Pay secrecy and effort provision.

Nosenzo, Daniele

I. INTRODUCTION

Pay secrecy norms are widespread in many organizations. A survey of 2,700 U.S. employees conducted in autumn 2010 found that almost 1 out of 2 employees face limitations to the extent to which they can access information about other employees' compensation. (1) Similarly, a recent survey of U.S. employers found that over one-third of private sector organizations have in place formal rules prohibiting employees to discuss their pay with co-workers (reported in Gely and Bierman 2003). A common argument used to justify the adoption of pay secrecy is that it helps minimize the negative effects arising when employees compare their pay with the pay of their co-workers (Colella et al. 2007). (2) Pay comparisons are in fact thought to be a key factor in employees' judgments of the fairness of their pay (e.g., Akerlof and Yellen 1990). Discovering the existence of pay differentials between oneself and co-workers who are in comparable positions within the firm may spur perceptions of unfairness and reduce job satisfaction (Card et al. 2012). This may ultimately lead to reduced work morale and lower effort. Whereas this conjecture finds considerable support among managers and compensation executives (e.g., Agell and Lundborg 1995; Bewley 1999; Campbell III and Kamlani 1997), there is limited direct evidence that the availability of pay comparison information has detrimental effects on effort behavior. (3) In this paper, I use laboratory experiments to provide a controlled test of this proposition.

The experiments reported in this paper are based on a multilateral version of the gift-exchange game (Fehr, Kirchsteiger, and Riedl 1993). In the standard bilateral gift-exchange game, a first-mover (the "employer") decides on the size of the gift ("wage") she sends to a second-mover (the "employee"), who can in turn reciprocate by choosing costly actions ("effort") that reward the first-mover. A typical result of bilateral gift-exchange game experiments is that employees are often willing to incur costs in order to reward employers who have treated them favorably (for reviews of the experimental literature see Fehr, Goette, and Zehnder 2009 and Charness and Kuhn 2011). In the multilateral version of the game used in this study, described in detail in Section II, the employer interacts with two employees at the same time, and pays a wage to each of them. Employees receive their wage and then independently choose an effort level. I observe effort choices in three different conditions. In a first condition, employees only learn their own wage while coworkers' wages remain undisclosed. I use this benchmark condition to assess how employees respond to given levels of the own wage under a wage secrecy regime. In two other conditions, I remove any limitation on the availability of information about other workers' wages: in these treatments employees are fully informed of the wage paid to co-workers when they choose effort. By comparing these "public wages" treatments with the "wage secrecy" treatment I can test whether the availability of pay comparison information is detrimental for effort provision.

The two "public wages" treatments differ in how wages are determined. In one treatment the employer can choose the two wages she pays to the employees, while in the other treatment the wage paid to one employee is mandated exogenously by the experimenter. This allows to study the effect of pay comparisons in settings where the employer has full discretion on the firm's wage structure, as well as in settings where an employer's wage policy is partly constrained by exogenous labor market regulations. Such exogenous constraints to firms' wage policies are commonplace in natural workplaces (e.g., minimum wage laws, centralized pay regulations, etc.), and previous experiments have shown that they can affect pay fairness considerations and effort behavior in important ways. (4) Differently from previous studies, which either focused on bilateral labor relations or studied multilateral relations where workers received no information about the treatment of co-workers, in this experiment I study how employees' effort is influenced by labor market regulations that affect the wage of their co-workers. Thus, I can study how these regulations affect pay fairness considerations through horizontal pay comparisons.

In Section III, I discuss behavioral considerations about potential treatment effects, and report the results of the experiment in Section IV. As in many other related gift-exchange game experiments, I observe a strong positive own wage-effort relation: employees reciprocate high wages with higher effort. Most importantly, the data support the proposition that the availability of pay comparison information can be detrimental for effort provision. Employees in the public wages treatments who are underpaid relative to the co-worker expend significantly less effort than employees in the wage secrecy treatment. In the treatment where one of the two wages is fixed exogenously, employees also react negatively to overpayment and expend lower effort even when wage inequality is to their advantage.

I discuss these results and the related literatures in Section V. This study contributes to a small but growing literature that uses multi-worker gift-exchange game experiments to examine pay comparison effects (Angelova, Guth, and Kocher 2012; Battling and von Siemens 2011; Charness and Kuhn 2007; Cohn et al. 2011; Gachter and Thoni 2010; Gachter, Nosenzo, and Sefton 2012b; Guth et al. 2001). (5) Only few of these studies, however, have compared regimes with secret and public wages, and the results are mixed: in some cases the availability of information about co-workers' wages has been found to systematically affect effort (Angelova, Guth, and Kocher 2012), whereas in other cases such effects are weak or absent (Charness and Kuhn 2007; Guth et al. 2001). Differently from these studies, where employees either differed in their productivity or in the duration of their employment contract, employees in the present experiment are ex ante symmetric, and unequal pay may be perceived as decidedly unfair. The results from the present experiment show that when pay inequalities have clear-cut implications for pay fairness judgments, pay disclosure can have detrimental effects on effort provision.

II. EXPERIMENTAL DESIGN AND PROCEDURES

A. The Experimental Game

The experiment is based on the following three-player game. At the outset of the game, an Employer is endowed with 22 [pounds sterling] from which she pays a wage to two Employees, labeled RED and BLUE. The wages [W.sub.RED] and [W.sub.BLUE] can take three values: 1 [pounds sterling], 4 [pounds sterling], or 7 [pounds sterling]. Employees are paid their wage and then select simultaneously an effort level [e.sub.i[member of]{RED.BLUE}] among three possible levels: low (-1), medium (0), or high (+1). Low effort costs an employee 1 [pounds sterling] and reduces the Employer's earnings by 4 [pounds sterling]. Medium effort is costless and does not affect the Employer's earnings. High effort costs an employee 1 [pounds sterling] and increases the Employer's earnings by 4 [pounds sterling]. Note that this effort technology, which is also used by Charness and Levine (2007), allows both rewarding (when employees choose high effort) and punishment (when employees choose low effort) of fair/unfair wage offers. This is different from the effort technology usually adopted in gift-exchange games where employees can only decide whether or not to reward the Employer, and allows fairness considerations to affect behavior through both positive and negative reciprocity. Also note that employees are ex ante symmetric as they do not differ in their productivity when the Employer sets the wages. After employees have chosen their efforts the game ends and earnings (in UK Pounds) are computed as:

[[pi].sub.ER] = 22 + 4 x([e.sub.RED] + [e.sub.BLUE]) - [W.sub.RED]

- [W.sub.BLUE] and [[pi].sub.i] = [w.sub.i] - [([e.sub.i]).sup.2]

for the Employer and for Employee i [member of] {RED, BLUE}, respectively.

In the experiment subjects played a one-shot version of the game. The game was described using a labor market framing, and was implemented using the strategy method (Selten 1967), that is, subjects had to specify complete strategies in the game-theoretic sense.

B. The Treatments

The experimental game was implemented in three different treatments which vary along two dimensions. The first dimension is whether wages are public (both employees learn both wages before making an effort choice) or secret (each employee learns only her own wage). The second dimension is whether the Employer can choose both wages she pays to the employees, or whether she can instead choose only one wage, while the other wage is fixed exogenously. In the experiment, subjects were assigned to treatments randomly in a between-subjects design. Table 1 provides an overview of the treatments used in the experiment.

In the SECRET treatment, the Employer chooses a wage for the RED Employee and a wage for the BLUE Employee. Each employee learns her own wage but is not informed of the wage that the Employer chooses for the coworker: thus, co-workers' wages are secret.

In the PUBLIC treatment, wages are also determined by the Employer, but they are public information and employees are informed of both wages before they choose effort.

In the PUBLIC/EXO treatment, wages are also public information, but the Employer chooses only one of the two wages she pays to the employees, namely the wage for the BLUE Employee. The wage for the RED Employee is instead determined exogenously by the experimenter. I conducted two versions of the PUBLIC/EXO treatment where the RED Employee's wage was either fixed equal to 1 [pounds sterling] (PUBLIC/EXO 1 [pounds sterling] sessions) or equal to 7 [pounds sterling] (PUBLIC/EXO 7 [pounds sterling] sessions). The level of the RED wage was mandated using a neutral language: in the instructions (reproduced in Appendix S1, Supporting Information) subjects were simply told that "the Employer must pay a 1 [pounds sterling] [7 [pounds sterling] in PUBLIC/EXO 7 [pounds sterling]] wage to the Red Employee, while he/she can decide on what wage (1 [pounds sterling], 4 [pounds sterling] or 7 [pounds sterling]) to pay to the Blue Employee." (emphasis in original).

