Do Incentives Destroy Voluntary Cooperation?
Transcript of Do Incentives Destroy Voluntary Cooperation?
Do Incentives Destroy Voluntary Cooperation?†
Simon Gächter*, Esther Kessler** and Manfred Königstein***
7 November 2010
PRELIMINARY VERSION, COMMENTS WELCOME!
Abstract
We investigate experimentally how explicit performance incentives in incomplete employment contracts interact with agents’ voluntary cooperation in one-shot and repeated gift-exchange experiments. If contracts are incentive compatible, agents choose their best-reply effort and there is no voluntary cooperation. By contrast, there is substantial voluntary cooperation if the contract is not incentive compatible. Experiencing incentive contracts reduces voluntary cooperation even after incentives are abolished. We then examine the robustness of these findings with regard to two variables: Experience with voluntary cooperation and the possibility of implicit incentives in repeated interactions. We find that incentive contracts have no lasting negative effects on voluntary cooperation if agents experience voluntary cooperation before being exposed to incentive contracts or if there are implicit incentives in repeated interactions. Implicit incentives increase voluntary cooperation strongly and make explicit incentives unnecessary or even counterproductive.
Key words: principal-agent games; gift-exchange experiments; incomplete contracts, explicit incentives; implicit incentives; repeated games; robustness.
JEL-Codes: M5, C7, C9
† This paper is part of the MacArthur Foundation Network on Economic Environments and the Evolution of Individual Preferences and Social Norms. Support from the EU-TMR Research Network ENDEAR (FMRX-CT98-0238) and from the Grundlagenforschungsfonds at the University of St. Gallen is gratefully acknowledged. We benefited from excellent research assistance by Christian Thöni and Eva Poen and from helpful comments in numerous workshops and seminars. Simon Gächter gratefully acknowledges the hospitality of the Institute for Advanced Studies at Hebrew University in Jerusalem while working on this paper. * University of Nottingham, School of Economics, Sir Clive Granger Building, University Park, Nottingham NG7 2RD, United Kingdom. Simon Gächter is also affiliated with CESifo and IZA. E-mail: [email protected]. **University College London, Institute of Child Health, Behavioural & Brain Sciences Unit, 30 Guilford Street, London WC1N 1EH, United Kingdom. E-mail: [email protected]. *** Universität Erfurt and IZA. Professur für Angewandte Mikroökonomie, Nordhäuser Str. 63, D-99089 Erfurt. E-mail: [email protected].
1. INTRODUCTION
Understanding the behavioral consequences of incentives is a fundamental goal in
economics. In this paper we contribute to this goal by investigating experimentally how
explicit and implicit incentives that give agents a selfish incentive to perform interact with
many people’s propensity to act against their self-interest. Numerous experiments and other
evidence in economics and the behavioral sciences demonstrates that many people are willing
to sacrifice their self-regarding motivations to benefit collective welfare (Fehr and
Fischbacher (2002); Camerer (2003); Gintis, et al. (2005); Bewley (2007); Henrich and
Henrich (2007)). Yet, in reality, many (potential) acts of voluntary cooperation do not take
place in isolation from situations that give people self-interested motivations to cooperate:
explicit and implicit incentives to cooperate are often present as well. The question we ask in
this paper is “What do explicit and implicit incentives imply for voluntary cooperation?”
There are at least two reasons why an answer to this question is important. First, on a
practical note, many contracts in reality are incomplete, which leaves important aspects
unregulated and therefore non-enforceable by third parties. Contractual incompleteness gives
agents an incentive to act opportunistically which may impair efficiency. Thus, many scholars
have argued that voluntary cooperation is necessary to ensure efficiency (Akerlof (1982);
Williamson (1985); Simon (1997); Bewley (1999); Bowles (2003); MacLeod (2007)).
Understanding interaction effects helps outlining the conditions under which voluntary
cooperation is likely to mitigate contractual incompleteness. Second, on a more fundamental
level, our question is on whether material interests and the ‘moral sentiments’ as expressed in
voluntary cooperation are separable, that is, whether incentives and voluntary cooperation are
independent of the levels of the other. If separability fails, incentives may be overused or
underused, and this has important implications for mechanism design (Bowles (2008);
Bowles and Hwang (2008); Bowles and Polanía Reyes (2010)).
We approach the separability issue on how incentives and voluntary cooperation
interact by means of laboratory experiments in the framework of gift-exchange games.1 The
gift-exchange game is a two-player game in which a principal (‘an employer’) offers a fixed
wage to an agent (‘an employee’). The agent can accept or reject the fixed wage offer. If the
agent accepts, he or she chooses an effort (or, equivalently, an output) level. Effort (output) is
costly for the agent and beneficial for the principal. Parameters are such that efficiency calls
1 We chose laboratory experiments for two reasons: (i) only the laboratory allows for the comprehensive investigation of all interaction effects we are interested in (Falk and Heckman (2009); Croson and Gächter (2010)) and (ii) controlling for selfish incentives, which will be crucial for our approach, is hardly feasible in the field.
1
for the maximal effort whereas selfish incentives induce the agent to provide the minimal
effort irrespective of the wage he or she has accepted (that is, there is no voluntary
cooperation). The results of numerous experiments refute this prediction and demonstrate the
relevance of voluntary cooperation – wages and effort are positively correlated even in one-
shot games.2 We will replicate this finding in a version of the gift-exchange game we call the
‘Trust game’. The results from the Trust games will provide the necessary benchmark for the
comparisons we are mainly interested in.
The explicit incentives take a simple form in our experiments: The principal offers a
contract that specifies a desired effort (= output) level and two compensation levels, one for
the case in which the agent delivers at least the contractually pre-specified output, and a
second (lower) compensation level if the agent’s output falls short of the desired output. We
design the set of feasible contracts such that the maximally enforceable output (by means of
incentive compatible contracts) is substantially less than the efficient output level. Thus, there
is room for efficiency-enhancing voluntary cooperation beyond the maximally enforceable
level. In general, we define voluntary cooperation as any effort the agent puts forward that
exceeds his or her selfishly-rational best-reply effort level. Our design also allows for an easy
distinction of incentive- and non-incentive compatible contracts; the latter are directly
comparable to Trust contracts which are non-incentive compatible by design.
One fundamental reason why separability may fail in this environment is the following:
Voluntary cooperation and incentive contracts operate on different psychological mechanisms
and contracts send messages that appeal to these mechanisms in a more or less coherent way.
The psychological sources of voluntary cooperation are motives like fairness and equity
concerns (Adams (1965); Akerlof (1982); Fehr and Schmidt (1999); Bolton and Ockenfels
(2000)), reciprocity (Rabin (1993); Levine (1998); Dufwenberg and Kirchsteiger (2004); Falk
and Fischbacher (2006)); let-down aversion (Dufwenberg and Gneezy (2000); Charness and
Dufwenberg (2006)); loyalty and goodwill (Simon (1991), Bewley (1999)); or social esteem
(Bénabou and Tirole (2006); Ellingsen and Johannesson (2008)). By contrast, explicit
incentives give the agent a self-regarding motive to perform. The general point is that
contracts convey messages about what the principal thinks the agent should do and what the
principal’s intentions are (Bénabou and Tirole (2003); Fehr and Rockenbach (2003); Falk and
Kosfeld (2006); Sliwka (2007); Bowles and Polanía Reyes (2010)). Trust contracts appeal to
the psychological sources of voluntary cooperation, whereas incentive-compatibly designed
2 For a recent overview of laboratory gift-exchange experiments see Fehr, et al. (2009). Evidence on gift exchange is not confined to the laboratory. See Gneezy and List (2006) and Falk (2007) for recent field experiments on gift exchange.
2
incentive contracts appeal to an agent’s rational self interest. Explicit incentive contracts that
are not designed incentive compatibly send an ambiguous message to the agent. Thus, the
predictions are that, on average, the amount of voluntary cooperation under Trust contracts
depends on the generosity of the contract: higher wages will lead to higher effort. By contrast,
incentive-compatible contracts will lead to incentive-compatible effort choices and little
voluntary cooperation beyond that, even if people would in principle be willing to provide
more effort. If this happens, we speak of a ‘short-run crowding out effect’. If the incentive
contract is not incentive compatible, we might observe ambiguous reactions: agents might
focus on the fact that the contract is an inconsistent appeal to the agent’s self-interest and
either shirk or respond to the perceived generosity of the contract.
A second reason why separability might fail is that explicit incentives change how
people interpret the nature of the relationship they are in (Bowles (1998); Gneezy and
Rustichini (2000); Bohnet, et al. (2001)). Since incentive contracts can be perceived as
appeals to self-interest, people might become more selfish. We test this argument by running
our experiments in two phases: in a first phase subjects act in an environment where incentive
contracts are possible and in a second phase the possibility of explicit incentives is removed:
only Trust contracts are feasible. If a spillover effect from the first phase to the second phase
exists and agents indeed shirk more after having been exposed to incentives, as compared to
having been exposed to Trust games in a control experiment, we speak of a ‘long-run
crowding out effect’.
A third reason for failure of separability might be related to how the explicit incentives
frame the decision for the agent (Liberman, et al. (2004); Salant and Rubinstein (2008);
Bowles and Polanía Reyes (2010); Dufwenberg, et al. (2010)). To study this aspect, we use
two incentive-equivalent forms of explicit incentives (from a selfishly-rational point of view):
in one set of games (called the ‘Bonus game’) principals can pay a bonus (‘a wage increase’)
if actual effort is at least as high as the desired effort, whereas in another set of games (the
‘Fine game’) principals can ask for a fine (a ‘wage deduction’) if the actual effort falls short
of the desired effort. A bonus might be perceived as a promise, and a fine as a threat, and
these perceptions might have different impacts on both short-run and long-run crowding out.3
Consistent with existing evidence (as surveyed in Bowles and Polanía Reyes (2010)),
our first set of experiments, reported in Section 4, provides evidence for all three sources of
3 There are theoretical arguments that frames in the context of social preferences might matter: Frames influence beliefs and beliefs influence motivations (see Dufwenberg, Gächter and Hennig-Schmidt (2010)). Empirically, the type of frames we study here may matter as well. In a natural field experiment using junior and senior academic economists as subjects, Gächter, et al. (2009) showed that labels akin to fine and bonus can have different behavioral consequences despite being the same in terms of the incentive structure.
3
failures of separability. Incentive-compatible contracts crowd out voluntary cooperation
almost entirely. If confronted with an incentive-compatible contract, agents choose their exact
best-reply effort in the majority of cases, under both Bonus and Fine contracts. By contrast,
both in the benchmark Trust games with no incentives and under non incentive-compatible
incentive contracts, we observe substantial voluntary cooperation. However, unlike in the
Trust games, the voluntary cooperation we observe under non-incentive compatible contracts
is unrelated to the offered wages; framing is unimportant for this result as well. We also
observe clear evidence for long-run crowding out: agents who have been exposed to incentive
contracts are significantly more likely to shirk in the subsequent Trust games.
After having established the importance of crowding out of voluntary cooperation in
our environment we investigate the robustness of our findings in two further sets of
experiments. Our second set of experiments, documented in Section 5, investigates the role of
experience with Trust contracts before being exposed to incentives. Thus, the first phase
consists of Trust games, the second phase of Bonus games or Fine games and the third phase
consists again of Trust games. We find again that incentive-compatible contracts lead to best-
reply effort choices in the majority of cases and hence very little voluntary cooperation. By
contrast to the first set of experiments, we find that (i) effort choices under non incentive
compatible contracts are now significantly positively correlated with the offered
compensation and (ii) the long-run crowding out effect has largely vanished. This result
demonstrates clearly that the perception of the contract is crucial for agents’ behavior: if
agents are experienced with Trust contracts, they can interpret the generosity of a non-
incentive compatible contract much better than when they lack this experience.
