Ontracts Airness AND Ncentives: E F A K K M. S
Ontracts Airness AND Ncentives: E F A K K M. S
Ontracts Airness AND Ncentives: E F A K K M. S
ERNST FEHR
ALEXANDER KLEIN
KLAUS M. SCHMIDT
Alexander Klein
Department of Economics
University of Munich
Ludwigstrasse 28
80539 Muenchen
Germany
alexander.klein@Lrz.uni-muenchen.de
Klaus M. Schmidt
Department of Economics
University of Munich
Ludwigstrasse 28
80539 Muenchen
Germany
klaus.schmidt@Lrz.uni-muenchen.de
A previous version of this paper was circulated under the title: Fairness, Incentives and Contractual
Incompleteness. We would like to thank Sam Bowles, Colin Camerer, Mathias Erlei, Dan Friedman, Gilat
Levy, Paul Milgrom, Massimo Motta, Jean Tirole and seminar participants at Bocconi University, the California
Institute of Technology, the University of California at Berkeley and at Santa Cruz, the University of
Copenhagen, Stanford University, the University of Toulouse, the University of Washington, the University
College London and the European University Institute for helpful comments and suggestions. Susanne
Kremhelmer provided excellent research assistance for the experiments discussed in this paper. Part of this
research was conducted while the third author visited Stanford University and he would like to thank the
Economics Department for its great hospitality. Financial support by Deutsche Forschungsgemeinschaft through
grant SCHM-1196/4-1 is gratefully acknowledged. Ernst Fehr also gratefully acknowledges support from the
Swiss National Science Foundation (project number 12-67751.02), the Ludwig Boltzmann Institute for Research
in Economic Growth and the Network on the Evolution of Preferences and Social Norms of the MacArthur
Foundation.
1. Introduction
This paper addresses the question of how concerns for fairness affect the actual and the optimal
choice of contracts. We conducted a series of experiments where principals could choose which type
of contract to offer to the agents. The optimal type of contract according to standard contract theory
proves to be far less efficient than this theory predicts, while contracts forecast to be very inefficient
if all agents are purely self-interested turn out to be superior. The experimental results suggest that
this reversal in contract efficiency is due to the existence of fair subjects, as they exert a decisive
impact on the incentive properties of different types of contracts. The principals in our experiments
seem to understand this quite well. A large majority of them chooses a contract that relies on fairness
as an enforcement device. Those who choose the contract predicted by standard contract theory do
very poorly. Moreover, we also observe intriguing interactions between explicit and implicit
incentives: the simultaneous use of both explicit and implicit incentives significantly weakens the
enforcement power of the implicit incentive. Therefore, the large majority of the principals prefer
contracts that rely solely on implicit incentives, i.e., they do not combine explicit with implicit
incentives. In the final part of the paper, we show that these results are largely consistent with a
simple model of fairness.
To better understand the nature of our results, consider one of our experiments in more detail.
Suppose that the principal wants to induce the agent to expend effort which is personally costly to
her. Both parties can observe effort, but the courts can only verify it if the principal invests in a
verification technology. If she makes this investment, she can offer an incentive contract to the
agent, which fines the agent for unsatisfactory performance. The problem with the explicit incentive
contract is that the verification technology is imperfect and the fine that can be imposed on the agent
is limited, so the highest effort level which can be implemented is positive but falls short of the
efficient level of effort. Alternatively, the principal can offer a bonus contract which does not rely
on effort verification and enforcement by third parties. Instead, the principal promises a non-binding,
voluntary bonus payment if the agents effort is satisfactory. This bonus contract is an implicit
contract because third parties do not enforce the principals promise.
Given that each principal interacts with each agent only once in the experiment, a selfish
principal would never pay the bonus. If it were common knowledge that all principals are selfish,
rational agents would choose the minimum effort level. Thus, standard contract theory forecasts that
the bonus contract is doomed to fail, while predicting the incentive contract to do much better. Yet,
the overwhelming majority of principals offered a bonus contract in our experiments. Even though
many principals did not pay the bonus, a substantial number of them made quite generous bonus
payments, inducing the agents to spend much more effort than under an incentive contract. Thus, the
bonus contract turns out to induce more efficient effort choices and, therefore, the principals
predominantly prefer the bonus contract relative to the incentive contract.
These results contradict standard contract theory based on the assumption that principals and
agents are solely interested in their own material payoffs. While this assumption may be an accurate
description of the behavior of many people, it is clearly wrong if applied to all people. In fact, there
is considerable evidence indicating that a substantial percentage of people also care about fairness
(see e.g. the surveys of Sobel 2002, Camerer 2003, and Fehr and Schmidt 2003). Our experiments
indicate that the principals' contract choices differ from those predicted by the self-interest model
because concerns for fairness strongly affect the incentive properties of the contracts.
However, we also conducted a second experiment in which the principal was restricted to
choose between an incentive contract and a trust contract. A trust contract offers a (generous) fixed
wage to the agent and asks him to return this favor by choosing a high effort level. Standard contract
theory again predicts that principals will choose the incentive contract. While the bonus contract
appeals to the fairness of the principal to reward high effort, the trust contract appeals to the fairness
of the agent to reciprocate a generous fixed wage. However, in contrast to the bonus contract the
trust contract did rather poorly. Many principals experimented with the trust contract, but on average
they incurred losses and eventually most of them shifted to the superior incentive contract.
The superiority of the bonus contract relative to the explicit incentive contract suggests that
fairness concerns might be responsible for the fact that principals often do not use explicit incentives
even though they are readily available.1 To examine this question, we conducted a third experiment
in which the principals could also propose a contract combining the explicit and the implicit
incentives. In principle, this contract could implement very powerful incentives because a shirking
agent incurs the cost of the fine and loses the bonus payment. Surprisingly, however, the use of the
explicit incentive is typically associated with significantly lower bonus payments, i.e., the explicit
incentive partially crowds out the bonus incentive. Therefore, pure bonus contracts are more efficient
than those combining explicit and implicit incentives. Moreover, the large majority of the principals
1
For example, a typical contract for a university professor does not make the salary directly contingent on easily
measurable and verifiable measures of performance such as citations, teaching ratings or the placement of Ph.D. students.
indeed choose pure bonus contracts, suggesting the relevance of fairness concerns for explaining the
absence of explicit incentives.
Our experimental evidence not only contrasts the viewpoint of standard contract theory, but
also constitutes a challenge for theories of fairness. For example, why does the incentive contract
outperform the trust contract as standard contract theory predicts while the bonus contract
surpasses the incentive contract, contradicting standard theory. How can the remarkable performance
difference between the trust and the bonus contract be explained as, after all, both contracts rely on
fairness as an enforcement device? Why do principals voluntarily forgo the opportunity to combine
the enforcement power of explicit incentives and implicit fairness-based incentives in favor of a
purely implicit incentive? We provide a unified interpretation of our results in terms of a simple
model of inequity aversion (Fehr and Schmidt, 1999) in the final part of the paper. We primarily
chose this model because of its tractability. Inequity aversion is a simple extension of the standard
self-interest model that takes the fact into account that some people are not only interested in their
own material payoff but also dislike inequity. The model implies that the incentive contract, which is
optimal when all actors are purely self-interested, is much less efficient when a share of people cares
about fairness. Furthermore, bonus contracts that would be very inefficient if all actors were selfish
achieve astonishingly high levels of efficiency when there are some fair-minded people. Thus, the
major predictions of the model are consistent with the observed qualitative pattern of contract
choices. In addition, the model makes surprisingly accurate quantitative predictions for the bonus
contracts and, if we choose plausible out-of-equilibrium beliefs, it can also account for the fact that
the principals prefer pure bonus contracts to those that combine explicit and implicit incentives.
However, the model also shows that the presence of some fair-minded people alone does not suffice
to implement efficient behavior. The incentive structure of the gift exchange is also very
important. We show that, as a general principle, the person who loses less from trusting the other
person should trust first. Thus, both our theoretical as well as our empirical results suggest that
concerns for fairness can and should be taken into account in the design of optimal incentive
schemes.
The model can also be used to illustrate some of the intricate and often surprising effects that
arise when some (but not all) people are fair-minded. A fair principal will pay the bonus if the agent
worked sufficiently hard. Thus, if the percentage of fair principals is not too small, a purely selfinterested agent will chose a high effort level as his expected return will be sufficiently large. This
makes it profitable for a selfish principal to mimic the contract the fair principals offer, in order to
benefit from the high effort levels of the selfish agents under a bonus contract without actually
paying the bonus. However, fair-minded agents strongly dislike being the sucker, i.e., they
experience additional disutility if they work hard but do not receive the bonus. If the principal cannot
credibly signal that she is going to pay the bonus, fair agents are not willing to work under a bonus
contract. While the existence of fair principals might induce selfish agents to perform well, the
presence of selfish principals encourages fair agents to provide little effort under a bonus contract.
Thus, too many fair agents can be detrimental to the efficiency of a bonus contract.
A large number of empirical papers have examined the effectiveness of different incentives
schemes over the last 10 15 years. This literature was surveyed in Prendergast (1999) and in
Chiappori and Salani (2003). Both survey papers conclude that incentives matter, i.e., agents
often seem to respond to changes in incentives in ways that are consistent with the predictions of
prevailing principal-agent models. However, both papers also report that the evidence for the
predictions of contract choices is much weaker. This is a main reason why we focused on the
principals contract choices between different types of contracts in our experiments. In this way, our
experiments may contribute to a better understanding of the forces determining which contracts
prevail.
Chiappori and Salani (2003) emphasize that problems of unobserved heterogeneity and
endogenous selection often complicate clean inferences about the incentive effects of contracts. In
fact, these problems result in an ambiguous interpretation of correlations between different contracts
and different behaviors. Do the contracts induce the corresponding behaviors or are the behavioral
differences across contracts the result of self-selection of heterogeneous individuals to different
contracts? This problem is in our view particularly severe in the context of fairness preferences
because there is little hope that non-experimental field data allow the control for such preferences.
Therefore, an experimental approach to these questions can offer additional insights. In our
experiments, for example, we had complete control of the selection of agents to contracts because
principals and agents were randomly matched. Thus, we can unambiguously infer the incentive
consequences of different types of contracts and how the principals responded to these incentive
effects.
