Do reputation concerns supress the reaction to being controlled?

D

O

R

EPUTATION

C

ONCERNS

S

UPRESS

THE

R

EACTION TO

B

EING

C

ONTROLLED?

O

ANA

B

OBICĂ, 10599290

MASTER THESIS, 15

ECTS

M

B

E

O

E

The paper experimentally analyzes the effect of the interaction between control and reputation on motivation, using a principal-agent game. The principal decides whether she wants to impose a minimum transfer from the agent, who afterwards decides on this transfer. The game is played for two rounds and reputation is induced by providing feedback to the principals about the agents’ choices in the first round. The effort enhancing reputation effect is confirmed, as well as the previously detected hidden costs of control (Falk & Kosfeld, 2006). Next to that, reputation also seems to slightly reduce the negative reaction to control and principals’, as well as agents’, behaviours seem to indicate direct and indirect reciprocity.

I. Introduction

The basic reason why the principal-agent problem is of interest is the conflict between the goals of the two: on one hand, the principal wants to get as much performance as possible from the agent while paying as little as possible for this performance and the latter wants to be rewarded with the highest possible wage in exchange for investing the minimum amount of effort. The agent will, therefore, try to maximize his net utility by shirking (Alchian & Demsetz, 1972). According to agency theory, the principal can prevent this from happening and increase the agent’s effort by monitoring him. A major drawback to this action is the crowding-out effect. Initially developed by Deci (1971) in connection with the cognitive evaluation theory, the motivation crowding-out theory suggests that an external intervention (monetary incentives, punishments) may impair the intrinsic motivation an agent initially has. Frey (1993) confronts the two theories and finds that agency theory holds and the disciplining effect dominates when the relationship between the principal and the agent is abstract, while the crowding-out effect is stronger in case of a personal relationship – in this situation, the agent is affected by the distrust the principal signals.

Trust is found to be a determinant factor for the outcome of the principal-agent relationship: as a factor that directly influences the agent’s decision, but also as an indicator of the general level of trustworthiness, as several papers experimentally studying potential actions and measures the principals can take in order to reduce shirking - Fehr et al. (1993), Fehr & Falk (2002), Fehr & List (2004), Falk & Kosfeld (2006) - find that the convictions on others’ trustworthiness may be self-reinforcing (Fehr, 2009). This interpretation of the previously mentioned experiments is confirmed by the fact that higher efforts are observed when the punishment/control decision is available for the principal but she refrains from using it, compared to the situation where there is no possibility for such an action. Even so, the results show that most principals choose to use the punishment/control action, the only exception being Falk & Kosfeld (2006), that find in every treatment that the majority of principals choose not to control.

Next to crowding-out intrinsic motivation, the principals’ shirk preventing actions are also considered to lead to a lower effort from the agents because they give information about the difficulty of the task – if incentives/punishments are necessary, then the task might be difficult or not very pleasant – or about the principal’s expectations: the agents might

believe more is expected from them and, by imposing a certain punishment/control action, the principals reveal lower expectations about the agents’ effort levels.

All of the above conclusions are relevant for organizations’ design. When deciding whether to control or not, whether to impose a fine or punishment or not, one has to take into account these implications. What previous studies haven’t taken into consideration is that the effort level an agent chooses is influenced by other interacting factors at the same time. There are many other forces that drive the agent’s decision and a very relevant one is reputation. It is more than reasonable to assume that reputation concerns are given strong consideration by the agent in a real-life situation before making any decision.

Servátka (2009) performs an experiment on reputation, social influence and identification effects in dictator games. He finds that reputation, causing indirect reciprocity, has the greatest effect of the three studied concepts on the dictator’s actions. The method used to induce reputation is used in a slightly altered way in the present paper.

Since the two factors – control and reputation concerns – have opposite effects, the following fact came into question: Do reputation concerns supress the reaction to being

controlled and if so, to what extent?

This particular matter has not been previously studied. Wernerfelt (1988) has a theoretical study on strategic monitoring’s impact on reputation building, one of the conclusions of relevance for the research question being that in the case of risk-neutral agents, the principal monitors in early stages, while the intensity of the agents’ effort might be lower or higher initially. The aspect that Wernerfelt (1988) ignores is the effect of being monitored on the agent – the monitoring decision is viewed only from the principal’s perspective.

The aim of the paper is, therefore, to study the interaction between these two factors, control and reputation, on the agent’s effort choices. As previously found, being controlled is expected to demotivate agents and lead to lower effort levels. Also, reputation concerns have been found to determine agents to make decisions closer to the principals’ interest (Gächter & Falk, 2002; Servátka, 2009). Thus, reputation concerns might diminish this effect and prevent the agents from expressing their discontent about being controlled. Seeing which of these two has the greater influence and how they interact is an interesting and previously not approached aspect of the principal-agent problem. Also, noticing the principals’ reactions to the agents’ effort choice gives some insight on whether indirect reciprocity is present and in

which cases. The paper is based on an experiment built on the one in Falk & Kosfeld (2006). In the first round, a principal that has no initial endowment chooses whether to impose a minimum transfer of 10 from an agent that disposes of 120 points, to whom she is randomly and anonymously matched. The agent decides how many points to transfer for each of the possible situation: when a minimum is imposed and when not. In the second round, the principals and agents are matched differently and the principal can also choose the agent’s initial endowment from the following values {30, 60, 120}. The other choices remain as in the first round. The procedure described above is for the control group. In the main treatment, there is an additional step: before choosing the agent’s initial endowment, the principal from the second round receives information on the agent’s choice from the first round. The whole of procedure is explained to the subjects before the experiment starts, this being an important aspect for inducing reputation concerns.

To begin with, the results completely reject the homo economicus theory, as expected. As Falk & Kosfeld (2006) found, control bears hidden costs, the effort levels being, on average, higher when there is no restriction imposed for all the situations. Also, these costs outweigh the benefits that imposing a minimum transfer brings, making the decision to control inefficient. This fact is generally true, certainly, for the current setting and will vary in real-life situations, being influenced by several other factors. A finding that does not come close to the Falk & Kosfeld's (2006) results concerns the principals’ decisions concerning control. Our experiment found that the majority of principals, between 72 and 78 percent, decide to impose a minimum transfer.

Regarding the actual contribution of the paper, information about the reaction between reputation concerns and control aversion is revealed, but the results are not completely clear. It is certain that agents choose higher effort levels when they make their decisions knowing that they will be revealed to the principals who will later decide their wages. Depending on whether the hidden costs are perceived as variable or fixed ones, the results either indicate that reputation does supress the manifestation of control aversion to some extent, either that control aversion is expressed the same irrespective of the reputation concerns, respectively. The agents’ reactions to control turn out to be heterogeneous: positive, neutral and, obviously, negative, responses are found, the last ones having a majority no matter the situation.