Our final set of experiments, which we report in Section 6, allows for both explicit and
implicit incentives. While in the first two sets of experiments we randomly changed pairings
between principal and agent to avoid confounds of crowding effects with strategic incentives,
we kept the pairs fixed in these experiments. This allows for implicit incentives coming from
repeated interactions with the same partner. Implicit incentives are arguably a very important
feature of many real life contractual relationships (e.g., employment relations). Our results
show a very strong positive effect of implicit incentives on voluntary cooperation. The
implicit incentives add to the voluntary cooperation we see in the one-shot experiment. There
is no long-run crowding out. In the repeated games, most contracts are designed in a non
incentive-compatible way. However, like in the first two sets of experiment, in those
contracts which are incentive compatible we find full crowding out of voluntary cooperation:
4
agents play their stage-game best replies in almost all instances. Thus, implicit incentives
cannot be added on top of explicit incentives.
While incentive-compatible contracts lead to incentive-compatible behavior we also
observe that non-incentive compatible contracts can be quite successful in inducing high
effort levels, in particular, when implicit incentives are present. We discuss this phenomenon
in detail in Section 6. Section 7 provides a concluding discussion.
2. DESCRIPTION OF GAMES AND BENCHMARK SOLUTIONS
2.1 The Games
We investigate three versions of principal-agent games based on the gift-exchange game
(Fehr, et al. (1993)). We summarize the games and parameters in Table 1. Each game consists
of three stages. The principal first offers a contract to the agent. In the Trust game the contract
comprises two components, a fixed wage w and a desired effort ed (effort can also be
interpreted as output in our games). The contract offer has to obey the restrictions
, in integers. 1 20 and -700 700de≤ ≤ ≤ ≤w
0
Second, the agent can accept or reject the offered contract. If he or she rejects the
contract, the game ends and both players earn nothing. If the agent accepts, he or she chooses
effort e in integers (where 1 ). The agent is not restricted by ed in his or her choice of
e. This reflects contractual incompleteness because ed is not enforceable. The stage game ends
after effort has been chosen.
2e≤ ≤
TABLE 1 GAMES AND PARAMETERS
Trust game Fine game Bonus game Wage Desired effort (=output) Performance incentive
w ∈ [-700, 700] ed ∈ [1,20]
w ∈ [-700, 700] ed ∈ [1,20]
f ∈ {0, 24, 52,80}
w ∈ [-700, 700] ed ∈ [1,20]
b ∈ {0, 24, 52,80} Agent’s payoff w – c(e) w – c(e) if e ≥ ed
w – c(e) – f if e < ed w – c(e) + b if e ≥ ed
w – c(e) if e < ed Principal’s payoff 35e – w 35e – w if e ≥ ed
35e – w + f if e < ed 35e – w – b if e ≥ ed
35e – w if e < ed Effort cost: c(e) = 7e – 7 Payoff if contract rejected: 0 for both
In the Trust game the principal’s profit is 35e – w and the agent’s profit is w – c(e) in
case of acceptance. Thus, effort induces a return that accrues to the principal and causes a
cost (disutility of effort) to be borne by the agent. The cost function is increasing and linear in
effort: c(e) = 7e – 7. Because w cannot be conditioned on effort, we refer to this game as the
5
‘Trust game’. It is a game of complete information. Each player knows the rules, including
the payoff functions for both players, and is informed about all choices made in the game.
In the Fine game the principal may punish the agent if effort falls short of ed. The
offered contract in the Fine game consists of w, ed and f, where f represents a fine (it can be
interpreted as a wage reduction). The principal can choose one of four fine levels: f ∈ {0, 24,
52, 80}. If e < ed, f is subtracted from the agent’s payoff and the principal’s wage bill is
reduced to w – f. If e ≥ ed, the fine is not imposed.
In the Bonus game the principal has the opportunity to reward the agent if the chosen
effort e is equal to or larger than the desired effort ed. Thus, the offered contract contains w, ed
and b, where b is a bonus (a wage increase) with b ∈ {0, 24, 52, 80}4. If e ≥ ed, the bonus is
added to the agent’s payoff and subtracted from the principal’s payoff. If e < ed, the bonus is
not due.
2.2 Benchmark Solutions for Rational Money Maximizers
Trust game: Since providing effort is costly and payment does not depend on effort, the
agent will choose e = emin = 1 irrespective of the offered wage. Anticipating minimal effort,
the principal offers the wage that just ensures the agent’s acceptance, namely w = 1 (or w = 0
if one assumes acceptance in case the agent is indifferent between acceptance and rejection).
These decisions result in payoffs of 34 money units for the principal and 1 money unit for the
agent. This solution is inefficient, since the efficient surplus is 567 at e = emax = 20.
Fine game and Bonus game: In choosing effort the agent has to consider two
alternatives, e = ed or e = 1. A higher effort than ed is suboptimal since it causes higher cost
without increasing payment. Conditional on e < ed, minimal effort e = 1 is best because fine
and bonus payments are independent of e. Hence, the optimal effort level is:
* if ( ) (1) ( ) or ( ) (1) ( ); 1 otherwise.
d d d de w c e w f c f c e w b c e w c b c ee
⎧ − ≥ − − ⇔ ≥ + − ≥ − ⇔ ≥= ⎨⎩
d
(1)
Notice that the best-reply efforts are the same in the Fine game and the Bonus game; any
behavioral difference for a given contract is therefore due to a framing effect.
The agent’s best reply function (1) is the incentive-compatibility constraint for the
principal’s contract design problem. For each level of f or b there exists a maximal level of
desired effort that satisfies f, b ≥ c(ed). The maximally enforceable effort is 12. Before
4 In our setup the bonus (as well as the fine) is enforceable. This is in contrast to Fehr, et al. (2007), where paying the bonus is at the discretion of the principal after the agent has chosen the effort.
6
choosing effort the agent has to accept an offered contract. This requires that the agent has
non-negative earnings (because the outside option is 0). With the parameters from Table 1 it
is optimal for the principal to set w such that the agent is just compensated for his or her effort
cost c(e*); furthermore, the solution to the principal’s problem is f, b = 80, ed = 12 and wf =
c(12) = 77 or wb = b – c(12)= –3 (where wf (wb) denotes the wage in the Fine (Bonus) game).
Accordingly, the agent will accept the contract and choose e = 12. The solution is more
efficient than the solution without incentives but it does not generate the maximal surplus. It
induces a rather asymmetric distribution of payoffs: 343 money units for the principal and 0
money units for the agent.
As we have just shown, the maximally enforceable effort under incentive-compatible
contracts is 12 in both the Fine game and the Bonus game. This feature is crucial for our
purposes because it allows for voluntary cooperation beyond what incentives can achieve.
Moreover, it reflects contractual incompleteness that characterizes many real life contracts,
even if some aspects of the contract can be contractually regulated. By allowing for different
fine and bonus levels (including zero) we give the principal the possibility to set the strength
of the incentives he or she wants to apply to the agent.5 Moreover, different combinations of
f, b, and ed that satisfy the incentive-compatibility constraint can induce different best-reply
efforts and this variation allows for a sharper test of whether agents choose best-reply efforts
than a more restricted (e.g., binary) set would have allowed for.
3. RESEARCH QUESTIONS, EXPERIMENTAL DESIGN AND PROCEDURES
3.1 Research Questions and Experimental Design
In the experiment the participants played two, or respectively three, blocks (‘phases’) of
ten games. Table 2 lists our research questions and all respective treatments. Table 2 also
indicates the respective matching procedure. In the first two sets of experiments subjects were
matched randomly in each round. In the third set of experiments subjects were matched with
the same player during the whole experiment (‘Partner’ matching). Within a treatment (the
rows) our experiments applied a within-subject design because subjects participated in
multiple phases. Across the eight different treatments we employed a between-subject design.
Subjects kept their role as principal or agent throughout the experiment.
The first treatment is TTT, which serves as our main benchmark. In these experiments
subjects play three phases where each phase comprises ten one-shot Trust games played in
5 We included zero because there is evidence that deliberately abstaining from using incentives when incentives could have been used induces more cooperation (Fehr and Rockenbach (2003)).
7
randomly matched pairs. If effort is higher than predicted according to the benchmark
solution (i.e., e > e*), we refer to this as ‘voluntary cooperation’. Based on previous
experiments on the gift-exchange game we predict that the fixed wage will be positively
correlated with effort.
Against this benchmark we conduct three sets of experiments that correspond to the
three main purposes of our experiments. The first set of experiments (panel A in Table 2)
aims at measuring (i) the impact of explicit incentives on effort choices; (ii) the role of
framing (fines vs. bonuses); and (iii) the short- and long-run crowding effects of voluntary
cooperation that are neither confounded with strategic incentives, nor with prior experience of
Trust contracts before being exposed to incentive contracts. Therefore, subjects play one-shot
experiments in two phases of ten periods each. In Phase 1 principals can design either Fine or
Bonus contracts, whereas in Phase 2 only Trust contracts are feasible. We refer to these
treatments as the FT or BT treatment.
TABLE 2
MAIN RESEARCH QUESTIONS, EXPERIMENTAL DESIGN, NUMBER OF SUBJECTS AND NUMBER OF MATCHING GROUPS
Treatment label Phase 1 (Period 1-10)
Phase 2 (Period 11-20)
Phase 3 (Period 21-30)
# Subjects # independent matching groups
0. Establishing a benchmark of voluntary cooperation.
TTT Trust Trust Trust 78 6 A. Establishing the effects of explicit incentives: Are there short-run and long-run crowding effects? Does the framing of incentives matter?
FT Fine Trust - 80 6 BT Bonus Trust - 78 6
B. Introducing incentives after experiencing Trust contracts: What is the impact of explicit incentives on short-run and long-run crowding effects? Does the framing of incentives matter?
TFT Trust Fine Trust 86 6 TBT Trust Bonus Trust 84 6
C. How do implicit incentives available in repeated interactions influence voluntary cooperation, the use of explicit incentives and short-run and long-run crowding effects? Does the framing of incentives matter?
TTT-R Trust Trust Trust 24 12 TFT-R Trust Fine Trust 36 18 TBT-R Trust Bonus Trust 34 17
The second set of experiments (panel B in Table 2) investigates the relevance of prior
experience of Trust contracts before being exposed to Incentive contracts. All subjects play
three phases of ten one-shot games with random matching. Phase 1 and Phase 3 consist of
Trust contracts, whereas Phase 2 allows for either Fine or Bonus contracts. We refer to these
experiments are TFT or TBT treatments. These experiments are interesting for two reasons.
8
First, we can investigate the impact of experience of Trust contracts on the robustness of
short- and long-run crowding. Second, it is also of practical relevance to understand what
happens to voluntary cooperation if incentive contracts are introduced into an existing Trust
environment.
The final set of experiments (panel C in Table 2) allows for both explicit and implicit
incentives. Participants play either the TTT, the TFT or the TBT experiments in fixed pairs
(‘Partner’ matching) for three phases of ten periods each. We refer to these experiments as
TTT-R, TFT-R and TBT-R, where the suffix ‘R’ stands for repeated interaction. Although
subjects are aware that they play finitely repeated games there are theoretical and empirical
reasons why there are implicit (that is strategic) incentives to cooperate: First, if selfishness
and rationality are not common knowledge cooperation can be sequentially rational (Kreps, et
al. (1982)). Second, boundedly rational play can also lead to cooperation (Selten and Stoecker
(1986)). Third, previous experimental evidence suggests that cooperation in the finitely
repeated gift-exchange game is higher than in one-shot games (Falk, et al. (1999)).
