Social Comparisons and Peer Effects with Heterogeneous Ability ∗ Aurélie Bonein† May 2016

Abstract When workers’ efforts are not perfectly contractible, the effort exerted by one worker may be influenced by his coworker’s effort. By considering workers of heterogeneous abilities, we investigate whether the display of workers’ efforts to coworkers influences wage and effort decisions, and we evaluate the overall impact on efficiency when workers differ in ability. We find that employers increase both wages and wage differences when the more able worker is observed. Observed workers increase their level of effort and reciprocity, and observer workers’ efforts are positively related to those of their coworkers. This strategic complementarity of effort is partially explained by inequity aversion. On average, displaying workers’ efforts to coworkers is detrimental to employer’s payoff but enhances efficiency, especially when the most valuable worker is observed.

Keywords: Heterogeneous ability, Gift-exchange game, Social comparison, Peer effect JEL Classification: C91, D03, J24, J31, J82 ∗ I am grateful to Klaus Abbink, Catherine Eckel, Olivier l’Haridon, David Masclet, Ragan Pétrie, participants at the ASFEE conference in Grenoble, at the ESA meeting in Copenhagen and members of the ANR Conflict for their helpful comments and suggestions that improve the paper a lot. I would like to thank Elven Priour for his excellent help in conducting the experiments. I gratefully acknowledge the hospitality of the University of Berkeley while working on this paper. An earlier version of this paper was circulated under the title “Wage transparency and effort comparison in the context of heterogeneous workers”. This work was supported by the French National Agency for Research (ANR Conflict ANR-08-JCJC-0105-01). All errors remain my own. † CREM - University Rennes 1. E-mail: [email protected]

1

Introduction

Although there is a long tradition in economics that emphasizes the role of financial incentives in workers’ efforts,1 the idea that these efforts are driven by more than financial incentives has received close scrutiny in recent experimental studies. These studies have shown that social pressure and peer effects (i.e., how being observed or observing others influences a focal worker’s behavior) might also be incentives in labor organizations (see Kandel and Lazear, 1992). In addition, social comparisons (i.e., how other workers’ outcomes influence a focal worker’s behavior) may have important implications in the labor market (see Cohn et al., 2014 for a recent discussion). In the present study, we consider workers with heterogeneous abilities, and we examine whether and to what extent the display of workers’ efforts to coworkers affects the financial incentives provided by employers and whether peer effects and social comparisons intervene in effort decisions. Social comparisons may arise in multi-worker settings in which workers frequently interact with one another. Some firms implement practices to favor such interactions: new physical spaces (open-plan offices, places to relax, etc.), workshop sessions or the use of new information technologies (email, chat, etc.). For instance, the human resources department of Google claims to have created the “perfect” environment to favor creativity and interactions among workers.2 Mackay (2007) also demonstrates that conversations within firms play a key role in workers’ efforts and in firm performance (see chapter 6). When workers’ effort levels cannot be enforced, this information may ultimately affect the behaviors of both employers and workers, even when there are no technical externalities across production (i.e., there is no teamwork). In addition, depending on the spatial orientation of workers in the production process, workers may have the opportunity to observe how their coworkers behave (for instance, the spatial orientation of cashiers permits workers to observe coworkers placed in front of them). Because it is impossible to prevent these interactions or the dissemination of information within firms, employers must assess the impact of observing a coworker’s effort on the worker’s own effort decisions. Although there is abundant survey and case-study evidence that has highlighted 1 See

Gneezy et al. (2011) and Gneezy and Rey-Biel (2014) for a discussion about monetary incentives. 2013, Google was named the best company to work for in America. See the original article at http: //www.cbsnews.com/news/inside-google-workplaces-from-perks-to-nap-pods/. 2 In

1

the importance of wage comparisons (see Bewley, 1999, for instance) and the impact of comparing effort (see Cialdini et al., 1990, for example) on worker behavior, until now, the studies that have examined this question have assumed that workers are identical. The purpose of this paper is to extend this research stream by examining, in the context of workers with heterogeneous abilities, whether and how the subsequent display of a worker’s effort to his coworker affects the workers’ behaviors and employers’ wage decisions. Under similar ability levels, employers have a priori no reason to discriminate among workers, and workers may easily imitate the effort exerted by their coworkers. Workers may perceive any wage difference as unfair, and such unfairness may have detrimental effects on employers’ payoff and efficiency.3 However, the story is more complex when workers differ in ability. In this setting, employers may take advantage - under some conditions - by favoring the most able workers, and the question of the spillover effects in effort remains open in this context. Answering this research question is of particular interest for labor organizations. Most often, workers differ from one another in terms of their abilities. This heterogeneity may result from differences in length of employment, human capital, educational attainment, qualifications, skills or (more broadly) personal backgrounds. If the ability level of the observed worker (respectively observer worker) impacts the level of total output (i.e., the sum of the efforts provided), it is important for employers to determine the optimal order of workers’ moves according to their respective ability when there is an opportunity to design the spatial orientation of workers in the production process. In the present paper we analyze whether the display of a worker’s effort to his coworker influences (i) the employer’s wage decisions, (ii) the behavior of heterogeneously skilled workers and (iii) the employer’s payoff and overall efficiency when there are no technical externalities across production. To that end, we conduct an experiment based on the gift-exchange game that was first developed by Fehr et al. (1993) to study labor relationships under contractual incompleteness in a laboratory setting.4 Each firm consists of an employer, a high-ability worker and a low-ability worker. Employers and workers are 3 Here

efficiency is simply understood as the sum of payoffs, not in the sense of Pareto efficiency, as in Engelmann and Strobel (2004). 4 See Fehr et al., 2009 for a recent review.

2

equally informed about each worker’s ability, and the effort-cost relationships are common knowledge. In the first stage of the game, the employer selects the wage he offers to each worker. In the second stage, workers learn their wage and that of their coworker; then, each worker chooses his level of effort. We implement three experimental treatments that vary only with respect to the information provided to the workers when they choose their effort levels. In the control treatment, the workers were unable to observe the effort exerted by their coworker. In the second treatment, the more able worker observed the effort decision of the less able worker in his firm before selecting his own effort level. The order of the workers’ moves was reversed in the third treatment. We find that enabling effort comparisons among heterogeneously skilled workers affects wage decisions by inducing employers to increase both (i) wages and (ii) the difference between the two workers’ wages when the effort exerted by the more able worker is displayed. This latter wage strategy increases the targeting effect toward the more able worker. Regarding worker behavior, although the observed workers and the observer workers have heterogeneous levels of ability, the observability of a coworker’s effort induces observed workers, especially the less productive ones, to increase their efforts and the degree of reciprocity (peer effect). Moreover, the efforts of observer workers are, on average, positively related to those exerted by their coworkers. This strategic complementarity of efforts is partially explained by inequity aversion. Finally, displaying a worker’s effort to his coworker has a detrimental effect on the employer’s payoff because employers pay higher wages that are not compensated by a sufficient increase in workers’ efforts. However, because the increase in workers’ payoffs is sufficiently high to compensate for the decrease in employers’ payoffs, displaying a worker’s effort to his coworker enhances overall efficiency, especially when the more able worker is observed. The paper is organized as follows. Section 2 reviews the related literature. The experimental design and our research hypotheses are presented in Section 3 before Section 4 presents the main results. Finally, Section 5 summarizes the findings and discusses their implications for labor organizations. Proofs can be found in the Appendix.

3

2

Background

Our paper is related to two strands of the literature. The first strand consists of experimental studies that address the issue of workers’ incentives under contractual incompleteness. In the moral hazard context, incentivizing workers to exert an expected effort is an important challenge for both employers and for firm performance. Akerlof (1982) and Akerlof and Yellen (1990) highlight that one of the key determinants of the effort exerted by workers is the wage received. They demonstrate a positive relationship between the wage received and the effort exerted by a worker as long as the worker’s wage falls below the “fair” wage (the fair wage-effort hypothesis). In a later study, Bewley (1999) emphasizes the importance of the fairness of a wage offer and argues that employers seem to believe that workers’ perceptions of fairness affect their productivity. The fairness of a wage offer depends not only on the level of the received wage but also on coworkers’ wages. This raises the prospect of the influence of wage comparisons among workers on (i) wage assignment, (ii) effort exerted and (iii) overall firm performance.5 Further, a horizontal dimension generated by peer effects must be considered in a multi-worker setting. It is established in the literature that individuals change their behaviors when they are observed (Hawthorne effect). Thus, the second strand of literature refers to the effects of peer pressure on productivity (Kandel and Lazear, 1992). Previous experiments in the laboratory (Falk and Ichino, 2006) and in the field (Mas and Moretti, 2009) have found that individuals enhance their efforts substantially based on concerns about how peers will view their efforts. Since these pioneering studies, some detrimental effects of peer pressure on productivity have also been noted (see Eriksson et al., 2009 for a laboratory experiment and Bellemare et al., 2010 for a field experiment, for instance). In general, it seems that the benefits of peer effects depend on the nature of the preferences that yield such effects and the context in which peer effects apply (see Sausgruber, 2009 for a review). The present paper sheds light on the interplay between these strands of the literature because peer effects in labor organizations may be added to the vertical relationship between workers and employers and to subsequent social comparisons. Workers frequently 5 See

Gross et al. (2015) for a recent experiment and dedicated literature review.