Reciprocity seems to be relevant for the principals who choose the agents’ wages after being informed by their previous effort choices, as well as for the agents’ effort choices after being informed on the initial endowments the principals chose for them. The first case is of indirect reciprocity, since the information does not affect the principal who receives it, but the previous one. Even so, wages seems to be quite sensitive to the agents’ effort choices, which is also a confirmation of reputation concerns having an actual foundation. The agents seem to reward, especially when they have no reputation concerns, the principals’ choices of not doing harm by choosing lower than maximum wages. For this case, reciprocity also seems to partially counterbalance the negative effects of control. However, the positive reactions do not seem to be enhanced through reciprocity.

The remainder of this paper is organized as follows. Section 2 presents the most important related literature along with a critical evaluation of it. Section 3 describes the experimental design and procedures and formulates the hypotheses. Section 4 presents summary statistics, the results of the experiment and discusses potential limitations. Section 5 summarizes the main findings and conclusions and some possible steps for future research.

II. Related literature

One of the first experiments involving the crowding-out of intrinsic motivation was performed by Deci (1971). College students solved a puzzle in three consecutive rounds. First, the subjects played without any further rule or instructions, freely. For the second session, a reward (payment) for solving the puzzle was introduced for the treatment group, while the control group did a repetition of the first session. The third round was the same for both treatment and control group and it involved no payment, just as the first one. The results show that in the third round the students from the treatment group spent less time trying to solve the puzzle than the ones from the control group. Since no other motivation was provided, it is assumed that the effort from the first and third rounds is completely owed to intrinsic motivation. The conclusion drawn from the results is that in the treatment group the introduction of the reward – extrinsic motivation – led to a decrease in the intrinsic motivation and the change was permanent, even after eliminating the reward in the third round. Taking this problem further, Gneezy & Rustichini (2000) did several treatments, varying the value of the reward. They found that the effect of the reward was not monotonic and that a higher reward was followed by a higher performance. Even so, the introduction of monetary incentives led to a lower performance compared to the situation of no compensation. The

explanation the authors find most plausible is that this reaction, in an incomplete contract situation, is the result of finding out new information and modifying the structure of the contract (by introducing the reward).

There is a plethora of studies on how motivation is influenced by incentive contracts, controlling decisions and fines. Most of these papers’ conclusions point in the same direction, but some are contradictory and others find there is not a single direction for this matter and by what factors the effect is driven one way or the other. Fehr & Gächter (2002) find that voluntary cooperation might be weakened by incentive contracts, by using two treatments based on Fehr et al. (1993)’s gift exchange game. One of them includes no performance incentives for contractual compliance and is used to determine the voluntary cooperation level as a baseline. The second one is identical to the base treatment, the only differentiating feature being that it also permits imposing a fine in case the contract is not respected. A powerful decrease in voluntary cooperation is observed in the second treatment and contracts including performance incentives turn out to be on average less efficient than the ones that do not include them. Agency theory is contradicted through these results, since incentive contracts were expected to be more efficient than contracts without incentives. It would initially seem that the decrease in voluntary cooperation is consistent with Fehr & Schmidt's (1999) inequity aversion model. After performing another treatment, identical with second one, the only difference being that the incentive is framed as a bonus and not as a fine the inequity aversion explanation is not as plausible anymore. The difference in voluntary cooperation levels between the second and third treatment is most probably owed to the way the two incentives are perceived: the fine as a hostile attitude, the bonus as an act of kindness. Fehr & Gächter's (2002) results are relevant to real-life settings, where the principals consider the agents’ voluntary cooperation very beneficial. Following a similar idea and line of thought, Fehr & Rockenbach (2003) consider that altruism based on fairness is a principal source of human cooperation and find that sanctions that show a selfish or greedy attitude almost completely destroy the altruistic cooperation. In both papers, the idea that economic incentives don’t generally have a negative effect from a motivational perspective and that it is sanctions that have an especially harmful influence on motivation is particularly stressed.

Falk & Kosfeld's (2006) experiment is also based on a principal-agent game, but what they aim for is to find the consequences of control on motivation. The setup is very simple: an agent makes a choice about a productive activity x which bears costs for him but increments the payoff of the principal. Before the agents decision about x, the principal has to

decide whether he wants to restrict the agent’s choice set or not by setting a minimum level x > 0. Imposing a minimum level of effort in the experiment can be considered an analogous for minimum working hours or minimum quality required in a real-life situation, while leaving the decision free to the agent is the same as no control mechanisms being present in the work place. The principal has no initial endowment, while the agent receives 120 points to start with. Since the cost of effort is c(x) = x, and it generates a 2x benefit to the principal, the payoffs are 120 , respectively 2 . Expecting the effect of control to vary depending on its level, the authors have three main treatments, in which the minimum level of effort the principal can impose x is 5 (C5 treatment), 10 (C10) or 20 (C20). The main result of the experiment is that for all the three treatments there are hidden costs of control and in each of the cases these costs exceed the advantages of restricting the minimum effort level. Average efforts of agents are significantly higher when they are not controlled in the C5 and C10 treatment and still higher, even if not significantly, in the C20 treatment. The authors also find heterogeneity between the subjects of each treatment. The ones that react negatively to being controlled are always the most (42-64%), but there are also plenty of positive (20-37%) or neutral reactions (16-21%) to the principal’s decision to control. The results concerning the principals’ behaviour is somewhat surprising, the majority choosing to trust in all the three treatments (74, 71 and 52 percent for C5, C10 respectively C20). All the other similar experiments confirmed the dysfunctional effects of incentives, as this paper does as well, but the majority of principals still chose those harmful actions (Fehr & Gächter, 2002; Fehr & List, 2004; Fehr & Rockenbach, 2003). Falk & Kosfeld (2006) justify this result by assuming that the principals anticipated the side-effects of controlling being stronger than the benefits this decision would bring.

Using the strategy method, the authors elicited the agents’ choices by asking them to decide on the level of effort for both possible situations: with or without control. In order to make sure this does not reduce the agents’ reaction to being controlled, a control treatment, SR10, was introduced. In this treatment, a specific response method is used and, thus, the agents make only one choice for their effort, after learning whether their principal decided to impose a minimum level of effort or not. Comparing the results of this treatment with the ones from C10, it results that agents’ choices are not significantly different when using the strategy method for either of the two possible situations – the principal trusting or monitoring. Also, a significant difference in the distributions between the two cases is found for the SR10 treatment.