We are now in a position to give a precise definition of crowding effects. Because
contracts can obey or violate incentive compatibility, we also have to distinguish these cases
in our definition of crowding effects.
Definition 1 (short-run crowding):
(a) Contracts are incentive-compatible (“IC”): A short-run crowding effect occurs in
Phase 1 of treatment FT, BT if * ,
, ,( 1) ( )d d
Trust Fine BonusICw e w e
e e e− ≠ − where e is the
average effort (over all ten periods) in Phase 1 of the Trust game and the Fine or
Bonus game, respectively.
(b) Contracts are not incentive-compatible (“NIC”): A short-run crowding effect occurs
in Phase 1 of treatment FT if , ,
( ) | 0dw e fTrust Fine
NICe e− ≠ , and in BT if
, ,( ) | 0dw e b
. Trust BonusNICe e− ≠
Thus, for establishing short-run crowding effects in our first set of experiments we
compare Phase 1 of FT, BT with Phase 1 of TTT, holding the offered contracts constant.
Analogously, measuring short-run crowding effects in TFT and TBT (or TFT-R and TBT-R)
requires comparing Phase 2 of TFT and TBT (or TFT-R and TBT-R) with Phase 2 of TTT (or
TTT-R), holding the offered contracts constant.
9
Establishing a long-run crowding effect requires measuring how under Trust contracts
the prior experience of Incentive contracts influences effort, compared to a prior experience
of Trust contracts:
Definition 2 (long-run crowding): Long-run crowding occurs if AFTER AFTER /
,( ) ≠ 0d
Trust Trust Trust Fine Bonus
w ee e− , where e indicates the average effort (across all
periods of the respective phase).
Thus, measuring long-run crowding effects requires comparing Phase 2 of TTT with
Phase 2 of FT, BT, and Phase 3 of TTT with Phase 3 of TFT, TBT, and TFT-R and TBT-R,
holding the offered contract constant.
3.2 Procedures
We conducted 20 sessions at the University of St. Gallen with a total of 500 participants
(first-year undergraduates of business, economics or law). We recruited our subjects from a
large data base of people who volunteered to participate in experiments. We drew a random
selection from the data base and invited them by email. In a typical session 28 participants
were present at the same time.
After arrival at the lab, participants first had to read instructions (see Appendix A). The
instructions were the same for principal and agent. Subjects also had to answer a set of
control questions to test their understanding of payoff calculations. The experiment did not
start before all participants had answered all questions correctly. All participants could ask
questions privately. Role assignment was random and fixed throughout the session. We
explained that all decisions would be anonymous during the whole experiment. A short
summary of the instructions was read aloud by the experimenter. At the beginning of each
session we told subjects that there would be different parts and that they would learn about
them one after the other.
The experiments were computerized and conducted with the software ‘z-Tree’
(Fischbacher (2007)). Participants were separated from each other by partitions and matched
anonymously via a computer network. In sessions with random matching (all treatments
except TFT-R and TBT-R and TTT-R) we formed two independent matching groups of 14
subjects each. Subjects were not informed about the matching groups but only that they
would be randomly matched with another player present in the room. Subjects also never
10
learned the identity of their opponent players. Each session lasted about two hours and the
subject's average earnings were about CHF 45 (about €30).
4. RESULTS I: GAUGING CROWDING EFFECTS
4.1 The TTT Benchmark: Voluntary Cooperation under Trust Contracts
In our benchmark we are in particular interested in how actual effort depends on the
offered wage and how this relationship changes over the up to thirty periods we have planned
for our main experiments. Based on previous evidence on closely related variants of our Trust
game (e.g., Fehr, Kirchsteiger and Riedl 1993; Fehr, Gächter and Kirchsteiger 1997; as well
as the psychological sources of voluntary cooperation discussed above) we predict that wage
and effort will be positively correlated. We expect these effects at least for the first ten
periods (Phase 1), since this is about the duration of most previous gift-exchange
experiments. The fact that we observe behavior in three phases allows us to investigate the
role of experience for the stability of voluntary cooperation. Figure 1 is a scatter plot of fixed
wage and effort choices in each of the three phases of treatment TTT; each dot is a single
observation.6
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Fixed wage
Trust
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Fixed wage
Trust
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Fixed wage
Trust
FIGURE 1. Relation between offered wage and actual effort in each of the three Phases of TTT.
6 Figure 1 shows the relation between actual effort and fixed wage for wages between 0 and 450. We had a few wages above 450 (up to wages of 700), but they comprised less than 2 percent of cases. For expositional clarity we therefore restrict our attention in this (and subsequent figures) to wages up to 450. In the econometric analyses we include all accepted contracts, not just those with wages up to 450. The lines in Figure 1 (as well as in all subsequent scatter plots) are locally weighted regressions of actual effort e on wage for e > 1.
11
In line with previous evidence Figure 1 shows extensive voluntary cooperation (e > e*)
and a clearly positive correlation between fixed wage and actual effort in a large number of
choices. The wage-effort relationship appears to be stable across all three phases.
While effort is increasing in fixed wages for a large fraction of the observations, there is
another fraction of observations with e = 1 independent of the wage. It seems that the
population of players can be partitioned into two subpopulations: a subpopulation that is
responsive to different levels of fixed wage and a subpopulation that does not respond to it.
This data pattern appears for other treatments of our experiment as well (see below). It
motivates our econometric model: instead of applying a simple tobit regression model we rely
on hurdle models (see Wooldridge (2002), p. 536). This allows for the possibility that the
decision for e > 1 versus e = 1 (the decision to pass the hurdle) follows different rules than the
choice of e conditional on e > 1. Thus, in a first step we estimate a probit regression for the
choice of e > 1 (value = 1) or e = 1 (value = 0), and in a second step we estimate a tobit
regression with an upper bound of 20 for e | e > 1. This is similar to the Heckman two-step
estimation procedure, just that in step two we apply a tobit regression rather than OLS to
account for censoring at e = 20. To account for conditioning on e > 1 choices we include the
inverse Mills ratio into the tobit regressions. We will use a hurdle approach for all estimations
reported below.
Table 3 shows the regression results for the three phases of TTT. The two explanatory
variables in Table 3 are the two elements of a contract offer in a Trust contract – the fixed
wage and the desired effort.
For Phase 1 Table 3 reports estimated coefficients for the influence of fixed wage of
0.003 in the probit regression and 0.086 in the tobit regression, with both being highly
statistically significant. This means that the probability of choosing above-minimal effort
increases in fixed wage and that effort conditional on e > 1 increases in effort as well.
Evaluating the two estimated models (probit and tobit) at mean values for all explanatory
variables predicts that minimal effort (e = 1) is chosen with a probability of 0.415. With
residual probability 0.585 players choose a higher effort, and this higher effort conditional on
e > 1 is predicted at 8.7. The two parts of the regression analysis can be put together to
provide a predicted value of effort E(e) with7
( ) ( ) ( ) ( )1 1 (1– 1 ) | 1 5.5E e prob e prob e E e e= = ⋅ + = ⋅ > = .
7 One might argue that the calculation of a predicted value of effort could have been done more simply by running a tobit regression with bounds at 1 and 20. However, given the distributions shown in Figure 1 our approach seems more adequate.
12
We refer to this as the total effect. Similarly, in all regression analyses below we may
consider the partial effects of an explanatory variable on prob(e=1) and e|e>1 separately or
combined (total effect).
TABLE 3 THE WAGE-EFFORT RELATIONSHIP IN THE THREE PHASES OF THE
BENCHMARK TTT EXPERIMENTS
Phase 1 Phase 2 Phase 3 probit
(e>1) tobit
(e|e>1) probit (e>1)
tobit (e|e>1)
probit (e>1)
tobit (e|e>1)
Fixed wage 0.003*** 0.086*** 0.006*** 0.054*** 0.005*** 0.050*** (0.000) (0.027) (0.002) (0.009) (0.001) (0.005) Desired effort 0.038*** 1.236*** -0.004 0.140 0.006 0.181 (0.013) (0.401) (0.020) (0.089) (0.027) (0.116) Inverse Mills ratio 39.156** 5.457*** 8.509** (17.006) (1.721) (3.546) Constant -1.015*** -56.877** -1.105*** -8.717*** -0.888*** -11.399** (0.208) (24.151) (0.175) (2.445) (0.147) (4.571) No. of obs. 316 142 284 116 281 123 LR chi2 57.42*** 111.83*** 31.42*** 313.33*** 88.29*** 1544.94*** probit indicates a probit model whether e>1 or e=1. tobit indicates a tobit regression on e>1 choices only (censored at 20). Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
Table 3 reports that fixed wages have a significantly positive effect on effort (partially
and combined) for phases 2 and 3 as well. Overall, the regression model supports our
impression from Figure 1 that (1) there is voluntary cooperation, (2) voluntary cooperation is
reciprocity based, and (3) voluntary cooperation is stable over time. These behavioral patterns
under Trust contracts will serve as basis for comparison in other treatments.8
4.2 Monetary Incentives and Short-run Crowding Effects
We now look at the Phase 1 data of treatments FT, BT to investigate the influence of
Fine contracts and Bonus contracts (see panel A in Table 2). Since subjects decide under
monetary incentives without having experienced Trust contracts before we refer to them as
inexperienced subjects.
Figure 2 is a scatter plot of actual effort choices against effort predicted by the
theoretical solution e* for FT (left panel) and BT (right panel). The size of dots in Figure 2 is
proportional to the number of underlying observations (a “bubble plot”). Figure 2 shows a
highly structured pattern which is very similar in both panels. Many observations cluster 8 Since desired effort is part of an offered contract our regression model also controls for desired effort. Table 3 reports a significant influence of desired effort in Phase 1, but not in later phases. Although we find that fixed wage and desired effort are positively correlated in all phases (all correlations exceed 0.70) we conclude that it is mainly the fixed wage level that drives behavior under Trust contracts. Desired effort plays a minor role.
13
exactly on the 45°-line (where e = e*). Provided the principal offers an incentive-compatible
contract (henceforth IC-contract; offered in 69.4 percent of all cases in FT, and in 67.6
percent of cases in BT), the agent chooses best-reply effort in 69.7 (71.7) percent of the cases
in FT and BT respectively.9 Given an IC-contract (e* > 1), if agents deviate from best-reply
effort, they tend to choose minimal effort (e =1). Effort levels at e = e* or e = 1 explain 93.1
percent of all effort choices. There is a crowding-out of voluntary cooperation, since only
very few effort choices exceed e* and, if contracts are not incentive-compatible (henceforth
NIC-contract; which implies e* = 1, like under Trust contracts), many agents choose a
substantially higher than minimal effort level (36.0 percent of the choices).
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 1 FT
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 1 BT
FIGURE 2. Relation between actual effort and best-reply effort in Phase 1 of FT (left panel)
and Phase 1 of BT (right panel). The size of dots is proportional to the number of underlying observations. The horizontal line at 12 indicates the maximally enforceable effort level under incentive-compatible contracts.
Although only positive deviations are feasible, it is nevertheless interesting to note that
effort above 12 – which according to equation (1) is the maximal level that can be enforced
via incentives – is relatively more frequent for e* = 1 than for e* > 1. Thus, NIC-contracts
may allow for effort levels that are largely infeasible with IC-contracts.
