Backscratching in Hierarchical Organizations

Backscratching in Hierarchical Organizations Abstract We experimentally investigate the role of reciprocity in sustaining the emergence of implicit collusive agreements in hierarchical organizations. We show that when an agent hires, on behalf of the principal, one worker out of two candidates: (i) low ability workers, being less entitled to be selected, are more likely to exert effort in a task that is exclusively beneficial to the agent; (ii) as a consequence, agents distort the hiring process in favor of low ability workers; and (iii) sharing a small part of the organization’s profits with the workers alleviates their effort distortion 1. Introduction The possibility that managers and bureaucrats buy the cooperation and the loyalty of their subordinates at the expense of the organization characterizes many hierarchical organizations (Edwards 1979). Employees who are promoted or hired based on their loyalty toward their supervisors, and not on their merit, are more likely to invest their time and attention to maintaining good relations with their boss, rather than improving their job performance (Pearce et al. 2000). Tirole (1986) highlights how, in a three-tier principal–supervisor–agent model, collusion between the supervisor and the agent can be sustained by a norm of reciprocity. Laffont (1988, 1990) generalizes the notion of moral hazard to include hidden gaming, defined as the “ability that some players may have to design and play games with other members of the hierarchy by which they benefit from others while they are not observable by the principal” (Laffont 1990: 302). In this paper, we provide the first experimental evidence that reciprocity sustains hidden gaming in hierarchical organizations. First, we show the emergence of backscratching between members at the lowest and at the intermediate level of a three-tier organization, at the expense of the principal. Agents, who select workers for a job on behalf of the principal, are more likely to hire candidates with lower ability. These candidates feel less entitled for the position and thus they are more prone to devote their effort in favor of the agent who selected them. Second, we design a compensation scheme which, without increasing the cost of the principal, limits the detrimental effect of backscratching reciprocity. In our experiment, we render a three-tier organization formed by one principal, one agent, and one worker. The worker has to be chosen from a pool of two candidates who differ in their ability.1 The principal, in contrast to the agent, is not able to distinguish the candidates’ ability and, therefore, delegates the hiring decision to the latter. Once employed, the worker receives a fixed wage and chooses a level of non-verifiable effort that can be exerted both in project X, which is beneficial for the principal and the agent, and in activity Y, that provides a private benefit solely to the agent. The joint payoff of the principal and the agent is maximized when all the effort of the worker is devoted to project X; however, the agent’s payoff is maximized when the worker’s effort is exerted in activity Y. Candidates differ in their abilities: for each level of effort exerted in project X the high ability candidate is more productive than the low ability one. When exerting effort in activity Y, the two candidates are equally productive. While it is public information that candidates have different abilities, only the agent is able to distinguish among them and this is precisely the reason why she is hired by the principal. Our aim is to capture a situation where the agent can exploit her position in order to induce subordinates to do certain activities that go beyond their formal job descriptions but give to her a personal benefit.2 In the Baseline treatment, the principal privately gives instructions to the agent about which candidate to hire: either the high or the low ability one. The agent selects one candidate and the hired worker chooses how much effort to exert in project X and in activity Y. The agent has to follow the principal’s instructions and thus cannot take any decision. In the Selection treatment, the agent is not forced to follow the instructions received by the principal. Moreover, the agent has the opportunity to privately suggest to the hired worker a level of effort to exert both in X and Y. Communication renders the agent’s intentions clear: the agent’s decision to select one of the two candidates may be aimed either at increasing the profit of the organization as a whole or at getting personal benefits from the employment relationship. Compared with the Baseline treatment, we observe a significant increase in the number of low ability candidates hired in the Selection treatment. Moreover, we provide evidence that such a hiring distortion is due to the fact that low ability workers exert more effort in activity Y and less effort in project X than the high ability ones. Agents do strategically exploit the reciprocal concerns of low ability workers who feel less entitled to get the job and thus are more grateful to the agent for being selected. We then examine whether distributing a small fraction of the profits to the workers is effective in limiting the emergence of backscratching reciprocity. The Baseline – Profit sharing and Selection – Profit sharing treatments replicate the design implemented in the corresponding Baseline and Selection treatments with the single difference that the worker receives a small fraction of the value generated in project X. In the Selection – Profit sharing treatment, we find that distributing a small share of the value of project X to the workers is effective in reducing the low ability workers’ effort distortion in favor of agents with respect to the Selection treatment. Moreover, when comparing the Baseline – Profit sharing to the Baseline treatment we find that workers, irrespectively from their ability, exert significantly higher effort in project X. Given that the share of the value of project X distributed to the worker is very small, we suggest that such a result is not simply driven by the monetary motive associated with the profit sharing payoff structure. It seems rather due to a positive effect on the worker’s attitude toward the firm: even the presence of a tiny link between payment and effort is shown to be relevant in shaping the way workers perceive the organization. 