C. Discussion of the Design

In this experiment, I study how the availability of information about co-workers' wages affects effort behavior in a setting where employees are ex ante symmetric in that they do not differ in their productivity. The advantage of this setup is that pay differentials between employees have a more straightforward interpretation in terms of pay fairness than in previous studies with asymmetric employees (Angelova, Guth, and Kocher 2012; Charness and Kuhn 2007). One may argue that the absence of productivity differences between employees might limit the potential for observing pay differentials in the experiment. While productivity differences constitute an important rationale for paying unequal wages, there are other reasons why employers may want to introduce pay differentials between employees. For example, employers may speculate that employees will only work harder if paid more than a co-worker. Moreover, I also note that unequal wages have been chosen relatively frequently in previous gift-exchange game experiments with symmetric employees (see, e.g., Gachter and Thoni 2010; Gachter, Nosenzo, and Sefton 2012b).

In the PUBLIC/EXO treatment, employers can only choose the wage for one employee, whereas the other wage is mandated exogenously. I use this treatment to study reactions to pay comparisons in environments where the wage of a portion of the workforce is regulated by external labor market institutions. Note that, because employees are ex ante symmetric in the experiment, in PUBLIC/EXO I study a setting where the portion of the workforce whose wage is not exogenously regulated has similar characteristics to the portion of the workforce whose wage is regulated. This may, for example, be the case when a firm's workforce is composed of a mixture of public and private sector employees, the former being paid a centrally regulated wage and the latter being employed at the market wage. (6)

Pay comparison effects in PUBLIC/EXO are studied focusing on two "extreme" levels of the co-worker's wage, 1 [pounds sterling] and 7 [pounds sterling]. These seem the most attractive cases to isolate pay comparison effects: it appears in fact unlikely that I would observe any effect of moderate amounts of wage inequality had I failed to observe reactions to larger amounts.

I am interested in observing how employees choose effort across situations that differ in the wage they receive and, in PUBLIC and PUBLIC/EXO, in the wage paid to the co-worker. In order to study how an employee behaves in different situations one could have subjects play repeatedly the experimental game using the direct response method. However, repeated play of the experimental game may introduce strategic confounds as well as confounds related to learning effects as behavior in different situations is necessarily observed at different times during the experiment. Moreover, repetition of the task does not guarantee the collection of a sufficient number of observations for all information sets in the game. To avoid these drawbacks, I use the strategy method to identify pay comparison effects. In principle, eliciting choices using the strategy method may produce different results than using the direct response method. In fact, most of the experimental literature directly comparing choices elicited with the strategy method and the direct response method find that the two elicitation methods do not lead to qualitatively different results (Brandts and Charness 2011). Gachter and Thoni (2010) and Gachter, Nosenzo, and Sefton (2012a) investigate this issue specifically in the context of trilateral gift-exchange game experiments and show that the two elicitation methods produce consistent results.

Note that the number of effort decisions that employees have to submit with the strategy method depends on the treatment they are playing, as they control a different number of information sets in different treatments. In PUBLIC employees control nine information sets, one for each wage combination that may be chosen by the Employer: thus, I collect nine effort choices from RED and BLUE Employees in the PUBLIC treatment. In PUBLIC/EXO the RED Employee's wage is exogenously fixed at either 1 [pounds sterling] or 7 [pounds sterling] depending on which version of the treatment is implemented. Thus, only three wage combinations are actually feasible, and vary in the wage the Employer chooses for the BLUE Employee. Thus, I collect three effort choices from BLUE and RED Employees in the two versions of the PUBLIC/EXO treatment. Finally, because in SECRET employees only learn their own wage and not the co-worker's wage, they control three information sets corresponding to the three wage levels that could possibly be paid to them by the Employer. Thus, I collect three effort choices from employees in the SECRET treatment.

In the instructions, the trilateral gift-exchange game was described to subjects using a labor market framing. The most common instruction framings used in laboratory gift-exchange game experiments are the labor market framing used here and the "buyer-seller" framing (e.g., Fehr, Kirchsteiger, and Riedl 1993), whereby buyers (employers) first choose product prices and sellers (workers) then choose product quality. Both frames are "non-neutral," although the "buyer-seller" frame is conceivably more "neutral" relative to the proposed research question. On the other hand, the labor market framing is more "natural" as it matches more closely the environment under study. Although it is ultimately an empirical question whether or not instruction framing matters in the gift-exchange game, it should be noted that previous studies have documented pay comparison effects using both the "buyer-seller" frame (e.g., Gachter and Thoni 2010) and the labor market frame (e.g., Angelova, Guth, and Kocher 2012). Moreover, Abbink and Hennig-Schmidt (2006) compare the effects of "neutral" and "loaded" instructions in a related "reciprocity game" (the bribery game), and find that instruction framing has little impact on choice behavior.

D. Experimental Procedures

The experiment was conducted at the University of Nottingham using the software z-Tree (Fischbacher 2007). Subjects were students from a range of disciplines recruited through the online recruitment system ORSEE (Greiner 2004). Twelve sessions with a total of 180 participants were conducted: I had 30 subjects participate in two sessions of the PUBLIC treatment, 30 subjects participate in two sessions of the SECRET treatment, and 120 subjects participate in eight sessions of the PUBLIC/EXO treatment, equally divided between its two versions, PUBLIC/EXO 1 [pound sterling] and PUBLIC/EXO 7 [pound sterling]. No subject took part in more than one session. The average age of participants was 20.7 years, and 52% of them were male.

All sessions used an identical protocol. Upon arrival, subjects were randomly seated at visually separated computer terminals and given a written set of instructions that the experimenter read aloud. Subjects were also given a set of Earnings Distributions tables (reproduced in Appendix S2 for the PUBLIC treatment) showing the Employer's and Employees' earnings for all possible combinations of efforts and wages. Subjects were then randomly assigned to a group and a role (Employer, RED Employee or BLUE Employee), and had to solve a set of control questions to corroborate their understanding of the experimental game. Subjects had to answer all questions correctly before the experiment could continue. The decision-making phase of the session consisted of a one-shot play of the relevant experimental game. All decisions were made anonymously as subjects could not learn the identity of the other people in the room they were matched with.

At the end of the experiment, subjects completed a post-experimental questionnaire. Subjects were then privately paid a 3 [pound sterling] show-up fee plus their earnings from the experimental game. Subject earnings, inclusive of the show-up fee, ranged from 3 [pound sterling] to 27 [pound sterling], with an average of 10.12 [pound sterling] and a standard deviation of 6.95 [pound sterling]. Sessions lasted about 50 minutes on average.

III. BEHAVIORAL PREDICTIONS

In the one-shot gift-exchange game used in the experiment a rational Employee who only cares about maximizing her own material payoff will choose medium effort, which minimizes effort costs, regardless of the wages paid by the Employer. Anticipating this, a rational and selfish Employer will offer employees the lowest possible wage in order to minimize wage costs. Thus, both in SECRET and PUBLIC low wages and medium efforts emerge in the subgame perfect equilibrium. In PUBLIC/EXO the Employer must pay an exogenously determined wage (either low or high) to the RED Employee, but she will always choose a low wage for the BLUE Employee. Both employees will choose medium wage in the subgame perfect equilibrium. Note that, according to standard predictions, pay comparison information has no effect on effort choices. Employees choose the effort level that minimizes effort costs regardless of the wage paid to the co-worker and regardless of whether this information is available (PUBLIC and PUBLIC/EXO treatments) or not (SECRET treatment).