Figure 3 looks closer at NIC-contracts by showing a scatter plot of chosen effort against
offered compensation, with offered compensation defined as w – f in Fine games and w in
Bonus games.10 Unlike Figure 1 we do not observe a clear positive correlation between effort
9 The variation in best-reply efforts visible in Figure 2 is mainly due to desired effort levels rather than fines and bonuses. In FT the percentages of cases of fines of 80, 52, 24, 0 are 85.5, 8.9, 4.0, and 1.6 percent, respectively. In BT percentages for the respective bonus levels are 73.4, 9.0, 12.8, and 4.8, respectively. These distributions are significantly different from each other (χ2-test, 26.250, p=0.000). 10 We look at offered compensations rather than the fixed wage only because NIC-contracts give agents a selfishly rational incentive to shirk in which case they have to pay the fine or they forego the bonus.
14
and offered compensation here. It appears that the presence of an NIC Bonus or Fine contract
destroys reciprocity, but it does not entirely crowd out voluntary cooperation because a
substantial number of effort choices are non-minimal.
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Offered compensation (w-f)
Phase 1 FT-NIC
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Offered compensation (w-f)
Phase 1 BT-NIC
FIGURE 3. Offered compensation and effort under NIC-contracts in Phase 1 of FT and BT.
The impressions we get from these figures are supported by the econometric model
reported in Table 4. It is a hurdle model that uses the Phase 1 data of FT, BT and TTT and
estimates the influences of fixed wage, monetary incentives and other variables for different
contract types (Trust contracts, IC-contracts and NIC-contracts) within a single regression
model. In a first step we estimate a probit regression for the choice of e > 1 (value = 1) or e =
1 (value = 0). The left column of Table 4 shows the estimation results. In a second step we
estimate a probit regression for the choice of e = e* (value = 1) or e ≠ e* (value = 0)
conditional on e > 1. And in a third step we estimate a tobit regression (with an upper bound
of 20) for effort conditional on e > 1 and e ≠ e*. In steps 2 and 3 we include the inverse Mills
ratio as regressor.11 The middle and the right column of Table 4 report the results.
11 Note that under incentive compatible contracts there are a few observations that choose e > 1 and e ≠ e*. For these observations step 3 should be conditioned on both, steps 1 and step 2. Since these are very few observations in the step 3 estimation, we think this is negligible and for simplicity we condition only on step 1.
15
TABLE 4
EXPLAINING PHASE 1 EFFORT CHOICES IN TREATMENTS TTT, FT AND BT
probit (e>1) probit (e=e*|e>1)
tobit (e|e>1, e≠e*)
Fixed wage (TTT) 0.003*** 0.022*** (0.001) (0.006) Desired effort (TTT) 0.038* 0.248*** (0.020) (0.084) Fixed wage (BT, IC) 0.007*** 0.019 (0.002) (0.015) Bonus b (BT, IC) 0.021*** 0.040*** (0.007) (0.008) Fixed wage (FT, IC) 0.005*** 0.016 (0.002) (0.011) Fine f (FT, IC) 0.016** 0.039*** (0.007) (0.009) Desired effort (BT/FT, IC) -0.143*** 0.019 (0.035) (0.062) Dummy IC 1.280*** 0.745 (0.440) (3.692) Fixed wage (BT, NIC) 0.002 0.010** (0.001) (0.005) Bonus b (BT, NIC) 0.011** 0.055*** (0.004) (0.013) Fixed wage (FT, NIC) 0.003** 0.007 (0.001) (0.007) Fine f (FT, NIC) -0.000 0.051* (0.005) (0.026) Desired effort (BT/FT, NIC) -0.082*** -0.110 (0.031) (0.266) Dummy NIC 0.982*** 7.064** (0.365) (2.938) Inverse Mills ratio -0.029*** -3.273 (0.354) (2.286) Constant -1.015*** -1.947*** 8.043 (0.148) (0.444) (5.471) No. of obs. 988 566 253 χ2=222.29*** χ2=233.44*** F=22.89*** Pseudo R2 0.190 0.645 0.066 probit (e>1)indicates a probit model whether e>1 or e=1. probit (e=e*|e>1) indicates a probit model whether e=e* (best-reply effort) or e≠e*. tobit (e|e>1, e≠e*) indicates a tobit regression with an upper bound of e=20. IC (NIC) indicates (non)incentive-compatible contracts. Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
The upper part of Table 4 shows the estimated effects of the variables fixed wage (TTT)
and desired effort (TTT) on behavior in Trust games.12 This reproduces the findings from
Table 3. The middle panel shows the estimated influences of IC Bonus and Fine contracts. For
12 Most variables in this table are defined as treatment-specific variables. For example, Fixed wage (Trust games) has value = fixed wage when treatment = TTT and value = 0 when treatment = BT or FT. Similarly for desired effort (Trust games). Fixed wage (BT, IC) has value = fixed wage when treatment = BT and if the contract is incentive compatible and value = 0 otherwise. And so on.
16
example, the significant coefficients 0.021 (Bonus) and 0.016 (Fine) reported in the left
column imply that prob(e > 1) is increasing in the size of the bonus (fine) offered in the
contract. The fixed-wage influence upon prob(e > 1) is positive as well and is estimated as
0.007 for fixed wage (BT, IC) and 0.005 for fixed wage (FT, IC). The influence of desired
effort (BT/FT, IC) is estimated as -0.143, meaning that asking for higher effort decreases
prob(e>1), ceteris paribus. All of these effects are plausible in sign and statistically
significant. In the middle column we find that the probability of choosing a best-reply effort
(provided e > 1) is significantly increasing in the bonus (0.040) and the fine (0.039) while
desired effort is insignificant. Furthermore, the third column reports results on effort
conditional on e > 1 and e ≠ e*. We find insignificant effects of fixed wages and an
insignificant coefficient of dummy IC (a variable that has value 1 for IC-contracts and 0
otherwise).
To understand the total effect of IC Bonus or Fine contracts on effort one has to consider
the partial effects shown in the middle panel together. This is illustrated in the left panel of
Figure 4 which is a predicted value graph based on the regression model. The predicted values
E(e) are calculated as
( ) ( ) ( ) )( ) ( )
1 1– * 1 ( | 1 and *
with 1 and * | 1 .
E e p p q e q E e e e e
p prob e q prob e e e
= ⋅ + ⋅ ⋅ + − ⋅ > ≠⎡⎣= = = = >
]
For calculating the predicted efforts we fix the desired effort at 12 and both the bonus
and fine at 80. Figure 4 shows predicted values for different contract types and for varying
compensation levels holding other influences constant.13
The right panel of Figure 4 shows that effort levels up to 12 can be implemented at
much lower compensation levels with incentive-compatible bonus or fine contracts than with
trust contracts. However, effort above 12 seems infeasible with IC-contracts. Increasing
compensation beyond 300 does not lead to substantially higher effort. This is different with
trust contracts were the regression model predicts further increases in effort even beyond 12.
Bonus contracts perform somewhat better than fine contracts, but the difference (framing) is
relatively minor.
The bottom panel of Table 4 reports the regression results for non-incentive compatible
bonus and fine contracts. The left panel of Figure 4 shows predicted value graphs analogous
to the right panel of Figure 4. While non-incentive compatible contracts by and large perform
worse than IC-contracts, there is substantial voluntary cooperation since predicted effort is
13 For Trust contracts and IC Fine contracts, compensation is simply equal to fixed wage. For an IC Bonus contract, compensation is equal to fixed wage plus bonus.
17
above the minimal level. Furthermore, effort is increasing in compensation (reciprocity-based
voluntary cooperation), but the influence of compensation is not as strong as within trust
contracts. We will further investigate the performance of NIC-contracts below.
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT FT-ICBT-IC
Phase 1 TTT, FT-IC, BT-IC
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT FT-NICBT-NIC
Phase 1 TTT, FT-NIC, BT-NIC
FIGURE 4. Predicted effort as a function of offered compensation (based on estimations in
Table 4, for ed = 12, and b, f = 80.
We summarize our findings in our first main result.
Result 1 (a) If it is possible to design a bonus or fine contract, most subjects in the role
of the principal design a contract that exhibits a non-zero bonus or fine. Most of the bonus
and fine contracts are incentive compatible (67.4 percent,) and IC-contracts induce exact
best-reply effort in the majority of cases (73 percent). Given an IC-contract we find no
voluntary cooperation. (b) IC-contracts can induce effort levels up to 12 quite effectively and
at lower compensation levels than Trust contracts and NIC-contracts. Yet, effort above 12 is
infeasible with IC-contracts while it is feasible with Trust contracts. (c) NIC-contracts are
relatively infrequent and perform worse than IC-contracts at low compensation levels, and
they are worse than Trust contracts at high compensation levels. Nevertheless there is some
voluntary cooperation. (d) The framing of incentives (bonus versus fine) is relatively
unimportant in particular when contracts are incentive compatible.
18
4.3 Long-Run Crowding Effects
Figures 5a and 5b now turn attention to the long-run crowding effect. Specifically we
investigate behavior under the condition of a Trust contract in Phase 2 after experiencing
Bonus or Fine contracts in Phase 1. Figure 5a shows the development of efforts over time and
Figure 5b provides a scatter plot of effort and fixed wages. The left-hand panel of Figure 5a
depicts the average effort observed in Phase 1, separately for TTT, BT and FT, and without
distinguishing whether the contract is incentive compatible or not. On average Bonus and
Fine contracts clearly lead to higher effort levels than Trust contracts. The right panel of
Figure 5a shows the development of effort in Phase 2, when there is no Bonus (Fine) contract
any longer. For BT and FT the average effort level drops from more than 6 to about 2. By
contrast, for TTT the average effort in Phase 2 is around 4.
Figure 5b shows scatter plots of effort (in Phase 2) against fixed wage for FT (left panel)
and BT (right panel). Visual comparison with Phase 2 of Figure 1 suggests that voluntary
cooperation is lower after the experience of Bonus (Fine) contracts. This is confirmed by the
regression model shown in Table 5 and the predicted value graphs in Figure 6 (based on
desired effort of 12). The regression model uses the Phase 2 data of BT, FT and TTT. It is
similar to the model shown in Table 3 except for introducing dummies for BT and FT.
Accordingly, having experienced Fine contracts in Phase 1 has a significantly negative effect
on prob(e >1) in Phase 2. Having experienced Bonus contracts also has a significantly
negative effect on e|e>1. Thus, voluntary cooperation in Phase 2 is lower after experiencing
Bonus (Fine) contracts rather than Trust contracts in Phase 1. The differences between FT and
BT are insignificant.14
02
46
810
12A
ctua
l effo
rt
1 2 3 4 5 6 7 8 9 10Period
TT FT BT
Phase 1 (period 1-10)
02
46
810
12A
ctua
l effo
rt
11 12 13 14 15 16 17 18 19 20Period
TT FT BT
Phase 2 (period 11-20)
FIGURE 5A. Mean effort levels over time in Phase 1 and 2 of TTT, FT and BT.
14 A test that compares the coefficients of ‘Dummy BT treatment’ with ‘Dummy FT treatment’ shows that they are not significantly different from each other (χ2-test, 0.00, p=0.964).
19
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Fixed wage
Phase 2 FT
05
1015
20A
ctua
l effo
rt
0 100 200 300 400Fixed wage
Phase 2 BT
FIGURE 5B. Wages and effort in Phase 2 of FT and BT.
TABLE 5
LONG-RUN CROWDING OUT (COMPARING PHASE 2 DATA IN FT, BT, AND TTT)
probit (e>1)
tobit (e|e>1)
Fixed wage 0.006*** 0.038*** (0.001) (0.012) Desired effort -0.008 0.097** (0.012) (0.047) Dummy BT treatment -0.295 -1.195* (0.210) (0.696) Dummy FT treatment -0.616*** -1.150 (0.135) (1.589) Inverse Mills ratio 2.346 (3.156) Constant -0.980*** -2.478 (0.135) (4.83) No. of obs. 876 215 LR chi2 123.57*** 159.89*** Pseudo R2 0.246 0.287
probit indicates a probit model whether e>1 or e=1. tobit indicates a tobit regression on e>1 choices only (censored at 20). Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
20
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT FTBT
Phase 2 TTT, FT, BT
FIGURE 6. Predicted effort in Phase 2 (based on estimates of Table 5 and ed = 12) as a function of wage after the
experience of trust or incentive contracts.