4

have the opportunity to observe not only the relationships between the employer and their coworkers but also their coworkers’ behavior, and both of these observations may impact a worker’s own behavior. To that end, we use a three-person gift-exchange game. To date, only a few experiments have extended the bilateral gift-exchange game to a multi-worker setting. Among those that do, some experiments have analyzed how effort decisions are influenced by a coworker’s wage (Charness and Kuhn, 2007; Nosenzo, 2010; Gächter and Thöni, 2010; Abeler et al., 2010; Greiner et al., 2011; Gächter et al., 2012; Cohn et al., 2014; Gross et al., 2015), but the results of these studies are not clear-cut. For instance, Charness and Kuhn (2007) find that workers are not influenced by a coworker’s wage, whereas Nosenzo (2010) observes that knowledge of coworkers’ wages has a detrimental effect on the effort exerted by a worker. One possible explanation for this discrepancy is based on workers’ abilities. In the experiment conducted by Charness and Kuhn (2007), workers have heterogeneous abilities, whereas workers are identical in the experiment developed by Nosenzo (2010). However, this explanation must be regarded with caution because a recent study by Gross et al. (2015) shows that heterogeneous workers may also be influenced by their coworkers’ wages. More precisely, Gross et al. (2015) find that high-ability workers reduce their efforts if they are not paid more than their lowability coworkers, but that the reverse is not true. Other studies have focused on how displaying a worker’s effort to coworkers affects workers’ efforts. Striking lessons can be drawn from these studies. Gächter et al. (2012) observe that reciprocity declines when effort is displayed. Gächter and Thöni (2015) also note that efforts are strategic complements (i.e., the effort exerted by a worker is positively related to the effort exerted by his coworker), although after the revision stage of effort decisions, conformity in effort tends toward low effort rather than high effort. Finally, Gächter et al. (2013) show that the complementarity of efforts is primarily explained by inequity aversion. However, the last three studies mentioned have two significant limitations that must be emphasized: (i) they assume that workers have identical abilities, and (ii) they do not analyze whether the observability of efforts among coworkers affects wage decisions. With the present study, we aim to contribute to this stream of literature by considering heterogeneously skilled workers within the same firm and by examining the influence of

5

the observability of workers’ efforts on wage and effort decisions and its overall impact on efficiency in this context.

3

Experimental design

This section presents the gift-exchange game employed in this study followed by the experimental conditions implemented. Next, it describes the experimental procedure. Finally, it presents our research hypotheses derived from our setting.

3.1

The three-person gift-exchange game

As is customary, we employ a labor market framework and create a small-scale laboratory replica of the interactions between employers and workers.6 The aim of our study is to analyze whether the display of a worker’s effort to his coworker influences (i) the employer’s wage decision and (ii) heterogeneously skilled workers’ behaviors. To this end, we implement a three-person gift-exchange game, which is the minimal setup to answer our research questions. The game consists of 10 periods. In each period, a threeperson firm is formed by randomly and anonymously matching an employer, a highability worker (H-worker) and a low-ability worker (L-worker).7 Ability level is randomly assigned to workers and is fixed throughout the entire game. In the experiment, ability is the result of good or bad luck and stems from different levels of effort exerted for a given cost. Thus, a difference in ability means that a given effort is less costly for Hworkers and more costly for L-workers. Although we use stated efforts rather than real efforts, modelling differences in ability in this way allows us to approximate a situation where differences in ability may arise from educational attainment, human capital, motivation or talent.8 For instance, two workers may cost the same during the working day, 6 Although

it remains an open question as to whether framing matters, some recent studies, such as Abbink and Hennig-Schmidt (2006), emphasize that framing does not matter. 7 To limit spurious bias during the experiment, the H-worker was called the “type A worker” and the L-worker was called the “type B worker”. 8 Whereas stated efforts may reduce the degree of realism, they are used instead of real efforts to induce, control and manipulate differences in ability, regardless of other personal characteristics that might affect real efforts. As noted by Charness and Kuhn (2011), a clear advantage of this method is that it makes it possible to know the cost of effort and, as a result, to precisely calculate what the equilibrium effort level should be under different behavioral hypotheses. In addition, results from the recent experiment of Charness

6

but if one is more talented than the other or benefits from higher educational attainment, the input of that worker may be higher than that of a comparable worker of lesser ability. Formalized in this way, the cost may represent the time spent at work, for example. Thus, if two workers differ in ability and spend the same time at work, the quantity of work exerted by the more productive worker will be higher than that of the less productive one.9 Thus, in our experiment, for a given cost, the H-worker’s effort is twice that of the L-worker (see Table 1). These effort-cost relationships are common knowledge. Table 1: Effort levels and associated costs for workers according to their ability Effort levels for H-workers Effort levels for L-workers Cost of efforts (in experimental points)

1 0.5 0

2 1 1

3 1.5 2

4 2 4

5 2.5 6

6 3 8

7 3.5 10

8 4 12

9 4.5 15

10 5 18

The game proceeds in two stages. In the first stage, the employer - knowing the effortcost relationship for each of his workers - determines the wage to offer to each worker. Each wage must be an integer between 0 and 100. He can offer the same wage to both workers or offer them different wages. To account for differences in ability, our payoff structures closely resemble those in Gächter et al. (2012). The employer’s payoff function is given by:

π E = v · ( el + eh ) − wl − wh

(1)

where v denotes the marginal value product of effort that is set equal to 10, regardless of the worker’s ability; el and eh represent the effort levels selected by the L-worker and the H-worker, respectively. Similarly, wl and wh represent the wages offered to the L-worker and the H-worker, respectively. The employer’s payoff function is linear and strictly increasing in efforts and decreasing in wages. et al. (2016) suggest that both real and stated efforts may yield similar results in laboratory experiments. 9 Of course, our specification leads to different ranges of efforts that workers can provide to employers. However, we believe that our design does not have a stronger influence on our results compared with a design in which both workers have the same range of efforts but a different range of costs. In addition, we believe that our specification makes the difference in ability more salient to workers and favors wages that are dependent upon ability. An alternative specification would be to use the same effort-cost relationship for both workers and to set different values for the parameter of productivity, as in Charness and Kuhn (2007) and Gross et al. (2015). However, in doing so, it might be more difficult for workers to perceive the difference in ability, and it is more likely to observe spillover effects in efforts.

7

In the second stage, workers learn their own wage and their coworker’s wage and choose the amount of effort to supply. In addition, in some experimental treatments (see the next subsection), a worker learns the effort level exerted by his coworker before choosing his own effort level. The workers must choose an effort level from those displayed in Table 1.10 The effort level chosen by an H-worker will be an integer between 1 and 10, whereas the effort level chosen by an L-worker will be a multiple of 0.5 between 0.5 and 5. The payoff functions are given by:

π l = wl − c ( el )

(2)

πh = wh − c(eh )

(3)

for the L-worker and

for the H-worker. c(el ) and c(eh ) denote the cost associated with the efforts exerted by the L-worker and by the H-worker, respectively. Regardless of ability, the worker’s payoff is strictly increasing in his own wage but decreasing in the cost of his effort. The function c(ei ) is strictly increasing and convex, with the minimum effort being costless (see Table 1). It is noteworthy that the effort produced by the worker costs him less than it benefits the employer (i.e., c(ei ) < v · ei , ∀i = {l, h}), which means that the marginal value product of effort is always higher than the marginal cost of effort, which makes maximum effort levels socially efficient. Finally, the payoff function of a worker is independent of the wage offered to his coworker and the level of effort his coworker exerts; thus, there is no earnings interdependence between workers. Overall, there are 10 periods of the game described above. At the end of each period, both the employer and workers learn their respective payoffs in the current period. Next, a new period begins in which the employers and workers are randomly reshuffled under the constraint that each player (employers and workers) is matched once with the same 10 The

term effort is used throughout this paper, but the expression “quantity of work” was employed in the experiment, which refers to the quantity of work the worker chooses to exert to provide the employer a certain level of output. See the instructions in the supplementary materials.

8

two opponents.11 This stranger design is common information. Because losses are possible for the employer (by paying high wages and receiving low effort levels from workers in response, i.e., wl + wh > v(el + eh )), all players begin the experiment with 400 points at a conversion rate of 50 points = 1.2 Euros.12 Finally, to avoid wealth effects and mitigate boredom in later periods, 4 out of 10 periods are randomly selected at the end of the experimental session for payment.

3.2

Experimental conditions

The game is implemented using a between-subjects design with three experimental treatments. Because we aim to examine, in the context of heterogeneous abilities, the impact of the display of workers’ efforts to coworkers on wage and effort decisions, it is essential that workers have heterogeneous abilities in all treatments. Consequently, to answer our research questions, the experimental treatments vary along a single dimension: the observability by a worker of his coworker’s effort decision. Before describing the experimental treatments, three points should be stressed. In all treatments, the following apply: (i) the effort-cost relationship of each worker, depending on his ability, is common knowledge; (ii) wages are public information within the firm, which means that workers have full information about their coworker’s wage at the time they select their effort levels; and (iii) the first stage of the game (i.e., the employer’s wage decision) is carried out under the same conditions. We implement three experimental treatments in these conditions. In the control treatment, workers have no information about their coworker’s effort choice when they choose their level of effort, i.e., effort decisions are secret (S treatment). In the second treatment, the L-worker selects his effort level first; then the H-worker observes his coworker’s effort choice and selects his effort level thereafter (L-H treatment). The order of moves is reversed in the third treatment (H-L treatment).13 Note that our 11 Because the players interacted in groups of 9 for 10 periods, in the last period, they were matched with the same players as those in period 1. However, workers can not anticipate this matching because they do not know the size of the matching group. This allows us to rule out any strategic considerations. 12 The methodology is similar to that adopted by Abeler et al. (2010) and Gächter and Thöni (2015). 13 Of course, the most natural way would be to allow a mutual observation of efforts, as Gächter and Thöni (2015) do in a context of homogeneous ability. However, such mutual observation would imply a revision stage in effort decisions (i.e., a first effort decision followed by mutual observation of efforts and then a revised effort conditional on the observed efforts). As a consequence, to keep the experimental design as simple as possible, we did not run an experimental treatment in which each worker can observe how his

9

design is such that workers move simultaneously in the S treatment while they move sequentially in the L-H and H-L treatments. It may therefore be a question whether the observed difference between the control treatment and the other two treatments are due to the information generated by the sequential moves or by sequentiality as such. According to standard game theory, if workers’ efforts are unobservable, a game in which workers move sequentially is strategically equivalent to a game in which they move simultaneously. Such equivalence has been questioned by several experimental studies mainly involving coordination problems (see, for example, Rapoport, 1997). In contrast, in our game, each worker has a dominant strategy corresponding to the minimum effort. As a consequence, according to Güth et al. (1998), there should not be any difference between simultaneous choices of effort and sequential choices of effort with no information (i.e., pseudo-sequentiality, Abele et al., 2004). In addition, the last setting would have the disadvantage of requiring two control treatments to take into account both the difference in ability and the order of moves. Thus, we implement a control treatment in which workers choose their level of effort simultaneously.