Previous work by Camerer and Weigelt (1988); Fehr, Kirchsteiger, and Riedl (1993); Berg,
Dickhaut, and McCabe (1995); and Fehr, Gchter, and Kirchsteiger (1997) indicated that fairness
concerns may play an important role in moral hazard contexts.2 However, these papers neither
studied the interaction between fairness concerns and explicit incentives nor how the principals
choose between explicit and implicit incentives. There have been several experimental studies in the
past few years which examined how the provision of explicit incentives affects the agents behavior
in a moral hazard context. DeJong, Forsythe, Lundholm and Uecker (1985) showed how different
institutional remedies, such as liability rules, mitigate the moral hazard problem. Schotter, Bull and
Weigelt (1987) study the effects of piece rates and tournament incentives, and Schotter and
Nalbantian (1997) examine the performance of various group incentive schemes. Chaudhuri (1998)
investigated the ratchet effect in a dynamic principal-agent experiment in which the principals chose
output contingent wages. Likewise, Gth, Klose, Knigstein, and Schwalbach (1998) examined a
multi-period principal agent game in which the principals could offer linear profit-sharing contracts.
Cooper, Kagel, Lo and Gu (1999) studied how Chinese students and managers respond to the
incentives underlying the ratchet effect. Keser and Willinger (2000); Gth, Knigstein, Kovacs, and
Zala-Mezo (2001); and Anderhub, Gchter, and Knigstein (2002) also studied the performance of
output-contingent wages in a moral hazard context.3
Many of the studies mentioned above find indications that concerns for fairness and
reciprocity affect the acceptance of explicit incentive contracts. However, the principals did not have
the choice between different types of contracts in these papers in particular, the choice between
explicit incentive contracts and implicit bonus contracts. It was this setting which enabled us to
identify the strength and the limits of the standard approach in contract theory by isolating conditions
under which the model's contract choice predictions are met and conditions under which these
predictions failed. Moreover, our fairness model explains why the predictions of the standard model
are correct if incentive contracts compete with trust contracts but are completely at odds with the
facts when bonus contracts become available. Our setting is also unique in the sense that it enabled
us to study the interaction between explicit and implicit incentives. Since we implemented the
possibility of choosing contracts which combined explicit and implicit elements we were able to find
the puzzling crowding out of implicit incentives through explicit incentives. We are not aware of any
other empirical work that addresses this question or that documented such a crowding out effect.
More recently, Charness and Dufwenberg (2003) pointed out that there are important interactions between social
preferences and communication opportunities in a moral hazard context. In their setting no explicit incentives are present.
3
There are also a few experiments on the effects of incentives in environments with adverse selection (see Cabrales and
Charness 2003 and the references therein).
The rest of this paper is organized as follows. Section 2 describes the principal-agent problem
that we used in the experiments. Section 3 discusses the experimental design and procedures and
Section 4 reports the results of the experiments. In Section 5 we offer a theoretical interpretation of
the experimental results. We show that a simple fairness model is largely consistent with the data and
can be quite useful in organizing and better understanding the data. Section 6 summarizes our main
results and concludes.
e < e * . The principal can impose a fine f on the agent if shirking (e e*) has been verified.
However, the agent cannot be punished arbitrarily harshly, i.e., the fine f is bounded above by f .
Let e denote the highest effort level such that p f c( e ), i.e., e is the highest effort level such that
it is more profitable for a risk neutral agent to choose this effort level than to shirk (choose e = e )
and to incur the expected punishment p f . We will call e the highest incentive compatible effort
level. To make the problem interesting, we assume that e < eFB.
The timing of events is as follows. At date 0, the principal decides whether to incur the
verification cost and offers a take-it-or-leave-it contract to the agent. If the agent rejects the offer,
both parties get their reservation utilities that we normalize to 0. If the agent accepts, he has to
choose e at date 1. At date 2, a random draw determines whether the agents effort is verifiable (in
case k has been invested). Then payoffs are realized and payments are made.
If the principal does not invest in the verification technology, she can only offer a contract
with a fixed wage w to the agent. If, however, she invests in the verification technology, she can
offer a contract ( w, e*, f ) that stipulates a wage w, a demanded effort level e*, and a fine f, to be
paid in case shirking (e<e*) is verified. Such a contract, relying on effort verification and explicit,
enforceable incentives, will be called Incentive Contract (IC). The agents (expected) monetary
payoff in an IC is given by MA = w c(e) if e e* and by MA = w c(e) pf if the agent shirked
(e<e*). The principals expected monetary payoff is defined by MP = v(e) w k in case of
e e* and by MP = v(e) w + pf k if the agent shirked.
From the point of view of traditional contract theory, the analysis of the (second best) optimal
contract is straightforward if we make the standard assumption that the principal and the agent both
want to maximize their material payoffs. If the agent is only offered a fixed wage, he has no
incentive to provide any effort above e = e . Thus, if the verification technology is not too expensive,
the principal will offer an IC based on the maximum feasible fine f and requiring the agent to
choose the highest incentive compatible effort level e . Furthermore, the principal will offer a wage
w = c( e ) which compensates the agent for his effort cost and leaves him indifferent whether to
accept the contract or not.
However, this analysis rests on the important assumption that both players are only interested
in their own material payoffs. To see the implications of this assumption, note that both parties can
observe effort. Thus, as an alternative to the above incentive contract, the principal could simply
ask the agent to put in e* > e and promise him a reward in return. This could be done in two
different ways:
(i)
With a Trust Contract (TC): The principal offers the agent an unconditional payment w >
c(e). In return, she asks the agent to put in effort e*>e. However, if the agent accepts a trust
contract, he cannot be forced to choose e = e*. The monetary payoff from a trust contract
(w,e*) is given by MA = w c(e), for the agent, and MP = v(e) w, for the principal, where e
is the agents actual effort level.
(ii)
With a Bonus Contract (BC): In a BC (w,e*,b*) the principal offers an unconditional base
wage w c(e) and asks the agent to expend effort e*>e. Furthermore, the principal
announces her intention to pay a bonus b* if the agent chooses e e*. However, neither the
agents effort nor the principals bonus payment is enforceable. If the agent accepts a bonus
contract, he chooses effort e at date 1. The agent is not obliged to choose e = e* but can
choose any e e. Then, at date 2, the principal is informed about e and chooses the actual
bonus b. The principal is not obliged to pay b = b* but can choose any b 0. A bonus
contract implies monetary payoffs MA = w c(e) + b, for the agent, and MP = v(e) w b,
for the principal.
Obviously, both the trust contract and the bonus contract are doomed to fail according to the selfinterest model. The agent knows that his wage with a trust contract is fixed independently of his
effort level. Therefore, he will choose e = e. The principal will never pay the promised bonus at date
2 in a bonus contract. Anticipating this, the agent will again choose e = e at date 1. Thus, the selfinterest model predicts that the IC will dominate both TCs and BCs, implying that a rational and selfinterested principal will never propose a TC or a BC.
If, however, principals and agents are not only self-interested but also motivated by concerns
for fairness and reciprocity, the outcome is less clear. By offering a generous trust contract, the
principal can appeal to the fairness of the agent, and the agent may indeed reciprocate by providing
e > e . If the agent is offered a bonus contract, he may choose a high effort level in order to appeal to
the fairness of the principal, and the principal may indeed reciprocate by paying a bonus voluntarily.
Thus, both the TC and the BC may be more efficient than the self-interest model predicts. Such a
change in the relative efficiency of the different contracts may then induce the principals to prefer a
TC or a BC over an IC. The question of whether TC and BC are more efficient than IC remains open,
however, and cannot be answered on the basis of general, qualitative notions of fairness or in the
absence of empirical evidence.
Therefore, we implemented a series of experiments in which the principals had the option of
choosing between IC, TC, and BC. In a first step, we studied how the existence of fairness concerns
affects the principals choice between TC and IC by implementing the Trust-Incentive (TI)
treatment, in which only a TC or an IC could be offered to the agents. In a second step, we examined
how the availability of a nonbinding bonus affects the principals relative preference for the incentive
contract. We implemented the Bonus-Incentive (BI) treatment for this purpose, where all three
contracts could be chosen. The BI treatment is thus well suited for examining whether the existence
of fairness concerns affects the relative efficiency of the different types of contracts and whether the
principals take such effects into account when they choose from the available contracts. The BI
treatment does not enable us, however, to examine whether the principals voluntarily forgo the
opportunity to use an available explicit incentive. This is due to the fact that they could only choose
one of the three available contracts, i.e., they could not combine the IC with the BC. We can only
argue that the principals voluntarily forgo the opportunity for using the incentive if the principals
prefer a pure BC or a pure TC over the combination of a BC with the explicit incentive. For this
reason, we implemented the extended Bonus-Incentive (EBI) treatment, in which TC, BC, IC, or a
combined incentive-bonus contract could be chosen. The EBI treatment is thus well suited for
examining the question of whether the principals voluntarily forgo the opportunity of using an
explicit incentive. Moreover, this treatment enables us to study the interaction between explicit and
implicit incentives if agents face both types of incentives simultaneously in a combined contract.
10
c(e)
10
13
16
20
An effort of e yields a gross profit v(e)=10e to the principal. If the principal invests in the
verification technology at cost k = 10, she can verify the agent's effort with probability p = 1/3. The
maximum fine the agent can be charged is bounded above by f = 13 . Note that in a first best world,
the total surplus would be maximized if the principal did not invest in verification and the agent
chose e=10 which would yield a total surplus of v(e) c(e) = 80. The principal is constrained to
choose w c(e*) in all types of contracts. This rules out losses for the agents if they meet their
contractual obligations. We imposed this constraint to ensure that loss aversion does not affect the
agents' behavior.
Given the parameters of the experiment, a self-interested agent who maximizes his expected
payoff can be induced to choose an effort level of at most 4 by imposing the maximum fine of 13.
Thus, if both parties are self-interested, the optimal incentive contract the principal offers stipulates f
10
= 13, e* = 4 and w = 4 which limits the agent to his reservation utility. In equilibrium, the monetary
payoffs are MA = 0 and MP = 26. If the principal were restricted to offer a fixed (unconditional)
wage, a self-interested agent would always choose e=1, so that the principal would offer w = 0; in
this case the monetary payoffs are MA = 0 and MP = 10.4
The experimental subjects were students of the University of Munich and the Technical
University of Munich (students of law, political science, engineering, etc.). We had 20-24 subjects in
each session, half of them randomly assigned to the role of the principal and half to that of the agent.
The two groups were located in separate rooms. All subjects had to read detailed instructions and to
solve several exercises before the experiment started, to ensure that all of them understood the rules
of the experiment. We had ten periods in each session. The agents were randomly matched with a
different principal in each period. The randomization procedure ensured that no agent interacted
more than once with the same principal. Thus, we had ten contracts with ten different contracting
partners for each subject in each experimental session.