Previous experiments with the gift-exchange game have found that, contrary to standard predictions, wages and efforts typically exceed the cost-minimizing levels. In particular, the experimental evidence points to the existence of a positive own wage-effort relation: the more generous the wage offered by the employer, the higher the effort employees are willing to expend, even in one-shot interactions where there are no future gains associated with reciprocal behavior (see, e.g., Fehr, Goette, and Zehnder 2009). These findings are typically interpreted as supportive of Akerlof and Yellen's (1990) fair wage-effort hypothesis whereby employees' effort depends positively on the perceived fairness of their pay. The existence of fairness concerns introduces the potential for treatment differences in effort provision because what constitutes a "lair pay" may not only depend on considerations about the absolute level of the own wage, but also on relative pay considerations. In particular, to determine how fairly they are being treated by the employer, employees may compare their pay with the pay of their co-workers, when this information is available. Observing salary differences between oneself and other employees who perform comparable work in the firm may lead employees to believe that wages have been set unfairly by the employer, and this will reduce their willingness to expend costly effort, ceteris paribus. (7)

In the SECRET treatment, employees are not informed about the wage that the employer pays to the co-worker. Thus, pay fairness judgments in this treatment cannot be directly based on relative pay comparisons, and must instead rely on considerations about the absolute level of the wage offered by the employer, as in standard bilateral gift-exchange games. (8) Thus, based on the evidence from previous gift-exchange game experiments, I expect to observe a positive own-wage effort relation in SECRET, whereby higher (fairer) wages are associated with higher effort (more rewarding and less punishment), whereas lower (unfair) wages are reciprocated with low effort (less rewarding and more punishment).

In the PUBLIC treatment, employees receive information about the wage paid to the coworker before making an effort decision. Thus, relative pay considerations can influence pay fairness judgments (and effort) in this treatment. In particular, if employees care about "horizontal" (i.e., employee-employee) pay equality and perceive as unfair wage configurations that entail disadvantageous or advantageous pay inequalities, I may expect to observe, ceteris paribus, lower effort (less rewarding, more punishment) when an employee is underpaid or overpaid relative to the co-worker than in the absence of relative pay information. Moreover, if employees are more sensitive to disadvantageous than advantageous inequalities (Loewenstein, Thompson, and Bazerman 1989), I may expect to observe stronger reactions when employees are underpaid than overpaid relative to the co-worker.

In PUBLIC/EXO employees also receive information about co-worker's wages, but these are mandated exogenously. It is not clear a priori how this may affect effort choices. On the one hand, the fact that co-workers' wages are not determined by the employer in PUBLIC/EXO may reduce the scope for relative pay comparisons, as only wages actively chosen by employers may be considered as relevant for pay fairness attributions. Indeed, Gachter and Thoni (2010) find reduced pay comparison effects in their "Non-intentional" treatment where a random device chooses employees' wages on behalf of the employer. Nevertheless, note that, differently from Gachter and Thoni, employers in the PUBLIC/EXO treatment do have some discretion over wages as they can set the level of the BLUE wage: hence the Employer is responsible for any wage inequality within the firm, as she can always treat employees symmetrically if she wishes to do so. Moreover, the RED wages in PUBLIC/EXO are not determined by a random mechanism: they are fixed exogenously by the rules of the experiment, and employees may view the realization of a random process differently from an exogenous wage mandated by the experimenter. (9) In fact, similarly to how the introduction of an exogenous minimum wage shifted workers' perceptions of what is a fair wage in previous experimental studies (e.g., Falk, Fehr, and Zehnder 2006), the exogenous intervention on RED wages may increase the prominence of coworkers' wages as a reference point for what constitutes a fair wage in an experimental firm, and thus promote relative pay considerations. (10)

IV. RESULTS

A large fraction of Employers in the experiment chose to pay the lowest possible wages to the employees: 60% of the Employers in SECRET and 50% of the Employers in PUBLIC paid two 1 [pound sterling] wages to their employees, while in PUBLIC/EXO 47% of the Employers chose to pay a 1 [pound sterling] wage to the BLUE Employee. The Employers who were in SECRET and PUBLIC and chose at least one non-minimal wage, paid either one low wage and one medium wage (20% in each treatment), or two medium wages (20% in each treatment), or two high wages (only one Employer in PUBLIC). In the sessions of PUBLIC/EXO where the RED wage was fixed at 1 [pound sterling] (7 [pound sterling]), 55% (30%) of the Employers chose a 4 [pound sterling] wage for BLUE, while 5% (15%) paid BLUE a 7 [pound sterling] wage.

In the remainder of the section, I examine employees' behavior in these wage combinations. In particular, I will compare the effort chosen by employees in the wage secrecy regime with effort in the public wages regimes. I start by comparing effort behavior across the SECRET and PUBLIC treatments. I then turn to a comparison of effort choices made in SECRET and PUBLIC/EXO. Because the focus of the study is on the impact of pay comparisons on own wage-effort reciprocal relations, only the effort decisions made by BLUE Employees are relevant to the analysis of effort in PUBLIC/EXO: since RED Employees' wage was set exogenously by the experimenter their effort responses cannot be interpreted as a form of reciprocation toward the employer, and I will therefore not use their responses in the following data analysis (RED Employees' effort in PUBLIC/EXO is analyzed in Appendix S3).

A. SECRET Versus PUBLIC

Figure 1 reports the proportions of low effort choices (punishments) and high effort choices (rewards) made by employees in SECRET and PUBLIC for different levels of the own wage. In PUBLIC, I also differentiate effort choices made when the co-worker's wage was low (1 [pound sterling]), medium (4 [pound sterling]), or high (7 [pound sterling]). Effort rates for SECRET are not disaggregated according to the co-worker's wage as employees could not condition their choices on this information in the experiment.

A first evident feature of Figure 1 is that in both treatments employees expend higher effort when they are paid a higher wage. Employees rarely reward the employer with high effort when the own wage is 1 [pound sterling], but the proportion of high effort choices when the own wage is 7 [pound sterling] varies from 20% to 45% depending on treatment and relative pay conditions. Conversely, there are virtually no low effort choices when the own wage is 7 [pound sterling], while a 1 [pound sterling] wage triggers punishment between 15% and 25% of the times. Thus, in these treatments, as in many other gift-exchange game experiments, a positive own wage-effort relation emerges. (11)

A second noticeable feature of Figure 1 is that, in general, employees expend more effort under the wage secrecy regime than when the co-workers' wages are public. In particular, note that, when employees in the PUBLIC treatment are underpaid relative to the co-worker (i.e., when the employee's own wage is 1 [pound sterling] and the co-worker's wage is 4 [pound sterling] or 7 [pound sterling], and when the employee's own wage is 4 [pound sterling] and the co-worker's wage is 7 [pound sterling]), employees never reward the employer with high effort, whereas punishment rates are between 15% and 25%. In contrast, in SECRET reward rates are between 10% and 20% for comparable own wage levels (i.e., when the employee's wage is 1 [pound sterling] and 4 [pound sterling]), and punishment rates are not higher than 15%. (12) Employees seem also to expend less effort in PUBLIC than in SECRET when they are overpaid relative to the co-worker (i.e., when the employee's own wage is 4 [pound sterling] and the co-worker's wage is 1 [pound sterling], and when the own wage is 7 [pound sterling] and the co-worker's wage is 1 [pound sterling] or 4 [pound sterling]). Reward rates in PUBLIC are between 15% and 20% for the wage combinations where an employee is overpaid relative to the coworker, whereas they vary from 20% to 45% in SECRET for comparable own wage levels. Punishment rates do not seem to vary markedly between treatments when employees in PUBLIC are overpaid. (13) Finally, there seem to be smaller differences in reward and punishment rates between SECRET and PUBLIC when employees in PUBLIC are paid the same wage as the co-worker. (14)

[FIGURE 1 OMITTED]

I further examine these effects by performing a regression analysis of effort behavior in the two treatments. In a first model (Model I), I regress effort on three dummy variables that assume value 1 for effort choices made in the PUBLIC treatment in wage combinations where an employee receives respectively the same wage as the co-worker (PUBLIC EQUAL PAY), a lower wage than the co-worker (PUBLIC UNDERPAID), or a higher wage than the co-worker (PUBLIC OVERPAID). Note that the baseline category is represented by effort choices made in the SECRET treatment. The model also includes two dummy variables for different levels of the own wage ("Own wage 1 [pound sterling]" and "Own wage 7 [pound sterling]," with the intermediate case where the own wage is 4 [pound sterling] as benchmark category), and dummy variables controlling for individual characteristics (gender and field of study). In a second model (Model II) I investigate pay comparison effects separately for the cases where an employee is paid a 1 [pound sterling], 4 [pound sterling], and 7 [pound sterling] wage. In Model II I measure pay comparison effects using dummy variables that distinguish among effort choices made in the PUBLIC treatment when the co-worker earns respectively 1 [pound sterling] (PUBLIC 1 [pound sterling]), 4 [pound sterling] (PUBLIC 4 [pound sterling]), or 7 [pound sterling] (PUBLIC 7 [pound sterling]). As before, the regressions also include controls for gender and field of study. Ordered logit estimations of the two regression models are reported in Table 2.