We summarize our findings on long-run crowding out in our second main result.
Result 2 (long-run crowding effect): The experience of Bonus or Fine contracts
reduces the extent of voluntary cooperation under subsequent Trust contracts. Framing of
incentives (bonus versus fine) is relatively unimportant for long-run crowding out.
Result 1 and 2 document short- and long-run crowding (out) of voluntary cooperation in
situations that are neither confounded with prior experiences of Trust contracts, nor with
strategic incentives in repeated interactions. Yet, these are precisely two interesting and
practically relevant situations, which we investigate in the next two sections. The goal is to
understand how Result 1 and 2 change if people are experienced with Trust contracts before
they are exposed to explicit performance incentives (Section 5), and if contractual relations
are embedded in an ongoing employment relationship which permits implicit incentives due
to the repeated nature of interactions (Section 6).
5. RESULTS II: THE ROLE OF EXPERIENCE WITH TRUST CONTRACTS FOR CROWDING EFFECTS
In this section we investigate the data of treatments TBT, TFT and TTT (see Table 2,
panel B) to address the question how the experience of Trust contracts before being exposed
to incentive contracts influences voluntary cooperation and the effectiveness of different
contract types. This is important because if subjects learn voluntary cooperation under Trust
21
contracts first, they might provide voluntary cooperation thereafter even under incentive
contracts. In this case short-run crowding out might be avoided. Similarly, long-run crowding
out when returning to Trust contracts might be avoided as well.
Furthermore, it is important to know how NIC-contracts perform with prior experience
of Trust contracts. NIC-contracts are psychologically quite interesting since the principal
neither sends a clear trust signal (as with a Trust contract), nor a clear signal that he relies on
straight pay for performance (as with an IC-contract). He uses a “stick or carrot” and at the
same time asks for more effort than the agent should provide if she were primarily interested
in “stick or carrot”. Thus a NIC-contract is neither a clear appeal to self-interest, nor an
unambiguous signal of the principal’s generosity and thereby an appeal to norms of
reciprocity. This ambiguity may be the reason for low effort by inexperienced subjects under
NIC-contracts. However, in principle a NIC-contract that offers a generous compensation
could be understood as an appeal to cooperation as well. Agents might learn to interpret NIC-
contracts this way, especially after experiencing Trust contracts.
The analysis in this section parallels the one in the previous section. Figure 7 illustrates
behavior in Phase 2 of TBT and TFT, that is, when monetary incentives are present
(analogous to Figure 2). It shows a bubble plot of actual effort against best-reply effort.15 The
results appear very similar to Figure 2. If the contract is incentive compatible and
consequently best-reply effort is above minimal effort (e* > 1), agents either choose e*
exactly or they deviate to e = 1. Compared to Figure 2 this pattern is even more pronounced
and clustering at e = e* for high levels of e* (levels 10, 11, 12) is particularly strong.
Together, e = e* choices and e = 1 choices comprise 98.6 percent of the decisions in
incentive-compatible contracts. Thus, given an IC-contract there is no voluntary cooperation.
If the contract is not incentive compatible and consequently best-reply effort is minimal
(e* = 1), effort is dispersed with e = 1 being the most frequent choice. Compared to Figure 2
we see more clustering at levels above 12 indicating that NIC-contracts perform better when
subjects are experienced with Trust contract.
15 In Phase 2 of TFT the percentages of cases of fines of 80, 52, 24, 0 are 78.8, 11.9, 5.8, and 3.5 percent, respectively. In Phase 2 of TBT percentages for the respective bonus levels are 83.1, 9.0, 5.6, and 2.3, respectively. These distributions are significantly different from each other (χ2-test, 12.392, p=0.006).
22
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 2 TFT
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 2 TBT
FIGURE 7. Relation between actual effort and best-reply effort in Phase 2 of TFT (left panel)
and Phase 2 of TBT (right panel). The size of dots is proportional to the number of underlying observations.
Table B1 in the Appendix reports an econometric model that is equivalent to the one
reported in Table 4 except that it relies on the Phase 2 data of TTT, TFT and TBT. Figure 8
shows predicted values of effort based on the regression model and for varying compensation
levels. Qualitatively we find about the same effects as outlined in Result 1: IC-contracts are
quite effective in implementing high effort for low compensation levels, but effort above 12 is
infeasible. Trust contracts can implement higher effort than IC-contracts at high
compensation levels. Yet, comparing Figures 8 and 4 we also see an important difference:
With experience Trust contracts are superior to IC-contracts for a much wider range of
compensation levels. Figure 8 displays that Trust contracts perform better than IC-contracts
for all compensation levels of about 270 and above. In Figure 4 Trust contracts predicted
higher effort than IC-contracts only for compensation levels of about 500 and above. While
experiencing Trust contracts in Phase 1 does not increase voluntary cooperation under IC-
contracts in Phase 2, it does increase voluntary cooperation under Trust contracts in Phase 2.
For NIC-contracts predicted value graphs based on Table B1 are displayed in Figure 8.
Similar to the analysis in Section 4.2 NIC-contracts are worse than Trust contracts at high
compensation levels. But NIC-contracts can induce a substantially higher effort than 12.
Moreover, at high compensation levels NIC-contracts induce higher effort than IC-contracts.
Thus, experiencing Trust contracts in Phase 1 increases reciprocity-based voluntary
cooperation and strengthens the performance of NIC-contracts.
23
02
46
810
1214
1618
20
Exp
ecte
d ef
fort
50 150 250 350 450 550 650Offered compensation
TTT TFT-IC
TBT-IC
Phase 2 TTT, TFT-IC, TBT-IC
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT TBT-NICTFT-NIC
Phase 2 TTT, TFT-NIC, TBT-NIC
FIGURE 8. Predicted effort as a function of offered compensation (based on estimations in Table B1, for ed = 12,
and b, f = 80.
To investigate how experience with Trust contracts affects long-run crowding out we
use the Phase 3 data of TTT, TBT and TFT to estimate a regression model reported in Table
B2 in the Appendix and illustrated in Figure 9. The model structure is equivalent to the one
reported in Table 5. Figure 9 looks strikingly similar to Figure 6 except that the differences
between the three shown curves have become smaller: the long-run crowding out effect has
become smaller.
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT TFTTBT
Phase 3 TTT, TFT, TBT
FIGURE 9. Predicted effort in Phase 3 (based on estimates of Table B2 and ed = 12) as a function of wage after
the experience of trust or incentive contracts.
24
We summarize the findings of our second set of experiments on the role of experience
with Trust contracts before being exposed to incentives in Result 3:
Result 3 (The role of experience of Trust contracts for crowding effects): (a) The
experience of Trust contracts in Phase 1 has no impact on behavior under IC-contracts in
Phase 2. Under IC-contracts agents choose their best-reply effort in the majority of cases and
there is no voluntary cooperation. (b) Yet, the experience of Trust contracts in Phase 1 does
increase the effectiveness of NIC-contracts in Phase 2. Under NIC-contracts reciprocity-
based voluntary cooperation is stronger than without prior experience of Trust contracts. (c)
The experience of Trust contracts in Phase 1 improves also the effectiveness of Trust
contracts in Phase 2. Trust contracts perform better than IC-contracts (NIC-contracts) at
compensation levels larger than about 270 (250), that is, Trust contracts are optimal for a
wide range of compensation levels. (d) The experience of Trust contracts in Phase 1 mitigates
long-run crowding out effects of monetary incentives.
So far we analyzed the data of random matching sessions. The main reason for this
design feature was to minimize strategic effects that might confound the crowding effects we
are interested in. However, in reality most employment relations are long-term relations
where strategic effects might influence the amount of voluntary cooperation we see. Long-
term relations permit implicit incentives and it is important to understand how implicit and
explicit incentives interact and influence crowding effects. The next section is devoted to this
analysis.
6. RESULTS III: THE BEHAVIORAL CONSEQUENCES OF IMPLICIT INCENTIVES
6.1 Short- and Long-Run Crowding Effects in the Presence of Implicit Incentives
To study the role of implicit incentives we compare the repeated game treatments
(partner-matching treatments) TFT-R, TBT-R and TTT-R with the respective random
matching treatments TFT, TBT and TTT. See Table 2C for further details. Figures 10a-c
display effort across periods for all treatments and all phases. The figures show that the
implicit incentives inherent in repeated games are a very strong motivator for high effort.
Every single repeated game curve is clearly above the respective random matching curve.
Under TTT-R (Figure 10a) average effort approaches almost the maximal (and efficient)
25
effort level. All time paths for repeated games show an endgame effect, that is, a decline of
effort in the last few periods of a phase. And finally, while the time paths for TFT and TBT
jump upwards in Phase 2 (the incentive phase), the corresponding time paths for TFT-R and
TBT-R show a less strong reaction upon the introduction of explicit incentives. This indicates
that explicit incentives are less important in repeated games. We will address these points
more formally in the statistical analyses below. 0
24
68
1012
1416
1820
Act
ual e
ffort
1 2 3 4 5 6 7 8 9 10Period
TTT TTT-R
Phase 1 (period 1-10)
02
46
810
1214
1618
20
11 12 13 14 15 16 17 18 19 20Period
TTT TTT-R
Phase 2 (period 11-20)
02
46
810
1214
1618
20
21 22 23 24 25 26 27 28 29 30Period
TTT TTT-R
Phase 3 (period 21-30)
FIGURE 10A. Mean effort over time in the one-shot games TTT and in the repeated games TTT-R.
02
46
810
1214
1618
20A
ctua
l effo
rt
1 2 3 4 5 6 7 8 9 10Period
TFT TFT-R
Phase 1 (period 1-10)
02
46
810
1214
1618
20
11 12 13 14 15 16 17 18 19 20Period
TFT TFT-R
Phase 2 (period 11-20)
02
46
810
1214
1618
20
21 22 23 24 25 26 27 28 29 30Period
TFT TFT-R
Phase 3 (period 21-30)
FIGURE 10B. Mean effort over time in the one-shot games TFT and in the repeated games TFT-R.
26
02
46
810
1214
1618
20
Act
ual e
ffort
1 2 3 4 5 6 7 8 9 10Period
TBT TBT-R
Phase 1 (period 1-10)
02
46
810
1214
1618
20
11 12 13 14 15 16 17 18 19 20Period
TBT TBT-R
Phase 2 (period 11-20)
02
46
810
1214
1618
20
21 22 23 24 25 26 27 28 29 30Period
TBT TBT-R
Phase 3 (period 21-30)
FIGURE 10C. Mean effort over time in the one-shot games TBT and in the repeated games TBT-R.
Figure 11 shows a bubble plot of effort against best-reply effort for the Phase 2 data of
TFT-R and TBT-R. The data pattern looks the same as in Figures 2 and 7, but the relative
sizes of bubbles have changed indicating that certain choices have become more frequent
while other have become less frequent. Specifically, the most frequent choice is maximal
effort under NIC-contracts (e* = 1). Furthermore, in the repeated games only 25.2 percent of
contracts are incentive compatible whereas with random matching (TBT, TFT) 74.5 percent
of contracts are incentive compatible.16 These changes may be understood by noting that in
repeated games, the implicit incentives of an ongoing relationship can substitute for explicit
incentives. Yet, while the implicit incentives lead to very high effort levels under NIC-
contracts, they do not increase effort under IC-contracts. As Figure 11 reveals, when the
contract is incentive compatible, agents choose their stage game best-reply effort in almost all
cases. Agents choose e = e* despite the fact that, as the NIC-contracts show, people in the
large majority of cases are willing to provide effort even beyond the maximally enforceable
effort level e = 12.