3.3

Procedure

Experimental sessions were conducted at the LABEX-EM, University Rennes 1. Participants were students with different educational backgrounds. The experiment was programmed and conducted using the Z-tree software (Fischbacher, 2007). Participants were invited using Orsee (Greiner, 2015). A total of 11 sessions were conducted, 3 for the S treatment and 4 for each of the two other treatments, with 18 participants per session. No subject had previously participated in a similar experiment, and nobody participated in more than one session, resulting in 198 participants. Before the experimental session started, participants were told the following: (i) there would be two independent experiments,14 (ii) the amount of money earned in the experiments would depend on their decisions and the decisions of others in their experimental group, and (iii) they would be paid the earnings they accrued in the coworker behaves. 14 The two experiments refer to the gift-exchange game experiment and the experiment dedicated to eliciting inequity aversion estimates.

10

two experiments. Participants were clearly informed that information about the earnings obtained in each experiment would be provided only at the end of the experimental session. We set this condition to reduce the potential spillover effects of earnings from one experiment to the next. Upon arrival, the subjects were randomly seated at visually separated boxes numbered from 1 to 18. For the gift-exchange game experiment, two groups of nine participants each were constructed (from box numbers 1-9 and from box numbers 10-18), with three employers, three L-workers and three H-workers per matching group. We thus took great care to ensure that the strategies and the history experienced by each participant were never contaminated and did not contaminate decision making within the other matching groups. A total of two independent observations per session were guaranteed by the fact that no information passed between the two matching groups. Participants were randomly allocated to one of the two nine-person matching groups. The computer then randomly allocated the roles of the participants, who were informed of their roles at the beginning of the first period, and they retained this role throughout all 10 periods.15 Neither during nor after the experiment were subjects informed about the identity of the other participants in the room with whom they were matched. To guarantee public knowledge, instructions regarding the gift-exchange game experiment were distributed and read aloud (see supplementary materials). All participants were required to answer several control questions to ensure that they understood the experimental procedures. In particular, they were required to calculate both the employers’ and workers’ payoffs in hypothetical exercises. Of course, this procedure may introduce some bias but to limit this possibility, the exercises reflected representative contingencies. Answers were privately checked and, if necessary, explained to the participants, and the experiment did not start until all participants had answered all of the questions correctly. Once the gift-exchange game experiment was completed, instructions for the experiment dedicated to eliciting inequity aversion estimates appeared on their screens, followed by control questions (see supplementary materials). At the end of the two experiments, the 15 To

ensure comparability across sessions and treatments, pairings within each matching group were randomly formed prior to the first session, and these pairings were used in all the sessions. Matching grids are available from the author upon request.

11

subjects learned their earnings and completed a brief post-experimental questionnaire to collect personal characteristics (i.e., age, gender, fields and level of study). Each session lasted up to 80 minutes, and the average participant earnings were 17 Euros.

3.4

Research hypotheses

Prior to stating our research hypotheses, we must establish the sub-game perfect Nash equilibrium. Standard economic theory disregards social comparisons and fairness motives and thereby assumes that individuals are motivated exclusively by their own material interests. Because selfish workers receive a guaranteed wage that is not contingent upon their effort decisions, they put forth the minimum amount of effort, which is costless. With the foregoing in mind, the employer will offer the minimum wage, i.e., 0, to both workers. The sub-game perfect Nash equilibrium is thus {wi = 0, ∀i = {l, h} ; el = 0.5; eh = 1}. Although displaying a worker’s effort to a coworker does not affect the equilibrium under standard assumptions of selfishness, it may have an impact if we consider the influence of peer pressure and social comparison. In the first step, we consider three hypotheses in which the players (both employers and workers) are not sensitive to wage differences, as previously highlighted by Charness and Kuhn (2007).16 Regarding employers, little or nothing is known about the effect of effort observability on wage decisions because the existing experimental studies focus on its impact on effort decisions in cases of workers with identical ability (Gächter et al., 2012, 2013, for instance). In the present study, we consider workers with differing abilities and we argue that if employers believe that workers will respond reciprocally (i.e., workers will exert a higher level of effort in response to being offered a higher wage), they should favor the more able worker because his effort is more valuable. Previous experiments have provided support for such behavior, which is known as the targeting effect or merit pay (Bandiera et al., 2007; Gross et al., 2015, respectively). Based on the assumption of a targeting effect in the S treatment, our study investigates what occurs when workers can observe their coworker’s behavior. If employers anticipate peer pressure effects and pos16 However,

the results from more recent papers on social comparisons in the three-person gift-exchange game discuss this point; see the related literature in Section 2.

12

itive spillover effects in efforts (see infra Hypotheses 2 and 3), these two effects should be sufficient to increase the incentives of both workers to exert a high level of effort so as to provide a high final payoff to employers. These two effects act as substitutes for financial incentives. Thus, optimal wages should be lower in the two treatments with effort observability compared with the S treatment. Hypothesis 1 – Higher wages should be offered to the H-workers, but the observability of a worker’s effort by his coworker should lead to lower wages for both workers. With regard to workers, as noted by Mittone and Ploner (2011), the roles of the observed worker and the observer worker must be distinguished. We first consider observed workers. Since the seminal study by Kandel and Lazear (1992) that highlighted the peer effect, numerous studies both in the lab and in the field have provided strong support to the peer effects in the context of identical workers (Eriksson et al., 2009; Mittone and Ploner, 2011, for example). Here, we examine whether the peer effect is robust to the introduction of heterogeneous abilities. By supposing that observed workers are influenced by the awareness that their effort will be revealed to their coworker (for instance, they may feel the pressure or the desire to be a good example, or simply knowing that they are being observed), a framework of observable efforts should favor an increase in the average level of effort supplied by the observed workers. If workers exert greater effort when their efforts are observed than when their efforts are not observed regardless of the wages offered by the employer, a stronger reciprocity will develop. We conjecture that peer pressure would occur in our setting, which is based on heterogeneous abilities. In line with the findings of Falk and Ichino (2006), we assume that the peer pressure effect should be stronger for L-workers because they are observed by the more able workers, in contrast to H-workers, who are observed by workers of lesser abilities, which might indicate that they feel less intense pressure.17 Hypothesis 2 – The observability of efforts should induce observed workers, especially the less 17 Although

Falk and Ichino (2006) examine the peer effect in a context of heterogeneous abilities, in their study, workers’ payoff is completely independent of the output while in our gift-exchange game, the worker’s payoff is strictly decreasing in the cost of his effort.

13

productive ones, to exert higher effort levels and reciprocity compared with those found in the S treatment. The third hypothesis refers to the effort decision of the observer workers. Observing coworkers’ efforts may affect workers’ effort decisions by influencing what workers perceive to constitute appropriate behavior. Some social spillover effects may arise in terms of effort, cost of effort, and/or their relative values (i.e., ei /wi or c(ei )/wi , ∀i = {l, h}). Recent experimental studies conducted under identical ability have found that efforts are strategic complements, i.e., the effort exerted by the observer worker is positively correlated to the effort supplied by the observed worker (see Gächter and Thöni, 2015 for instance). An interesting question concerns the robustness of this result when we consider workers with differing abilities. We conjecture that effort levels should be positively correlated (i.e., efforts are strategic complements) in the context of heterogeneous abilities. Even if workers differ in ability, the effort exerted by an observed worker may be perceived as an example or as an appropriate behavior that induces the observer worker to act similarly. In other words, the level of generosity displayed by the observed worker through his effort decision and his degree of reciprocity should positively influence those of the observer worker. Hypothesis 3 – The efforts exerted by observer workers should be positively related to those exerted by their coworkers. Until now, we have restricted social comparisons to effort comparisons. However, players may be sensitive to - or they may believe that other firm members are sensitive to - wage differences that may affect their decision. For instance, peer pressure and positive social spillover effects (Hypotheses 2 and 3, respectively) may be lower if workers believe that the wages offered are unfair. Given that wages, and consequently wage differences, are common knowledge in all experimental treatments, the question is whether the observability of workers’ effort induces the observer worker to exert a level of effort that aims to reduce the final payoff difference. We focus on the observer workers because employers and observed workers may only make some conjectures regarding the decisions of the observer workers and the resulting final payoffs. In what follows, we are particu14

larly interested in inequity aversion as a potential explanation for the observer worker’s effort decision. This focus on inequity aversion is driven by previous experimental evidence of inequity aversion as an explanation for workers’ effort decisions in a context of homogeneous ability. For instance, Gächter et al. (2013) demonstrate that the positive social spillover effect they find in the context of identical ability is better explained by a model of distributional social preferences, such as Fehr and Schmidt (1999)’s model, rather than a desire to comply with social norms.18 However, while one can easily understand why inequity aversion may explain the observer worker’s decision in case of identical workers, in which the exerted effort can suppress or reduce inequalities of payoffs that can be perceived as unfair, the question of interest is whether inequity aversion can also explain the effort exerted by observer workers when we consider heterogeneous ability. In our experiment, a worker’s ability is due to good or bad luck and is not the result of individual skill. As a consequence, differences in ability may be less accepted than if they result from hard work and efforts to obtain higher skills. Consequently, effort decisions in line with inequity aversion may occur in order to reduce such unfairness. To test the relevance of inequity aversion, we refer to the model developed by Fehr and Schmidt (1999). This model assumes that individuals are concerned about fairness with respect to final payoff distributions in addition to their own material payoffs. Generally speaking, individuals may exhibit aversion toward both disadvantageous inequity (they are worse off than others) and advantageous inequity (they are better off than others).19 In our three-person gift-exchange game, observer workers may be concerned with two dimensions of fairness: vertical fairness, which refers to the relationships between them and the employer, and horizontal fairness, which involves relationships among workers. Strictly speaking, if the observer workers have Fehr-Schmidt preferences, they should adjust their levels of effort up or down to minimize payoff inequities, and the strength of the adjustment should be related to their individual sensitivities to payoff inequities. Hypothesis 4 – Fehr-Schmidt observer workers should choose the level of effort that minimizes 18 Note

that we do not aim to consider all social preference models to test their respective explanatory power. For this purpose, see Gächter and Thöni (2015). 19 Presentation of the model and formal predictions are derived in Appendix 1.