After each period, the subjects had to compute their own payoff and that of their partner. The
outcome of each period remained strictly confidential in order to rule out the possibility of reputation
building, that is, each principal-agent pair only observed what happened in their own relationship.
They did not observe the contracts offered by the other subjects in the room. Nor did they observe
their current partner's past behavior. Furthermore, the matching was random and anonymous, i.e., the
subjects identity was never revealed to the other players. Finally, the subjects collected their total
monetary payoffs privately and anonymously at the end of the session. Each session lasted between
two and two and a half hours. A complete set of the instructions for all our experiments can be found
on our webpage.5
We conducted seven experimental sessions. We implemented the TI treatment in Sessions 1
and 2, where the principals could choose between TC and IC. In Sessions 3 and 4, we applied the BI
treatment in which the principal could choose TC, IC, or BC. Finally, we conducted three sessions
(S5 S7) in which the principals could also offer a contract combining the available explicit
incentive with the nonbinding announcement of a bonus payment.
Note that the agent is indifferent whether to accept or to reject this contract. Because wages are discrete, a second
equilibrium exists in which the principal offers a wage that is one token higher. This increases the agents payoff while
decreasing the principals payoff by 1.
5
The full set of all our experimental instructions, in the original German and translated into English, are available at
http://www.vwl.uni-muenchen.de/ls_schmidt/experiments/incomplete_contracts/index.htm .
11
4. Experimental Results
4.1. The Trust-Incentive (TI) Treatment
We present the results of the TI-treatment in this section, where principals could choose between a
trust contract (w,e*) and an incentive contract (w,e*,f). We observed a total of 195 contractual
choices in sessions S1 and S2. Ten incentive contracts and two trust contracts were rejected meaning
that, in total, the agents made 183 effort choices. Our first result concerns the principals contract
choices.
Result 1(a): A clear majority of the contracts in the TI-treatment are incentive contracts and the
share of incentive contracts increases substantially over time.
(b) The average effort of the agents and the average payoff of the principals are higher under the
incentive contracts.
Figure 1 and the following numbers support Result 1a: 135 (69 percent) of the 195 offered contracts
are incentive contracts while only 60 contracts (31 percent) are trust contracts. However, these
numbers fail to demonstrate the strong time trend in the share of incentive contracts shown in Figure
1. While slightly less than 50 percent of the proposed contracts were incentive contracts in the first
period of the experiment, this fraction never fell below 70 percent beginning in period 4 and
exceeded 80 percent of all contracts in the final three periods. Although 71 percent of the principals
tried the trust contract at least once, only 33 percent did so in more than three periods. This indicates
that most principals experimented somewhat with the trust contract until settling for the incentive
contract.
12
0.5
0.4
0.3
0.2
0.1
0
1
10
Period
Figure 2: Average effort and average demanded effort in the Trust-Incentive treatment
7
average
dem anded effort
incentive contracts
6
5
average
dem anded effort
trust contracts
average effort
incentive contracts
3
2
1
0
1
Period
10
13
Figure 2, depicting the evolution of average effort levels (and average demanded effort levels)
over time for both contract types, illustrates Result 1b. The figure shows that the average effort is
higher in almost all periods in the incentive contracts. Moreover, the fraction of trust contracts is
already small in those periods in which average effort is somewhat higher in the trust contract,
meaning that this is driven by very few observations. The effort difference between ICs and TCs is
significant (p = 0.028, Mann Whitney test). This difference in effort levels is also associated with
differences in the principals payoffs. On average, the principals earned a payoff of 0.87 when they
proposed an incentive contract and 2.4 when they proposed a trust contract. These payoff
differences are, however, not statistically significant (p > 0.59, Mann Whitney test).
Viewed from the perspective of the self-interest model, the rather low profits resulting from the
incentive contracts are surprising because recall from Section 3 the predicted profit is MP = 26.
Moreover, it is also surprising that there is such a strong trend towards the incentive contracts in
view of the small payoff differences between the incentive and the trust contracts. Why did the
principals have such a strong preference for incentive contracts if these contracts performed so
poorly? The next result shows that the distinction between incentive compatible and non-incentive
compatible ICs is crucial in this context.
Result 2(a): Although most incentive contracts stipulate the maximal fine, the majority of incentive
contracts violate the no-shirking condition because the principals demand too high effort levels. In
the majority of the cases, non-incentive compatible ICs induce the agents to shirk fully, implying
negative payoffs for the principals.
(b) Incentive compatible ICs are, however, associated with significantly positive payoffs for the
principals because the agents shirk much less in these contracts. The large payoff difference between
incentive compatible and non-incentive compatible ICs is associated with a strong increase in the
share of incentive compatible ICs over time.
Our data supports R2a as follows: the average fine is 12.3, closely approximating the maximal fine
of 13. However, the no-shirking condition, pf c(e*), is violated in 79 (58.5 percent) of the 135
incentive contracts, i.e., principals demanded too high effort levels. Figure 2 also illustrates this fact,
showing that the average demanded effort level in the incentive contracts persistently exceeds the
maximal enforceable effort of e* = 4. We present the agents effort behavior and the principals
14
payoffs for incentive compatible ICs, non-incentive compatible ICs and TCs in Table 2, which shows
that non-incentive compatible ICs are associated with a high rate of shirking and rather low payoffs
for the principals. The last row of Table 2 indicates that there are 79 non-incentive compatible
contracts, that only 1 of these contracts is rejected, and that the agents shirked fully by choosing the
minimal effort level of e = 1 in 48 (62 percent) of the accepted contracts. This high rate of shirking
has the consequence that the non-incentive compatible ICs cause on average a loss of 7.6 for the
principals (see shaded area in last row of Table 2).
Figure 2 not only indicates that the average demanded effort in the ICs is too high relative to
the enforceable effort level but also shows that the demanded effort level declines over time. The
average demanded effort level in period 1 is close to e* = 6 while it is only slightly above the
incentive compatible level of e* = 4 in the final period. This suggests that the share of incentive
compatible ICs increases over time. In fact, only 10 percent of all ICs are incentive compatible in
period one while this amount already exceeds 64 percent of all ICs in period ten. The profit
differences between incentive compatible and non-incentive compatible ICs provide a natural
explanation for this strong time trend. The shaded areas in the last row of Table 2 show that the
average profit in the incentive compatible ICs is 8.6, which is much larger than the loss of 7.6 in the
non-incentive compatible contracts. Thus, while incentive compatible ICs are considerably more
profitable than trust contracts, the non-incentive compatible ICs are less profitable than the trust
contracts.6 The strong profit differences between the incentive compatible ICs and the TCs also
explains why the share of trust contracts strongly declines over time.
There are two reasons why the incentive compatible ICs are more profitable than the nonincentive compatible ICs. First, the principals pay far lower wages when they offer incentive
compatible contracts. Second, although the principals pay less when they offer incentive compatible
ICs, shirking is much less frequent in these contracts. Table 2 shows that the wage is above w = 10
in all 79 offered ICs that are not incentive compatible while in the majority of the incentive
compatible ICs (in 29 of 56 cases) the wage is strictly below w = 10. This suggests that the
principals attempted to elicit reciprocal effort choices from the agents when they proposed nonincentive compatible contracts. Recall, however, that these attempts frequently failed. This contrasts
These differences are statistically significant according to a Mann-Whitney test (p = 0.005 when the TC is compared to
the incentive compatible IC; p = 0.036 when the TC is compared to non-incentive compatible IC).
15
sharply with those contracts that meet the no-shirking condition (see last row of Table 2). The agents
shirk in only 12 (26 percent) of the 47 accepted incentive contracts.7
Wage
Offer No. reje e<e ee
Non-Incentive Compatible
Incentive Contracts
Ps
No. reje
ct
payoff of
Offers
e=
1
e>
1
Trust Contracts
Ps
payoff
No.
of
Offers
reje
ct
e=
1
e>
Ps
1 payoff
n.a.
17
15
3.7
of
Offers
ct
29
15
8.5
26
19
9.8
33
20
12
-1.4
13
-1.0
high
20 < w
-20.0
46
28
18
-12.0
30
13
17
-6.4
All
56
12
35
8.6
79
48
30
-7.6
60
37
21
-2.4
low
w < 10
mediu
m
10w
20
Remark: n. a. means that no entries are available for the respective cells. The sum of the column indicating the number of
contracts in which the agent shirked (e < e*) and in which the agent met the demanded effort (ee*) for the incentive
compatible contracts yields the total number of accepted contract offers. Likewise, the sum of the column indicating how
often the agents chose e = 1 and how often they chose e 1 for the non-incentive compatible ICs and the trust contracts
gives us the total number of accepted contracts.
Table 2 also indicates that when trust contracts were offered, the principals paid relatively
high wages in 30 of the 60 trust contracts the principals offered a wage above 20. Thus, the strong
decrease in trust contracts and non-incentive compatible ICs over time caused a decreasing trend in
wages over time. The principals offered average wages well above 20 during the first few periods,
when the share of incentive compatible ICs was still low. The average wage decreased, however,
strongly over time and reached a level of 11.9 in period ten. The strong time trend in the share of
7
Note that the number of accepted contracts is given by the sum of the two effort columns. For example, for the
incentive compatible contracts this sum is given by the 12 contracts with e < e* plus the 35 contracts with e e*. In all
16
incentive compatible contracts and the average wage suggests that, initially, the principals tried to
elicit non-incentive compatible effort levels by paying generous wages but, as these attempts failed,
they converged slowly towards incentive compatible ICs.
There is a further noteworthy feature in Table 2. For the trust contracts the principals payoff
is decreasing in the offered wage. The principals earn 3.7 for wages below w = 10, the payoff
declines to MP = 1.0 for wages in the middle interval (10 w 20), and further diminishes to MP =
6.4 for high wages (w > 20).8 A similar relation holds for the non-incentive compatible ICs, where
earnings amount to MP = 1.4 in the middle interval while corresponding to MP = 12.0 for high
wages. We summarize this payoff pattern in
Result 3: Increasing the generosity of the wage offer as an attempt to induce non-incentive
compatible effort levels decreases the principals average payoff.
Results 1 3 in the TI treatment are interesting because many observed qualitative data patterns are
consistent with the predictions of the self-interest model. This model predicts that the principals offer
incentive compatible ICs by imposing the maximum fine and demanding the maximum enforceable
effort level. This contract preference is based on the prediction that both the trust contract and nonincentive compatible ICs will induce shirking of the agents, implying that high wages will be
associated with losses. In fact, the principals' contract choices do converge towards the prediction of
the self-interest model and as Result 3 shows the payment of high wages is indeed not profitable.