Starting with Model I, I confirm the existence of a positive own wage-effort relation in the experiment: receiving a 7 [pound sterling] wage rather than a 4 [pound sterling] wage increases the odds of expending higher effort by almost 200%, ceteris paribus. The effect is significant at the 1% level. In contrast, receiving a 1 [pound sterling] wage reduces the odds of expending higher effort by about 63%, and the effect is significant at the 5% level. Thus, own wages are a powerful determinant of effort in the experiment.

The regression also shows that being in the PUBLIC treatment generally reduces effort provision. The effect is statistically significant only for the case where employees in PUBLIC are underpaid relative to the co-worker: the odds of expending higher effort are 66% lower for an underpaid employee in PUBLIC than for an employee in SECRET, and the effect is significant at the 5% level. (15) Being overpaid relative to the co-worker also reduces the odds of expending higher effort (by 52%), but the effect falls short of conventional significance levels (p = .237). (16) Receiving the same wage as the co-worker has instead a smaller effect on the odds of expending effort (the percentage change in the odds is -28%) and the effect is not statistically significant (p = .535). (17,18)

Model II shows that the negative impact of pay comparisons on effort is driven by the case where an employee is paid a 4 [pound sterling] wage and is paired with a co-worker paid a 7 [pound sterling] wage: the effect is strong (the odds of expending higher effort in this wage combination are 88% lower than in SECRET) and highly significant (p = .001). These results compare with those reported by Gachter and Thoni (2010), who also find stronger pay comparison effects when employees receive a medium wage and are paired with a highly paid co-worker. In both experiments, this is the only case where reactions to underpayment can be detected because in the other wage combinations where employees are underpaid they receive the lowest wage, and this already induces low effort.

Among the controls for individual characteristics included in the regressions, the gender dummy enters significantly in Model I and in two regressions of Model II: men expend lower effort than women, and the difference is significant at the 5% level. This result compares with findings on second-mover's behavior in related trust and sequential prisoner's dilemma games, where men are sometimes found to act more selfishly than women (e.g., Croson and Buchan 1999), although the effect is not always significant (e.g., Cox and Deck 2006).

Overall, these findings show that employees expend more effort under a wage secrecy regime than under a regime where information about coworkers' wages is public and they are underpaid relative to the co-worker. Being overpaid or being paid the same wage as the co-worker does not affect effort decisions relative to the case where wages are secret. (19)

B. SECRET Versus PUBLIC/EXO

I now turn to effort in PUBLIC/EXO. Figure 2 reports the proportions of low and high effort choices made by BLUE Employees in PUBLIC/EXO for different levels of the own wage. I distinguish between effort chosen in sessions where the co-worker's wage was low or high. For comparison, I also include effort choices made in SECRET. As in Figure 1, I do not disaggregate effort in SECRET according to the co-worker's wage.

As in PUBLIC and SECRET, also in PUBLIC/EXO I observe a positive own wage-effort relation: irrespective of the wage paid to the co-worker, higher own wage levels decrease the frequency of low effort and increase the frequency of high effort (in fact, high effort is only chosen when the own wage is 7 [pounds sterling]). (20)

A second important feature of Figure 2 is that also in PUBLIC/EXO the availability of pay comparison information appears to have a negative impact on effort choices relative to SECRET. In particular, note how employees who are in PUBLIC/EXO and are underpaid relative to the co-worker (i.e., in the wage combinations where the co-worker's wage is 7 [pounds sterling] and the employee's own wage is 1 [pounds sterling] or 4 [pounds sterling]) never choose high effort and choose instead low effort between 20% and 30% of the times. In contrast, in SECRET employees who receive a low or medium wage reward employers between 10% and 20% of the times. Low effort is never chosen in SECRET when employees are paid a medium wage, and is chosen 15% of the times in response to a low own wage. (21) Overpayment has also a negative impact on effort choices in PUBLIC/EXO. When employees in PUBLIC/EXO receive a 4 [pounds sterling] wage and the co-worker's wage is 1 [pounds sterling], they never expend high effort and choose low effort 5% of the times. In SECRET the reward rate is instead 20% when employees receive a 4 [pounds sterling] wage. Similarly, when employees in PUBLIC/EXO receive a 7 [pounds sterling] wage and the co-worker's wage is 1 [pounds sterling], the reward rate is 15%, whereas in SECRET 45% of the employees reward the employers when they receive a 7 [pounds sterling] wage. (22) Finally, being paid the same wage as the co-worker does not affect effort choices much relative to SECRET. (23)

[FIGURE 2 OMITTED]

Table 3 reports a regression analysis of effort behavior in PUBLIC/EXO and SECRET. Similarly to the analysis performed in the previous sub-section, in a first model (Model I) I regress effort on a set of dummy variables measuring whether an employee in PUBLIC/EXO is paid the same wage (PUBLIC/EXO EQUAL PAY), a lower wage (PUBLIC/EXO UNDER-PAID), or a higher wage than the co-worker (PUBLIC/EXO OVERPAID). The model also includes two dummies for different levels of the own wage ("Own wage 1 [pounds sterling]" and "Own wage 7 [pounds sterling]") and the usual controls for individual characteristics. Model II analyzes pay comparison effects separately for each possible level of the own wage. The regressions contain two dummy variables measuring the different levels of the co-worker's wage (PUBLIC/EXO 1 [pounds sterling] and PUBLIC/EXO 7 [pounds sterling]), and dummy variables controlling for gender and field of study.

Model I confirms the existence of a positive own wage-effort relation in the experiment: increasing (decreasing) an employee's wage from 4 [pounds sterling] to 7 [pounds sterling] (1 [pounds sterling]) increases (reduces) the odds of expending higher effort, and the effects are highly significant. Turning to pay comparison effects, the dummies controlling for the cases where employees in PUBLIC/EXO are underpaid or overpaid are both highly significant in Model I. Being underpaid relative to the co-worker reduces the odds of exerting higher effort by 86% and the effect is significant at the 1% level. Overpayment has a smaller negative impact on the odds of choosing higher effort (the percentage change in the odds is -74%), and the effect is also significant at the 1% level. (24) Being in the PUBLIC/EXO treatment has no significant effect on effort when there is no wage inequality. (25) Model II shows that, as in the PUBLIC treatment, also in PUBLIC/EXO the negative effects of pay comparisons are driven by employees receiving a medium wage. Underpayment and overpayment do not have instead a significant impact on effort when the own wage is 1 [pounds sterling] or 7 [pounds sterling]. Contrary to the results reported in Table 2, the gender dummy is insignificant in all regressions reported in Table 3, showing that overall I do not observe a clear gender effect in the experiment.

Overall, the results from the PUBLIC/EXO treatment confirm that the availability of pay comparison information can be detrimental for effort provision. Relative to the treatment where wages are secret, employees expend lower effort when wages are public if there is wage inequality in their experimental firm, even if co-workers' wages are not chosen by the employer but are mandated exogenously. In PUBLIC/EXO negative pay comparison effects are observed both when an employee is underpaid and overpaid relative to the co-worker. If employees are paid the same wage as their co-workers, there is instead no systematic difference between the wage secrecy and public wages regimes. (26)

V. DISCUSSION AND CONCLUSIONS

Pay secrecy is often justified on the ground that letting employees engage in pay comparisons might have detrimental consequences for effort provision: pay satisfaction and work morale may be damaged if employees discover (apparently arbitrary) differences between their own pay and the pay of co-workers who are in comparable positions within the firm. This might in turn lead employees to reduce their effort at work. This study reports an experiment that examines whether the availability of pay comparison information has in fact a negative impact on effort behavior. Using a multilateral version of the gift-exchange game, I compare effort choices made by employees in a "pay secrecy" treatment and in treatments where coworkers' wages are instead public information. The two "public wages" treatments vary in how the wages paid to the employees are determined: in one treatment both wages are chosen by the employer, while in another treatment one of the two wages is chosen by the experimenter. This latter treatment is a stylized representation of workplaces where employers are partially constrained in their wage policy by external labor market regulations.