15

payoff inequities. To examine whether inequity aversion explains the decisions observed, we conduct a second experiment after the three-person gift-exchange game experiment to elicit individual estimates of inequity aversion to thereby control for it. To this end, we follow the experimental design of Blanco et al. (2011).20

4

Results

To test our hypotheses, the results are presented in four steps. First, we focus on the employers’ decisions. Second, we examine the observed workers’ behaviors, and third, we examine the behaviors of the observer workers. To conclude, we analyze the overall effect of displaying a worker’s effort to his coworker on employer’s payoff and efficiency.21 The following analysis pools all the data because no significant learning effect, especially on effort decisions, has been found.22 Finally, it is noteworthy that (1) statistical tests are conducted at the matching group level, unless the contrary is explicitly mentioned, and (2) all results are supported by parametric analyses and robustness checks are reported in the supplementary materials.

4.1

Employers’ behaviors

We first analyze whether the future observability of a worker’s effort by his coworker affects the upstream wage decision, and if so, in what way. A prerequisite to test our first hypothesis concerns the existence of a targeting effect in the S treatment. From the descriptive statistics about wage decisions reported in Table 2, we observe that the wage offered to the H-workers is significantly higher than that offered to the L-workers (twosided T-test: p = 0.0001). This result corroborates the existence of a targeting effect in the S treatment. 20 See

Appendix 2 for details. the main purpose of our experiment involves decisions made in the three-person gift-exchange game, the results regarding Experiment 2 are reported, for exposition purposes, in Appendix 2. 22 See the supplementary materials. 21 Because

16

Table 2: Descriptive statistics for employers’ wage offers by treatment

H-worker wage L-worker wage Wages difference

S 18.07 (20.45) 14.05 (17.72) 4.02 (15.10)

Treatments L-H 22.90 (25.08) 17.89 (20.26) 5.01 (14.62)

H-L 25.97 (27.17) 18.65 (20.27) 7.32 (19.57)

Note: The table reports the mean of the variables in experimental points and the standard deviation in parentheses.

We now turn to the impact of the observability of workers’ effort. We first note that both the observed and the observer workers receive higher wages in the two considered treatments: H-workers receive higher wages when their effort decisions are observed than when both workers’ efforts are hidden (two-sided T-test: p = 0.0405). Similarly, they receive higher wages when they observe the effort decision of their coworkers compared with the S treatment (two-sided T-test: p = 0.0758). We find the same for L-workers both when they observe their coworker (two-sided T-test: p = 0.0936) and when they are observed by their coworker (two-sided T-test: p = 0.1000). These findings contradict our Hypothesis 1 because the observability of a worker’s effort by his coworker induces employers to increase, on average, their wage offers. Figure 1: Wage decisions by treatment

Note: The size of the dots is proportional to the number of observations. The larger symbol in the bottom left corners corresponds to wh = wl = 1. The 45-degree line corresponds to equal wage decisions.

17

Second, we find that the increase in workers’ wages has an impact on wage differences. Even if both wages increase in the H-L treatment compared with the S treatment, the increase in the H-worker’s wage is stronger than that of the L-worker, which increases the difference in the two workers’ wages (see Table 2; two-sided T-test: p = 0.0868). This larger discrepancy between wages is also observed in the right panel of Fig.1 which depicts the wage decisions in each treatment. For ease of examination, non-cooperative wage equilibrium decisions (i.e., wh = wl = 0) are not reported. This evidence indicates that the wage difference is even more pronounced when the H-worker’s efforts will be displayed to his coworker. However, this finding no longer holds when the L-worker acts as the leader (two-sided T-test: p = 0.4602). Thus, only the future observability of the H-workers’ efforts by L-workers induces employers to target their wages toward the more able workers (i.e., H-workers) to urge them to exert high effort levels. These results require discussion. While we expected that the observability of a worker’s effort by his coworker would induce employers to decrease their wage offers, thus anticipating that the peer effect and positive social spillover effects would play an incentive role, we find the opposite. One possible explanation is the following. Employers provide stronger financial incentives to both workers to encourage a high level of effort, to limit any detrimental effects, such as the responsibility alleviation effect (i.e., when a worker exerts a low level of effort because he has observed the high level of effort exerted by his coworker) and to promote high levels of effort.23 This explanation is supported by the significant wage difference observed in the H-L treatment. In this setting, H-workers choose their level of effort first. By exerting a high level of effort, they may induce Lworkers to provide a high level of effort thereafter, which will increase the employer’s final payoff. In addition, the H-workers cannot be influenced by their coworkers’ (possibly lower) efforts. Such explanations suppose that employers believe that H-workers will respond reciprocally and that they will not be negatively affected by the wage difference. We summarize our findings for employers in Result 1. Result 1. On average, the subsequent display of a worker’s effort to his coworker induces employers to increase (i) the wages offered to both workers and (ii) the difference 23 However,

such an increase will benefit the employer only if the increase in effort is sufficiently high.

18

in wages only when the more able worker’s effort is displayed to his coworker.

4.2

Observed workers

The question of interest is whether observed workers are influenced in their effort decisions by the display of their effort to their coworker who differs in ability. To address this issue, we will first examine the level of effort exerted by observed workers and then the degree of reciprocity.

4.2.1

Peer effect on effort

From Table 3, we note that, for a given ability, the workers who are observed exert greater effort, on average, than the workers in the S treatment do. This result holds regardless of worker ability.24 This first evidence of the peer pressure effect may be explained by the higher wage received in the context of a worker’s observable effort. Table 3: Summary statistics for workers

wh wl eh el c(eh ) c ( el )

S 18.07(20.45) 14.05(17.72) 1.99(1.63) 0.93(0.87) 1.38(2.49) 1.23(2.92)

Treatments L-H 22.90(25.08) 17.89(20.26) 1.83(1.85) 1.26(1.13) 1.26(3.13) 2.30(3.92)

H-L 25.96(27.17) 18.65(20.27) 2.65(2.54) 1.22(1.15) 2.63(4.49) 2.26(4.06)

Note: The table reports the mean of the variables in experimental points and the standard deviation in parentheses.

To examine whether being observed induces observed workers to increase their effort level compared with the S treatment while controlling for the increase in wages between the two experimental treatments, we conduct double-censored Tobit regression analyses on the effort decisions to account for the lower and upper bounds of effort. The explanatory variables are the wages of both workers, the socio-demographic characteristics of the participants, and their individual estimates of inequity aversion. To control for the wage difference between workers, we introduce the difference in wages by distinguishing between disadvantageous (i.e., the worker’s wage is lower than his coworker’s wage) 24 All

two-sided T-tests are significant at the 5% level.

19

and advantageous wage differences (i.e., the worker’s wage is higher than his coworker’s wage). For an observed worker i associated with a coworker j, it results two explanatory variables: A disadvantageous wage difference equal to w j − wi if wi < w j and 0 otherwise, and an advantageous wage difference equal to wi − w j if wi > w j and 0 otherwise. To examine the influence of being observed, we consider decisions in the S treatment and those in the treatment in which the worker is observed (e.g., the H-L treatment for the effort decision of the H-worker). In the reported results, the S treatment is used as the reference. We include individual fixed-effects and period fixed effects. Standard errors are clustered at the group level and account for the intra-group correlation in the error term over the 10 periods. The results are reported in Table 4 (columns 1 & 2 for L-workers and columns 4 & 5 for H-workers).25 As usually observed in gift-exchange game experiments, we observe a positive and significant relationship between the wage received and the effort exerted (positive reciprocity). More interestingly, we find that being observed induces workers to exert on average a higher level of effort, regardless of the ability of the observed worker. In addition, we note that workers who exhibit a strong aversion toward disadvantageous inequity (measured through α) exert, on average, a lower level of effort, while those who exhibit a strong aversion toward advantageous inequity (measured through β) exert, on average, a higher level of effort. Finally, we observe that the larger the advantageous difference in wages is and the more averse observer workers are to advantageous inequity, the higher their exerted effort is (column 5). This result holds only for the more able worker. Because efforts are costly, such behaviors may be consistent with a desire to reduce the inequality in payoffs that can be perceived as unfair. H-workers benefit on average from higher wages thanks to their higher ability, which is due to good luck and not the result of their skill (see Table 2). H-workers who express strong aversion to advantageous inequity may exert high levels of effort to correct this unfairness. 25 Note that due to the inconsistency of the decisions made by some participants in the second experiment, we were unable to estimate their individual parameters of inequity aversion. As a consequence, these subjects have been excluded from the estimations that follow (i.e., 4 participants in the S treatment, 3 participants in the L-H treatment and 2 participants in the H-L treatment). However, all of the results are qualitatively similar if we introduce these subjects and exclude inequity aversion parameters from the estimations. See the supplementary materials for the robustness checks.

20

Table 4: Estimations for the effect of the effort observability on observed workers efforts Dependent variables

Worker wage Coworker wage Being observed(=1 if yes) α β Dis. wage difference Adv. wage difference α × dis. wage difference β × adv. wage difference Constant Socio-demographic controls Ind. fixed effects Time fixed effects Log pseudolikelihood Pseudo R-square N Left-censored observations Right-censored observations

L-workers S & L-H treat. Effort Reciprocity (1) (2) (3) 0.055*** (0.011) 0.014** -0.004 (0.007) (0.003) 0.571*** 0.756*** 0.452*** (0.092) (0.240) (0.025) -0.293*** -0.338*** -0.103*** (0.031) (0.043) (0.016) 1.914*** 0.410* -0.399*** (0.219) (0.385) (0.029) 0.012 (0.013) 0.136* (0.074) 0.006 (0.007) -0.152 (0.121) 0.336 1.488*** 0.107*** (0.356) (0.393) (0.035) Yes Yes Yes Yes Yes Yes Yes Yes Yes -261.0440 -337.5394 -259.7678 0.3774 0.1950 0.3125 350 350 228 217 217 – 4 4 –

H-workers S & H-L treat. Effort Reciprocity (4) (5) (6) 0.128*** (0.010) 0.003 -0.004** (0.011) (0.002) 0.593*** 1.286*** 0.055*** (0.184) (0.479) (0.006) -1.380*** -2.260*** -0.013*** (0.134) (0.150) (0.004) 4.501*** 3.835*** -0.109*** (0.351) (1.393) (0.037) 0.032 (0.061) 0.067** (0.032) -0.003 (0.023) 0.152** (0.065) -2.943*** 0.183 0.215*** (0.492) (0.759) (0.026) Yes Yes Yes Yes Yes Yes Yes Yes Yes -350.4230 -402.6170 -348.3350 0.3280 0.2279 0.4485 360 360 253 230 230 – 5 5 –

Notes: ∗∗∗ , ∗∗ , ∗ denote significance at the 1%, 5% and 10% level, respectively. Robust standard errors adjusted for clustering at the group level are in parentheses. Socio-demographic controls include dummies for gender, first year student or not, economic studies or not and whether participants have a job activity. All F-tests performed on the socio-demographic controls are significant at the 1% level. α and β correspond to individual estimates of inequity aversion, see Appendix 2 of the manuscript.

