current Business, by the U.S. Department of
Commerce. Bureau of Economic Analysis. 71, no. 12 (December): 43-59.
Turkey. Devlet Istatistik Enstitusu. 1992. 1990 census of population: Administrative

division. Ankara: T. C. Basbakanlik, Devlet Istatistik Enstitusu.
Turkey. Devlet Istatistik Enstitusu. 1994. Statistical yearbook of Turkey 1993. Ankara:
Devlet Istatistik Enstitusu.
Tversky, Amos, and Kahneman, Daniel. 1974. Judgment under uncertainty: Heuristics and
biases. Science 185, no. 4157 (September 27): 1124-31.
United Kingdom. Humberside County Council. County emergency plan. Quoted in United
Nations. Office of United Nations Disaster Relief Co-ordinator (UNDRO).
Preparedness Aspects. Disaster Prevention and Mitigation: A Compendium of Current
Knowledge, vol. 11. New York: United Nations, 1984.
United Nations. 1993. Trends in international distribution of gross world product. National
Accounts Statistics Series X, no. 18, special issue. New York: United Nations.
United Nations Centre for Human Settlements. 1995. Compendium of human settlements
statistics 1995. 5th issue. New York: United Nations.
United Nations Centre for Regional Development (UNCRD). 1995. Comprehensive study of
the Great Hanshin earthquake. Research Report Series, no. 12. Nagoya, Japan:
UNCRD.
United Nations. Department for Economic and Social Information and Policy Analysis.
Population Division. 1995a. The challenge of urbanization: The world’s large cities.
New York: UN.
266 References
United Nations. Department for Economic and Social Information and Policy Analysis.
Population Division. 1995b. World urbanization prospects: The 1994 revision. New
York: UN.
United Nations. Department of Humanitarian Affairs. 1994. DHA News 9-10 (May/August).
United Nations Development Program. 1990-95. Human development report, 1990-1995

editions. New York: Oxford University Press.
United Nations. Office of United Nations Disaster Relief Co-ordinator (UNDRO). 1979.
Economic aspects. Disaster Prevention and Mitigation: A Compendium of Current
Knowledge, vol. 7. New York: United Nations.
Preparedness aspects. Disaster Prevention and Mitigation: A Compendium of Current
Knowledge, vol. 11. New York: United Nations.
Social and sociological aspects. Disaster Prevention and Mitigation: A Compendium
of Current Knowledge, vol. 12. New York: United Nations.
United Nations. Secretariat for the International Decade for Natural Disaster Reduction. May
1996. "Disaster Reduction for Sustainable Development” project proposal.
Unpublished.
U.S. Bureau of the Census. 1930-1994. County and city data book¸ 1930, 1940, 1943,
1950, 1960, 1970, 1975, 1980, 1986, 1990, and 1994 editions. Washington, D.C.:
Government Printing Office.
U.S. Bureau of the Census. 1993a. 1990 Census of housing. Detailed housing
characteristics. Metropolitan areas. Washington, D.C.: Government Printing Office.
U.S. Bureau of the Census. 1993b. 1990 Census of population and housing. Guide; Part B.
Glossary. Washington, D.C.: Government Printing Office.
U.S. Bureau of the Census. 1993c. 1990 Census of population and housing. Supplementary
reports. Metropolitan areas. Washington, D.C.: Government Printing Office.
U.S. Bureau of the Census. 1995. Statistical abstract of the United States 1995.
Washington, D.C.: Government Printing Office.
U.S. Geological Survey. 1994. The next big earthquake in the Bay Area may come sooner
than you think. Are you prepared? Newspaper insert.
References 267
U.S. Office of Disaster Assistance. 1990. Disaster history: Significant data on major
disasters worldwide 1900-1989. Washington, D.C.: OFDA.
Varley, A., ed. 1994. Disasters, development and environment. London: Wiley.
Vega, Rosa Alicia Torres. 1992. Reedition history of seismic design code in Peru. Bulletin of
the International Institute of Seismology and Earthquake Engineering 26: 405-11.
Wang, Fu, Christian P. Mortgat, Zifa Wang, and Janet L. Boley. 1996. Seismic hazard analysis
of Mexico. In Proceedings of the eleventh world conference on earthquake
engineering in Acapulco, Mexico, June 23-28, 1996, by the International Association
of Earthquake Engineering, paper no. 1972. [CD-ROM]. n.p.: Elsevier.
Watabe, Makoto. 1983. Summary of current seismic code requirements in the world. In
Seismic risk and its use in code formulation: Proceedings of the review meeting of
the US-Japan cooperative research on generalized seismic risk analysis and
development of a model seismic format, in Honolulu, Hawaii, March 1983, edited
by H. C. Shah and H. Shibata, no. 16. Honolulu: sponsored by the National Science
Foundation and Japan Society for the Promotion of Science.
Webb, R., and G. Fernandez Baca, ed. 1993. Peru en numeros 1993. Lima, Peru: Cuanto
S.A.
Wenger, Dennis. 1978. Community response to disaster: Functional and structural alterations.
In Disasters: Theory and research, ed. E. L. Quarantelli, Sage Studies in International
West, Carol Taylor. 1996. Indirect economic impacts of natural disasters: Policy implications of
recent research and experience. In The Northridge earthquake. Analyzing economic
impacts and recovery from urban earthquakes: Issues for policy makers.
Proceedings of a conference in Pasadena, California,
October 10-11, 1996, by the Earthquake Engineering Research Institute and Federal
Emergency Management Agency.
Windeler, Don, Risk Management Solutions, Inc. 1995-97. Interviews by author, Menlo
Park, California.
World Bank. 1980. World tables 1980. 2nd ed. Baltimore: Published for the World Bank by
the Johns Hopkins University Press.
Youd, T. L., and Perkins, D. M. 1978. Mapping of liquefaction induced ground failure
potential. Journal of the Geotechnical Engineering Division, American Society of
268 References
Civil Engineers 104, no. 4: 433-46. Quoted in Risk Management Solutions, Inc. 1997.
Hazus97: Earthquake loss estimation methodology. Technical Manual. Prepared for
National Institute of Building Sciences. Menlo Park, California.
Zegarra, Luis Antonio. 1987. Peruvian aseismic code. Bulletin of the International Institute
of Seismology and Earthquake Engineering 22: 125-37.
References 269