In order to analyse the agents’ reactions to being controlled an additional control group was created, where the control decision was exogenously induced (EX10). The agents’ choices of effort when their action set was restricted was significantly higher when the control was exogenous, which proves that it is the reaction to the principal’s decision that demotivates the agents. One of the suggested reasons for this is that controlling reveals the principal’s expectations being lower than the agent might have thought were he trusted. This could, indeed, justify the agents choosing higher levels of effort when being allowed to decide freely. Another important aspect is revealed by the questionnaire administered to the agents concerning their feelings when having a minimum transfer imposed by the principal. The answers were classified in six categories and also by the agent’s reaction to control. The agents who reacted positively or neutrally were mostly understanding of the principal’s decision, the next most frequent feeling being distrust. In the case of the agents who reacted negatively, lack of autonomy and distrust were felt the most, by around 50 percent of the subjects. The main conclusion of this paper is that explicit incentives can backfire and that this matter should be considered when making decisions about control and monitoring, since there are chances the side-effects overcome the benefits. Having a somewhat isolated setup, it cannot be decided generally whether it is better to trust or to control, the extent of the benefits and costs depending on various factors.

Frey (1993) has a theoretical study that determines that one of these factors is how personal the relationship between the principal and the agent is. Therefore, the author concludes that the presence of a psychological contract between the principal and the agents makes a difference in whether increased monitoring is perceived as a signal of distrust or not. The crowding-out effect, thus, takes place when the principal and the agent have a personal relationship. It is then when the agents are affected by the show of distrust and they feel that their dedication is not appreciated. At the opposite end, for a non-personal principal-agent relationship, such as a competitive market setting, the disciplining effect of monitoring is expected to prevail, as agency theory predicts.

Sliwka (2007) develops a theoretical model according to which the belief one expresses has a great influence on the counterpart’s choice. So if a principal trusts, he influences the agent’s view on the typical behaviour of the reference group leading him to believe the social norm is not being selfish. As most agents are influenced by social norms, they will act according to them as well. Thus, an agent who has been trusted will most probably adapt to the information he got about the social norm and prove himself trustworthy,

while an agent who’s been given information about the group being selfish, by a principal who chose not to trust, will also adapt and confirm the selfishness. Another interesting observation in this paper is about Falk & Kosfeld's (2006) explanation concerning distrust aversion. While they argue that the agents respond with lower effort levels because they dislike not being trusted by the other player (the principal), Sliwka (2007) highlights the fact that the experiment was anonymous and, therefore, principals and agents did not meet. When deciding whether to control or not, the principals don’t know the agents they are paired with and their decision is then made based on their beliefs on the distribution of types of agents. Thus, it might be reasonable to assume that the reaction of the agents is also partly owed to the information they get about the reference group through the principal’s decision.

Supporting a similar idea, Fehr & List (2004) find that trust reinforces trustworthy behaviour. Their experiment is based on a combination of a transfer in an altered trust game and a possibility to choose a sanction if the back-transfer is lower than the requested one. The subjects, CEOs and students, can also abstain from using the sanction. As found by the previously discussed papers, imposing a threat leads to less trustworthy behaviour and, in this case, lower back-transfers. The direct effect and impact is supposedly the same for an agent whether there is no possibility for a sanction or if there is one, but the principal decides not to use it. Despite this, agents transfer significantly more when the punishment threat is available but the principal decides not to use it. Therefore, refraining from imposing the sanction is perceived as a trusting act and reinforces a trustworthy attitude from the agent. In spite of this, most principals decide to use the punishment threat, but CEOs significantly less than students, which leads them to higher levels of efficiency.

Next to incomplete contracts, an important characteristic of most economic relationships is the fact that they are constituted of repeated interactions. Gächter & Falk (2002) study the interaction between reciprocity and repeated games, having two treatments based on a gift-exchange game. The first treatment, the baseline, consists of ten one-shot games in which every pair is different and it is used to determine the intensity of reciprocity. In the second treatment, the same team plays the one-shot game ten times, allowing for reciprocity but also for repeated game effects. The results show a significantly positive wage-effort relationship, which translates into reciprocity, is present in both treatments. However, for the repeated interaction treatment this relationship is significantly steeper. From this, it can be inferred that some of the selfish individuals that wouldn’t normally reciprocate have been “disciplined” because of reputation concerns.

Wernerfelt's (1988) paper analyses reputation, monitoring and effort, but from a rather different perspective compared to the present work. First of all, it is a theoretical study that looks at the effect of monitoring on reputation building. Secondly, the monitoring decision is viewed only from the principal’s perspective, without considering the individual impact that it has on the agent. The agents’ decisions are only analysed regarding their concerns about reputation building, the reaction to the monitoring decision of the principal not being taken into account. One of the most important conclusions concerns risk-neutral agents, whom the principal monitors in early stages, while the intensity of the agents’ effort might be lower or higher initially. In the case of risk-averse agents, Wernerfelt (1988) claims that they work harder in the incipient stages, without being able to tell whether the principal will monitor earlier in the game or not.

Servátka's (2009) experiment on reputation, social influence and identification effects in dictator games comes to the conclusion that reputation, causing indirect reciprocity, has the greatest effect of the three studied concepts on the dictator’s actions. The paper supports the idea that reputation information leads to indirect reciprocity. Also, the part of the experimental design concerning triggering reputation concerns serves as a model for the present paper.

III. Methodology

The Game

The experiment consists of two treatments, each of them comprising two rounds of Falk & Kosfeld's (2006) two-stage principal-agent game. The first round is exactly the same as their C10 treatment. Principals and agents are randomly and anonymously matched in pairs. With an initial endowment of 120 points, an agent chooses a level x for a productive activity that costs him c(x) = x and brings the principal a benefit of 2x, therefore always having a marginal cost that is lower than the marginal benefit of effort. Having no initial endowment, the principal can decide, before the agent makes his effort choice, whether he wants to impose a minimum level of effort for the agent or not. The payoffs are, thus, 120 for the agent and 2 for the principal. This level is fixed at 10 points and enforcing it would restrict the agents action set to {10, 11, 12, …, 120}, while being allowed to decide freely gives an agent the possibility choose any of the following values {0, 1, 2, …, 120}. Obviously, allowing the agent to decide freely would not guarantee a positive payoff, while imposing the minimum transfer of 10 would ensure a payoff of 20. Also, the agents’

choices were elicited through the strategy method, which involved them making an effort choice for each possible situation before being informed about their principal’s decision. Because of the limited number of available subjects for the experiment, I assumed, without actually introducing an analogous treatment, that the same conclusions from Falk & Kosfeld's (2006) SR10 treatment would hold and, thus, the agents’ choices would not be significantly different if they came as specific responses. For the same reason, the EX10 treatment results were assumed to hold without checking again that the agents’ reactions to being controlled are significantly stronger when the decision comes from the principal rather than exogenously.