In all treatments I observe a strong positive own wage-effort relationship: employees reciprocate high wages with high effort, as it has also been found in many previous gift-exchange game experiments (e.g., Charness 2004; Fehr, Kirchsteiger, and Riedl 1993; Fehr, Gachter, and Kirchsteiger 1997; Gachter and Falk 2002). Most importantly, I find that the availability of pay comparison information damages effort provision. Employees in the "public wages" treatments who learn that they are underpaid relative to an ex-ante symmetric co-worker expend significantly less effort than employees in the pay secrecy treatment. In the treatment where the employer is forced to pay an exogenously fixed wage to one of the two employees, I observe negative reactions to pay comparisons also in the case where an employee is overpaid relative to the co-worker. In both "public wages" treatments I do not find significant pay comparison effects when employees are paid equal wages.

These findings contribute to a recent experimental literature on the effects of pay comparisons on effort behavior. Most of the previous studies have investigated pay comparison effects by examining how employees' effort decisions change across situations that differ in the wages paid to their co-workers. Results from these experiments are mixed: some studies find that co-workers' wages have weak or no impact on effort behavior (Bartling and von Siemens 2011; Gachter, Nosenzo, and Sefton 2012b), whereas other studies document the existence of systematic pay comparison effects (Abeler et al. 2010; Cohn et al. 2011; Gachter and Thoni 2010). Importantly, these studies only include treatments where co-workers' wages are public. It is unclear that observing reactions to co-workers' wages in "public wages" settings implies that effort will be higher under wage secrecy. For example, under wage secrecy employees may suspect unfair treatment and reduce their effort as a consequence (e.g., Lawler 1967). This study shows that this is not actually the case. The availability of pay comparison information and the discovery of arbitrary pay differentials lead to significantly lower effort relative to a setting where co-workers' wages are kept secret. Effort under pay secrecy is instead not significantly different from the effort expended in a "public wages" setting if employers treat employees equally.

Three studies have previously examined settings where co-workers' wages are public or secret. Guth et al. (2001) study a setting where a principal has to design separate contracts for two agents who differ in productivity. They compare a treatment where agents only learn their own contract with a treatment where agents also learn the contract offered to the other agent. While principals in their experiment anticipate the existence of pay comparison effects and offer less asymmetric contracts in the treatment where contracts are public information, there is only weak evidence that pay comparisons actually affect agents' behavior. Charness and Kuhn (2007) use a multilateral version of the gift-exchange game where one employer interacts with two differently productive employees, and find that pay comparisons have negligible effects on effort behavior. Angelova, Guth, and Kocher (2012) study a three-person game where a principal is matched with a permanent agent, who is in a long-term relationship with the principal, and a temporary agent, who is matched with a different principal in each period. They find that when agents are only informed of their own contracts principals discriminate against temporary agents by offering them less favorable contracts. Discrimination is reduced when contracts are public as temporary agents reduce their effort when they are discriminated against. (27) An important difference between these studies and the present study is that in this experiment employees do not differ in their productivity or in the duration of their employment contract. The existence of asymmetries between employees may justify pay differentials and attenuate perceptions of unfairness. In contrast, in the present study it is straightforward to associate unequal pay with unfair treatment. Under such circumstances I find that pay disclosure can be detrimental for effort provision.

The present study also suggests that labor market policies and institutions can play an important role in shaping pay fairness perceptions and work behavior. Previous studies have documented the effects of these regulations on vertical fairness considerations (i.e., fairness between an employer and the employees). For example, Falk, Fehr, and Zehnder (2006) show that introducing external regulations on the lower bound of the possible wage offers made by an employer influences what workers perceive to be a fair wage and affect their behavior. The results of this study suggest that labor market policies and institutions can affect pay fairness perceptions also through considerations of horizontal fairness (i.e., fairness among coworkers). I show that regulating the wage that employers pay to a portion of the workforce significantly affects the behavior of the portion of the workforce which is not directly affected by these regulations. Indeed, the fact that I observe negative pay comparison effects for both underpayment and overpayment in the treatment where one of the wages is exogenously regulated by the experimenter suggests that horizontal fairness concerns may be even stronger in the presence of external regulations, possibly because they increase the prominence of coworkers' wages as a reference standard for what constitutes a fair wage in a given environment.

Overall, this study shows that the disclosure of pay information can have detrimental effects on effort provision. Relative to the pay secrecy setting employees' efforts are significantly lower when employees observe disadvantageous pay inequalities, can be lower in the presence of advantageous pay inequalities, and are not higher when employees learn that they receive the same wage as their co-worker. These results may be viewed as lending support to the use of pay confidentiality rules in workplaces. Employers should prefer to adopt pay secrecy rules whenever there are salary differences between employees that may be perceived as unfair. In the absence of such salary differences employers may be indifferent to the use of pay secrecy clauses as disclosing payments has little effect on effort provision when employees are paid equal wages.

On this respect, it should be noted that whether or not pay inequalities will be actually seen as unfair is likely to depend on the details of the work environment. In the stylized setting studied in this paper pay differentials have rather clear-cut implications for pay fairness judgments. In the absence of observable asymmetries between workers, paying different upfront wages to different employees is unfair. However, in other settings pay inequalities may actually be perceived as fair and may thus be effort-enhancing. For example, Abeler et al. (2010) study a multi-worker gift-exchange game where employers choose wages after having observed employees' effort decisions. In one treatment, employers have to pay equal wages to the employees by design. In a second treatment employers can pay different wages to different employees. They find that employees who are paid equal wages expend significantly lower effort than employees in the treatment where employers can pay individualized wages. The reason for this effect is that in the latter treatment employers use wage differentials to compensate for observed differences in performance, for example by offering higher wages to the harder-working employees. Thus, unequal wages can actually be used to enforce equity and fairness, whereas equal wages may engender unfairness. An implication of these considerations is that, in settings where employer-employees interactions are not one-shot (as in the present experiment) but are repeated, pay inequalities that are based on past performance differentials (when these are observable) may actually promote fairness and effort provision.

ABBREVIATION

DEFRA: Department for Environment, Food, and Rural Affairs

doi: 10.1111/j.1465-7295.2012.00484.x

REFERENCES

Abbink, K., and H. Hennig-Schmidt. "Neutral Versus Loaded Instructions in a Bribery Experiment." Experimental Economics, 9(2), 2006, 103-21.

Abeler, J., S. Altmann, S. Kube, and M. Wibral. "Gift Exchange and Workers' Fairness Concerns--When Equality Is Unfair." Journal of the European Economic Association, 8(5), 2010, 1299-324.

Agell, J., and P. Lundborg. "Theories of Pay and Unemployment--Survey Evidence from Swedish Manufacturing Firms." Scandinavian Journal of Economics, 97(2), 1995, 295-307.

Akerlof, G. A., and J. L. Yellen. "The Fair Wage-Effort Hypothesis and Unemployment." Quarterly Journal of Economics, 105(2), 1990, 255-83.

Angelova, V., W. Guth, and M. Kocher. "Co-employment of Permanently and Temporarily Employed Agents." Labour Economics, 19(1), 2012, 48-58.

Bartling, B., and F. A. von Siemens. "Wage Inequality and Team Production: An Experimental Analysis." Journal of Economic Psychology, 32, 2011, 1-16.

Bewley, T. Why Wages Don't Fall in a Recession. Cambridge, MA: Harvard University Press, 1999.

Brandts, J., and G. Charness. "Do Labour Market Conditions Affect Gift Exchange? Some Experimental Evidence." Economic Journal, 114(497), 2004, 684-708.

--. "The Strategy Versus the Direct-Response Method: A First Survey of Experimental Comparisons." Experimental Economics, 14(3), 2011, 375-98.

Campbell III, C. M., and K. S. Kamlani. "The Reasons for Wage Rigidity: Evidence from a Survey of Firms." The Quarterly Journal of Economics, 112(3), 1997, 759-89.

Card, D., A. Mas, E. Moretti, and E. Saez. "Inequality at Work: The Effect of Peer Salaries on Job Satisfaction." American Economic Review, 102(6), 2012, 2981-3003.

Charness, G. "Attribution and Reciprocity in an Experimental Labor Market." Journal of Labor Economics, 22(3), 2004, 665-88.