4.2.2

Peer effect on reciprocity

The next step consists of analyzing whether the strength of reciprocity exhibited by observed workers is also higher when their effort will be displayed compared with the S treatment. To this end, Table 5 reports the proportion of reciprocal workers and the strength of reciprocity by experimental treatment and the order of moves. Because only some workers in each matching group can be considered reciprocal, the Spearman rank correlation coefficients reported in Table 5 are computed at the individual level. They measure the strength of the relationship between the wage received and the effort exerted for the considered sample (i.e., all observed workers in a given treatment or depending on their ability). Table 5 shows that the proportion of reciprocal workers is significantly higher for workers who are observed compared with the S treatment (χ2 = 42.2400, p =

21

0.001 for H-workers and χ2 = 86.5906, p = 0.001 for L-workers); the strength of reciprocity exhibited by observed workers is also higher compared with the S treatment. Table 5: Relationship between the wage received and the effort exerted S ρ Prob % of reciprocal workers

0.5661 < 0.001 66.66

ρ Prob % of reciprocal workers

0.6657 < 0.001 61.11

ρ Prob % of reciprocal workers

0.6456 < 0.001 72.22

Treatments L-H Overall 0.6211 < 0.001 64.58 H-workers 0.5400 < 0.001 33.33 L-workers 0.7643 < 0.001 95.83

H-L 0.6091 < 0.001 52.08 0.7440 < 0.001 66.66 0.5979 < 0.001 37.50

Notes: ρ is the Spearman correlation coefficient and Prob the associated probability. Reciprocal workers are those for whom the Spearman rank correlation coefficient between the wage received and the effort exerted is significant at the 5% level.

This finding is supported by the result of the parametric analysis reported in Table 4. In columns 3 and 6, we examine the determinant of the level of reciprocity exhibited by observed workers. Reciprocity is defined as the ratio between the exerted effort and the received wage. The reported estimates underline that strong inequity averse workers are less reciprocal. More interestingly, we find that being observed is a strong and positive determinant of the level of reciprocity. A stronger reciprocity can be ascribed to an awareness of being observed by their coworker. Even with no reputational considerations at work, a significant proportion of observed workers may feel pressure to set a good example by exhibiting stronger reciprocity than they might exhibit if their efforts were not displayed. Finally, we note that the effect is even more pronounced for the less able workers. This latter result concur with those reported in Table 5 that underlines a higher increase in the proportion of observed reciprocal workers and a higher increase in reciprocity for L-workers than for H-workers when compared with the S treatment. This result is consistent with our Hypothesis 2 and the previous results of Falk and Ichino (2006), who show that low-ability workers are more sensitive to pressure from their peers

22

than high-ability workers. These findings are summarized below. Result 2: Despite differences in ability, being observed by their coworker induces the observed workers - particularly the less productive ones - both to increase their effort levels and to be more reciprocal, on average.

4.3

Observer workers

4.3.1

Social spillover effects

Regarding observer workers, we examine whether the observer workers are influenced by the effort of their coworker when they choose their level of effort. To examine this question, Fig. 2 depicts this relationship for two wage ranges to account for the positive relationship between a worker’s wage and his effort level. Simple regression lines are also drawn in Fig. 2 to depict the sign of the relationship between efforts. Fig. 2 demonstrates that the observer worker’s effort is positively related to his coworker’s effort, which applies to both H-workers (left panel of Fig. 2) and L-workers (right panel of Fig. 2). The Spearman rank correlation coefficients corroborate these observations (ρ = 0.4412, p < 0.001 in the L-H treatment and ρ = 0.3886, p < 0.001 in the H-L treatment). Figure 2: Relation between workers’ effort and coworkers’ effort

To strengthen these observations, we conduct several econometric regressions. First, 23

we perform double-censored Tobit regressions on effort decisions by controlling for the wages of both types of workers, coworker’s effort, the socio-demographic characteristics of participants and their individual estimates of inequity aversion. The Tobit estimates account for the efforts being left-censored by the minimum effort and right-censored by the maximum effort. Standard errors are clustered at the group level and account for the intra-group correlation in the error term over the 10 periods. The results are reported in Table 6.26 Consistent with the positive reciprocity highlighted in previous studies, the wage is a positive and strong predictor of the effort exerted for both types of workers. In addition, we note the negative and significant impact of L-workers’ wage on H-workers’ effort, which underline the feelings of jealousy experienced by H-workers regarding their coworkers’ wages. High wages offered to L-workers tend to discourage H-workers from working at high effort levels, all else being equal. The feelings of jealousy that are associated with relatively low efforts exerted by the L-workers may explain why observer H-workers exert lower levels of effort compared with those exerted in the S treatment (see Table 3). Moreover, the results are also clear regarding the link between efforts. We note that the observed worker’s effort has a positive impact on the effort decision made by the observer worker. This finding holds regardless of worker ability (columns 1 & 2 for Lworkers and columns 5 & 6 for H-workers). Therefore, there are positive social spillover effects in efforts, such that the efforts are strategic complements (Hypothesis 3). These positive social spillover effects are also observed on the level of reciprocity. In columns 4 and 8, we report the estimates of an OLS regression in which the dependent variable is the degree of reciprocity. Reciprocity is defined as the ratio between the worker’s effort and his own wage. We find, on average, a strong and positive relationship between the degree of reciprocity for both types of workers. These observations provide additional support for the strategic complementarity of efforts and the positive social spillover effect. This finding means that despite differences in workers’ abilities, observer workers are on average influenced by the effort exerted by their coworkers. Thus, the observer worker may consider that the observed worker’s effort is “the right thing to do” or is an 26 Robustness

checks are provided in the supplementary materials. Note that the results are qualitatively similar if we exclude the individual parameters of inequity aversion, which allows us to consider the entire sample of workers in the estimations.

24

example to follow; accordingly, observing his coworker’s high level of effort induces the observer worker to behave the same way. Therefore, it may be important for a firm that the observed worker is a high performer. Table 6: Estimations for the relationship between workers efforts Dependent variables

Wage Partner’s wage Partner’s effort Partner’s reciprocity α β Dis. temporary payoffa Adv. temporary payoffb α × Dis. temporary payoff β × Adv. temporary payoff Constant Socio-demographic controls Ind. fixed effects Time fixed effects Prob > F Log pseudolikelihood Pseudo or adjusted R-square N Left-censored observations Right-censored observations

L-worker in the H-L treatment Effort Reciprocity (1) (2) (3) (4) 0.032*** 0.029** (0.012) (0.012) 0.006 (0.009) 0.231** 0.348*** (0.108) (0.129) -3.404** 0.374*** (1.479) (0.059) -0.951*** -1.588*** -0.486*** -0.003 (0.088) (0.231) (0.063) (0.006) 6.157*** 6.870*** 4.818*** 0.053** (0.561) (0.972) (0.364) (0.031) -0.018* (0.009) 0.021 (0.032) -0.024*** (0.009) 0.051 (0.058) -0.296 0.364 0.767 0.067* (0.847) (0.704) (0.828) (0.033) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 0.000 0.000 0.000 – -186.892 -190.052 -167.998 – 0.3172 0.3056 0.2826 0.6554 220 220 164 154 137 137 85 – 6 6 5 –

H-worker in the L-H treatment Effort Reciprocity (5) (6) (7) (8) 0.134*** 0.122*** (0.001) (0.000) -0.043*** (0.001) 0.242*** 0.708*** (0.022) (0.019) 2.526*** 0.791*** (0.037) (0.107) -5.120*** -4.543*** -4.933*** 0.034*** (0.016) (0.016) (0.013) (0.008) 6.234*** 4.070*** 4.811*** -0.059 (0.044) (0.039) (0.028) (0.214) -0.017*** (0.003) 0.118*** (0.002) -0.243*** (0.044) 0.098*** (0.004) -3.703*** 0.914*** -4.687*** 0.124 (0.027) (0.026) (0.018) (0.296) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 0.000 0.000 0.000 – -127.288 -140.009 -119.516 – 0.4561 0.4018 0.4295 0.6374 210 210 149 144 157 157 97 – 2 2 2 –

Notes: ∗∗∗ , ∗∗ , ∗ denote statistical significance at the 1%, 5% and 10% level, respectively. Clustering errors at the group level are in parentheses. Socio-demographic controls include dummies for gender, first year student or not, economic studies or not and whether participant has a job activity. All F-tests performed on the socio-demographic controls are significant at the 1% level. α and β correspond to individual estimates of inequity aversion, see Appendix 2.a Dis. temporary payoff means that the observer worker receives a lower wage than the final payoff obtained by the observed worker. b Adv. temporary payoff means that the observer worker receives a higher wage than the final payoff obtained by the observed worker.

However, even if efforts are strategic complements regardless of workers’ ability, a noticeable difference depending on ability must be highlighted. We note that for L-workers, the more their coworkers are reciprocal, the lower the level of effort that they exert (column 3). Conversely, for H-workers, the more their coworkers are reciprocal, the higher the effort level that they exert (column 7). An explanation may be the following. All else being constant, most of the time the L-workers receive a lower wage than the H-workers do (see the central panel of Fig. 1.) Although disadvantaged, L-workers exhibit strong 25

reciprocity (see the bottom panel of Table 5; 95.83% of L-workers are reciprocal when they are observed). Such reciprocity induces H-workers, who are favored by employers, to exert a high level of effort. The high level of effort supplied allows them to reward the employer for his wage offer, and acting differently may be perceived as inappropriate by H-workers. This last result attenuates the negative impact of the L-worker’s wage on the H-worker’s effort (column (1)). Result 3: Despite differences in ability, observer workers’ efforts are positively influenced by their coworker’s efforts. This result holds regardless of the ability of the observer worker. There are positive social spillover effects in effort and reciprocity.