16 The reason for this striking difference is not in the design of fines and bonuses but in the desired effort levels. In the TFT and TBT experiments the average desired effort level is 10.8 (11.3) in Phase 2 of TFT (TBT), whereas it is 16.3 (14.9) in Phase 2 of TFT-R (TBT-R). The distribution of fines and bonuses is as follows: In TFT-R the percentages of cases of fines of 80, 52, 24, 0 are 58.9, 15.4, 13.7, and 12.0 percent, respectively. In TBT-R percentages for the respective bonus levels are 70.5, 15.1, 11.4, and 3.0, respectively. These distributions are significantly different from each other (χ2-test, 11.166, p=0.011).
27
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 2 TFT-R
14
812
20A
ctua
l effo
rt
1 4 8 12Optimal effort (best reply)
Phase 2 TBT-R
FIGURE 11: Relation between actual effort and best-reply effort in Phase 2 of TFT-R and TBT-R.
Based on the repeated game data, Tables B3 and B4 in Appendix B report regression
models analogous to those reported in Tables B1 and B2 for random matching data.17 Figures
12a and 12b provide predicted value graphs for varying compensation levels (and assuming b,
f = 80, ed = 12).
The left panel of Figure 12a highlights that in repeated games Trust contracts perform
better than IC-contracts at most compensation levels. Only for very low compensation levels
IC-contracts perform somewhat better. The right panel of Figure 12a shows that at low
compensation levels (up to about 270) trust contracts perform substantially better than NIC-
contracts. However, at larger compensation levels NIC-contracts outperform Trust contracts
(and IC-contracts). Thus, NIC-contracts – i.e., using “stick or carrot” and asking for an effort
that is higher than e* – is a good strategy if compensation overall is sufficiently high. If
compensation is too low, this strategy causes a substantial loss in effort.
Figure 12b illustrates long-run crowding effects. Differences between the three curves
can hardly be detected. Consequently if subjects experience Trust contracts and implicit
incentives there is no long-run crowding out of voluntary cooperation anymore.
17 In the regressions for repeated games we excluded the data of the final three periods of each phase to downplay influences of endgame effects.
28
02
46
810
1214
1618
20Ex
pect
ed e
ffort
50 150 250 350 450 550 650Offered compensation
TTT-R TFT-R-ICTBT-R-IC
Phase 2 TTT-R, TFT-R-IC, TBT-R-IC
02
46
810
1214
1618
20Ex
pect
ed e
ffort
50 150 250 350 450 550 650Offered compensation
TTT-R TFT-R-NICTBT-R-NIC
Phase 2 TTT-R, TFT-R-NIC, TBT-R-NIC
FIGURE 12A: Predicted effort as a function of offered compensation (based on estimations in Table B3, for ed = 12, and b, f = 80.
02
46
810
1214
1618
20E
xpec
ted
effo
rt
50 150 250 350 450 550 650Offered compensation
TTT-R TFT-RTBT-R
Phase 3 TTT-R, TFT-R, TBT-R
FIGURE 12B: Predicted effort in Phase 3 (based on estimates of Table B4 and ed = 12) as a function of wage after
the experience of trust or incentive contracts.
We collect the findings on the repeated games in our final main result.
Result 4 (The role of implicit incentives): (a) Implicit incentives inherent in repeated
interactions have a very strong behavioral impact on voluntary cooperation under Trust
contracts and under NIC-contracts. (b) Repeated interactions hardly influence behavior
under IC-contracts. IC-contracts induce stage game best-reply effort. This is effective at low
compensation levels, but it forgoes efficiency increases at higher compensation levels which
are feasible with Trust contracts and NIC-contracts. (c) NIC-contracts induce very low efforts
at low compensation levels but very high effort levels at high compensation levels. (d) There
is no long-run crowding out effect any longer.
NIC-contracts perform worse than IC-contracts and Trust contracts under conditions of
inexperienced subjects and random matching. But NIC-contracts perform best with
29
experience and partner matching and if compensation is sufficiently high. While it is not that
surprising that experience and partner matching increase effort, the question is why these
factors have relatively more impact on behavior under NIC-contracts than under IC-contracts
and Trust contracts. We address this question in the following subsection.
6.2 The Surprising Effectiveness of NIC-Contracts
We believe that behavior in our experimental games is driven by two mechanisms – the
pay for performance mechanism and the trust and reciprocity mechanism. Above we provided
clear evidence that both mechanisms exist. It is quite plausible that individuals consider both
aspects at the same time.
If inexperienced agents face a Bonus or Fine contract, they focus on the pay for
performance character of the interaction. The trust and reciprocity mechanism (a positive
wage-effort correlation) – which characterizes behavior under Trust contracts, is hampered as
Figure 3 illustrates. With experience of Trust contracts and in repeated games trust and
reciprocity become more and more important even under Bonus and Fine contracts. Figure 13
supports this claim. This figure displays scatter plots of effort against offered compensation
for NIC-contracts in Phase 2 of treatments TFT, TBT, TFT-R and TBT-R.
05
1015
20A
ctua
l effo
rt
0 200 400Offered compensation
TFT-NIC
05
1015
20A
ctua
l effo
rt
0 200 400Offered compensation
TBT-NIC
05
1015
20A
ctua
l effo
rt
0 200 400Offered compensation
TFT-R-NIC
05
1015
20A
ctua
l effo
rt
0 200 400Offered compensation
TBT-R-NIC
FIGURES 13: Offered compensation and effort under non-incentive-compatible contracts in Phase 2 of TFT,
TBT, TFT-R and TBT-R.
30
In the repeated game trust and reciprocity, strengthened by implicit incentives, are more
important than explicit incentives. From this viewpoint high offered compensation might
induce reciprocally high effort for NIC-contracts as well as for IC-contracts. But if one notes
in addition that voluntary cooperation usually does not mean that effort is chosen above
desired effort but is chosen equal to desired effort or below (see Table 6), one understands
why NIC-contracts outperform IC-contracts at high compensation levels. Ceteris paribus,
NIC-contracts ask for higher effort than IC-contracts.
TABLE 6 ACTUAL AND DESIRED EFFORT IN TTT-R, TFT-R AND TBT-R
IC-contracts (TFT-R, TBT-R)
NIC-contracts (TFT-R, TBT-R)
Trust contracts (TTT-R)
cases # percent # percent # percent
e > ed 1 1.89% 87 10.13% 8 2.40%
e = ed 44 83.02% 355 41.33% 133 39.94%
1 < e < ed - - 296 34.46% 174 52.25%
e = 1 (ed > 1) 8 15.09% 121 14.09% 18 5.41%
Sum 53 100% 859 100% 333 100%
A remaining question is why NIC-contracts outperform Trust contracts (for high
compensation levels in repeated games) even though the latter seem an even stronger,
unambiguous appeal to trust and reciprocity than the former. A reason may be seen in the
higher frequency of choices e with 1 < e < ed under Trust contracts than under NIC-contracts
(see Table 6). By choosing effort between minimal and desired effort the agent may fine-tune
the distribution of payoffs between principal and agent. This is more difficult in NIC-
contracts than in Trust contracts, since in NIC-contracts undercutting desired effort implies
that the entire fine is to be paid or that the entire bonus is lost. Consequently, under NIC-
contracts, if the agent has decided to provide higher than minimal effort at all, it makes little
sense to marginally undercut desired effort. In turn, choosing effort at the desired level or
above is a stronger cooperative signal than choosing effort above the minimum but below
desired effort. The higher relative frequency of choices e ≥ ed under NIC-contracts might
therefore facilitate and improve cooperation relative to trust contracts.
Our view is supported by Figure 14 which displays prob(e > 1) in the left panel and
e|e>1 in the right panel according to the regression model (see Table B3). For high
31
compensation levels we see that prob(e > 1) is about equal but e|e>1 is higher for NIC-
contracts. 0
.1.2
.3.4
.5.6
.7.8
.91
Pro
babi
lity
e>1
50 150 250 350 450 550 650Offered compensation
TTT-R TFT-R-NICTBT-R-NIC
Phase 2 TTT-R, TFT-R-NIC, TBT-R-NIC
0
24
68
1012
1416
1820
e|e>
1
50 150 250 350 450 550 650Offered compensation
TTT-R TFT-R-NICTBT-R-NIC
Phase 2 TTT-R, TFT-R-NIC, TBT-R-NIC
FIGURE 14: Behavior in NIC-contracts in the presence of implicit incentives. LEFT PANEL: Probability to choose
non-minimal effort; RIGHT PANEL: predicted effort conditional on non-minimal effort.
Overall we conclude that NIC-contracts perform better than IC-contracts in repeated
games and for high compensation levels since they give more leverage to trust and
reciprocity. And NIC-contracts perform better than Trust contracts because Bonus and Fine
contracts reduce the frequency of undercutting desired effort. While NIC-contracts have been
shown as quite effective for high compensation levels they are risky since effort is rather low
and much lower than under IC-contracts and Trust contracts when compensation is small.
7. CONCLUDING DISCUSSION
Within the experimental study presented here we used a comprehensive experimental
design to investigate whether and under what conditions incentives and social preferences are
separable ({Bowles, 2008 #3734}; {Bowles, 2008 #3419}; {Bowles, 2010 #5503}). One part
of our results confirms and strengthens previous findings of other studies, and another part of
the results refines or extends the research on principal-agent experiments.
We confirm that both Trust contracts and explicit incentive contracts are highly
effective in inducing high effort. Under Trust contracts agents either choose minimal effort or
respond with reciprocally higher effort the higher the offered compensation is. This is
evidence for reciprocity-based voluntary cooperation, and this sort of behavior is clearly
visible under conditions of random matching as well as partner matching. Under incentive
compatible Bonus or Fine contracts agents either choose minimal effort or best-reply effort.
32
There is a strong crowding-out effect in the sense that under IC-contracts agents
almost never exceed their best-reply effort level. Surprisingly this is the case even with
partner matching where one could have expected that reciprocity induces higher effort
choices. While implicit incentives in repeated games do not improve cooperation under IC-
contracts, they do so under Trust contracts. Remember that we deliberately designed the stage
games such that best-reply behavior predicts a maximally implementable effort level of 12
leaving room for efficiency-enhancing increases in effort. Such increases occur systematically
under Trust contracts – especially in repeated games – but not at all under IC-contracts. IC-
contracts perform well in implementing best-reply effort at relatively low compensation
levels. But they cannot induce higher than best-reply effort even at large compensation levels.
Whether incentives are provided in the form of a bonus or a fine (framing of incentives) was
shown to have a relatively minor influence.
The effectiveness of Trust contracts in inducing reciprocally high effort is only to a
small degree weakened when the participants have experienced a phase of incentive
contracting before. We called this the “long-run crowding-out effect” and showed that it
unambiguously exists for subjects who are inexperienced with Trust contracts but long-run
crowding out is substantially reduced by experience with Trust contracts.
We were surprised by the good performance of NIC-contracts under partner matching.
At high compensation levels these contracts led to rather high effort levels. There are two
things we found surprising: first, that NIC-contracts may perform better than IC-contracts,
and second, that NIC-contracts may perform better than Trust contracts. We provide some
speculative explanations for these effects.