Charness, G., and P. Kuhn. "Does Pay Inequality Affect Worker Effort? Experimental Evidence." Journal of Labor Economics, 25(4), 2007, 693-723.

--. "Lab Labor: What Can Labor Economists Learn from the Lab?," in Handbook of Labor Economics, Vol. 4A, edited by O. Ashenfelter and D. Card. Amsterdam: Elsevier B.V., 2011, 229-330.

Charness, G., and D. I. Levine. "Intention and Stochastic Outcomes: An Experimental Study." Economic Journal, 117(522), 2007, 1051-72.

Clark, A., D. Masclet, and M. C. Villeval. "Effort and Comparison Income." Industrial & Labor Relations Review, 63(3), 2010, 407-26.

Cohn, A., E. Fehr, B. Herrmann, and F. Schneider. "Social Comparison in the Workplace: Evidence from a Field Experiment." IZA Discussion Paper No. 5550, 2011.

Colella, A., R. Paetzold, A. Zardkoohi, and M. Wesson. "Exposing Pay Secrecy." Academy of Management Review, 32(1), 2007, 55-71.

Cox, J. C., and C. A. Deck. "When Are Women More Generous Than Men?" Economic Inquiry, 44(4), 2006, 587-98.

Croson, R., and N. Buchan. "Gender and Culture: International Experimental Evidence from Trust Games." American Economic Review, 89(2), 1999, 386-91.

Danziger, L., and E. Katz. "Wage Secrecy as a Social Convention." Economic Inquiry, 35(1), 1997, 59-69.

Falk, A., E. Fehr, and C. Zehnder. "Fairness Perceptions and Reservation Wages--The Behavioral Effects of Minimum Wage Laws." Quarterly Journal of Economics, 121(4), 2006, 1347-81.

Fehr, E., S. Gachter, and G. Kirchsteiger. "Reciprocity as a Contract Enforcement Device: Experimental Evidence." Econometrica, 65(4), 1997, 833-60.

Fehr, E., L. Goette, and C. Zehnder. "A Behavioral Account of the Labor Market: The Role of Fairness Concerns." Annual Review of Economics 1(1), 2009, 355-84.

Fehr, E., G. Kirchsteiger, and A. Riedl. "Does Fairness Prevent Market Clearing--An Experimental Investigation." Quarterly Journal of Economics, 108(2), 1993, 437-59.

Fischbacher, U. "z-Tree: Zurich Toolbox for Ready-Made Economic Experiments." Experimental Economics, 10(2), 2007, 171-78.

Gachter, S., and A. Falk. "Reputation and Reciprocity: Consequences for the Labour Relation." Scandinavian Journal of Economics, 104(1), 2002, 1-26.

Gachter, S., D. Nosenzo, and M. Sefton. Forthcoming. "Peer Effects in Pro-social Behavior: Social Norms or Social Preferences?" Journal of the European Economic Association, 2012a.

--. "The Impact of Social Comparisons on Reciprocity." Scandinavian Journal of Economics, 2012b. doi: 10.1111/j.1467-9442.2012.01730.x

Gachter, S., and A. Riedl. "Moral Property Rights in Bargaining with Infeasible Claims." Management Science, 51(2), 2005, 249-63.

Gachter, S., and C. Thoni. "Social Comparison and Performance: Experimental Evidence on the Fair Wage-Effort Hypothesis." Journal of Economic Behavior & Organization, 76(3), 2010, 531-43.

Gely, R., and L. Bierman. "Pay Secrecy/Confidentiality Rules and the National Labor Relations Act." University of Pennsylvania Journal of Labor and Employment Law, 6, 2003, 120-56.

Glasnapp, D., and J. Poggio. Essentials of Statistical Analysis for the Behavioral Sciences. Columbus, OH: Merrill, 1985.

Greiner, B. 2004. "An Online Recruitment System for Economic Experiments," in Forschung und wissenschaftliches Rechnen GWDG Bericht 63, 79-93. Gottingen, Germany: Gesellschaft fur Wissenschaftliche Datenverarbeitung.

Guth, W., M. K6nigstein, J. Kovacs, and E. Zala-Mezo. "Fairness within Firms: The Case of One Principal and Many Agents." Schmalenbach Business Review, 53, 2001, 82-101.

Hennig-Schmidt, H., B. Rockenbach, and A. Sadrieh. "In Search of Workers' Real Effort Reciprocity--A Field and a Laboratory Experiment." Journal of the European Economic Association, 8(4), 2010, 817-37.

Lawler, E. E. "Secrecy about Management Compensation: Are There Hidden Costs?" Organizational Behavior and Human Performance, 2(2), 1967, 182-89.

Loewenstein, G., L. Thompson, and M. Bazerman. "Social Utility and Decision Making in Interpersonal Contexts." Journal of Personality and Social Psychology, 57(3), 1989, 426-41.

Maximiano, S., R. Sloof, and J. Sonnemans. "Gift Exchange in a Multi-Worker Firm." Economic Journal, 117(522), 2007, 1025-50.

Owens, M. F., and J. H. Kagel. "Minimum Wage Restrictions and Employee Effort in Incomplete Labor Markets: An Experimental Investigation." Journal of Economic Behavior & Organization, 73(3), 2010, 317-26.

Selten, R. 1967. "Die Strategiemethode zur Erforschung des eingeschrankt rationalen Verhaltens im Rahmen eines Oligopolexperimentes," in Beitrage zur experimentellen Wirtschaftsforschung, edited by H. Sauermann. Tubingen, Germany: J.C.B. Mohr (Paul Siebeck), 136-68.

Van Reenen, J., and C. Propper. "Can Pay Regulation Kill? Panel Data Evidence on the Effect of Labor Markets on Hospital Performance." Journal of Political Economy, 118(2), 2010, 222-73.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Appendix S1. Instructions.

Appendix S2. Earnings Distributions Tables.

Appendix S3. RED Employees' effort in PUBLIC/EXO.

Table S1. Earnings distributions resulting from the employees' effort choices in SITUATION I, i.e., when the Employer chooses a 1 [pounds sterling] wage for the Blue Employee and a 1 [pounds sterling] wage for the Red Employee.

Table S2. Earnings distributions resulting from the employees' effort choices in SITUATION II, i.e., when the Employer chooses a 1 [pounds sterling] wage for the Blue Employee and a 4 [pounds sterling] wage for the Red Employee.

Table S3. Earnings distributions resulting from the employees' effort choices in SITUATION III, i.e., when the Employer chooses a 1 [pounds sterling] wage for the Blue Employee and a 7 [pounds sterling] wage for the Red Employee.

Table S4. Earnings distributions resulting from the employees' effort choices in SITUATION IV, i.e., when the Employer chooses a 4 [pounds sterling] wage for the Blue Employee and a 1 [pounds sterling] wage for the Red Employee.

Table S5. Earnings distributions resulting from the employees' effort choices in SITUATION V, i.e., when the Employer chooses a 4 [pounds sterling] wage for the Blue Employee and a 4 [pounds sterling] wage for the Red Employee.

Table S6. Earnings distributions resulting from the employees' effort choices in SITUATION VI, i.e., when the Employer chooses a 4 [pounds sterling] wage for the Blue Employee and a 7 [pounds sterling] wage for the Red Employee.

Table S7. Earnings distributions resulting from the employees' effort choices in SITUATION VII, i.e., when the Employer chooses a 7 [pounds sterling] wage for the Blue Employee and a 1 [pounds sterling] wage for the Red Employee.

Table S8. Earnings distributions resulting from the employees' effort choices in SITUATION VIII, i.e., when the Employer chooses a 7 [pounds sterling] wage for the Blue Employee and a 4 [pounds sterling] wage for the Red Employee.

Table S9. Earnings distributions resulting from the employees' effort choices in SITUATION IX, i.e., when the Employer chooses a 7 [pounds sterling] wage for the Blue Employee and a 7 [pounds sterling] wage for the Red Employee.

Table S10. Effort behavior of RED Employees in PUBLIC/EXO.

Figure S1. Low and high effort rates: RED Employees in PUBLIC/EXO. Notes: The Figure omits the category "medium effort," thus % low effort + % high effort = 100% - % medium effort. Bars are based on choices made by 20 RED Employees in each version of PUBLIC/EXO.

(1.) Source: Survey of Economic Security (2010) conducted by the Institute for Women's Policy Research (http://www.iwpr.org/press-room/press-releases/paysecrecy-and-paycheck-fairness-new-data-shows-paytransparency-needed).