4.3.2

Inequity aversion

By selecting his effort and the associated cost, the observer worker has the opportunity to increase or decrease the gap in final payoffs. In the following, we focus only on the equality of the workers’ final payoffs, mainly because few observations are characterized by equal final payoffs for each member of the firm. The level of his effort might be driven by inequity aversion concerns.27 To illustrate the prominence of such behavior, Fig. 3 displays the relationship between a worker’s payoff and that of his coworker in each treatment.28 For ease of observation, effort decisions associated with the null wage for both workers have been omitted because in this case, workers necessarily exert the lowest level of effort. From Fig. 3, we note that 36.66% of workers’ choices in the L-H treatment and 36.25% in the H-L treatment aim to equalize the final payoffs, whereas the vast majority of decisions maintain the advantage of the H-worker.29 Eliciting individual inequity aversion parameters from the decisions made in Experiment 2 helps us explore the relationship between the sensitivity to inequity aversion and the level of effort exerted in greater detail. To this end, two situations must be considered 27 Gächter et al. (2013) demonstrate the influence of inequity aversion on the effort decision of the observer worker in the context of identical abilities. 28 Even if efforts are not displayed in the S treatment, the final payoffs obtained in this treatment are depicted for purposes of comparison with the other two treatments. 29 For purposes of comparison, the corresponding frequency in the S treatment is 29.44%, whereas the wage difference, i.e., before workers’ efforts, is smaller than that found in the L-H and H-L treatments. However, if we exclude effort decisions that equalize workers’ payoffs, the payoff dispersion is higher in the H-L treatment than in the S treatment.

26

if we restrict our attention to payoff inequity among workers.30 In the first situation, the observer worker, j, receives a lower wage than the final payoff obtained by the observed worker i, (i.e., w j < πi with πi = wi − c(ei )). In this case, the observer worker is worse off than his coworker, and the disadvantageous inequity cannot be suppressed. Therefore, the greater the worker’s aversion is to such inequity (i.e., high α), the lower the level of effort he will exert to avoid widening the final payoff gap. This negative relationship is confirmed by the Spearman rank correlation coefficient: ρ = −0.5836, p = 0.0765 for L-workers.31 This finding is supported by individual estimates for sensitivity to disadvantageous inequity reported in columns (1-3) and (5-7) in Table 6, which demonstrate that, on average, the observer workers who suffer from a strong aversion to disadvantageous inequity choose low levels of effort. In addition, we observe from Table 6 that (1) the higher the disadvantageous inequity is (Dis. temporary payoff equal to πi − w j in Table 6), the lower the effort is for observer workers (columns 2 and 6), and (2) for a given disadvantageous inequity, the greater the worker’s aversion is to such inequity, the lower the level of effort he will exert (α× Dis. temporary payoff in Table 6). Because effort is costly, workers exert a low level of effort to avoid widening the final payoff gap. These observations are in line with inequity aversion preferences. 30 If

we consider the payoff inequity among the 3 members of the firm, it results 4 conceivable cases. In Appendix 1, we present the results regarding the relationship between the sensitivity to inequity aversion and the level of the effort exerted by the observer worker in each of these 4 cases. 31 We only have 3 effort levels for H-workers in this situation; therefore we cannot perform this test.

27

Figure 3: Relation between workers’ final payoffs

Note: The size of the dots is proportional to the number of underlying observations. The dashed line is the 45-degree line corresponding to equal final payoffs.

In the second situation, the observer worker, j, receives a higher wage than the final payoff obtained by the observed worker i, (i.e., w j > πi with πi = wi − c(ei )). In this case, the observer worker is better off than his coworker and has the opportunity to close this gap by exerting a high and costly level of effort. In this case, the more sensitive the observer worker is to advantageous inequity (i.e., high β), the higher the level of effort. The Spearman rank correlation coefficients corroborate this strong and positive relationship for both types of workers (ρ = 0.7857, p = 0.0362 for L-workers and ρ = 0.8167, p = 0.0072 for H-workers). Further, the estimates for sensitivity to advantageous inequity reported in columns (1-3) and (5-7) in Table 6 are positive and highly significant, which indicates that the observer workers who exhibit a strong aversion to advantageous inequity choose, on average, high - and so costly - effort levels. Moreover, for H-workers, we observe from Table 6 that (1) the higher the advantageous inequity is (Adv. temporary payoff equal to w j − πi in Table 6), the higher the effort is for observer workers (column 6), and (2) it is all the more true that the H-observer worker exhibits a strong aversion to advantageous inequity (β× Adv. temporary payoff in Table 6). The fact that these last two observations hold only for H-workers is consistent with the allocation of ability in our experiment: Workers’ ability is due to good or bad luck and not the result of 28

individual’s skill. As a consequence, being favored may be perceived as unfair, and effort decisions that are in line with inequity aversion allow workers to reduce this unfairness (because effort is costly, exerting a high level of effort allows a worker to decrease the advantageous payoff gap). Result 4: Despite having workers with differing abilities, efforts exerted by observed workers and the strategic complementarity of efforts may be explained by inequity aversion preferences.

4.4

Employers’ payoff and efficiency

To conclude, the results of our experiment provide a basis to investigate whether it is beneficial for a firm to allow workers to observe their coworkers’ behaviors and whether one order of moves is better than another. For that purpose, in the final part, we analyze the impact of displaying workers’ efforts to coworkers on total output, employers’ payoffs and efficiency.32 An examination of Table 3 reveals that the effort levels are higher in the experimental treatments with observable efforts regardless of the order of moves (twosided T-tests: = 0.0332 for H-workers between the S and H-L treatments and p < 0.001 for L-workers between both the S and H-L and the S and L-H treatments). The single exception is the observer H-workers who put forth similar levels of effort to those exerted by H-workers in the S treatment (two-sided T-test: p = 0.4646) which leads them to experience lower costs than L-workers in the L-H treatment. However, in all treatments, the display of the worker’s effort to his coworker increases the total output (i.e., the sum of workers’ efforts): 13% increase in the L-H treatment and 42% increase in the H-L treatment. Nonetheless, because employers offer higher wages when workers can observe their coworkers’ behavior (Result 1), these settings will benefit the employer only if the increase in total output is sufficiently high. 32 Here

efficiency is simply understood as the sum of payoffs, not in the sense of Pareto efficiency, as in Engelmann and Strobel (2004).

29

Table 7: Final payoffs, total output and efficiency

πh πl πE Total outputa Efficiencyb

S 16.69(19.01) 12.82(16.10) −4.79(23.88) 2.73(2.20) 24.72(19.18)

Treatments L-H 21.64(23.36) 15.61(17.94) −9.88(30.08) 3.09(2.53) 27.37(19.83)

H-L 23.33(24.10) 16.42(18.67) −5.93(29.43) 3.87(3.16) 33.82(24.89)

Note: a Total output corresponds to the sum of workers efforts. b Efficiency corresponds to the sum of final payoffs obtained by each member of the firm. The table reports the mean of variables in experimental points and the standard deviation in parentheses.

Table 7 shows that the employer’s payoff is lower in the experimental treatments in which workers’ efforts are observed by their coworkers. This result means that, in this context, workers do not sufficiently increase their effort levels to compensate for the higher wages that employers offer. The decrease in employers’ payoffs is only significant in the L-H treatment (see Table 7; two-sided T-test: p < 0.0001). The following explanation can be provided. From Table 3, we note that in the L-H treatment, observed L-workers increase their efforts relative to the S treatment, but observer H-workers decrease their efforts. Further, only 33/33% of H-workers are reciprocal in this setting (see the central panel of Table 5). Consequently, the overall workers’ efforts are insufficient to compensate for the cost borne by the employer as a result of the high level of wages that he is offering. It follows that allowing workers to observe their coworkers’ behavior has a detrimental effect on employers’ payoffs, primarily because employers pay higher wages that are not compensated by a sufficient increase in efficiency. Put differently, encouraging workers’ interactions might actually cause the incentive effect set by the employer to backfire and may be detrimental for his payoff thereafter. Conversely, workers’ final payoffs are higher, on average, in settings with observability than in the S treatment. This result holds regardless of the order of moves and the workers’ ability (all two-sided T-tests are significant at the 5% level for H-workers and at the 10% level for L-workers). Because the increase in workers’ payoffs is sufficiently high to compensate for the decrease in employers’ payoffs, displaying a worker’s effort to his coworker has efficiencyenhancing effects in the H-L treatment (two-sided T-test: p = 0.0067). It follows that 30

when employers have the opportunity to design the spatial orientation of workers in the production process and when they cannot prevent worker interactions, they should favor an orientation that allows the more able workers to be observed because this orientation enhances both total output and efficiency. Result 4: On average, showing a worker’s effort to his coworker is detrimental to the employer’s payoff, but it enhances efficiency when the more able worker is observed.