However, there are also some aspects of the data that violate the predictions of the self-interest
model. If the principals offer incentive compatible ICs, they earn substantially less than predicted.
Recall that according to the prediction they should earn MP = 26; in fact they earn only 8.6. This
misprediction has three reasons: (i) although the principals reap a larger share of the surplus, they
rarely extract the whole rent from the agents, i.e., they still pay substantial wages. The principals
offer on average 35% of the surplus that occurred if the worker met e = e* in the incentive contracts.
Even if they propose incentive compatible ICs, the offered surplus is on average 31 %. (ii) The
agents reject wages below w = 10 in 25% of the cases. (iii) The agents shirk by choosing the
minimal effort in roughly 1/3 of the incentive compatible contracts. All three reasons suggest a role
12 cases with e < e* the agents chose e = 1.
A simple OLS-regression of effort on wages also confirms this result, yielding e = 1.08 + 0.04 w + where denotes
the error term. The t-value for the constant is 3.11 and the t-value for the coefficient on w is 3.65. According to this
regression, effort significantly increases with wages but the increase is associated with a marginal loss: A wage increase
17
for fairness concerns. Likewise, the fact that the principals initially expressed a strong preference for
trust contracts or non-incentive compatible ICs suggests that they attempted to elicit generous effort
choices from the workers. However, the prevailing fairness motives were apparently not strong
enough to render these contracts more efficient than the incentive compatible ICs. It remains to be
seen whether fairness concerns can overturn the contract predictions of the self-interest model if the
principals can announce a bonus payment.
Figure 3 presents the evidence for R4a. Trust contracts do not appear in this figure because they were
never chosen. The figure shows the evolution of the share of bonus and incentive contracts over time.
87 percent of all contracts are already bonus contracts in period one. The share of bonus contracts
drops slightly below 80 percent in periods three to five because a few principals experimented with
the incentive contract in these periods. However, the share of bonus contracts is roughly 90 percent
from period six onwards and even approaches 96 percent in the final period. There can thus be little
doubt that principals strongly prefer the bonus contract.
by 10 units raises effort only by 0.4 and, hence, the expected revenue increases only by 4 units.
18
0 .5
10
P e rio d
To examine the reasons for this preference, we compare the average effort level in bonus and
incentive contracts (see Figure 4). The figure shows that the average effort is considerably higher in
the bonus contracts in nine out of ten periods.9 While the average effort in the incentive contracts is
generally between e = 2 and e = 3, effort in the bonus contracts is, in general, above e =5. This
difference across contract types is highly significant (p < 0.001, Mann Whitney Test). Figure 4 also
indicates that agents efforts in the bonus contracts are somewhat below the demanded effort level
but the gap between actual and demanded effort levels is much smaller than in the incentive
contracts. In fact, as in the TI-treatment, many incentive contracts are not incentive compatible. This
is indicated by the fact that the demanded average effort always exceeds e* = 4. The large effort
differences between the contracts are also translated into large profit differences. Principals average
profit from bonus contracts, taken over all ten periods, is 27 tokens while the incentive contract
generates an average loss of 9 tokens. The average profit from bonus contracts is always above 20
tokens in each of the ten periods while the incentive contract causes losses in six of the ten periods.
The exception is period ten, where the effort difference is negligible. However, there was only one incentive contract in
period ten, so that this data point has little relevance for the overall comparison.
19
In view of these large profit differences it is no longer surprising that principals exhibit a strong
preference for bonus contracts.
7
6
average
demanded effort
incentive contracts
5
4
average effort
bonus contracts
3
2
average effort
incentive contracts
1
0
1
10
Period
There is an interesting difference in the performance of the incentive contracts across the TI
and the BI treatment. The principals who offer incentive contracts in the BI treatment demand nonincentive compatible effort levels in all periods while they learned to make incentive compatible
contracts over time in the TI treatment. Thus, it seems that the strong superiority of the bonus
contract inhibited a learning process in favor of incentive compatible ICs. Note, however, that the
average profit in the bonus contracts is more than three times higher than that in the incentive
compatible ICs in the TI treatment. This suggests that in view of the superiority of the bonus
contracts it simply did not pay to learn to make incentive compatible ICs in the BI treatment.
20
The higher effort level in the bonus contracts implies a higher surplus. To what extent did the
agents receive part of this increase in the surplus relative to the incentive contracts? On average
agents earned an income of 14.4 in the incentive contracts while in the bonus contracts their payoff
was 17.8. Thus, agents received a small part of the surplus increase while the principals reaped the
bulk of the increase. This shows that the option to pay a bonus yields a substantial efficiency increase
and causes sizable changes in the distribution of the surplus.
Why does the bonus contract prove to be vastly superior to the incentive contract? Our next
result shows that the key for understanding this result lies in the principals bonus payments.
Result 5: The principals devote a substantial part of the agents compensation to bonus payments.
Moreover, the average bonus increases strongly with respect to the effort level so that non-minimal
effort choices are profitable for the agents.
Figure 5a and 5b support R5. Figure 5a shows the average wage offered in both the incentive and the
bonus contracts; in addition the figure presents the average bonus payments in the BCs. The average
wage in the BCs remains in the vicinity of w = 15 throughout the whole experiment and the bonus
payment amounts to b = 10.4. Thus, the principals pay roughly 40% of the agents compensation in
the form of a bonus. However, this bonus payment strongly depends on the agent's effort (see Figure
5b). If the agent provides low effort at e = 1 or e = 2 the average bonus is zero, but the bonus
approaches b = 30 for high effort levels. The positive slope of the bonus-effort schedule is also
confirmed by the following regressions that related the bonus payment to the agents effort e, the
demanded effort e*, the base wage w, and the announced bonus b* (see Table 3).
21
Figure 5a: Average wages and average bonus over time in the BI treatment
40
30
25
20
15
10
5
0
1
10
Period
35
Average Bonus
30
25
20
15
10
5
0
1
Effort
10
22
(1)
(2)
(robust standard
errors)
(robust standard
errors & clusters)
Constant
-5.58***
(1.94)
-5.58**
(2.59)
Effort
2.86***
(0.17)
2.86***
(0.33)
0.33
(0.27)
0.33
(0.46)
-0.30***
(0.09)
-0.30*
(0.16)
0.12*
(0.06)
0.12
(0.08)
198
198
Bonus payments
Demanded effort
wage
Announced bonus
No. of observations
0.57
0.57
Adjusted R2
Table reports the coefficients of OLS regressions. Robust standard
errors are in parentheses. ***, ** and * indicate significance at the
1%, 5% and 10% level, respectively.
Table 3 reports the results of two regressions with the associated robust standard errors. The
first is a simple OLS regression. In the second regression we treated the observations of individual
principals as separate clusters because they may not be independent of each other. Thus, in the
second regression the standard errors are based on the assumption that the bonus payments are
independent across different principals but we allow for dependent observations within each cluster.
The assumption that the bonus payments are independent across principals is reasonable because a
principal could never observe what the other principals did. Moreover, since the bonus payment is
23
the final choice in each period, the principals can respond to all previous actions that occurred in the
match of that period.10
Both regressions in Table 3 show that an increase in the effort level by one unit increases the
expected bonus payment significantly by 2.86 tokens. Note that this is higher than the marginal cost
of effort for all effort levels e 7, i.e., a rational and selfish agent chooses an effort level of e = 7 if
he faces this bonus-effort relation. The impact of the demanded effort level is small and not
significant, indicating that e* is considered to be cheap talk. The fixed wage enters the regression
with a significantly negative sign, suggesting that if the actual wage increases by 1 token, the
principal will reduce the bonus payment by 0.3 tokens on average. The announced bonus enters
significantly with a positive, but very small coefficient. An increase in the announced bonus by 10
tokens increases the average actual bonus by only 1.2 tokens. Thus, it seems that principals feel
somewhat but not excessively committed to their bonus announcements and that the effort level is
the major determinant of the principals bonus choice.
Although the principals respond, on average, quite strongly to increases in the effort level, it
is important to notice that there are big differences in individual behavior. In those 162 contracts
where the agents chose a non-minimal effort level (e > 1), the principals did not pay any bonus at all
in 34 cases (21%). Among those principals who did pay a bonus, many paid very little even if the
agent selected a high effort level. However, there were also many principals who reciprocated high
effort levels very generously.
Taken together, the TI and the BI treatments show that the principals strongly prefer the
bonus contract. If this contract is not available, they prefer the incentive over the trust contract. The
same ranking holds in terms of the average effort and the average surplus associated with the three
types of contracts. These facts are puzzling from the viewpoint of the self-interest model. Recall that
this model captures important qualitative aspects of the TI treatment quite accurately. Although there
are several hints in the data suggesting that fairness concerns play a role in the TI treatment, these
concerns are apparently too weak to overturn the basic prediction that the principals prefer the
incentive contract relative to the trust contract. However, the mere addition of the possibility of
announcing and paying a nonbinding bonus which represents a completely innocuous change from
the viewpoint of the self-interest model suddenly transforms fairness concerns into a powerful
10
To check the robustness of our results we also conducted Tobit regressions. All variables that are significant in the
OLS regressions are also significant in the Tobit regressions and all variables that are insignificant in the OLS
regressions remain insignificant in the Tobit regressions.
24
determinant of principals contract choices: many principals reward generous effort levels with
generous bonus payments and thus create powerful incentives for effort provision. We will provide a
unified explanation for this puzzle with a fairness model that is based on the assumption of
heterogeneous fairness preferences in Section 5 below.
Before doing so, we want to emphasize a further important aspect. A general prediction of
traditional contract theory is that contracts should depend on all verifiable signals that contain
statistical information about the agents action or type (see Holmstrm 1982, Laffont and Tirole
1993). However, actual contracts frequently specify important obligations of the contracting parties
in vague terms, and they do not explicitly tie the parties monetary payoffs to measures of
performance that would be available at a relatively small cost. Thus it seems that many contracts are
left deliberately incomplete. In our BI treatment the principals expressed a strong preference for a
fairly incomplete bonus contract in which bonus payments were only vaguely and implicitly tied to
the agents performance. This indicates that principals did not want to use the explicit incentive
contract if they had to give up a vaguely defined, but nonetheless powerful, implicit incentive. In
reality, however, implicit and explicit incentives can be combined. Thus, the question arises whether
the principals would also forgo the opportunity to implement an explicit incentive if they could
combine it with an implicit one. If we can show that the principals prefer a purely implicit incentive
over a combination of explicit and implicit incentives, we indeed have direct evidence that they
voluntarily do not utilize the explicit incentive. Moreover, since fairness concerns are a decisive
force behind the bonus contracts in our context, we also have evidence that these concerns are
important for the absence of explicit incentives.