The second round is very similar to the gift exchange GE10 control treatment from Falk & Kosfeld (2006). Principals and agents are matched, without any pair being repeated. The only difference compared to the first round is that, in this case, the principal also gets to decide a wage w for the agent (the initial endowment) from the following values {30, 60, 120}, the agent’s payoff consequentially depending on this decision. The possibility to impose a minimum transfer of 10 points remains. The payoffs are, thus, for the agent and 2 for the principal. The possible action sets for the agents are, in this case, {0, 1, 2, …, w} when the principal decides to allow them to decide freely on the level of effort and {10, 11, 12, …, w} when a minimum level of 10 points is imposed by the principal.

Control Treatment

This treatment is used to set a baseline for the agents’ attitude when being controlled. Both first and second round take place just as described above, with the pairs being changed in the second round. All the details, including the changing of the pairs, were completely clear to the subjects when the experiment started.

Main Treatment

Reputation concerns are induced in this treatment. To make the following explanation clear, we will look at things from an agent’s (A) perspective. P1 will be the principal the agent (A) is matched with in the first round, while P2 will be the principal from the second round. In the second round, the principal (P2) is informed about the decision the agent he is currently matched with (A) took in the first round. Since the agents need to make two choices, the one that is relevant to the outcome, and implicitly to the payoffs, is determined by the principal’s (P1) decision. Therefore, if the principal from the first round (P1) decides to control, the principal from the second round (P2) is informed about the agent’s (A) decision from the first round concerning the situation when a minimum transfer is

imposed. If P1 decided to allow the agent to choose freely the amount he wants to transfer, then P2 will be informed about the agent’s choice in the first round for the case where his action set is not restricted. While this was carefully explained to the agents, the principals are not aware of the agents having to actually make two decisions in each round, and therefore they were only be told that they will learn the agent’s choice from the first round, but not P1’s decision. A very important aspect of the design is properly explaining to the agents before the experiment starts about the informing process at the beginning of the second round. Since this is how reputation concerns were induced, it was crucial to make sure the agents were aware of the fact that their initial endowment from the second round would be set by the principal that was informed about their choice in the first round. Which information would be transmitted was clearly explained in the instructions (see Appendix).

There were several options to be considered concerning the information provided to the principal in the second round. For instance, informing the principal about the agent’s both choices would have been a possibility. On one hand, this option would have had the advantage of providing more complete information and a more appropriate basis to judge the agent’s choice on, but also the disadvantage of revealing to the principals the fact that the agents make their choices before knowing whether they are controlled or not. A method to avoid revealing this detail would have been to inform the principal on the agent’s choice together with the choice of the principal from the first round. This way, comparisons would also have been more accurate, but the feedback the principals got might have led to more than reputation concerns for the agents: the choice another principal previously made might lead the present principal’s decision. The last possibility was to inform the principals about only one of the choices (under control or under trust), with the obvious comparability advantage. On the other hand, this would have implied revealing to the principals the method of elicitation for the agents’ choices, as well as only creating reputation concerns for one of the agents’ choices and not for both.

The alternative was also considered for changing the matching of the principals and agents for the second round, but having different pairs in the second round fitted better the situation. Keeping the same pairs in the second round would have led to a more “personal” relationship between the agents and the principals. This would have enhanced the reputation concerns, on one hand, but it might have also lead to a small number of very different observations, considering the limited number of available subjects. Having a different pairing for the second round is expected to reduce the heterogeneity of the observations.

Procedures

For both treatments, each principal-agent game, formed by the two rounds, was played one-shot. All experiments were run paper and pencil. Subjects were students at the “Anghel Saligny” Technical High School in Bucharest, Romania, and were aged 16-18. None of them had participated in an experiment before. A total of 96 subjects participated in the experiment, 48 as principals and 48 as agents. Each subject participated in only one treatment, in only one of the roles. At the beginning of each session, each subject was randomly attributed his role: participant A (agent) or participant B (principal), which they kept throughout the experiment. All the instructions and decision sheets were formulated as neutrally as possible, therefore the subjects only identified themselves as participant A/B and not as principal/agent (see Appendix).

The main treatment was organized in three sessions, with 10, 22, respectively 18 subjects, while the control treatment only had two sessions, one with 22 subjects and the other with 24. The average duration of the sessions was 50 minutes and subjects earned, on average, RON 14.68 (approximately EUR 3.26)1_{, plus a one-week voucher to a gym in the same}

neighbourhood as the high school, as a show-up gift. The gift was not announced previously and therefore no self-selection problems should arise from this perspective.

After being attributed one of the roles, principals and agents were sent to different but neighbouring rooms and sat far from one another. They were given the set of instructions according to their roles and, also, a set of control questions. The procedure and payoffs were common knowledge to both principals and agents and it was made clear that they are not calculated differently for a certain role. The only information that the principals were not aware of was the fact that the agents make effort choices for both potential situations, control and no control, before being informed about the principals’ actual decisions. After being read aloud the instructions for the whole experiment, the subjects were also presented a summary of them – after being explained all the rules, it was important to make sure the order of events was also clear for everybody, especially for the subjects from the main treatment. Before starting, the subjects were asked to solve the control questions. The experiment only began when all the subjects had answered correctly to each control question.

1 _{The minimum hourly wage in Romania is approximately EUR 4.48, while the average hourly wage is around}

Verbal instructions were provided each step of the experiment. When the first round started, principals were given decision sheet 1B (see Appendix) and were asked to decide whether they want to impose a minimum transfer of 10 points, while the agents received decision sheet 1A (see Appendix) and made their transfer choices for the case when they would be allowed to decide freely, as well as for when the minimum amount imposed by the principal was 10. The first round was identical for both treatments.

For the control group, the second round began with the principals filling in the decision sheet 2Bc with the amount they chose as an initial endowment for the agents. In the treatment group, decision sheet 2Bt was provided to the principals. This sheet asked for the same choice as the one in the control group, but also provided information about the agent’s choice in the first round. After that, the agents received a decision sheet (2A) for their effort choice that also contained the information about their initial endowment and the principals received a new decision sheet 1B and chose whether or not they wanted to impose a minimum transfer. After the second round ended, all subjects received a sheet informing them about their own outcomes of the two rounds.