The first effect, that NIC-contracts induce higher effort than IC-contracts at large
compensation levels, may be caused by agents choosing usually no larger than the desired
effort and by principals being too cautious. Principals ask for best-reply effort instead of
higher than best-reply effort, that is, they do not rely enough on reciprocity.
The second effect, that NIC-contracts induce higher effort than Trust contracts at large
compensation levels, is surprising since we view e.g. a fine contract (that asks for a higher
than best-reply effort and threatens a fine) as a less cooperative signal than a Trust contract
that offers equal compensation and asks for equal effort. While we view a Trust contract a
stronger cooperative signal being sent from the principal to the agent, many agents send a
weaker signal of cooperation to the principal under Trust contracts than under NIC-contracts.
Choosing lower than the desired effort is more frequent under Trust contracts and this may be
interpreted by the principal as a weaker signal of cooperation.
33
We are not the first to investigate issues of separability.18 However, to our knowledge
no study has investigated several failures of separability of implicit and explicit incentives in
one comparable design. This approach allows us to gauge which effects are of first-order
importance, and which ones are of second-order importance. Our experiment shows that best-
reply effort behavior and reciprocity-based voluntary cooperation are both first-order effects
in the sense that they are sizeable and important across all three sets of experiments. Framing
effects and long-run crowding out are much less robust and may hence be considered of
second-order importance.
In conclusion, given that both voluntary cooperation and incentives are of first-order
importance, it is not always the best idea to rely on an IC-contract in principal-agent
relationships. Such contracts guide behavior very strongly (“you get what you pay for”), but
they also unambiguously crowd out any (reciprocity-based) voluntary cooperation, even for
very high compensation levels and even in repeated games with otherwise strong implicit
incentives.
18 For overviews of important findings see Frey and Jegen (2001); Fehr and Falk (2002); Bowles (2008);Bowles and Polanía Reyes (2010).
34
Appendix A: Instructions Here we document the instructions of the Trust game and the Fine game used in our TFT experiment. The instructions in the other treatments were adapted accordingly. The instructions were originally written in German.
The experiment in which you participate today is joint research with the Humboldt-University Berlin. It is financed by several science foundations. During the experiment your income will be calculated in points. In the beginning you get a lump-sum of 1500 points. It is possible that some of your decisions lead to losses. You will have to finance them out of the gains from your other decisions, or, if necessary out of your lump-sum. However, you can always make decisions that avoid any losses. The exchange rate of points into Swiss Francs is:
1 Point = 0.6 Rappen.
At the end of the experiment all points which you have earned will be summed up, exchanged into Swiss Francs and paid out to you in cash. Please note that during the experiment communication is not allowed. If you have any questions, please raise your hand. We will answer your questions individually.
Instructions 1. Introduction In this experiment you will learn about a decision problem that involves two persons. The persons will be called participant X and participant Y. All participants in this experiment are allocated into two groups: the group of the participants X and the group of the participants Y. After the experiment has started you can see on your computer's display whether you are participant X or participant Y.
At the beginning you will be randomly matched with a participant of the other group. You will make your decisions at the computer. Your decisions will be transmitted via computer to the participant of the other group. This participant will only get informed about your decision. He will never learn about your name or your participant - number, i.e. your decisions remain anonymous.
2. An overview of the experiment It may help you to understand the decision situation if you think about the following scenario. Participant X decides in the role of a "firm". The firm engages an employee (participant Y), who's work effort produces some period return. Y can choose his work effort freely in each period. Below we will explain what work effort means and how the period return comes about. A higher effort leads to a higher period return, but it also causes costs that Y has to bear. Y's payment is determined in an employment contract. The employment contract consists of a fixed wage defined by X and a "desired effort". The fixed salary has to be paid by participant X to participant Y regardless of the period return. Thus, each period consists of three stages:
1. Participant X proposes in accordance with the rules a employment contract including the fixed salary and the
"desired effort".
2. Participant Y decides to accept or reject the contract.
3. Y chooses his effective effort. The desired effort of X is not binding for Y. Afterwards X and Y will be paid according to the rules. There are 10 periods. You will be matched in each period randomly with another person.
35
3. The experimental details 3.1 Employment contract: The proposal of participant X
At the beginning of each period a employment contract will be determined. For the employment contract the following rules hold: The proposed employment contract consists of two components: a fixed wage and a desired effort. Participant X can design the contract according to the below-mentioned rules.
• The contract can contain a positive or a negative fixed salary. If the fixed salary is positive, this means that participant Y gets the wage form participant X, regardless of the period return. A negative fixed wage means that Y has to pay that amount to X, regardless of the period return.
• The proposed employment contract is only valid if participant Y accepts the labour contract. If Y accepts the contract, then Y decides about his effective work effort. X's desired work effort is not binding for Y. The participant Y can choose an effective work effort, which is higher, equal or lower than the desired effort.
• For the contract design the following rules hold: −700 ≤ fixed wage ≤ 700
1 ≤ desired work effort ≤ 20
All numbers must be integer.
In designing the contracts all combinations that are compatible with these rules are possible! To make the rules clear to you, we depict the screen that will be shown to X at the beginning of period 1:
On this screen (as well as in all other screens in which you have to make a decision) you see the current period number on top left and the remaining time on top right. Participant X makes up his proposed employment contract on this screen. 3.2 Employment contract: Acceptance of the contract
After participant Y has received the proposal of the contract, he decides whether he accepts or rejects the contact.
36
3.3 Work effort of participant Y
After the acceptance of the contract Y determines his work effort. The desired work effort stated by participant X in the contract is not binding for participant Y. The effort is expressed as a number. In the enclosed table all possible work effort (all integer numbers between 1 and 20) as well as the produced returns are given. The table also contains the costs of the work that Y has to bear. The higher the work effort, the higher is the return, but also the costs of the work effort.
The screen of participant Y is shown below.
3.4 Period payoffs and end of period After participant Y has entered his work effort into the computer, the period gains will be calculated and shown on the display. The following cases result for the calculation of the profits:
Profit of X: Profit of Y:
Y rejects the contract: Zero Zero
Y accepts the contract: Period return of the effective work effort
– fixed wage Fixed wage – cost of the effective work effort
Please note: For the profit only the effective work effort is relevant. After this-screen the period is finished and the next one starts. On the whole there are 10 periods.
37
Effort, period return of the work effort and costs of the work effort for Y:
Work effort : Period return of the work effort costs of the work effort for Y
1 35 0 2 70 7 3 105 14 4 140 21 5 175 28 6 210 35 7 245 42 8 280 49 9 315 56
10 350 63 11 385 70 12 420 77 13 455 84 14 490 91 15 525 98 16 560 105 17 595 112 18 630 119 19 665 126 20 700 133
Profit of Y: Fixed wage − costs of the effective work effort
Profit of X: Period return of the effective work effort – fixed wage
Profit of Y and X by rejection of the contract of Y: Zero
Only the effective work effort is relevant for the calculation of the profits! __________________________________________________________________________________________
Information about the new experiment
The new experiment also consists of 10 periods. In this experiment, too, you are matched randomly with another person in each period. Again you don’t get to know the others person's identity. As before all decisions are anonymous. The only change compared to the first experiment is given by an additional parameter that X can offer in the proposed employment contract. In addition to the fixed salary and the desired effort participant X determines a potential deduction from wages. This is the only difference to the previous experiment. The potential deduction from wages is enforced if Y chooses a work effort that is lower than the desired effort of X. If Y choose a work effort which is higher or equal than the desired effort than the deduction from wages isn’t enforced. There are four possible levels of potential deduction from wages. The deduction from wages can be either 0 or 24 or 52 or 80. The deduction from wages will only be enforced if the effective effort is lower than the desired effort!
For the contract design the following rules hold:
−700 ≤ fixed wage ≤ 700
Potential deduction from wages: either 0 or 24 or 52 or 80
1 ≤ desired work effort ≤ 20
All numbers must be integers. In designing the contract all combinations that are compatible with these rules are possible!
To make the rules clear to you, we depict the screen that will be shown to X at the beginning of the first period.
38
The profits are calculated as follows: Profit of X: Profit of Y:
Y rejects the contract:
Zero Zero
The effective work effort is higher or equal than the desired work effort.
Period return of the effective work effort – fixed wage Fixed wage – costs of the effective work effort
The effective work effort is lower than the desired work effort:
Period return of the effective work effort – fixed wage
+ deduction from wages
Fixed wage –deduction from wage – costs of the
effective work effort
The process of this experiment is identical with the previous experiment.
39
APPENDIX B: SUPPORTING ANALYSES
Measuring incentive effects and short-run crowding out effects in TFT and TBT The regression model reported in Table B1 is very similar to the one reported in Table 4 in the paper. The
only difference is that we include a variable “Individual cooperation level” to measure the average effort chosen by the individual agent in Phase 1. We included this to control for an individual (fixed effect) inclination to provide high effort.
TABLE B1 EXPLAINING PHASE 2 EFFORT CHOICES IN TREATMENTS TTT, TFT AND TBT
Probit (e>1)
Probit (e=e*|e>1)
Tobit (e|e>1, e≠e*)
Fixed wage (Trust games) 0.006*** 0.040*** (0.001) (0.005) Desired effort (Trust games) -0.011 0.086 (0.014) (0.110) Individual cooperation level 0.080*** -0.171*** -0.411** (0.015) (0.027) (0.172) Fixed wage (Bonus games, IC) 0.000 0.003 (0.002) (0.003) Bonus b (Bonus games, IC) 0.049*** 0.060*** (0.009) (0.017) Fixed wage (Fine games, IC) 0.006*** -0.000 (0.002) (0.003) Fine f (Fine games, IC) 0.045*** 0.070*** (0.009) (0.016) Desired effort (Bonus/Fine games, IC) -0.276*** -0.056 (0.053) (0.094) Dummy IC 0.695* 19.974*** (0.416) (3.704) Fixed wage (Bonus games, NIC) 0.006*** 0.024*** (0.002) (0.009) Bonus b (Bonus games, NIC) 0.002 0.010 (0.008) (0.029) Fixed wage (Fine games, NIC) 0.005*** 0.017*** (0.002) (0.006) Fine f (Fine games, NIC) -0.002 0.054 (0.008) (0.042) Desired effort (Bonus/Fine games, NIC) -0.098** 0.525 (0.042) (0.354) Dummy NIC 1.099*** -0.472 (0.386) (1.349) Inverse Mills ratio -0.34 3.449 (0.342) (1.165)*** Constant -1.392*** -1.153 -8.257 (0.134) (0.274)*** (2.760)*** No. of obs. 981 619 231 χ2=291.68*** χ2=327.38*** F=295.28*** Pseudo R2 0.285 0.900 0.149 Phase 2 data of TTT, TFT and TBT. Probit (e>1) indicates a probit model whether e>1 or e=1. Probit (e=e*|e>1) indicates a probit model whether e=e* (best-reply effort) or e≠e*. Tobit (e|e>1, e≠e*) indicates a Tobit regression with an upper bound of e=20. Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
40
Measuring long-run crowding out in TFT and TBT The regression model reported in Table B2 is the same as the one reported in Table 5 in the paper except
for adding the variable “Individual cooperation level” (see Table B1).