A U E D R I: N Rban Arthquake Isaster ISK Ndex

Uploaded by

Copyright:

Available Formats

A U E D R I: N Rban Arthquake Isaster ISK Ndex

Uploaded by

Document Information

Original Title

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

A U E D R I: N Rban Arthquake Isaster ISK Ndex

Uploaded by

Copyright:

Available Formats

Department of Civil and Environmental Engineering

AN URBAN EARTHQUAKE DISASTER RISK INDEX

Report No. 121

The John A. Blume Earthquake Engineering Center

©1997 The John A. Blume Earthquake Engineering Center

Table of Contents .................................................................................................................vii

1.1. Anticipated benefits of the EDRI .............................................................................2

Chapter 2. Key Features of the Earthquake Disaster Risk Index.............................................9

2.1. Definition of earthquake disaster risk .......................................................................9

Chapter 3. Factor Identification and Conceptual Framework................................................27

Table of Contents vii

Chapter 4. Indicator Selection.............................................................................................63

4.1. Criteria for indicator selection................................................................................64

viii Table of Contents

4.3.5. Emergency Response and Recovery Capability factor................................99

Chapter 5. Mathematical Combination...............................................................................107

5.1. General approach ...............................................................................................110

Chapter 6. Data Gathering and Evaluation..........................................................................145

6.1. Data sources.......................................................................................................145

Chapter 7. Sensitivity Analysis...........................................................................................159

Chapter 8. Presentation and Interpretation.........................................................................193

8.1. Goals of the presentation.....................................................................................193

Chapter 9. Future Work and Conclusions..........................................................................215

9.1 Future work .........................................................................................................215

A. Selected Previously Proposed Definitions of Disaster ..........................................221

xii List of Tables

List of Figures xiii

xiv List of Figures

Imagine a landscape without people, buildings, or any human influence. Suppose an

Chapter One: Introduction 1

The EDRI represents a novel approach to earthquake risk assessment. It is more

1.1. Anticipated benefits of the EDRI

2 Chapter One: Introduction

Chapter One: Introduction 3

4 Chapter One: Introduction

1.2. Six-step development procedure

1. Factor identification and conceptual framework development

Chapter One: Introduction 5

XV1: Seismic code indicator 1.5

XR1: Planning indicator

Hazard Exposure Vulnerability External Context Emer. Resp. & Recov.

Step 3. Mathematical combination Step 6. Presentation and interpretation

Figure 1.1. Six-step EDRI development procedure

6 Chapter One: Introduction

4. Data gathering and evaluation

6. Presentation and interpretation

Chapter One: Introduction 7

1.3. Organization of the dissertation

8 Chapter One: Introduction

2.1. Definition of earthquake disaster risk

Chapter Two: Key Features 9

10 Chapter Two: Key Features

Chapter Two: Key Features 11

12 Chapter Two: Key Features

I. Type of losses • Total for the city overall

Figure 2.1. Impact parameters

Chapter Two: Key Features 13

14 Chapter Two: Key Features

2.2. Objectives of the EDRI