Behavioural Predictions

The assumptions made about the players’ preferences can lead to different predictions about the outcomes of the experiment. First, let us assume common knowledge of selfishness and rationality, the standard situation. Agents that only care about maximizing their own income will always choose the minimum level of effort x. When allowed to decide freely, their choice of effort will be 0 and for the case when the principal decides to control, they will choose a transfer of x = 10, the minimum imposed amount. In this situation, agents can be considered opportunistic and preventing them from shirking by restricting their choice will undoubtedly lead to a higher income for the principal.

The homo economicus assumption has been, however, disputed repeatedly. According to Camerer (2003), individuals do not normally act selfishly, but according to their social preferences. Experiments involving dictator games find that positive amounts are transferred most of the times and Fehr & Schmidt (1999) explain this through people’s inclination to care for efficiency and equity. Based on this theory and the positive transfers dictators usually make, it is highly likely that most of the agents will choose an effort level above the minimum required. Some agents will have strong preferences for equity and will, therefore, choose to transfer amounts above the minimum potentially required, x = 10. Agents

with weaker social preferences will choose positive effort levels, but there are chances these are still below the minimum that the principal has the chance to enforce. Also according to the social preferences theory, it could be predicted that agents will not care whether they are controlled or not. If this were to be true, it would always be optimal for the principals to control and, thus, increase the transfers of the agents with weak social preferences.

When taking into consideration the previously discussed literature and theories about intrinsic motivation being crowded-out by incentives, predicting outcomes and detecting the optimal strategies becomes more difficult and unclear. There are, mainly, two ways through which the demotivation of the agents might occur in this experiment as a cause of the principal’s control. The first one is the signal of distrust that a principal sends when she decides to restrict the agent’s choice of effort. As previously discussed, trusting reinforces trustworthy behaviour and it also represents a sign of respect. On the other hand, when the principal decides to control and, thus, impose a minimum level for the transfer, she gives information to the agents about her expectations. Not controlling allows the agent to make his own impression about these expectations. If this impression was higher than the requirements the principal decides to express, the agent might reconsider his choice by adapting to these requirements. If the agents act accordingly to this behaviour, the principal’s optimal strategy is the opposite than under the previous assumptions: not controlling would lead to better outcomes. All this also depends on whether the level she can enforce is higher or lower than what the agent would decide freely, so, as mentioned above, it is more difficult to determine the optimal strategy in this case. However there are a few predictions that can be made about the results of the experiment.

Hypothesis 1: Effort levels in the first round will be higher for the treatment group,

compared to the control group.

For the agents in the treatment group, the relevant choices from the first round will be reported at the beginning of the second one to the principals. After being informed, the principals will make their decisions about the wages w/initial endowments of the agents in the second round. Being aware of this, agents will consider reputation when making their choices of effort in the first round. Since in the control group no information is transmitted to the principals about first round effort choices of the agents, it is expected that the agents in the treatment group will choose higher effort levels, considering the influence these choices might have on their initial endowments from the second round.

Hypothesis 2: Higher effort levels in the first round will be rewarded with higher initial endowments, while lower effort levels will punished with lower initial endowments (in the treatment group).

The rational decision concerning the initial endowment would be for all principals to provide the maximum possible amount. Despite this, there are several other aspects that will influence these decisions. For instance, indirect reciprocity is also expected: principals in the second round are expected to provide higher initial endowments for agents that chose higher effort levels in the first round, even though these implied transfers to a different principal themselves. Of course, assessing whether the effort choice in the first round will be subjective, depending on each principal’s expectations and beliefs, and imprecise, because they will not be informed whether that effort choice was made under a restriction or not and also because of having no term for comparison (no information about the average effort levels in the first round).

Something that cannot be predicted but will be interesting to follow is the effect of reciprocity. Higher effort levels of the treatment group in the first round will lead to higher wages chosen by the principals. These higher wages will be further rewarded with higher effort levels. When reporting efforts from the second round to the ones in the first round for each treatment, will a better evolution be found for the treatment group or not? In other words, does reciprocity intensify the agents’ motivation or not?

IV. Results

Summary statistics

Table 1 contains information about the agents’ effort choices (mean, standard deviation, median) separately by treatment/control group, round and principal’s choice. The first observation confirms what the other studies have found previously: the homo economicus theory does not hold. Not only the average effort is above from 0 in the case when the principal trusts and above 10 when she controls, but the average efforts turn out to be higher in the situation of trust, even if it implies a lower minimum transfer for the agents. In the case when the principals trust, average effort levels vary from 23.00 in the second round of the treatment group to 29.92 in the first round of the same group. When the principals decided to impose a minimum amount to be transferred, average choices of the agents were between 14.16 in the second round of the treatment group and 19.88 in the first

round of the same group. It can be, thus, observed that the treatment group has the lowest averages of the experiment in the second round, as well as the highest ones in the first one. The smallest variations were found in the effort choices of the agents for the situation when they were being controlled, in the first round of the control group, as well as in the second round of the treatment group, which indicates less heterogeneity in the agents’ reaction to control in these situations.

Table 1. Agents’ effort choices, by group, round and principal’s decision

Treatment Group (25 observations) Control Group (23 observations) Round 1 Control Average 19.88 14.26 Std. Dev. 8.20 4.73 Median 18 14 Trust Average 29.92 24.30 Std. Dev. 16.32 14.65 Median 36 20 Round 2 Control Average 14.16 17.95 Std. Dev. 4.26 8.81 Median 13 15 Trust Average 23.00 24.39 Std. Dev. 10.98 13.92 Median 23 25

Also, for each of the situations the average effort level that the agents chose seems to be higher in the situation of no control. The average effort difference between the trust and control cases starts from 6.44, for the second round of the control group, rises to 8.84 in the second round of the treatment group, going up to 10.04 for both groups in the first round. This observation hints to the presence of hidden costs of control, but the variation in the average effort levels and in the difference between the two cases, depending on the round and group will also receive attention later in the paper.

Table 2 presents the principals’ decisions concerning control for each round of the two groups. Unlike the agents’ decisions, the principals’ choices are very different to the ones Falk & Kosfeld (2006) got. In every situation, principals decided to impose a minimum amount in more than 70 percent of the cases, as found in most other similar studies (e.g. Fehr & List, 2004).

Table 2. Average control rate, by group and round

Treatment Group (25 observations) Control Group (23 observations) Round 1 72.00% 73.91% Round 2 76.00% 78.26% Results

Figure 1a-d displays the cumulative distributions of agents’ choices for each group and round. The four graphs also confirm the existence of hidden costs of controlling the agents. For every value of effort above the minimum enforceable amount x > x, the number of trusted agents who chose to transfer at least x is higher than the number of agents who made the same choice when being controlled. Had there been no reaction to being controlled, the distribution for the values of x > x for the case when the principal chooses to control would have been the same to the one when she trusts the agents. That doesn’t prove to be true for any of the groups and rounds.