5

Conclusion

Despite the undeniable difference in ability among workers within a firm and the opportunity for a worker to observe his coworker’s behavior before deciding upon his own effort, we are unaware of other attempts to experimentally study how displaying a worker’s effort to his coworker affects both employers’ and workers’ decisions in this context. The results obtained in our experiment reveal that showing a worker’s effort to his coworker leads to an increase in both wages and wage differentials, particularly when the more able worker is observed. Two potential explanations can be provided. First, employers offer higher wages in the hope that workers are less selfish (i.e., they exert higher effort levels) when they are observed by their coworker. This result is even more remarkable when the more able worker cannot be negatively influenced by the effort exerted by his lower-productive coworker. Second, employers offer higher wages to avoid the detrimental effects (such as the responsibility alleviation effect) that stem from the observability of coworkers’ efforts. Future studies that elicit employers’ beliefs and motives could shed light on these explanations. Regarding workers’ efforts compared with situations in which the coworker’s effort is unknown, those who are observed supply a higher level of effort, on average, and a greater strength of reciprocity, which are consistent with a peer pressure effect. For observer workers, the level of effort they exert reflects a strategic complementarity between efforts, i.e., workers’ efforts are positively correlated with those exerted by their coworkers. One explanation for the strategic complementarity of efforts highlighted here relates to inequity aversion: more than one-third of workers ad31

just their effort levels up or down to equalize workers’ final payoffs, and the strength of these adjustments is directly related to the individual parameters of inequity aversion. Overall, allowing a worker to observe his coworker’s behavior has a detrimental effect on employer’s payoff, because employers pay higher wages that are not compensated by a sufficient increase in workers’ efforts. However, because the increase in workers’ payoffs is sufficiently high to compensate for the decrease in employers’ payoffs, displaying a worker’s effort to his coworker has efficiency-enhancing effects, especially when the more able worker is observed. These findings have important implications for optimal workplace organization. Specifically, when employers have the opportunity to design the spatial orientation of workers in the production process and when they cannot prevent workers interactions, they should favor the observability of the more able workers because this setting generates a strong efficiency-enhancing effect. However, employers have to keep in mind that regardless of the spatial orientation, a setting with observable efforts is detrimental to employer’s payoff relative to a setting without observability. Our experiment provides initial insights into how allowing the observability of efforts among workers who differ in ability impacts (1) wages and efforts decisions and (ii) consequently, the efficiency of the firm. Nonetheless, two main limitations in the present study should be underlined. We have explored inequity aversion as an explanation for efforts exerted by observer workers. An alternative explanation might be related to compliance with social norms, as suggested by Gächter and Thöni (2015). If the observed worker’s effort is perceived as a norm to follow, then a higher level of effort exhibited by the observed worker will result in higher effort from the observer worker. This explanation is consistent with Mittone and Ploner (2011), who emphasize that the positive relationship between efforts may originate either from pure preferences for conformity among workers or from a combination of learning about social norms and a desire to comply with these social norms. Thus compliance with social norms may explain the positive relationship among workers’ efforts. A fruitful avenue for future research might be to investigate in more detail the role of social norms in these settings. To conduct such research, substantial changes in our experimental design will be necessary. A second limitation involves the attribution of ability. In our experiment, workers were randomly

32

selected to be high- or low-ability workers. Being a low-ability worker may be perceived as unfair by participants, which may subsequently affect their decisions. Further, to increase the external validity of our results and thereby inform human resource policy, it would be interesting to test the robustness of our findings when ability is determined through the actual skill of participants, such as based on the results of a real-effort stage in the spirit of Gross et al. (2015).

References Abbink, K., Hennig-Schmidt, H., 2006. Neutral versus loaded instructions in a bribery experiment. Experimental Economics 9, 103–121. Abele, S., Bless, H., Ehrhart, K.M., 2004.

Social information processing in strategic

decision-making: Why timing matters. Organizational Behavior and Human Decision Processes 93, 28 – 46. Abeler, J., Altmann, S., Kube, S., Wibral, M., 2010. Gift exchange and workers’ fairness concerns: When equality is unfair. Journal of the European Economic Association 8, 1299–1324. Akerlof, G., 1982. Labor contracts as a partial gift exchange. Quarterly Journal of Economics 97, 543–569. Akerlof, G.A., Yellen, J.L., 1990. The fair wage-effort hypothesis and unemployment. Quarterly Journal of Economics 105, 255–283. Bandiera, O., Barankay, I., Rasul, I., 2007. Incentives for managers and inequality among workers: Evidence from a firm-level experiment. Quarterly Journal of Economics 122, 729–773. Bellemare, C., Lepage, P., Shearer, B., 2010. Peer pressure, incentives, and gender: An experimental analysis of motivation in the workplace. Labour Economics 17, 276–283. Bewley, T., 1999. Why wages don’t fall during a recession. Cambridge, MA:Harvard University Press. 33

Blanco, M., Engelmann, D., Normann, H.T., 2011. A within-subject analysis of otherregarding preferences. Games and Economic Behavior 72, 321–338. Charness, G., Cobo-Reyes, R., Lacomba, J.A., Lagos, F., Perez, J.M., 2016. Social comparisons in wage delegation: experimental evidence. Experimental Economics 19, 433– 459. Charness, G., Kuhn, P., 2007. Does pay inequality affect worker effort? experimental evidence. Journal of Labor Economics 25, 693–723. Charness, G., Kuhn, P., 2011. Lab labor: What can labor economists learn from the lab, in: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, pp. 229–330. Cialdini, R., Reno, R., Kallgreen, C., 1990. A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places. Journal of Personality and Social Psychology 58, 1015–1026. Cohn, A., Fehr, E., Herrmann, B., Schneider, F., 2014. Social comparison in the workplace: Evidence from a field experiment. Journal of the European Economic Association 12, 877–898. Engelmann, D., Strobel, M., 2004. Inequality Aversion, Efficiency, and Maximin Preferences in Simple Distribution Experiments. The American Economic Review 94, 857– 869. Eriksson, T., Poulsen, A., Villeval, M.C., 2009. Feedback and incentives: Experimental evidence. Labour Economics 16, 679–688. Falk, A., Ichino, A., 2006. Clean evidence on peer effects. Journal of Labor Economics 24, 39–58. Fehr, E., Goette, L., Zehnder, C., 2009. A behavioral account of the labor market: The role of fairness concerns. Annual Review of Economics 1, 355–384. Fehr, E., Kirchsteiger, G., Riedl, A., 1993. Does fairness prevent market clearing? an experimental investigation. Quarterly Journal of Economics 108, 437–459.

34

Fehr, E., Schmidt, K.M., 1999. A theory of fairness, competition, and cooperation. Quarterly Journal of Economics 114, 817–868. Fischbacher, U., 2007. Z-tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10, 171–178. Gächter, S., Nosenzo, D., Sefton, M., 2012. The impact of social comparisons on reciprocity. Scandinavian Journal of Economics 114, 1346–1367. Gächter, S., Nosenzo, D., Sefton, M., 2013. Peer effects in pro-social behavior: Social norms or social preferences? Journal of the European Economic Association 11, 548– 573. Gächter, S., Thöni, C., 2010. Social comparison and performance: Experimental evidence on the fair wage-effort hypothesis. Journal of Economic Behavior & Organization 76, 531–543. Gächter, S., Thöni, C., 2015. Peer effects and social preferences in voluntary cooperation. Journal of Economic Psychology 48, 72–88. Gneezy, U., Meier, S., Rey-Biel, P., 2011. When and why incentives (don’t) work to modify behavior. Journal of Economic Perspectives 25, 191–210. Gneezy, U., Rey-Biel, P., 2014. On the relative efficiency of performance pay and non contingent incentives. Journal of the European Economic Association 12, 62–72. Greiner, B., 2015. The online recruitment system orsee - a guide for the organization of experiments in economics. Journal of the European Economic Association 1, 114–125. Greiner, B., Ockenfels, A., Werner, P., 2011. Wage transparency and performance: A realeffort experiment. Economics Letters 111, 236–238. Gross, T., Guo, C., Charness, G., 2015. Merit pay and wage compression with productivity differences and uncertainty. Journal of Economic Behavior & Organization 117, 233– 247.

35

Güth, W., Huck, S., Rapoport, A., 1998. The limitations of the positional order effect: Can it support silent threats and non-equilibrium behavior? Journal of Economic Behavior & Organization 34, 313 – 325. Kandel, E., Lazear, E.P., 1992. Peer pressure and partnerships. Journal of Political Economy 100, 801–817. Mackay, A., 2007. Motivation, Ability and Confidence Building in People. Routledge. Mas, A., Moretti, E., 2009. Peers at work. American Economic Review 99, 112–145. Mittone, L., Ploner, M., 2011. Peer pressure, social spillovers, and reciprocity: An experimental analysis. Experimental Economics 14, 203–222. Nosenzo, D., 2010. The Impact of Pay Comparisons on Effort Behavior. Discussion Papers 2010-03. The Centre for Decision Research and Experimental Economics, School of Economics, University of Nottingham. Rapoport, A., 1997. Order of play in strategically equivalent games in extensive form. International Journal of Game Theory 26, 113–136. Sausgruber, R., 2009. A note on peer effects between teams. Experimental Economics 12, 193–201.

36

Appendix Appendix 1. Predictions derived from the inequity aversion model of Fehr and Schmidt (1999) in the three-person gift-exchange game case and experimental results

1. Predictions In the following, we derive the predictions of the Fehr and Schmidt (1999)’s model for the more able worker (H-worker). But it is worth pointing out that the same predictions hold for the less able worker. A Fehr Schmidt H-worker maximizes the following utility function: 1 Uh (π ) = πh − αh [max (π E − πh , 0) + max (πl − πh , 0)] 2 1 − β h [max (πh − π E , 0) + max (πh − πl , 0)] 2

(4)

with h denoting the H-worker, l the L-worker, E the employer and π represents the vector of payoffs. αh measures how much H-worker dislikes disadvantageous inequity and β h how much he dislikes advantageous inequity. Fehr and Schmidt (1999) assume αh ≥ β h so that member h suffers more from disadvantageous inequity than advantageous inequity. Second, they assume 0 ≤ β h < 1. The constraint 0 ≤ β h excludes individuals who prefer being better off than others, while β h < 1 rules out individuals who are willing to “burn” their own money to reduce advantageous inequity. Recalling that the employer’s payoff function is:

π E = v ( el + eh ) − wl − wh

(5)

and those of workers:

= wl − c ( el )

(6)

πh = wh − c(eh )

(7)

πl

37

for the L-worker and the H-worker, respectively. The derivative of Uh (π ) with respect to the level of effort exerted eh is:33    ∂c(eh ) ∂c(eh )  − 0.5β (− 10 − 2 ) −  h  ∂eh ∂eh        − ∂c(eh ) − 0.5αh (10 + 2 ∂c(eh ) ) ∂Uh ∂eh ∂eh =       ∂c(eh ) ∂c(eh ) ∂c(eh ) ∂eh  + 0.5β − 0.5α 10 + −  h h ∂eh ∂e ∂e    h    h    ∂c(eh ) ∂c(eh )  − ∂eh − 0.5αh ∂eh − 0.5β h −10 − ∂c∂e(ehh )

if πh > πl and πh > π E ( A) if πh < πl and πh < π E

( B)

if πh > πl and πh < π E

(C )

if πh < πl and πh > π E ( D )

The above expressions (eq. 8) are directly related to the marginal cost of the exerted effort and the sensitivity to the difference in payoffs. Thus, according to the payoff inequity and the degree of sensitivity, the H-worker is induced to increase or decrease his effort. We assume that the sensitivity to the difference in payoffs does not depend on the role of players (for example a greater sensitivity to payoffs differences with coworker than employer). The four situations are depicted in Fig. 4. To plot Fig. 4, we approximate the effort-cost functions as follows: c (eh ) = 0.1061e2h + 0.8333eh − 1.0667

(9)

c (el ) = 0.4242e2l + 1.6667el − 1.0667

(10)

with eq. (9) for the H-worker and eq. (10) for the L-worker. For exposition, we set wh = wl = 25.34 33 The 34 The

results are symmetric if we consider the L-worker. same figure holds for wh 6= wl .