TABLE B2
LONG-RUN CROWDING OUT (COMPARING PHASE 3 DATA IN TFT AND TBT WITH TTT)
Probit (e>1)
Tobit (e|e>1)
Fixed wage 0.006*** 0.041*** (0.001) (0.009) Desired effort -0.014 0.117 (0.017) (0.073) Dummy TBT treatment -0.363 0.224 (0.246) (0.890) Dummy TFT treatment -0.598*** 0.905 (0.213) (0.822) Individual cooperation level 0.166*** 0.351 (0.018) (0.260) Inverse Mills ratio 2.781 (2.451) Constant -1.689*** -6.292 (0.247) (4.747) No. of obs. 929 300 LR chi2 279.34*** 125.54*** Probit (e>1) indicates a probit model whether e>1 or e=1. Tobit (e|e>1) indicates a Tobit regression on e>1 choices only (censored at 20). Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
41
Measuring incentive effects and short-run crowding out in TFT-R and TBT-R The regression model reported in Table B3 is very similar to the one reported in Table 4 in the paper and
in Table B1. In contrast to Tables 4 and B1 we did not run the Probit regression to estimate prob(e = e*|e > 1). Since there were almost no deviations (see Figure 11 in the paper) this probability is virtually equal to 1.
TABLE B3 EXPLAINING PHASE 2 EFFORT CHOICES IN TREATMENTS TTT-R, TFT-R AND TBT-R
Probit (e>1)
Tobit (e|e>1)
Fixed wage (Trust games) 0.006 0.020 (0.005) (0.015) Desired effort (Trust games) -0.016 0.685** (0.077) (0.323) Individual cooperation level 0.079** 0.425*** (0.039) (0.151) Fixed wage (Bonus games, IC) 0.010*** (0.002) Bonus b (Bonus games, IC) 0.008 (0.007) Fixed wage (Fine games, IC) 0.011*** (0.003) Fine f (Fine games, IC) -0.010 (0.007) Desired effort (all games, IC) -0.159** (0.069) Dummy IC 1.672** 12.535*** (0.779) (1.735) Fixed wage (Bonus games, NIC) 0.071*** (0.016) Bonus b (Bonus games, NIC) 0.110*** (0.028) Fixed wage (Fine games, NIC) 0.080*** (0.016) Fine f (Fine games, NIC) -0.018 (0.028) Desired effort (all games, NIC) -0.438 (0.317) Dummy NIC 0.366 -1.851 (0.945) (2.639) Inverse Mills ratio 6.725** (3.147) Constant -0.176 -7.125*** (0.621) (2.480) No. of obs. 289 265 χ2 55.23*** 31.91*** Pseudo R2 0.333 0.216
Phase 2 data of TTT-R, TFT-R and TBT-R. Probit (e>1) indicates a probit model whether e>1 or e=1. Tobit (e|e>1) indicates a Tobit regression on e>1 choices only (censored at 20). Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
42
Measuring long-run crowding out in TFT-R and TBT-R The regression model reported in Table B4 is analogous to the ones reported in Tables 5 and B2.
TABLE B4 LONG-RUN CROWDING OUT (COMPARING PHASE 3 DATA IN TFT-R AND TBT-R WITH TTT-R)
Probit (e>1)
Tobit (e=e*|e>1)
Fixed wage 0.009*** 0.042*** (0.002) (0.010) Desired effort 0.002 0.570*** (0.034) (0.219) Dummy TBT-R treatment -0.572 -0.964 (0.541) (0.89) Dummy TFT-R treatment -0.615 0.153 (0.527) (0.971) Individual cooperation level -0.046 0.103 (0.03) (0.09) Inverse Mills ratio 10.965*** (3.995) Constant 0.313 -9.265*** (0.614) (2.922) No. of obs. 304 270 LR chi2 57.24*** 77.39*** Pseudo R2 0.510 0.287
Probit indicates a probit model whether e>1 or e=1. Tobit indicates a Tobit regression on e>1 choices only (censored at 20). Robust standard errors in parentheses; * p< 10%; ** p< 5%; *** p< 1%.
43
REFERENCES
ADAMS, J. S. ed. (1965): Inequity in Social Exchange. New York: Academic Press. AKERLOF, G. A. (1982): "Labor Contracts as Partial Gift Exchange," Quarterly Journal of
Economics, 97, 543-569. BÉNABOU, R., and J. TIROLE (2003): "Intrinsic and Extrinsic Motivation," Review of
Economic Studies, 70, 489-520. — (2006): "Incentives and Prosocial Behavior," American Economic Review, 96, 1652-1678. BEWLEY, T. (1999): Why Wages Don’t Fall in a Recession. Cambridge: Harvard University
Press. BEWLEY, T. F. (2007): "Fairness, Reciprocity, and Wage Rigidity," in Behavioral Economics
and Its Applications, ed. by P. Diamond, and H. Vartiainen. Princeton: Princeton University Press, 157-188.
BOHNET, I., B. S. FREY, and S. HUCK (2001): "More Order with Less Law: On Contract Enforcement, Trust, and Crowding," American Political Science Review, 95, 131-144.
BOLTON, G. E., and A. OCKENFELS (2000): "Erc: A Theory of Equity, Reciprocity, and Competition," American Economic Review, 90, 166-93.
BOWLES, S. (1998): "Endogenous Preferences: The Cultural Consequences of Markets and Other Economic Institutions," Journal of Economic Literature, 36, 75-111.
— (2003): Microeconomics: Behavior, Institutions, and Evolution. Princeton: Princeton University Press.
— (2008): "Policies Designed for Self-Interested Citizens May Undermine “the Moral Sentiments”: Evidence from Economic Experiments," Science, 320, 1605-1609.
BOWLES, S., and S.-H. HWANG (2008): "Social Preferences and Public Economics: Mechanism Design When Social Preferences Depend on Incentives," Journal of Public Economics, 92, 1811-1820.
BOWLES, S., and S. POLANÍA REYES (2010): "Economic Incentives and Social Preferences as Complements or Substitutes," Mimeo, Santa Fe Institute.
CAMERER, C. F. (2003): Behavioral Game Theory: Experiments in Strategic Interaction. Princeton: Princeton University Press.
CHARNESS, G., and M. DUFWENBERG (2006): "Promises and Partnership," Econometrica, 74, 1579-1601.
CROSON, R., and S. GÄCHTER (2010): "The Science of Experimental Economics," Journal of Economic Behavior & Organization, 73, 122-131.
DUFWENBERG, M., S. GÄCHTER, and H. HENNIG-SCHMIDT (2010): "The Framing of Games and the Psychology of Play," CeDEx Discussion Paper 2010-16, University of Nottingham.
DUFWENBERG, M., and U. GNEEZY (2000): "Measuring Beliefs in an Experimental Lost Wallet Game," Games and Economic Behavior, 30, 163-182.
DUFWENBERG, M., and G. KIRCHSTEIGER (2004): "A Theory of Sequential Reciprocity," Games and Economic Behavior, 47, 268-298.
ELLINGSEN, T., and M. JOHANNESSON (2008): "Pride and Prejudice: The Human Side of Incentive Theory," American Economic Review, 98, 990-1008.
FALK, A. (2007): "Gift Exchange in the Field," Econometrica, 75, 1501-1511. FALK, A., and U. FISCHBACHER (2006): "A Theory of Reciprocity," Games and Economic
Behavior, 54, 293-315. FALK, A., S. GÄCHTER, and J. KOVACS (1999): "Intrinsic Motivation and Extrinsic Incentives
in a Repeated Game with Incomplete Contracts," Journal of Economic Psychology, 20, 251-284.
FALK, A., and J. J. HECKMAN (2009): "Lab Experiments Are a Major Source of Knowledge in the Social Sciences," Science, 326, 535-538.
44
FALK, A., and M. KOSFELD (2006): "The Hidden Costs of Control," American Economic Review, 96, 1611-1630.
FEHR, E., and A. FALK (2002): "Psychological Foundations of Incentives," European Economic Review, 46, 687-724.
FEHR, E., and U. FISCHBACHER (2002): "Why Social Preferences Matter - the Impact of Non-Selfish Motives on Competition, Cooperation and Incentives," Economic Journal, 112, C1-C33.
FEHR, E., L. GOETTE, and C. ZEHNDER (2009): "A Behavioral Account of the Labor Market: The Role of Fairness Concerns," Annual Review of Economics, 1, 355-384.
FEHR, E., G. KIRCHSTEIGER, and A. RIEDL (1993): "Does Fairness Prevent Market Clearing? An Experimental Investigation," Quarterly Journal of Economics, 108, 437-459.
FEHR, E., A. KLEIN, and K. M. SCHMIDT (2007): "Fairness and Contract Design," Econometrica, 75, 121-154.
FEHR, E., and B. ROCKENBACH (2003): "Detrimental Effects of Sanctions on Human Altruism," Nature, 422, 137-140.
FEHR, E., and K. M. SCHMIDT (1999): "A Theory of Fairness, Competition, and Cooperation," Quarterly Journal of Economics, 114, 817-68.
FISCHBACHER, U. (2007): "Z-Tree: Zurich Toolbox for Readymade Economic Experiments," Experimental Economics, 10, 171-178.
FREY, B. S., and R. JEGEN (2001): "Motivation Crowding Theory," Journal of Economic Surveys, 15, 589-611.
GÄCHTER, S., H. ORZEN, E. RENNER, and C. STARMER (2009): "Are Experimental Economists Prone to Framing Effects? A Natural Field Experiment," Journal of Economic Behavior & Organization, 70, 443-446.
GINTIS, H., S. BOWLES, R. BOYD, and E. FEHR eds. (2005): Moral Sentiments and Material Interests. The Foundations of Cooperation in Economic Life. Cambridge: MIT Press.
GNEEZY, U., and J. A. LIST (2006): "Putting Behavioral Economics to Work: Testing for Gift Exchange in Labor Markets Using Field Experiments," Econometrica, 74, 1364-1985.
GNEEZY, U., and A. RUSTICHINI (2000): "A Fine Is a Price," Journal of Legal Studies, 29, 1-17.
HENRICH, J., and N. HENRICH (2007): Why Humans Cooperate: A Cultural and Evolutionary Explanation. Oxford: Oxford University Press.
KREPS, D., P. MILGROM, J. ROBERTS, and R. WILSON (1982): "Rational Cooperation in the Finitely Repeated Prisoners' Dilemma," Journal of Economic Theory, 27, 245-252.
LEVINE, D. K. (1998): "Modeling Altruism and Spitefulness in Experiments," Review of Economic Dynamics, 1 3, 593-622.
LIBERMAN, V., S. M. SAMUELS, and L. ROSS (2004): "The Name of the Game: Predictive Power of Reputations Vs. Situational Labels in Determining Prisoner’s Dilemma Game Moves," Personality and Social Psychology Bulletin, 30, 1175-1185.
MACLEOD, W. B. (2007): "Reputations, Relationships, and Contract Enforcement," Journal of Economic Literature, 45, 595-628.
RABIN, M. (1993): "Incorporating Fairness into Game-Theory and Economics," American Economic Review, 83, 1281-1302.
SALANT, Y., and A. RUBINSTEIN (2008): "(a,F): Choice with Frames," Review of Economic Studies, 75, 1287-1296.
SELTEN, R., and R. STOECKER (1986): "End Behavior in Sequences of Finite Prisoners-Dilemma Supergames - a Learning-Theory Approach," Journal of Economic Behavior & Organization, 7, 47-70.
SIMON, H. (1991): "Organizations and Markets," Journal of Economic Perspectives, 5, 25-44. SIMON, H. A. (1997): Administrative Behavior. A Study of Decision-Making Processes in
Administrative Organizations. New York: Free Press.
45
46
SLIWKA, D. (2007): "Trust as a Signal of a Social Norm and the Hidden Costs of Incentive Schemes," The American Economic Review, 97, 999-1012.
WILLIAMSON, O. (1985): The Economic Institutions of Capitalism. New York: Free Press. WOOLDRIDGE, J. M. (2002): Econometric Analysis of Cross Section and Panel Data.
Cambridge: MIT Press.