In Figure 1c, for instance, representing the first round of the control group, 52 percent of the agents who were allowed to decide freely chose effort levels above 20, while only 4 percent of the controlled agents transferred amounts of 20 or more. On the other hand, 30 percent of the agents whose choices were restricted chose the minimum level of 10 but only around 23 percent of the agents who were trusted transferred 10 or less. The conclusions are similar for the other three situations, since they all display the agents’ negative reactions to being controlled.

In order to determine whether control does have an effect on effort, the distributions of the choices with and without control were compared, following Falk & Kosfeld's (2006) method. First, the distribution of effort choices for the case when the principal trusts was modified so that all effort choices below 10 were set equal to 10. Assuming there had been no effect of control, the modified distributions would have been the same. This hypothesis is

.1 .2 .3 .4 .5 .6 .7 .8 .9 1 C umu la tive fre qu en cy 5 10 15 20 25 30 35 40 45 0 50 55 Effort Control Trust

Treatment Group - Round 2

Figure 1b. Cumulative distribution of agents’ effort choices for the treatment group, round 2

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 C um u la tiv e fr e qu en cy 0 5 10 15 20 25 30 35 40 45 50 55 Effort

Treatment Group - Round 1

.1 .2 .3 .4 .5 .6 .7 .8 .9 1 0 C um u la tiv e fr e qu en cy 0 5 10 15 20 25 30 35 40 45 50 55 Effort

Control Group - Round 1

Figure 1c. Cumulative distribution of agents’ effort choices for the control group, round 1

0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 C umu la tiv e fr e qu en cy 0 5 10 15 20 25 30 35 40 45 50 55 Effort Control Trust

Control Group - Round 2

rejected and the shifts in the distributions turn out to be significant for each group and round, according to the Wilcoxon signed rank test (p = 0.0035 for the first round and p = 0.0001 for the second round of the treatment group; p = 0.0005 for the first round and p = 0.0021 for the second round of the control group)2. Evidently, some benefits come from restricting the agents to choose effort levels below 10. According to the information Table 1 gives about the agents’ choices in each round and group, it can be concluded that the higher average transfer for the situation with no control compared to the one when the principal decides to control shows that these benefits are lower than the costs that the decision to control bears. Also, the averages for the situation when the agents are being trusted are from 36 to 70 percent higher than when the principals decide to control them.

We will now turn to comparing average effort choices of agents throughout rounds and treatment groups. In the first round, for the treatment group, average effort choices of the agents are different from the ones in the control group both when principals decide to control or to trust, even if only significantly so for the situation of control (Mann-Whitney test, p = 0.1825 for the agents who were not controlled, p = 0.0054 for the agents who were imposed a minimum amount to transfer). As previously mentioned, the only difference between the procedures of the treatment and control group was the induction of reputation concerns, by telling the agents from the beginning that the principals in the second round will be informed about the decisions they take in the first round.

Therefore, hypothesis 1 is confirmed: reputation concerns lead to higher effort levels from the agents, both when being controlled and when being allowed to decide freely. The absolute average difference between the trust and control situations is exactly the same for the two groups (29.92 – 19.88 = 10.04 = 24.30 – 14.26) and the distributions of these differences are very similar for the treatment and control group (Mann-Whitney test, p = 0.9260). Even so, it is important to mention that this difference represents a higher percentage from the average effort choices in the control group (41.32%) than in the treatment group (33.56%). Thus, we could conclude that reputation not only increases the average amount the agents transfer, but it also diminishes the reaction to being controlled. But this only holds if we assume the reaction to being controlled as a variable cost. Were we to consider it a fixed cost, then the conclusion would be that reputation concerns only increase the average effort of the agents, but have no influence on the intensity of their reactions when being controlled.

Table 3 contains information about the principals’ decisions concerning the initial endowments of the agents for the second round. The initial endowments the principals from the control and the treatment group choose are significantly different (Mann-Whitney test, p = 0.0352), with higher averages for the control group. For the second round, the main difference between the two groups is the fact that only principals in the treatment group receive information about the decisions the agents for which they are deciding the endowment took in the first round.

Table 3. Principals’ average decisions concerning the agents’ initial endowments in the second round

Treatment Group (25 observations) Control Group (23 observations) Round 2 Average 67.20 91.30 Std. Dev. 32.73 37.82 Median 60 120

In the control group, principals make the endowment decision without receiving any information. In the treatment group, the average reported effort was 22.56, with a standard deviation of 11.54. Considering the fact that the initial endowments set by the principals were chosen from a set of three values, {30, 60, 120}, performing a linear regression to determine how much their choices were influenced by the information they received on the agent’s effort choices in the first round would not be appropriate. A more suitable approach would be to run an ordered logit regression in order to determine whether the information had a significant influence on the principals’ choices of initial endowments. The results of this regression show that the coefficient of the information is significantly different from zero (z = 3.07, p = 0.002). since the coefficients resulting from an ordered logit regression cannot be interpreted as regular OLS coefficients, the probabilities of receiving an initial endowment of 30, 60 and 120 for each level of effort reported to the principals have been predicted and are shown in Table 4. For the lowest reported effort level, 4, the possibility of receiving the lowest initial endowment, 30, is of 77.55 percent, while the possibility of receiving an initial endowment of 120 is less than 1 percent. Overall, the probability of a low initial endowment decreases as the reported effort level increases. For the middle value, 60, the probability increases with the reported effort level of 22 and it starts decreasing as the effort choice grows higher. The probability of receiving the highest initial endowment, 120, raises constantly with

the reported effort level. These predicted probabilities are the upright confirmation of the second hypothesis.

Since 72 percent of the principals decided to control in the first round, 72 percent of the choices reported in the second round were the ones agents made for the situation when

Table 4. Probability of receiving each potential initial endowment, by reported effort levels

Treatment group, Round 2 (25 observations)

Effort Frequency

Initial Endowment probability

30 60 120 4 1 0.7755 0.2168 0.0077 10 3 0.5608 0.4186 0.0206 12 1 0.4782 0.4934 0.0284 15 3 0.3578 0.5963 0.0459 17 1 0.2856 0.6515 0.0628 18 2 0.2530 0.6736 0.0733 20 3 0.1955 0.7051 0.0993 22 1 0.1485 0.7183 0.1332 23 2 0.1288 0.7177 0.1535 26 1 0.0824 0.6878 0.2298 28 1 0.0606 0.6458 0.2937 32 1 0.0321 0.5212 0.4467 36 1 0.0168 0.3727 0.6105 38 1 0.0121 0.3019 0.6860 40 1 0.0087 0.2386 0.7527 41 1 0.0074 0.2103 0.7823 51 1 0.0014 0.0489 0.9497

they are controlled and this explains the relatively low average effort that the principals were informed on. From the difference between the distributions of initial endowments for the two groups, it is reasonable to assume that the principals thought the choices were too low and therefore responded with lower wages to them. The slight increase in the control rate in the second round, 4 percentage points, can, thus, be justified as some kind of insurance against the agents’ reaction to lower than maximum (120) wages.