25
outweigh the verification costs of k = 10 that are associated with the implementation of the explicit
incentive.
All the other contracts were still available in the EBI treatment, i.e., the principals could also
offer a (pure) trust contract (i.e., f = b* = 0), a (pure) bonus contract or a (pure) incentive contract
(i.e., f > 0, b* = 0). We conducted three sessions (S5 S7) in the EBI treatment. In total we observed
339 contract offers in this treatment. 5 pure bonus contracts and 8 combined contracts were rejected,
leaving 326 accepted contracts. The following result informs us about the principals contract
preferences in this setting.
Result 6: We observe neither trust contracts nor incentive contracts in the EBI treatment. Roughly
2/3 of all contract offers are pure bonus contracts and 1/3 are combined contracts. In the final
periods the share of pure bonus contracts even surpasses 70%.
Period
10
26
Support for R6 is provided by Figure 6 and the following numbers. 229 of the 339 contracts (67.6%)
are pure bonus contracts, 110 contracts (32.4%) are combined contracts. Figure 6 also indicates that
the principals prefer the pure bonus contract in each period. From period 3 onwards, there is a
slowly increasing trend in favor of the pure bonus contract which peaks in period 10 at a share of
roughly 75 percent. Thus, the clear majority of principals voluntarily forgo the opportunity to
implement an explicit incentive.
The strong preference for the pure bonus contract is puzzling because the enforcement power
of implicit and explicit incentives can be used in the combined contract. Our next result shows,
however, that this potential advantage of the combined contract does not generate significantly
higher effort levels:
Result 7: Effort is not significantly higher in the combined contracts than in the pure bonus
contracts. Hence, pure bonus contracts are more efficient. The agents largely reap these efficiency
gains.
R7 is supported by Figure 7a and 7b and the following facts. The average effort is 5.8 in combined
contracts and 5.3 in pure bonus contracts. This small difference is not significant (p > 0.10, Mann
Whitney test) and does not suffice to outweigh the implementation costs for the explicit incentive.11
This fact is also indicated by Figure 7a and 7b. In Figure 7b we plotted the principals profit from
both types of contracts. The figure shows that the principals earned more in most periods when they
offered pure bonus contracts. In addition, their payoff from combined contracts exhibits considerably
higher volatility, suggesting that the variance of profits was higher when combined contracts were
offered. In fact, the standard deviation of the principals payoff in the combined contract is 26.75
while in the pure bonus contract it is only 23.68. If we average over all periods, profits are 24.7 in the
pure bonus contracts and 24.0 in the combined contracts. This difference is not significant (p > 0.99
Mann Whitney test). Figure 7b shows, however, that the agents earned considerably higher incomes
in pure bonus contracts. Except for the final period, the agents average income was roughly 20 units
in the pure bonus contracts while they earned 15 units or less most of the time in the combined
contracts. If we average over all periods, the agents earn 19.2 in the pure bonus contracts and 12.5 in
the combined contracts. This difference is significant (p < 0.001) according to a Mann Whitney test.
11
The average demanded effort level is virtually identical across contracts 7.30 for the pure bonus contracts and 7.22
for the combined contracts.
27
Figure 7a: The principals profits over time in the Extended Bonus-Incentive treatment
50
P ro fits in p u re b o n u s c o n tra c ts
P ro fits in c o m b in e d c o n tra c ts
45
40
35
30
25
20
15
10
5
0
1
10
P e rio d
Figure 7b: The agents income over time in the Extended Bonus-Incentive treatment
40
30
25
20
15
10
5
0
1
Period
10
28
Due to the considerably higher earnings of the agents, total earnings are higher when pure bonus
contracts are offered. Under the pure bonus contracts total earnings are 43.9 while they are only 36.5
under the combined contracts. This difference is again significant (p = 0.0012, Mann Whitney test).
These results suggest that the principals prefer pure bonus contracts because they generate
slightly higher average profits and are less risky than combined contracts. Combined contracts
require an initial investment of k = 10 which does not pay off because effort increases only
insignificantly if workers face the threat of being fined. Thus, the key behind the preference for the
bonus contract is that the combined contract does not raise effort sufficiently. Our next result
indicates a plausible reason for this fact.
Result 8: In the combined contracts principals reward high effort levels less generously than in pure
bonus contracts. Thus, combined contracts provide lower implicit incentives.
Support for R8 is provided by Figure 8 below and the associated regressions. Figure 8
unambiguously indicates that the principals paid a higher average bonus in the pure bonus contracts
for any non-minimal effort level. Moreover, the difference is quite large for effort levels above e = 6
on average the principals paid more than 10 tokens less in the combined contracts. This difference
in the bonus payments is also a major source of the difference in the agents income across contracts
because the average bonus in the pure bonus contract is 10.92 while in the combined contract it is
only 6.16. Note that this difference in the average bonus payments cannot be due to different effort
levels because the actual distribution of effort is roughly identical across contract types.
29
30
25
20
15
10
0
1
10
Effort
We also conducted multivariate regressions to examine whether the relationship between
effort and bonus payments still holds if we control for other variables. In addition to effort we used
the demanded effort e*, the wage w, the announced bonus b* and interactions between these
variables and a dummy indicating the pure bonus contract as regressors. The dummy is denoted by D
and takes on a value of 1 if the observation is from a pure bonus contract. The results of the
regression analyses are presented in Table 4.
30
(1)
(2)
(3)
(4)
(5)
pure
combined
all
all
all
bonus
contracts
contracts
contracts
contracts
payments
contracts
Constant
-3.21
(2.07)
- 6.01
(4.14)
-3.76**
(1.64)
-3.83**
(1.63)
-3.83*
(2.03)
3.03***
(0.160)
1.61***
(0.24)
1.79***
(0.18)
1.58***
(0.22)
1.58***
(0.34)
0.19
(0.36)
0.97
(0.66)
0.42
(0.28)
0.75*
(0.44)
0.75
(0.56)
-0.18***
(0.08)
0.29**
(0.14)
-0.21***
(0.06)
-0.29**
(0.13)
-0.29*
(0.15)
-0.01
(0.04)
0.04
(0.09)
0.01
(0.03)
0.03
(0.08)
0.03
(0.10)
1.13***
(0.18)
1.46***
(0.26)
1.46***
(0.40)
D demanded
effort
-0.50
(0.48)
-0.50
(0.49)
D wage
0.11
(0.15)
0.11
(0.17)
D announced
bonus
-0.03
(0.08)
-0.03
(0.09)
Effort
Demanded
effort
wage
Announced
bonus
(clusters)
D effort
No. of
observations
224
102
326
326
326
Adjusted R2
0.64
0.32
0.593
0.596
0.596
Table reports the coefficients from OLS regressions. Robust standard errors are in
parentheses. D denotes a dummy variable for the pure bonus contracts. ***, ** and *
indicate significance at the 1%, 5% and 10% level, respectively.
31
The first important result is that the effort coefficient is sizeable and highly significant in all
regressions. However, a comparison between regression (1) and (2) shows that the effort coefficient
is much larger under a pure bonus contract. An increase in the effort level by one unit in a pure
bonus contract increases the bonus by roughly 3 units which is consistent with the results we
observed in the BI treatment. The bonus increase is only 1.6 units on average in the combined
contracts. Regression (3) shows that this difference is highly significant as indicated by the
coefficient on D effort. This result remains robust when we introduce further interaction variables
in regression (4). All the other variables show no significant interaction with the pure bonus dummy.
Another noteworthy effect in Table 4 is that a higher base wage always reduces the bonus payments
significantly. In regression (5) of Table 4 we checked the robustness of our results with regard to the
clustering of individual principals. As already explained in the context of Table 3 the regression with
clusters controls for potential dependencies in the bonus payments of individual principals.
Regression (5) shows that this additional control does not affect the significance of the effort and the
D effort coefficient.12
This evidence leaves little doubt that principals who choose combined contracts reward high
effort less generously. It seems that the use of the explicit incentive crowds out implicit incentives to
some degree and thus weakens the enforcement power of the combined contracts. To examine how
the crowding out of implicit incentives affects the agents incentives, the fact that the principals
chose the maximum fine in almost all combined contracts but also demanded effort levels that were
not incentive compatible must also be taken into account. The average fine is given by 11.99 and,
among the 110 combined contracts that were offered, only 6 contracts stipulated an incentive
compatible effort level of e* = 4. Thus, the combined contracts clearly also relied on the enforcement
power of the implicit bonus incentive. This interpretation receives further support if we compare the
actual average effort that was enforced in the incentive contracts in the TI or the BI treatment with
the effort actually enforced in the combined contracts of the EBI treatment. Recall that the principals
in the TI and BI treatment could enforce on average e = 2.51 and e = 2.1, respectively, with the
explicit incentive alone, while they enforced e = 5.8 in the combined contracts. Therefore, a large
part of the enforcement power of combined contracts must be due to the implicit incentive. However,
the fact that the combined contracts rely on the implicit incentive also suggests that variations in the
steepness of the bonus-effort relationship are associated with variations in the enforcement power of
12
We also conducted Tobit regressions. Relative to the OLS regressions the same variables remain, and no other
32
these contracts. Therefore, the less generous reward policy of the principals in the combined
contracts provides a natural explanation for the fact that the combined contracts did not elicit
significantly higher effort levels than the pure bonus contracts.
5. A Unified Interpretation
The major puzzle posed by the evidence in Section 4 is that the incentive contract outperforms the
trust contract while the bonus contract outperforms the incentive contract. Why are fairness concerns
too weak to render the trust contract an effective tool but strong enough to render the bonus contract
very effective? In addition, we also have to explain the remarkable result that principals
predominantly choose the pure bonus contract even if they could combine the enforcement power of
explicit and implicit incentives.
In view of the importance of fairness concerns in our experiments, it is natural to seek an
explanation of these puzzles in the context of recently developed fairness models. We will show that
the fairness approach is indeed capable of explaining the pattern of contract choices observed in
Section 4; sometimes the approach even provides surprisingly precise quantitative characterizations
of the observed behavior. In the following, we apply the theory of inequity aversion of Fehr and
Schmidt (1999) to our experiments because it captures important aspects of fairness driven behavior
in a tractable way and is consistent with the outcomes of many different classes of experimental
games.13 The choice of this theory does not mean that other theories of social preferences (e.g.