(2.) Danziger and Katz (1997) discuss an alternative reason why firms may benefit from discouraging employees to discuss their salaries. In their model, pay secrecy reduces the ability of employees to locate other higher-wage firms in the market, thus reducing labor mobility and making risk-shifting contracts feasible.

(3.) In fact, some studies argue that the lack of pay comparison information may be detrimental: pay secrecy may actually lead employees to suspect that they are treated unfairly and thus damage morale (e.g., Lawler 1967).

(4.) For example, Falk, Fehr, and Zehnder (2006) show that introducing a nonbinding wage guideline in a previously unregulated experimental labor market shifts employees' perceptions of what constitutes a fair wage and increases their reservation wages considerably. The effect is even stronger if the same wage level chosen for the wage guideline is set as a (binding) "minimum wage." On the impact of minimum wages on effort in gift-exchange games see also Brandts and Charness (2004) and Owens and Kagel (2010).

(5.) Maximiano, Sloof, and Sonnemans (2007) also study a multi-worker gift-exchange game. However, in their experiment all employees are paid the same wage by design.

(6.) For example, Van Reenen and Propper (2010) discuss the case of the UK National Health Service, relying both on "permanent" public sector nurses, whose wages are regulated at the national level, and "agency" nurses with non-regulated wages. The existence of separate pay negotiations for employees working at similar jobs across different civil service departments may also give rise to situations of "exogenous" pay comparisons: for example, in 2005 employees at the UK Department for Environment, Food, and Rural Affairs (DEFRA) took industrial action over pay gaps resulting from separate pay negotiations for DEFRA employees and employees at its associated agencies (see, e.g., http://news.bbc.co.uk/1/hi/uk/4363684.stm).

(7.) Note that here I am focusing on pay equality as a criterion to assess pay fairness. However, in some circumstances employees may perceive equal pay as unfair (see, e.g., Abeler et al. 2010). I return to this point later in Section V.

(8.) However, differently from standard bilateral gift-exchange games, fairness considerations in SECRET may also depend on employees' beliefs about the (unobservable) wages paid to the co-worker.

(9.) However, Charness (2004) find little difference between a treatment where wages are mandated by the experimenter and a treatment where wages are selected by a random device.

(10.) This may be particularly true for settings where the exogenous constraints contain value-laden elements that may speak to workers' entitlements, as in the case of minimum wages or union-negotiated wage rates (see Gachter and Riedl 2005 on the importance of entitlements for economic outcomes). On this respect, note that I opted for a conservative approach to mandate the RED wage levels in PUBLIC/EXO rather than for more value-laden interventions.

(11.) Using Wilcoxon signed-ranks tests I find statistically significant differences in effort across contingencies where the own wage is 1 [pound sterling] or 7 [pound sterling], both in SECRET (p = .030) and PUBLIC for any level of the co-worker's effort (all p [less than or equal to] .034). I further analyze this own wage-effort relation in the regression analysis of Table 2.

(12.) Treating each subject as an independent observation and using tests of equality of proportions (Glasnapp and Poggio 1985), I find that reward (punishment) rates are significantly higher (lower) in SECRET when the own wage is 4 [pound sterling] than in PUBLIC when the own wage is 4 [pound sterling] and the co-worker's wage is 7 [pound sterling] (p = .035 for rewards; p = .072 for punishments). I do not detect significant differences in reward and punishment rates of SECRET and PUBLIC for the other cases of underpayment (p > .146).

(13.) A series of tests of equality of proportions detect marginally significant differences in reward rates between SECRET when the own wage is 7 [pound sterling] and PUBLIC when the own wage is 7 [pound sterling] and the co-worker's wage is 1 [pound sterling] or 4 [pound sterling] (p = .091 in both comparisons). 1 do not find significant differences between treatments for reward rates when the own wage is 4 [pound sterling] and the co-worker's wage is l [pound sterling], or punishment rates in any comparison (p > .146).

(14.) Using tests of equality of proportions, I cannot detect any significant difference in punishment or reward rates between SECRET when the own wage is 1 [pound sterling], 4 [pound sterling], or 7 [pound sterling] and PUBLIC when the own wage and the co-worker's wage are respectively 1 [pound sterling], 4 [pound sterling], or 7 [pound sterling] (p > .311).

(15.) Moreover, according to a Wald test, underpayment has a significant negative effect on effort in PUBLIC relative to the case where employees receive the same wage as the co-worker (p = .022).

(16.) Comparing the effects of pay comparison information within PUBLIC, I find that overpayment does not reduce effort significantly relative to equal payment according to a Wald test (p = 0.119).

(17.) The fact that all three PUBLIC treatment dummies enter the regression with a negative sign raises the question whether disclosing co-workers' wages has a generally negative effect on effort provision regardless of relative pay conditions. To examine this, I ran an additional regression where I replaced the three PUBLIC treatment dummies in Model I with one dummy variable assuming value 1 for observations from the PUBLIC treatment. The coefficient of the dummy variable is negative but statistically insignificant (p = 0.173) suggesting that I do not observe a general "public information effect" in PUBLIC.

(18.) An interesting question is whether the negative effects of pay comparisons are driven by changes in rewarding or punishing behavior. To address this question I perform separate logit regressions for the cases where an employee punishes or rewards the employer using a slightly modified version of Model I. in the modified specification I replace the dummy variables PUBLIC OVERPAID and PUBLIC UNDERPAID with one variable which assumes value 1 when the employee is paid a different wage than the co-worker, and value 0 otherwise. This modification is necessary because I never observe rewards when an employee is underpaid in PUBLIC (see Figure 1). The regressions show that being paid differently than the co-worker significantly reduces rewarding (p = .042), but does not increase the propensity to punish (p = .655). The complete regression results are available upon request.

(19.) The negative effects of pay comparisons reflect on realized total earnings per firm. The 10 firms in SECRET earned on average 22.4 [pounds sterling] from the experiment, whereas the average earnings of the 10 firms in PUBLIC were 20.9 [pounds sterling]. The difference is just insignificant according to a Mann-Whitney rank-sum test (p = .134).

(20.) Using Wilcoxon signed-ranks tests I find statistically significant differences in employees' effort across wage combinations where the own-wage is 1 [pounds sterling] or 7 [pounds sterling] for both possible levels of the co-worker's wage (p [less than or equal to] .026).

(21.) Using tests of equality of proportions, I find that reward (punishment) rates are higher (lower) in SECRET when the own wage is 4 [pounds sterling] than in PUBLIC/EXO when the own wage is 4 [pounds sterling] and the co-worker's wage is 7 [pounds sterling] (p = .035 in both comparisons). I do not find significant differences in reward or punishment rates across the two treatments when the own wage is 1 [pounds sterling] and, in PUBLIC/EXO, the co-worker's wage is 7 [pounds sterling] (p > .146). I also note that underpayment leads to similar effort choices in PUBLIC/EXO and PUBLIC (see Figure 1). In fact, I do not find any significant difference in reward or punishment rates between the two "public wages" treatments in any possible comparison (all p > .677).

(22.) Reward rates in SECRET when the own wage is 4 [pounds sterling] or 7 [pounds sterling] are higher than reward rates in PUBLIC/EXO when the own wage is 4 [pounds sterling] or 7 [pounds sterling] and the co-worker's wage is 1 [pounds sterling] (tests of equality of proportions, p = .035 and p = .038, respectively). I do not find significant differences in punishment rates (p > .311). Reward rates are generally lower in PUBLIC/EXO than in PUBLIC for overpaid employees. In fact, I find a significant difference in reward rates between the two "public wages" treatments when the own wage is 4 [pounds sterling] and the co-worker's wage is 1 [pounds sterling] (p = .072), suggesting that overpayment may result in higher effort when employers control both employees' wages. I do not find, however, significant differences between PUBLIC and PUBLIC/EXO in reward rates when the own wage is 7 [pounds sterling] and the co-worker's wage is 1 [pounds sterling] (p = .677), or in punishment rates (p > .311).

(23.) There are no significant differences in reward or punishment rates between SECRET when the own wage is 1 [pounds sterling] or 7 [pounds sterling] and PUBLIC/EXO when both the own wage and the co-worker's wage are 1 [pounds sterling] or 7 [pounds sterling] (tests of equality of proportions, p > . 146). I also do not find any difference in reward or punishment rates between PUBLIC/EXO and PUBLIC (p > .311).