The decrease in the average effort in the second round, for the treatment group is significant for the situation of control, but not for when the agents are allowed to decide freely (Mann-Whitney test, p = 0.0064 and p = 0.1620, respectively. The only difference between the treatment and control group in the first round is reputation. The differences between the first and second round of the treatment group are the elimination of reputation concerns in the second round and, also, the lower average initial endowment. Since the average initial endowment in the second round of the treatment group is almost half compared to the one in the first round, lower average effort levels would have been expected. Nevertheless, the average effort levels are very similar for the first round of the control group and the second round of the treatment group. This indicates that the reduction in the initial endowment, together with the elimination of reputation didn’t have as strong of an effect as it would have been predicted, since their joint effects are as strong as the effect of reputation that we observe by comparing the average effort levels of the two groups in the first round.

For the control group, the average effort does not change in the second round for the case of no control, even if the initial endowment is significantly different compared to the first round (Mann-Whitney test, p = 0.001), a lower average in the second round. Also, the average transfer of the agents who were controlled increases in the second round, even if not significantly so (Mann-Whitney test, p = 0.3387). A possible explanation for the fact that the average transfers not only did not decrease, but even increased in case of control is reciprocity. Even if they had the possibility to choose lower initial endowments, 14 of the principals chose the maximum amount of 120, 5 chose the middle option of 60 and only 4 principals decided on the minimum initial endowment. Just as Fehr & Rockenbach (2003) and Fehr & List (2004) found, having the choice to do harm and abstaining from it induces rewards and reciprocity, compared to the situation where doing harm is not an option. This effect seems not only to increase the general choice of effort, but even diminished the reaction to being controlled, reducing the average difference between the transfers in the trust condition and the ones in the control condition from 10.04 to 6.44.

Table 5 shows the types of reactions to control the agents have in every round of every group, in absolute and relative frequencies, and their average choices dependent on the principal’s control decision. As the previous results would predict, most agents react negatively to being controlled: 64 and 76 percent in the first, respectively the second round of the treatment group and 69 and 61 percent in the first and, respectively second round of the control group. The average effort values for the agents who reacted negatively reflect the same effects that we observed on the overall effort averages: reputation concerns lead to higher transfers in the first round of the treatment group, compared to the control group. Also, due to reciprocity, the average effort levels in the second round of the control group are higher, despite the lower average initial endowment. The same things can be observed concerning the agents who are neutral to the principal’s decision to control: average transfer amounts are the highest in the first round of the treatment group and in the second round of the control group. However, only between 8 and 17 percent of the agents reacted neutrally. The agents who reacted positively are the ones that are selfishly motivated. For the treatment

Table 5. Agents’ reactions to control, by group and round

Treatment Control

Negative Neutral Positive Negative Neutral Positive

Round 1 No. of agents 16 2 7 16 2 5 Share 0.64 0.08 0.28 0.69 0.09 0.22 Average effort when controlled 18.06 28.0 21.71 13.9 17.5 14.0 Average effort when trusted 37.50 28.0 13.14 31.5 17.5 4.0 Round 2 No. of agents 19 2 4 14 4 5 Share 0.76 0.08 0.16 0.61 0.17 0.22 Average effort when controlled 14.9 12.5 11.5 17.4 25.0 14.0 Average effort when trusted 27.6 12.5 6.5 31.4 25.0 4.2

group 28 percent of the agents reacted positively in the first round and 16 percent in the second round; in the control group, 22 percent of the agents reacted positively to being controlled both in the first round, as well as in the second one. The average effort levels that they chose are only slightly above the minimum, be that 0 when the principal trusts or 10 when she decides to control. The only exception is the first round of the treatment group, where reputation concerns seem to be incentivizing them to choose effort levels that are comparable to the other agents’.

V. Discussion and conclusion

There are a few aspects because of which the results of the paper should be taken with caution. First of all, there are two assumptions that were made for the experiment without actually being tested, because of the limited number of available subjects and financial resources. The assumptions were based on the conclusions drawn from two of Falk & Kosfeld's (2006) control treatments: the SR10 treatment and the EX10 treatment. SR10 was used to check whether the agents’ decisions are significantly different when they make their transfer decisions after being informed about the principal’s choice concerning control. Thus, agents chose how much effort to put in after knowing exactly in which situation they find themselves, unlike the main treatments, where one choice had to be made for each possibility and only one of the two became relevant, after the principal made her choice. The results showed that agents don’t behave significantly different in this situation, which confirmed the validity of the method used for the main treatments. The same method was also used for the experiment in the present paper, assuming that the subjects would not react differently if they knew the principals’ decisions before making their effort choices.

The second control treatment, EX10, had the purpose of verifying that it is the decision of the principal that leads the agents’ choices of effort, and not the constraint itself. The results showed that, indeed, agents react strongly to the constraint when it comes as a principal’s decision rather than as an outside rule. The principals made no decision in this treatment and the minimum imposed transfer was given to the agents exogenously. Thus, the agents only had to choose the amount of effort for the situation with a constraint. This aspect can also represent a potential limitation for Falk & Kosfeld's (2006) results, since only having one decision to make might not make the comparison with the results of their main treatment, C10, as relevant. When asked to make two choices at the same time, the agents might consider that there is supposed to be a difference between the two choices and make one them

reporting to the other. In this situation, this line of thought is excluded from the agents’ perspective when only asking them to make one choice. A proper way to ensure the comparability of the EX10 and C10 treatments would have been to also ask the agents to make an effort decision when no constraint would be imposed, which would have almost been a basic dictator game, with the exception of the fact that the principals received their choice of transfer doubled. It is, therefore, not completely secure to assume that the agents in our experiment only reacted to the principals’ decision and not also to the fact of being restricted, especially since even Falk & Kosfeld's (2006) control treatment concerning this matter might be slightly flawed. Overall, the results from the part of the experiment that overlapped theirs are quite similar, the only exception being the percentage of principals who decided to control. Due to using the strategy method as elicitation procedure, this exception is not relevant for the agents’ choices. It should, thus, be rather safe to assume that the results of both treatments would hold for the present paper as well, with some reserve still.