Bolton and Ockenfels 2000, Levine 1998, Falk and Fischbacher 1999, Charness and Rabin 2002,
Cox and Friedman 2002, Dufwenberg and Kirchsteiger, in press) may not also be able to rationalize
the data.14 A main reason for applying the Fehr-Schmidt model is tractability. In addition, we want to
emphasize that it is not our aim to explain the dynamic pattern of the data over time. Instead, we
33
focus on the robust behavioral regularities that emerge in all treatment conditions in the final few
periods.
The theory of Fehr and Schmidt (1999) has two main tenets: First, it assumes that some
people are not only concerned about their own material payoff but also care about inequity or, in our
context, inequality.15 Second, the theory acknowledges that people differ. Some people are very
much concerned about inequality and have a high willingness to pay in order to reduce it, while
others only care about their own material payoff. In the two-player case, the utility function of
inequity averse (fair) players is given by
Ui(x) = xi - i max{xj - xi,0} - i max{xi - xj,0},
i {1,2}, i j, where x=(x1,x2) denotes the vector of monetary payoffs and i i, 0 i < 1. In this
utility function, the term weighted with i measures the utility loss that stems from inequality to is
disadvantage, while the term weighted with i measures the loss from advantageous inequality. For
sufficiently high values of i and i players with this utility function want to achieve equality. If the
inequality is to their disadvantage, they are prepared to incur costs in order to reduce the payoff of
their opponent. If the inequality is to their advantage, they are willing to spend resources in order to
benefit the other player.
We use a simplified version of this theory. Following Fehr and Schmidt (1999), we assume
that there are 60 percent self-interested types (i = i = 0) and 40 percent inequity-averse types. In
addition, we assume i, i > 0.5, i.e., the inequity averse subjects are willing to share the surplus of a
contract equally and reject offers that give them less than 25 percent of the surplus.16 The restriction
to two types for our games is quite natural because what matters is whether a player wants to
equalize payoffs. The assumption that 40 percent of all players are prepared to equalize payoffs is
derived from the distribution of types calibrated in Fehr and Schmidt (1999) with experimental data
on the ultimatum game. Fehr and Schmidt used this distribution to explain the experimental results in
many different classes of games, so we want to use it for this game as well. However, the qualitative
15
There is no generally accepted notion of fairness, but probably all fairness definitions imply that equals should be
treated equally. In our experiments, the subjects enter the laboratory as equals. They have no information about their
opponents and do not know with whom they trade. Thus, it seems natural to define equality in these very simple
environments as the reference point for a fair payoff distribution.
16
See Fehr and Schmidt (1999) for a more extensive discussion of the experimental evidence on the distribution of
inequity averse types. When Fehr and Schmidt calibrate their model to explain the quantitative evidence in the different
games they use four different types, but in the aggregate these assumptions imply that 40 percent of subjects exhibit i
i>0.5 and that 60 percent exhibit 0.5>i i.
34
results that follow are robust to changes in this distribution, as long as the share of inequity averse
types is at least 33 percent but not larger than 60 percent.
On the basis of these assumptions, our principal agent problem can be analyzed using
standard game theoretic tools. The full analysis is not difficult but somewhat lengthy and therefore
relegated to an appendix that can be found on our webpage.17 Here, we want to report the main
predictions that follow from this analysis and expound on them. First, we focus on the trust contract.
Proposition 1 [Trust contracts]: Increasing the wage in a trust contract increases the effort of the
inequity-averse agents, but, on average, the effort increase is too small to make a wage increase
profitable for the principal.
To see the intuition for this proposition consider an inequity-averse (i.e., fair) agent who
accepted a generous trust contract. He will choose an effort level that equalizes the monetary payoff
of the principal with his own monetary payoff:
MP = 10e - w = w - c(e) = MA
Using the implicit function theorem, we get
de
2
=
dw 10 + c' (e)
Thus, e increases with w for an inequity-averse agent, but, if the fraction of inequity-averse agents in
the population is q = 0.4, then an increase of w by 1 token increases average effort by at most e =
0.4(2/11) = 0.07 which increases the principals gross profit by at most 100.07 = 0.7 tokens. Hence,
a wage increase does not pay off for a selfish principal.18 What about the fair principal? A generous
wage will not pay off in monetary terms, and it will generate inequality to the principals
disadvantage whenever a selfish agent chooses e=1. Hence, a fair principal will not pay a higher
wage either. The reason why the trust contract does not work is that the percentage of fair agents is
simply too small. This percentage would have to be at least 60 percent in order to make it profitable
17
35
for the selfish principals to offer generous wages that induce the fair agents to choose e > 4.19 Recall
from Table 2 in Section 4.1. that in a trust contract higher wages are indeed associated with lower
payoffs for the principals. Proposition 1 neatly rationalizes this fact. Next we turn to the analyses of
the incentive contracts.
Proposition 2 [Incentive Contracts]: Both the selfish and the fair principals stipulate the maximal
fine, f = 13, and demand the maximal incentive compatible effort level, e* = 4 in the optimal
incentive contract. However, the selfish principals offer a wage w = 4 that gives the agents none of
the surplus while fair principals offer w = 17 which distributes the surplus equally. Selfish agents
accept and obey these contracts. The fair agents accept and obey generous contracts with w 17
and choose e 4, where e is increasing with w. Fair agents reject, however, w = 4. If 4 < w < 17,
they will either reject or shirk even if the contract is incentive compatible.
We already know from Section 3 that a selfish principal would offer (w=4, e*=4, f = 13 ) if
all agents were selfish. With some fair agents, the selfish principal runs the risk that the agent will
reject her unfair offer. Thus, she may want to increase w in order to increase the probability that her
offer will be accepted. However, it turns out that as long as q < 0.6, increasing the wage does not pay
off. Furthermore, increasing w somewhat, but not up to the level of full equality, is dangerous,
because the agent may accept the contract but consider it unfair and shirk. Therefore, (w=4, e*=4,
f = 13 ) is optimal for a selfish principal. Note that the contracts of the selfish principals will be
rejected with probability 0.4 due to the existence of fair agents. This explains the frequent rejection
of low wage offers and why the optimal incentive contract becomes less efficient than predicted by
the self-interest model. An inequity-averse principal offers (w=17, e*=4, f = 13 ) which shares the
surplus equally if the agent chooses e = 4. The maximum fine is necessary to induce the selfish
agents to choose e = 4. A fair principal could also pay a higher wage in order to induce the fair
agents to choose e > 4 (which is no longer incentive compatible), but, for the same reasons as in the
trust contract, this strategy will lose money in expectation.
Note that for e > 4, c(e) 2. Hence, the marginal revenue of a unit increase in wages at e = 4 equals 10q(2/12) which
exceeds 1 for q > 12/20 = 6/10. An effort level e 4 can be implemented with an incentive contract at a lower risk of
suffering from inequality to the principal. Note also, that even if q > 0.6, the inequity averse principals need not offer
generous wages because they may still be afraid to suffer from the inequality caused by the selfish agents.
19
36
According to Proposition 1, trust contracts are very ineffective because it is too costly for the
principals to offer generous wages to induce non-minimal effort levels. In contrast, both the fair and
the selfish principal can enforce e = 4 in the optimal (accepted) incentive contracts. Hence, for the TI
treatment, in which the principals can choose between trust and incentive contracts, the following
proposition holds:
Proposition 3 [TI-Treatment]: (a) Both types of principals prefer incentive to trust contracts.
(b) Incentive contracts are more efficient and give a higher monetary payoff to the principal because
they elicit, on average, a higher effort level than trust contracts.
Thus, the main conclusion from the model of inequity aversion is the same as from the self-interest
model: incentive contracts outperform trust contracts. However, the inequity aversion model is
consistent with several observations in the TI treatment that are not consistent with the self-interest
model. First, it explains why low wage offers are frequently rejected, or, if they are accepted, why
agents often choose e=1 even if the contract is incentive compatible. Second, it predicts correctly
that incentive contracts are frequently associated with generous wages between 10 and 20 (offered by
fair principals). Finally, it offers an explanation as to why many agents choose effort levels larger
than 1 in response to generous wage offers in trust contracts and in incentive compatible incentive
contracts.
We now turn to the analysis of the BI treatment. For this purpose, we first examine the
equilibrium in a setting in which the principals can only choose a bonus contract. This gives us a
characterization of the equilibrium bonus contract. Then we compare this contract with the optimal
incentive contract from Proposition 2, which gives us the prediction for the case where the principals
can choose between a bonus and an incentive contract.
The analysis of the bonus contract is a little more complicated than the analysis of the TC or
the IC. The problem is, that the principal moves twice, first when he offers the contract and second
when he chooses which bonus to pay. Thus, the agents may take the contract offer as a signal about
the principals type and update the probability that a bonus will be paid. However, it can be shown
that no separating equilibrium exists in this signaling game and that both types of principals must
offer the same bonus contract in equilibrium. To see this suppose that the inequity-averse and the
selfish principal offer different contracts. In this case, the agents know from the contract offer
whether they face a fair principal or a selfish principal. If they face a fair principal they choose e =
37
10 because this principal pays a bonus that distributes the surplus equally. If they face a selfish
principal they choose e = 1. Hence, a selfish principal always wants to mimic the contractual offer of
the fair principal.
At the last stage of the game it is obvious that a selfish principal will not pay a bonus while a
fair principal pays a bonus that equalizes payoffs:
10e - w - b = w + b - c(e)
Using the implicit function theorem, we get
db 10 + c' (e)
=
de
2
If agents believe that there are q percent fair principals who choose b in this way while (1-q) percent
of the principals are self-interested and choose b=0, the expected monetary payoff of the agent as a
function of e is given by
MA(e) = q [w + b(e)-c(e)] + (1-q) [w-c(e)]
Differentiating with respect to e yields
dM A
db
10 + c' (e)
= q
c ' (e ) = q
c ' ( e)
de
de
2
This expression is positive if q is large enough compared to c'(e). Recall that, according to the cost
schedule in Table 1, 1 c'(e) 4. For c'=1, the critical value for q is 2/11 0.18, for c' = 2 it is 0.33,
for c' = 3 it is 0.46 and for c' = 4 it is 0.57. Hence, in a pooling equilibrium, where the agents believe
that they face a fair principal with probability q = 0.4, selfish agents will choose the maximal effort
level for which the marginal effort cost does not exceed 2, that is, they choose e = 7.
It is important to note that the theory implies that only self-interested agents choose
e = 7, hoping that they will be rewarded with a generous bonus payment by the principal. Fair agents
choose e = 1 or e = 2 depending on the wage offered (the fair agent chooses e = 2 iff w 10 ). The
reason for this interesting implication is that a fair agent suffers more than a self-interested agent if
he meets a selfish principal who does not pay the bonus. Hence, even if the selection of e = 7 is
profitable from a monetary perspective, a fair agent prefers e = 1 or e = 2 in order to ensure that
equality prevails and to avoid the disutility from disadvantageous inequality when the bonus is not
paid. Thus, the presence of fair principals induces selfish agents to choose high effort levels while
the presence of selfish principals induces the fair agents to provide low effort levels. This is an
38
interesting example of the sometimes surprising effects that arise in a heterogeneous population with
fair and selfish subjects.