38

(8)

Figure 4: Location of equal payoffs in the three-person gift-exchange game

From Fig. 4, above the line π E = πh , the H-worker obtains a higher final payoff than the employer. Similarly, to the left of the line πl = πh , the H-worker obtains a higher final payoff than the L-worker. The intersection of these two lines creates four conceivable cases. • In the first case (A), the H-worker is better off than the other two players. In this case, the derivative is positive (∂Uh /∂eh > 0) if β h >

∂c(eh )/∂eh . (∂c(eh )/∂eh )+5

This means

that a H-worker with a sufficiently high β will increase his effort - that is costly to be closer to the employer’s payoff (an increase in effort leads to an increase in the employer’s payoff and a decrease in the H-worker’s payoff so that the payoffs difference diminishes) and to the low-ability worker’s payoff. • Regarding the symmetric situation (H-worker is worse off than the others - case of both disadvantageous inequities (B)), the derivative is unambiguously negative

(∂Uh /∂eh < 0, ∀α). Thus, the H-worker will reduce his effort to reduce the gap between his payoff and the payoffs of other members of his firm. While conclusions drawn from the first two cases (i.e., A and B) seem natural, it is more difficult without calculations to postulate the change in exerted effort when both advantageous and disadvantageous inequities are at work. 39

• When the H-worker has the potential to earn more than his coworker but his payoff is lower than that of the employer (case C), the H-worker will unequivocally decrease his effort (∂Uh /∂eh < 0, ∀α ≥ β), assuming, as Fehr and Schmidt do, that the worker is more concerned with being behind than being ahead. • When the H-worker has an advantageous inequity with respect to the employer and a disadvantageous inequity with respect to his coworker (case D), a worker with a low β (i.e., β h
∂(c(eh )/∂eh )(2+αh ) , (∂c(eh )/∂eh )+10

∀α ≥ β) will exert high effort (∂Uh /∂eh > 0).

2. Results If we consider the 4 conceivable cases described above, the results obtained are the following. 1. Case A of both advantageous inequities: In accordance with our theoretical prediction, there is a positive relationship between the sensitivity to advantageous inequity (β) and the level of exerted efforts. The Spearman rank correlation coefficient is ρ = 0.7857, p = 0.0362 for L-workers and ρ = 0.8000, p = 0.0096 for H-workers. 2. Case B of both disadvantageous inequities: There are only few occurrences of this case so that the Spearman rank correlation coefficient can be computed neither for L-workers and H-workers. However, we can observe on average low level of efforts in this setting for both L-workers (0.6032) and H-workers (1.0126). These results are in line with a low level of effort exerted in order to not increase the gap in final payoffs. 3. Case C of disadvantageous inequity with respect to the employer and advantageous inequity with respect to his coworker: We can observe on average low level of efforts in this setting for both L-workers (0.5833) and H-workers (1.0375). These results are in line with a low level of effort exerted in order to not increase the gap in

40

final payoffs. In addition, the assumption α ≥ β is satisfied for 97.40% of L-workers and 98.75% of H-workers. 4. Case D of advantageous inequity with respect to the employer and disadvantageous inequity with respect to his coworker: In accordance with our theoretical prediction, there is a positive relationship between the sensitivity to advantageous inequity (β) and the level of exerted efforts. The level of efforts exerted is high, especially for the L-workers (1.5232 on average) while we have only few observations for H-workers in this setting (mean exerted effort 1.25). The assumption α ≥ β is satisfied for 75.58% of L-workers and 83.33% of H-workers.

Appendix 2. Experiment 2: Inequity aversion 1. Elicitation After the first experiment, we conducted a second experiment that aimed at estimating the individual parameters of inequity aversion, following Fehr and Schmidt (1999)’s model. This model assumes that the utility of player i may be written as:



Ui = xi − αi max x j − xi , 0 − β i max xi − x j , 0

(11)

where xi is the monetary payoff of player i, x j is the monetary payoff of player j, αi is the parameter for disadvantageous inequity of player i and β i is the parameter for advantageous inequity of player i. We followed the procedure of Blanco et al. (2011), whereby subjects make decisions in two different games: an ultimatum game using the strategy method and a modified dictator game. In each game, subjects do not learn their role (for example, proposer or responder in the ultimatum game) until the end of the game. More precisely, the ultimatum game is used to elicit the individual parameter of disadvantageous inequity, αi . In this game, the proposer must divide 20 points between himself and the responder. Next, the responder must decide whether to accept or reject the proposition. Note that all subjects decided first as a proposer and second as a responder. To avoid any feedback and to elicit the complete strategy of responders, we used the 41

strategy method; that is, responders must decide whether to accept or reject any of the 21 possible distributions (ranging from (20, 0) to (0, 20); see Fig. 5). The estimation of αi is obtained through the decisions of the responder i and corresponds to the switch point between rejecting and accepting the distribution.

Figure 5: Table for responder’s choices in the ultimatum game Regarding the individual parameter of advantageous inequity, β i , we used a modified version of the dictator game. In this game, each subject must make decisions as a proposer by choosing between two distributions: a non-egalitarian one (20, 0) and an egalitarian one ( xi , xi ), for 21 possibilities (ranging from (0, 0) to (20, 20); see Fig. 6). The estimate of the advantageous inequity parameter, β i , corresponds to the switch point from the unfair distribution (20, 0) to the egalitarian one ( xi , xi ).

42

Figure 6: Table for proposer’s choices in the modified dictator game To avoid any order effects, in half of the experimental sessions, the ultimatum game was played before the modified dictator game, and we reversed the order in the other half. We applied this setting to each experimental treatment. Outcomes of these two games are known at the end of Experiment 2. Moreover, subjects knew that they would be paired with a different participant in these two games, a participant who was also different from their partner in the gift-exchange game, to rule out reputation and retaliation (or acknowledgment) effects.

2. Results of Experiment 2 Decisions made in the two games of Experiment 2 enable the selection of subjects with consistent preferences, i.e., subjects switch at some point (if at all) from choosing the left column to choosing the right column but they do not switch back. Overall, out of 198 participants, 176 (88.88%) behave consistently in both games. This result is in accordance with Blanco et al. (2011) who find 84.72%. In Table 8, we summarize the distribution of the advantageous and disadvantageous inequity parameters of all consistent subjects. Next, we present these distributions depending on (1) the role of players in the gift-exchange game and (2) the experimental treatment.

43

Table 8: Distribution of inequity aversion parameters By role

By treatment

All

Employers

H-workers

L-workers

Secret

L-H

H-L

α < 0.4 0.4 ≤ α < 0.92 0.92 ≤ α < 4.5 4.5 ≤ α

40.34% 26.14% 25.00% 8.52%

37.88% 28.79% 22.73% 10.61%

44.44% 20.37% 27.78% 7.41%

39.29% 28.57% 25.00% 7.14%

37.78% 22.22% 28.89% 11.11%

38.46% 29.23% 24.62% 7.69%

43.94% 25.76% 22.73% 7.58%

β < 0.235 0.235 ≤ β < 0.5 0.5 ≤ β

27.27% 15.34% 57.39%

25.97% 15.58% 44.16%

20.29% 10.14% 47.83%

19.72% 11.27% 47.89%

22.22% 22.22% 55.56%

18.46% 12.31% 69.23%

39.39% 13.64% 46.97%

Note: The theoretical distribution of Fehr and Schmidt (1999) is 30%, 30%, 30%, 10% for α and 30%, 30%, 40% for β. The empirical distribution found by Blanco et al. (2011) is 31%, 33%, 23%, 13% for α and 29%, 15%, 56% for β in their experiment.

Because this experiment replicates the one of Blanco et al. (2011), it is interesting to compare our results with theirs and with the theoretical distribution assumed by Fehr and Schmidt (1999). Computation of Chi-square goodness-of-fit tests indicate no significant differences between our distributions and that assumed by Fehr and Schmidt (1999) and that observed by Blanco et al. (2011), for both advantageous and disadvantageous inequities. More importantly, because this experiment was conducted after the gift-exchange game, and the elicited parameters of inequity aversion were used in the data analysis of results obtained in the gift-exchange game, we need to check whether the type of players or the information provided has not biased the results of Experiment 2. Again, from the Chi-square tests, we note that all comparisons (i.e., between experimental treatments and between roles of players) fail to be significantly different from each other. We conclude that the information provided and the roles in the gift-exchange game had no impact on the choices made in Experiment 2. Finally, the implementation of the two games (i.e., the modified dictator game and the ultimatum game) makes it possible to elicit the joint distribution of the α and β parameters. Fig. 7 depicts both individual parameters that are widely distributed in the subject pool. The assumption of a positive correlation between αi and β i , as assumed by Fehr and Schmidt (1999), is confirmed (Spearman rank correlation coefficient: ρ = 0.3045, p < 0.0001), while it contradicts the result of Blanco et al. (2011) (ρ = −0.03, p = 0.820). Moreover, 53.41% of subjects’ decisions confirm the hypothesis of αi ≥ β i , which is close to the findings of Blanco et al. (2011) who observe 62.29%. The corresponding data points 44

lie above the α = β line in Fig. 7. Figure 7: Distribution of inequity aversion parameters for consistent choices

Result. Subjects exhibit various degrees of inequity aversion. The positive relationship between both inequity aversion parameters is corroborated and a little more than half of the subjects’ decisions confirm the assumption of αi ≥ β i .

45