Another aspect that should be taken into consideration is the fact that, in the second round, the decision the principal makes does not affect his final outcome directly. It is, indeed, not very realistic to consider that the wage the principal offers bears no direct cost to her. As a consequence, the rational choice for all principals would be to give all the agents the maximum initial endowment of 120, since this would increase their chances of getting a higher payoff by ensuring the agents the highest amount to transfer from. Despite this, rationality seems to be taken over by the instinct to punish choices from the first round that are considered unfair by giving the respective agents endowments below 120. Even in the control treatment, where principals receive no information on the agents’ previous choices, there are some principals who do not give the maximum initial endowment, but significantly less than in the treatment group. Judging by these results, this design problem might not be as serious as it initially seems.

A very important conclusion to be drawn from the results is that the decision whether to control or not must be carefully considered before actually implementing it. Our results show that the costs of control outweigh the benefits. Of course, this might not always be the case, but it is certainly a possible outcome and, therefore, this possibility should be excluded when making such a decision. Also, there are several factors that might push the outcomes of this choice in one direction or the other. Reputation is a very relevant aspect of the labour relation in most cases and studying its interaction with control offered some more insight about the potential consequences of the decision of control.

Being considered in general a strong enforcement device for incomplete contracts, reputation also turns out to be helpful in supressing the negative reactions to control. Thus, reputation is not only found to lead to higher levels of effort, but agents in the treatment group also diminished their choice of effort for the case when they were controlled by a smaller percentage, compared to the ones in the control group. This represents the confirmation that reputation is a factor whose influence should be accounted for when considering the benefits and costs of control.

Keeping the same pairs of principals and agents throughout the experiment and maybe adding more rounds could have been also been interesting and would have, maybe, revealed more information about a long-term relationship between agents and principals and their reactions in these conditions. The principals who choose the agents’ wages after being informed by their previous effort choices, as well as agents who make effort choices after being informed on the initial endowments the principals chose for them show signs of reciprocity. The agents’ weakness for the principal’s “kind” act of abstaining from a harmful action is another result, which, next to all the others, should be taken into account when designing the labour relation.

The purpose of extending Falk & Kosfeld's (2006) experiment by introducing reputation was creating a setting that is closer to real-life situation. This work undoubtedly brings some new information, but there are still many aspects, the nature of the relationship between the principal and the agents for example, whose interaction with control and reputation would give even more insight into this problem. Of course, the more factors considered at the same time, the more difficult to disentangle their contributions to the results.

References

Alchian, A.A. & Demsetz, H., 1972. Production, Information Costs, and Economic Organization. The American Economic Review, 62(5), pp.777–795.

Camerer, C., 2003. Behavioral Game Theory: Experiments in Strategic Interaction, Princeton University Press.

Deci, E.L., 1971. Effects of externally mediated rewards on intrinsic motivation. Journal of

Personality and Social Psychology, 18(1), pp.105–115.

Falk, A. & Kosfeld, M., 2006. The Hidden Costs of Control. American Economic Review, 96(5), pp.1611–1630.

Fehr, E., 2009. ON THE ECONOMICS AND BIOLOGY OF TRUST. Journal of the

European Economic Association, 7(2/3), pp.235–266.

Fehr, E. & Falk, A., 2002. Psychological foundations of incentives. European Economic

Review, 46(4-5), pp.687–724.

Fehr, E. & Gächter, S., 2002. Do incentive contracts undermine voluntary cooperation? , (April 2012).

Fehr, E., Kirchsteiger, G. & Riedl, A., 1993. Does Fairness Prevent Market Clearing? An Experimental Investigation. The Quarterly Journal of Economics, 108(2), pp.437–459. Fehr, E. & List, J.A., 2004. The Hidden Costs and Returns of Incentives-Trust and

Trustworthiness among CEOs. Journal of the European Economic Association, 2(5), pp.743–771.

Fehr, E. & Rockenbach, B., 2003. Detrimental effects of sanctions on human altruism.

Nature, 422(6928), pp.137–40.

Fehr, E. & Schmidt, K.M., 1999. A Theory of Fairness, Competition, and Cooperation. The

Quarterly Journal of Economics, 114(3), pp.817–868.

Frey, B.S., 1993. Does Monitoring Increase Work Effort? The Rivalry with Trust and Loyalty. Economic Inquiry, 31(4), p.663.

Gächter, S. & Falk, A., 2002. Reputation and Reciprocity: Consequences for the Labour Relation. The Scandinavian Journal of Economics, 104(1), pp.1–26.

Gneezy, U. & Rustichini, A., 2000. Pay Enough or Don’t Pay at All*. Quarterly Journal of

Economics, 115(3), pp.791–810.

Servátka, M., 2009. Separating reputation, social influence, and identification effects in a dictator game. European Economic Review, 53(2), pp.197–209.

Sliwka, D., 2007. Trust as a Signal of a Social Norm and the Hidden Costs of Incentive Schemes. American Economic Review, 97(3), pp.999–1012.

Stock, J.H. & Watson, M.W., 2011. Introduction to Econometrics, Addison-Wesley.

Wernerfelt, B., 1988. Reputation, monitoring, and effort. Information Economics and Policy, 3(3), pp.207–218.

Appendix

1. Instruction for control group, participants A

You are now participating in an economic experiment for a master’s thesis.

Please read the following instructions carefully. The instructions will provide you with all the information required for participation in the experiment. Please ask for assistance if there is something that you do not understand. Your question will be privately answered at your desk. Communication is strictly prohibited during the experiment.

You will each receive a one-day voucher for Westlife Fitness&Aerobic to thank you for your participation. Additionally, you can earn money over the course of the experiment by collecting points. All of the points which accrue to you over the course of the experiment will be converted to lei. Please note that: 1 point = 30 bani.

The experiment

In this experiment, each participant A is associated with a participant B in a group of two. There will be two rounds and each participant will keep his role for the entire experiment. In the second round, different groups will be created. No participant knows with whom he is associated, meaning that all decisions are made anonymously.

You are participant A. First round

Participant A receives an amount of 120 points at the beginning of this round. Participant B receives no points.

Participant A’s decision

Participant A can decide how many points he wants to transfer to participant B. The experimenter doubles each point which A transfers to B. Thus, each point which A transfers to B reduces A’s income by one point and increases B’s income by two points.

The formula for calculating income is as follows:

Participant A’s income: 120 – transfer

Participant B’s income: 0 + 2*transfer

The following examples will clarify the income formulae:

Example 1: A transfers 0 points to B. The incomes are then 120 for A and 0 for B. Example 2: A transfers 20 points to B. The incomes are then 100 for A and 40 for B. Example 3: A transfers 80 points to B. The incomes are then 40 for A and 160 for B.