There are many pooling equilibria in this game that differ in the unconditional base wage
(and therefore also in the surplus-sharing bonus to be paid ex post). If we impose the mild condition
on out of equilibrium beliefs that higher wage offers are not taken as a signal that the principal is
more likely to be selfish, then the set of pooling equilibrium outcomes shrinks to a singleton in which
all principals offer w=15.20 If the selfish agent chooses e=7, then a bonus of 25 just equalizes
payoffs. However, this bonus is paid only by the fair principals, so the expected bonus is 0.425=10.
The fair agents choose e=2; therefore, the average effort level is 0.67 + 0.42=5. We summarize our
results regarding the bonus contract in
Proposition 4 [Bonus contract]: (a) No separating equilibrium exists in which the selfish principal
offers a different contract from that which the fair principal offers.
(b) If a higher wage offer of the principal is not interpreted as a signal that the principal is more
likely to be selfish, then a unique pooling equilibrium outcome exists in which both types of
principals offer w=15. The selfish agent chooses e=7 and is rewarded by the fair principal with a
bonus of 25, while the selfish principal does not pay a bonus. The fair agent chooses e=2 and neither
type of principal pays a bonus.
Proposition 4 shows that the average effort level is higher in a bonus contract than in an optimal
incentive contract. Moreover, the selfish principals reap the benefits of this high effort level without
actually paying the bonus. Thus, the selfish principals can exploit the fact that there are fair
principals. It is, therefore, obvious that they earn more by offering bonus contracts. However, the fair
principals also earn more in the bonus contract because they share the surplus in the incentive and
the bonus contract equally but in the latter the surplus is higher. Thus we get:
Proposition 5 [BI-Treatment]: Both types of principals prefer the bonus relative to the incentive
contract.
20
Note that in a pooling equilibrium with bonus contracts wages cannot exceed 15. For w > 15,the surplus-sharing bonus
b(7) = 40 w < 25 and, hence, the expected bonus payment is 0.4b(7) which is less than the effort cost of c(7) = 10.
Therefore, the selfish agent will no longer be willing to provide e = 7.
39
Propositions 4 and 5 are in sharp contrast to the self-interest model, but they explain the
experimental results of the BI Treatment surprisingly well. Our fairness model not only predicts that
bonus contracts outperform incentive contracts; it also offers remarkably precise quantitative
predictions of the data. Recall from Figure 5a that the average wage is roughly 15 in the bonus
contracts while the average bonus is roughly 10. This is exactly the prediction provided in
Proposition 4. Moreover, the actual effort level in the bonus contracts is, on average, 5.2; the
predicted average effort level according to Proposition 4 is e = 5. In addition, the model predicts that
some principals pay a bonus while others dont a fact that we emphasized already in Section 4.2.
The above analysis also shows why the bonus contract is so much better than the trust
contract, even though both contracts rely on fairness as an enforcement device. Under a bonus
contract, the agent has to trust first by expending more than the minimum level of effort. Note that
even if the agent engages in the maximum effort of 10, the cost of this effort is relatively small
(c(10)=20). Thus, even if the principal does not pay the bonus, the agent does not lose too much. On
the other hand, under a trust contract the principal has to pay a generous wage upfront. In order to
induce a fair agent to expend the efficient level of effort (e=10) she has to offer w=60, i.e., the
principal risks a loss of 50 if the agent is not trustworthy. Therefore, it is much more risky for the
principal to appeal to the agents fairness than the other way round. This suggests that, as a general
principle, the player who loses less from trusting the other person should trust first.
Finally, we consider the EBI treatment briefly. The major fact that needs to be explained in
this treatment is the strong preference in favor of the pure bonus contract. Moreover, this preference
is probably associated with the fact that the effort is not significantly larger in the combined contract,
so that it is not worthwhile to pay the verification cost for the explicit incentive. The surprisingly low
enforcement power of the combined contract relative to what seems, in principle, enforceable is
also related to the much lower bonus incentives in this contract. Recall that the principals paid much
lower bonuses in the combined contracts.
Our fairness model is capable of rationalizing the principals preference for the pure bonus
contract. If the agents believe that those principals who propose a combined contract are more likely
to be selfish and, hence, pay no bonus, the bonus incentive in the combined contract is weakened
decisively. In fact, it is easy to show that if the agents take a combined contract offer as a signal that
the principal is likely to be selfish and if a higher wage is taken as a signal that the principal is more
likely to be fair, then the only equilibrium outcome in the EBI treatment is that the pure bonus
40
contract of Proposition 4 will be offered. However, this is not the only equilibrium. Other pooling
equilibria also exist and, in particular, there is an equilibrium in which all principals choose the
combined contract. In such an equilibrium the principals are capable of enforcing almost the
maximal effort level because they combine the power of the explicit and the implicit incentive in an
optimal way: both the selfish and the fair principals offer a combined contract (w=18, e*=9, f=13,
b*=30) which is accepted by both types of agents. The selfish agent chooses e = 9 while the fair
agent chooses e = 3. The selfish principal does not pay a bonus while the fair principal pays b = 30 if
e = 9 and b = 0 if e = 3.
To assess the plausibility of the different equilibria, one has to make a judgment about the
plausibility of the out-of-equilibrium beliefs that support the different equilibria. Here the fact that
those principals who offered the combined contract indeed paid lower bonuses becomes important.
This means that the agents had a strong reason to believe that these principals are more selfish. Once
this belief was established among the agents, the principals have an incentive for choosing the pure
bonus contracts. Therefore, we believe that the out-of-equilibrium beliefs, which support the pure
bonus equilibrium are the more reasonable ones.
6. Conclusions
Our experiments have shown that fairness concerns may have important consequences for the
optimal provision of incentives. Incentive contracts that are optimal when there are only selfish
actors perform less well when some agents are concerned about fairness. On the other hand, implicit
bonus contracts that cannot work when all actors are selfish provide powerful incentives and become
superior when there are also fair-minded players. Our results indicate that the principals understand
that fairness matters and predominantly choose the superior bonus contract that relies on fairness as
an enforcement device. Moreover, the principals even prefer the pure bonus contract over a contract
that combines the explicit incentive with the implicit bonus incentive. This result suggests that
fairness concerns may be one important reason why principals often do not use explicit incentives
although they are readily available. In this context we also observe intriguing interactions between
the explicit and the implicit incentive the use of the explicit incentive substantially weakens the
enforcement power of the bonus incentive. We conjecture that this is an important reason for the
principals preference in favor of the pure bonus contract.
41
There are several other points that deserve to be emphasized. First, the principals converge
towards the most efficient contract in the set of available contracts. This observation is important
because the efficiency principle provides the basis for much of modern contract theory. However,
it remains to be seen whether this observation extends to other more complicated environments.
Second, it is important to remember that only some subjects are concerned about fairness. A
considerable percentage of subjects also seems to be only interested in their own material payoff.
Whether fairness motives provide a good enforcement device depends on the percentage of fair
persons in the population and on the strategic situation in which the subjects interact. We have
shown that fairness concerns are too weak for contract enforcement in a setting where the trust
contract competes with the incentive contract while they are strong enough if the bonus contract
becomes available. This asymmetry in the impact of fairness concerns is due to the fact that it is less
costly for the agent to trust in the bonus contract than for the principal to trust in the trust contract.
Third, our theoretical results show that simple and tractable models of fairness can yield interesting
and non-obvious insights into the problems of contract choice and incentive provision. Our fairness
model is consistent with the major qualitative patterns in the data. In addition, it provides
surprisingly accurate quantitative predictions of the details of the bonus contract.
Finally, our experiments and the theoretical analysis show that the presence of fair types does
not automatically provide a solution to every contracting problem and may sometimes even
exacerbate incentive and contracting problems. Fair types are much more afraid of exploitative
situations in which the other party may take advantage of them. The reason is that they do not only
value their material payoffs but they also value the fairness of the opponents behavior and the equity
of the final outcome. For example, our theoretical analysis shows that self-interested agents respond
more strongly to the implicit incentives provided by a bonus contract than fair agents. The reason is
that in case that the principal does not reward the agent with a bonus a fair agent experiences
additional disutility from the unfairness of the behavior of the principal while the selfish agent
only suffers from the reduced material payoff. This shows that the presence of fair players may
complicate the task of incentive provision because in addition to the conventional incentive
compatibility constraints the fairness compatibility of the contract also has to be taken into
account.
To conclude, our experiments show that concerns for fairness have an important impact on
the actual and the optimal choice of contracts. Traditional contract theory has neglected these effects,
42
but they have to be taken into account if we want to fully understand the functioning of real world
contracts and the associated incentives schemes. Our theoretical analysis shows that it is possible to
model these effects explicitly and to develop richer models that may become part of a new paradigm
of behavioral contract theory.
43
References
Anderhub, Vital, Gchter, Simon and Knigstein, Manfred, 2002. Efficient Contracting and Fair
Play in a Simple Principal-Agent Experiment. Experimental Economics, 5, 5-27.
Berg, Joyce, Dickhaut, John W., and McCabe, Kevin A., 1995. Trust, Reciprocity, and Social
History. Games and Economic Behavior 10, 122-142.
Bolton, Gary E. and Ockenfels, Axel, 2000. A Theory of Equity, Reciprocity and Competition.
American Economic Review 100, 166-193.
Cabrales, Antonio, and Charness, Gary, 2003. Optimal Contracts, Adverse Selection, and Social
Preferences: An experiment. Mimeo, University of California at Santa Barbara.
Camerer, Colin and Weigelt, Keith (1988). Experimental Tests of a Sequential Equilibrium
Reputation Model, Econometrica, 56, 1-36.
Camerer, Colin F., 2003. Behavioral Game Theory: Experiments in Strategic Interaction, Princeton,
Princeton University Press.
Charness, Gary and Dufwenberg, Martin, 2003. Promises and Partnerhsips, Mimeo, University of
California at Santa Barbara.
Charness, Gary, and Rabin, Matthew, 2002. Understanding Social Preferences with Simple Tests.
Quarterly Journal of Economics, 117, 817-869.
Chaudhuri, Ananish, 1998. The Ratchet Principle in a Principal-Agent Game with Unknown Costs:
An Experimental Analysis, Journal of Economic Behavior and Organization, 37, 291-304.