(24.) Note, however, that effort choices under underpayment or overpayment do not differ significantly from effort expended in PUBLIC/EXO under pay equality according to the Wald tests reported in Table 3.

(25.) As with the PUBLIC treatment (see Footnote 17), I ran an alternative specification of Model 1 where I replaced the three PUBLIC/EXO dummies with one dummy variable assuming value 1 for observations from the PUBLIC/EXO treatment, regardless of relative pay conditions. The coefficient of the dummy variable is negative and significant at the 1% level, suggesting that disclosing information about coworkers' wages has a generally detrimental effect on effort provision in PUBLIC/EXO. I also performed separate logit regressions with dummy-dependent variables for punishment and reward using the modified version of Model I discussed in Footnote 18 (replacing the two dummy variables PUBLIC/EXO OVERPAID and PUBLIC/EXO UNDERPAID with one variable taking value 1 when the employee's wage is different from the co-worker's). The propensity to reward is significantly reduced when employees are paid a different wage than the co-worker (p = .002), whereas the propensity to punish does not increase significantly (p =. 167). The complete regression results are available upon request.

(26.) Realized earnings per firm in PUBLIC/EXO are lower than SECRET (21.2 [pounds sterling] vs. 22.4 [pounds sterling]), but the difference is statistically insignificant according to a Mann-Whitney rank-sum test (p = .429).

(27.) No pay comparison effects are instead found in the real effort experiment by Hennig-Schmidt, Rockenbach, and Sadrieh (2010). However, employees in their experiment are also unwilling to provide high effort in return for high own wages. Subsequent laboratory experiments (where no pay comparisons were possible) revealed that this was due to the lack of surplus information, which limited the scope for employees to develop adequate fairness attributions. Clark, Masclet, and Villeval (2010) use bilateral gift-exchange games that vary in whether employees receive information about the wages offered in four other firms in the market (i.e., they study inter-firm pay comparisons). They find that effort is significantly affected by how the own wage is ranked relative to others' wages.

DANIELE NOSENZO, I thank the Editor and two anonymous referees for useful comments. I have received helpful comments from Robin Cubitt, Armin Falk, Simon Gachter, Martin Sefton, Lise Vesterlund, seminar participants at the Tinbergen Institute Amsterdam, and participants at the 2009 ESA European Regional meeting. Funding from the University of Nottingham and the Leverhulme Trust (ECF/2010/0636) is gratefully acknowledged. I am grateful for the hospitality of the University of Pittsburgh while working on this paper.

Nosenzo: School of Economics, University of Nottingham, University Park, Nottingham NG7 2RD, UK. Phone +44(0)115 84 67492, Fax +44(0)115 95 14159, Email Daniele.Nosenzo@nottingham.ac.uk

TABLE 1
Overview of Treatments

 Both determined One determined
Wages by the Employer exogenously

Public PUBLIC {PUBLIC/EXO 1 [pounds sterling]
 PUBLIC/EXO 7 [pounds sterling]
Private SECRET --

TABLE 2
Effort Behavior in SECRET and PUBLIC

 Model II

 Own wage = 1
 Model I [pounds sterling]

Own wage 1 [pounds sterling] -0.63 ** (0.511) --
Own wage 7 [pounds sterling] 1.94 *** (0.392) --
PUBLIC EQUAL PAY -0.28 (0.525) --
PUBLIC UNDERPAID -0.66 ** (0.481) --
PUBLIC OVERPAID -0.52 (0. 615) --
PUBLIC 1 [pounds sterling] -- -0.15 (0.980)
PUBLIC 4 [pounds sterling] -- -0.35 (0.908)
PUBLIC 7 [pounds sterling] -- -0.50 (0.911)
1 if male -0.73 ** (0.532) -0.86 ** (0.806)
1 if studies Social Sciences 0.15 (0.511) 0.34 (0.889)
 (incl. Economics)
N 240 80
Pseudo-[R.sup.2]: 0.155 0.121

 Model II

 Own wage = 4 Own wage = 7
 [pounds sterling] [pounds sterling]

Own wage 1 [pounds sterling] -- --
Own wage 7 [pounds sterling] -- --
PUBLIC EQUAL PAY -- --
PUBLIC UNDERPAID -- --
PUBLIC OVERPAID -- --
PUBLIC 1 [pounds sterling] -0.51 (0.937) -0.56 (0.752)
PUBLIC 4 [pounds sterling] -0.52 (0.784) -0.56 (0.737)
PUBLIC 7 [pounds sterling] -0.88 *** (0.643) -0.15 (0.728)
1 if male -0.75 ** (0.703) -0.62 (0.689)
1 if studies Social Sciences -0.19 (0.740) 0.30 (0.569)
 (incl. Economics)
N 80 80
Pseudo-[R.sup.2]: 0.121 0.059

Wald tests:
PUBLIC EQUAL PAY = PUBLIC UNDERPAID p = 0.022
PUBLIC EQUAL PAY = PUBLIC OVERPAID P = 0.119
PUBLIC OVERPAID = PUBLIC UNDERPAID p = 0.440

Notes: Ordered logit regressions. The dependent variable is effort.
Robust standard errors in parentheses, adjusted for intragroup
correlation (individuals are used as independent clustering units).
The results are displayed as percentage changes in the odds of
expending higher effort. In Model I the reference subject type is:
in the SECRET treatment, receiving a medium wage of 4 [pounds
sterling], female, and studying in a field other than Social
Sciences. In Model II the reference subject type is: in the SECRET
treatment, female, and studying in a field other than Social
Sciences. Significance levels: * 10%; ** 5%; *** 1%.

TABLE 3
Effort Behavior in SECRET and PUBLIC/EXO

 Model II

 Own wage = 1
 Model I [pounds sterling]

Own wage 1 [pounds sterling] -0.70 *** (0.407) --
Own wage 7 [pounds sterling] 3.67 *** (0.546) --
PUBLIC/EXO EQUAL PAY -0.57 (0.554) --
PUBLIC/EXO UNDERPAID -0.86 *** (0.657) --
PUBLIC/EXO OVERPAID -0.74 *** (0.485) --
PUBLIC/EXO 1 [pounds sterling] -- -0.55 (0.906)
PUBLIC/EXO 7 [pounds sterling] -- -0.73 (0.916)
1 if male 0.28 (0.398) 0.32 (0.639)
1 if studies Social Sciences 0.35 (0.391) 0.14 (0.689)
 (incl. Economics)
N 180 60
Pseudo-[R.sup.2] : 0.185 0.038

 Model II

 Own wage = 4 Own wage = 7
 [pounds sterling] [pounds sterling]

Own wage 1 [pounds sterling] -- --
Own wage 7 [pounds sterling] -- --
PUBLIC/EXO EQUAL PAY -- --
PUBLIC/EXO UNDERPAID -- --
PUBLIC/EXO OVERPAID -- --
PUBLIC/EXO 1 [pounds sterling] -1.00 *** (0.605) -0.71 (0.829)
PUBLIC/EXO 7 [pounds sterling] -1.00 *** (1.028) -0.39 (0.728)
1 if male 0.69 (0.755) 0.01 (0.557)
1 if studies Social Sciences 1.63 (0.900) 0.08 (0.622)
 (incl. Economics)
N 60 60
Pseudo-[R.sup.2] : 0.269 0.039

Wald tests:
PUB/EXO EQ. PAY = PUB/EXO UNDERP. p = 0.116
PUB/EXO EQ. PAY = PUB/EXO OVERP. p = 0.397
PUB/EXO OVERP. = PUB/EXO UNDERP. p = 0.345

Notes: Ordered logit regressions. The dependent variable is effort.
Robust standard errors in parentheses adjusted for intragroup
correlation (individuals are used as independent clustering units).
In PUBLIC/EXO only data from BLUE Employees is used in the
regressions. Results are displayed as percentage changes in the
odds of expending higher effort. In Model I the reference subject
type is: in the SECRET treatment, receiving a medium wage of 4
[pounds sterling], female. and studying in a field other than
Social Sciences. In Model II, the reference subject type is: in the
SECRET treatment, female, and studying in a field other than Social
Sciences.

Significance levels: * 10%; ** 5%; *** 1%.