Chiappori, Pierre-Andre and Salani, Bernard, 2003. Testing Contract Theory : A Survey of Some
Recent Work. In: M. Dewatripont, L.P. Hansen, S. Turnovski, Advances in Economic
Theory, Eigth World Congress of the Econometric Society, Vol. 1, Cambridge: Cambridge
University Press, 115-149.
Cooper, David J., Kagel John H., Lo W. and Qing Liang Gu (1999). Gaming against Managers in
Incentive Systems: Experiments with Chinese Students and Chinese Managers, American
Economic Review, 89, 781-804.
Cox, Jim and Friedman Daniel, 2002, A Tractable Model of Reciprocity and Fairness, Mimeo,
University of Arizona at Tucson.
44
DeJong, Douglas V., Forsythe Robert, Lundholm Russel J., and Uecker Wilfred C., 1985. A
Laboratory Investigation of the Moral Hazard Problem in an Agency Relationship, Journal
of Accounting Research 23, 81 120.
Dufwenberg, Martin and Kirchsteiger, Georg, in press. A Theory of Sequential Reciprocity.
Games and Economic Behavior.
Falk, Armin and Fischbacher, Urs, 1999. A Theory of Reciprocity. Working Paper No 6, Institute
for Empirical Research in Economics, University of Zrich.
Fehr, Ernst, Kirchsteiger, George and Riedl, Arno, (1993). Does Fairness Prevent Market Clearing?
An Experimental Investigation, Quarterly Journal of Economics 58, 437-460.
Fehr, Ernst, Gchter, Simon and Georg Kirchsteiger, 1997. Reciprocity as a Contract Enforcement
Device. Econometrica 65(4), 833-860.
Fehr, Ernst and Schmidt, Klaus M., 1999. A Theory of Fairness, Competition and Co-operation.
Quarterly Journal of Economics 114, 817-868.
Fehr, Ernst and Schmidt, Klaus M., 2003. Theories of Fairness and Reciprocity: Evidence and
Economic Applications, in: M. Dewatripont, L.P. Hansen, S. Turnovski, Advances in
Economic Theory, Eigth World Congress of the Econometric Society, Vol. 1, Cambridge:
Cambridge University Press, 208-257.
Gth, Werner, Klose, Wolfgang, Knigstein, Manfred and Schwalbach, Joachim, 1998. An
Experimental Study of a Dynamic Principal-Agent Relationship. Managerial and Decision
Economics, 19, 327-341.
Gth, Werner, Knigstein Manfred, Kovacs Judith, and Zala-Mezo Eniko, 2001. Fairness within
Firms: the Case of One Principal and Multiple Agents, Schmalenbach Business Review 53,
82 101.
Holmstrm, Bengt, 1982. Moral Hazard in Teams. Bell Journal of Economics 13, 324-340.
Keser, Claudia and Willinger, Marc, 2000. Principals Principles when Agents Actions are
Hidden, International Journal of Industrial Organization 18, 163-185.
Laffont, Jean J. and Tirole, J., 1993. A Theory of Incentives in Procurement and Regulation,
Cambridge MA: The MIT Press.
45
Levine, David, 1998. Modeling Altruism and Spitefulness in Experiments, Review of Economic
Dynamics 1, 593 - 622.
Prendergast, Canice, 1999. The Provision of Incentives in Firms. Journal of Economic Literature,
37, 7-63.
Rabin, Matthew, 1993. Incorporating Fairness into Game Theory and Economics. American
Economic Review, 83(5), 1281-1302.
Schotter, Andrew, Bull, Clive and Weigelt, Keith, 1987. Tournaments and Piece Rates: An
Experimental Study, Journal of Political Economy, 95, 1-33.
Schotter Andrew and Nalbantian Haig, 1997. Productivity under Group Incentives: An
Experimental Study, American Economic Review 87, 314 340.
Sobel, Joel (2002). Social Preferences and Reciprocity, mimeo, University of California, San Diego.
1165 Helge Berger and Carsten Hefeker, One Country, One Vote? Labor Market Structure
and Voting Rights in the ECB, March 2004
1166 Clemens Fuest and Martin Kolmar, A Theory of User-Fee Competition, March 2004
1167 Friedrich Schneider and Robert Klinglmair, Shadow Economies around the World:
What Do We Know?, April 2004
1168 Horst Raff and Nicolas Schmitt, Exclusive Dealing and Common Agency in
International Markets, April 2004
1169 M. Hashem Pesaran and Allan Timmermann, Real Time Econometrics, April 2004
1170 Sean D. Barrett, Privatisation in Ireland, April 2004
1171 V. Anton Muscatelli, Patrizio Tirelli and Carmine Trecroci, Can Fiscal Policy Help
Macroeconomic Stabilisation? Evidence from a New Keynesian Model with Liquidity
Constraints, April 2004
1172 Bernd Huber and Marco Runkel, Tax Competition, Excludable Public Goods and User
Charges, April 2004
1173 John McMillan and Pablo Zoido, How to Subvert Democracy: Montesinos in Peru,
April 2004
1174 Theo Eicher and Jong Woo Kang, Trade, Foreign Direct Investment or Acquisition:
Optimal Entry Modes for Multinationals, April 2004
1175 Chang Woon Nam and Doina Maria Radulescu, Types of Tax Concessions for
Attracting Foreign Direct Investment in Free Economic Zones, April 2004
1176 M. Hashem Pesaran and Andreas Pick, Econometric Issues in the Analysis of
Contagion, April 2004
1177 Steinar Holden and Fredrik Wulfsberg, Downward Nominal Wage Rigidity in Europe,
April 2004
1178 Stefan Lachenmaier and Ludger Woessmann, Does Innovation Cause Exports?
Evidence from Exogenous Innovation Impulses and Obstacles, April 2004
1179 Thiess Buettner and Johannes Rincke, Labor Market Effects of Economic Integration
The Impact of Re-Unification in German Border Regions, April 2004
1180 Marko Koethenbuerger, Leviathans, Federal Transfers, and the Cartelization
Hypothesis, April 2004
1181 Michael Hoel, Tor Iversen, Tore Nilssen, and Jon Vislie, Genetic Testing and Repulsion
from Chance, April 2004
1182 Paul De Grauwe and Gunther Schnabl, Exchange Rate Regimes and Macroeconomic
Stability in Central and Eastern Europe, April 2004
1183 Arjan M. Lejour and Ruud A. de Mooij, Turkish Delight Does Turkeys accession to
the EU bring economic benefits?, May 2004
1184 Anzelika Zaiceva, Implications of EU Accession for International Migration: An
Assessment of Potential Migration Pressure, May 2004
1185 Udo Kreickemeier, Fair Wages and Human Capital Accumulation in a Global
Economy, May 2004
1186 Jean-Pierre Ponssard, Rent Dissipation in Repeated Entry Games: Some New Results,
May 2004
1187 Pablo Arocena, Privatisation Policy in Spain: Stuck Between Liberalisation and the
Protection of Nationals Interests, May 2004
1188 Gnter Knieps, Privatisation of Network Industries in Germany: A Disaggregated
Approach, May 2004
1189 Robert J. Gary-Bobo and Alain Trannoy, Efficient Tuition Fees, Examinations, and
Subsidies, May 2004
1190 Saku Aura and Gregory D. Hess, Whats in a Name?, May 2004
1191 Sjur Didrik Flm and Yuri Ermoliev, Investment Uncertainty, and Production Games,
May 2004
1192 Yin-Wong Cheung and Jude Yuen, The Suitability of a Greater China Currency Union,
May 2004
1193 Ins Macho-Stadler and David Prez-Castrillo, Optimal Enforcement Policy and Firms
Emissions and Compliance with Environmental Taxes, May 2004
1194 Paul De Grauwe and Marianna Grimaldi, Bubbles and Crashes in a Behavioural Finance
Model, May 2004
1195 Michel Berne and Grard Pogorel, Privatization Experiences in France, May 2004
1196 Andrea Galeotti and Jos Luis Moraga-Gonzlez, A Model of Strategic Targeted
Advertising, May 2004
1197 Hans Gersbach and Hans Haller, When Inefficiency Begets Efficiency, May 2004
1198 Saku Aura, Estate and Capital Gains Taxation: Efficiency and Political Economy
Consideration, May 2004
1199 Sandra Waller and Jakob de Haan, Credibility and Transparency of Central Banks: New
Results Based on Ifos World Economicy Survey, May 2004
1200 Henk C. Kranendonk, Jan Bonenkamp, and Johan P. Verbruggen, A Leading Indicator
for the Dutch Economy Methodological and Empirical Revision of the CPB System,
May 2004
1201 Michael Ehrmann, Firm Size and Monetary Policy Transmission Evidence from
German Business Survey Data, May 2004
1202 Thomas A. Knetsch, Evaluating the German Inventory Cycle Using Data from the Ifo
Business Survey, May 2004
1203 Stefan Mittnik and Peter Zadrozny, Forecasting Quarterly German GDP at Monthly
Intervals Using Monthly IFO Business Conditions Data, May 2004
1204 Elmer Sterken, The Role of the IFO Business Climate Indicator and Asset Prices in
German Monetary Policy, May 2004
1205 Jan Jacobs and Jan-Egbert Sturm, Do Ifo Indicators Help Explain Revisions in German
Industrial Production?, May 2004
1206 Ulrich Woitek, Real Wages and Business Cycle Asymmetries, May 2004
1207 Burkhard Heer and Alfred Mauner, Computation of Business Cycle Models: A
Comparison of Numerical Methods, June 2004
1208 Costas Hadjiyiannis, Panos Hatzipanayotou, and Michael S. Michael, Pollution and
Capital Tax Competition within a Regional Block, June 2004
1209 Stephan Klasen and Thorsten Nestmann, Population, Population Density, and
Technological Change, June 2004
1210 Wolfgang Ochel, Welfare Time Limits in the United States Experiences with a New
Welfare-to-Work Approach, June 2004
1211 Luis H. R. Alvarez and Erkki Koskela, Taxation and Rotation Age under Stochastic
Forest Stand Value, June 2004
1212 Bernard M. S. van Praag, The Connexion Between Old and New Approaches to
Financial Satisfaction, June 2004
1213 Hendrik Hakenes and Martin Peitz, Selling Reputation When Going out of Business,
June 2004
1214 Heikki Oksanen, Public Pensions in the National Accounts and Public Finance Targets,
June 2004
1215 Ernst Fehr, Alexander Klein, and Klaus M. Schmidt, Contracts, Fairness, and
Incentives, June 2004