2. Related Literature There is a vast experimental literature on gift exchange games, initiated by Fehr et al. (1993) and followed by many applications both in the laboratory (Fehr et al. 1997, 1998a, 1998b; Fehr and Falk 1999; Charness 2004; Eriksson and Villeval 2012) and in the field (Gneezy and List 2006; Bellemare and Shearer 2009; Hennig-Schmidt et al. 2010; Kube et al. 2013). In the field, the effectiveness of gift exchange seems to depend on the details of the environment and, in particular, it could be enhanced when targeted toward reciprocal employees (Englmaier and Leider 2012). These studies all focus on two-tier settings and test the positive effect of reciprocity in limiting opportunistic behaviors of workers. The novelty of our approach consists in testing the gift exchange hypothesis in three-tier organizations. By mean of an experiment we show that reciprocity may damage hierarchical organizations, when their members use it as an enforcement device to acquire personal illegitimate benefits. A second branch of literature related to our paper refers to the distortions in the hiring process due to favoritism within organizations. Several studies have pointed out that organizations’ performance is usually negatively affected when candidates’ evaluation is not based on their ability (Levine et al. 2010; Kramarz and Thesmar 2013). This can happen when candidates are hired or promoted on the basis of subjective rather than objective criteria.3 Managers may indeed favor workers according to their social connections (Bandiera et al. 2009) and personal preferences when objective evaluations of workers’ performance are not available (Prendergast and Topel 1993), or may favor those who engage in ingratiatory behavior regardless of their objective ability (Robin et al. 2014). With respect to these studies, we look to an additional motivation of the distortion of the hiring process: the attempt to induce reciprocal behavior in less entitled workers by favoring them in the selection, in the hope of receiving future benefits.4 The dark side of reciprocity has already been analyzed by scholars in other disciplines. Studies in Social Psychology underline how recipients of (unsolicited) gifts feel indebted toward the gift givers and are more likely to “return the favor,” once requested. These studies show how individuals can trigger reciprocity in order to gain an unfair advantage (Cialdini 1996). In Organizational Science backscratching is identified as (vertical) cronyism and indicates a favoritism of the superior toward subordinates (as, e.g., the assignment of promotion, bonus, pay rise, or better job) based on criteria different than merit in exchange for the latter’s personal loyalty (Khatri and Tsang 2003). Reciprocity has also been shown to be a key feature in sustaining corruption agreements, which cannot be enforced by third parties (Abbink et al. 2002; Abbink 2004; Barr and Serra 2009).5 We are not aware of any experimental study looking at the emergence of implicit collusion sustained by reciprocity in hierarchical organizations. In a recent study, Malmendier and Klaus (2017) analyze the emergence of a dark side of reciprocity in a different context, a client–producer relationship. They find that when a decision maker has to buy a product on behalf of a client, and two producers compete to sell the product, the possibility of one producer sending a small gift to the decision maker increases the probability that the recipient chooses the gift-giver’s product, even if favoring the gift-giver will damage the client. Compared with Malmendier and Klaus (2017), there are two main differences in our design. First, in our experiment workers hired by agents can reciprocate toward them without damaging their principal, and therefore there is not necessarily a tension between reciprocating a gift and fulfilling a duty. Second, our design allows us to investigate the role of subjective entitlement within organizations. Following Schlicht (1998: 24), we define entitlements as “subjectively perceived rights that go along with a motivational disposition to defend them”. In our setting, the low ability candidate is less entitled to be selected for the job compared with the high ability one. The relevance of subjective entitlement has been displayed by recent papers showing that gifts offered by employers to workers who belong to relatively disadvantaged groups and/or to the lower part of the performance distribution are likely to elicit more gratitude (Baron 2013; Montinari et al. 2016). Our findings are in line with Kolm (2006) evidencing how the beneficiaries of (unsolicited) gift tend to feel “morally indebted” toward the gift giver. Moreover, our contribution highlights the crucial role of subjective entitlement in activating this feeling. A third branch of literature related to our work analyzes how to reduce the negative effects of internal collusion. Bac (1996) studies how hierarchical structures affect the level of corruption in an organization; Thiele (2013) suggests decreasing the incentives for employees while increasing managers’ compensations. Chang and Lai (2002) investigated the impact of social norms on supervisors’ corrupt behavior showing that, when in the presence of corruption, paying supervisors more than workers limits workers’ slack. We show that reward systems do not only provide monetary incentives but also affect workers’ attitudes toward the corporate culture, ultimately affecting their productivity. Using a profit sharing compensation scheme to increase the organizational performance, via positive changes in employees’ attitude, has already been suggested (Osterman 1994; Knez and Simester 2001; Coyle-Shapiro et al. 2002; Heywood et al. 2005) and is a well-known phenomenon in the economics literature (Kerr and Slocum 1987). 3. The Model The model analyzes a three-tier organization, formed by a Principal (P), an Agent (A), and a worker. The worker is selected from a pool of two candidates, a low ability worker (L worker) and a high ability worker (H worker). The principal needs a worker to carry out a project but, being unable to distinguish between the ability of the two candidates, he hires an informed agent and pays her a fixed compensation to select one of the two candidates. Once selected, the worker receives a fixed compensation and chooses how much effort x≥0 to exert in the project valuable to the principal and how much effort y≥0 in a different activity that benefits the agent, with the overall effort resulting from the sum of the two effort components: e=x+y. The (monetary) cost of effort c(e) is a differentiable, strictly increasing and convex function in the overall amount of effort e exerted by the worker, with c(0)=0. The principal’s monetary payoff is equal to πP=(x+φt), where φt is the ability of worker t∈{H, L}, with φH>φL≥0. The (hired) worker’s monetary payoff is equal to πt=mW−c(e), where mW > 0 is the compensation paid to the worker which does not depend on his ability. The unemployed candidate receives a monetary payoff equal to zero. The agent’s monetary payoff is equal to πA=mA+y, with mA > 0. We assume that mA > mW which implies that if a worker with ability t∈{H, L} exerts a total amount of effort equal to zero, then πA>πH=πL>πP. We assume that individuals may exhibit other regarding preferences (ORPs) and therefore their utility may not coincide with their monetary payoff.6 Specifically, the utility function of individual i∈{P, A, H, L}, who belongs to the organization, is given by: Ui=πi+I(∑j≠i(θiρi,jγπj−(1−θi)(αmax {(πj−πi),0}+βmax {(πi−πj),0}))). (1)   I∈{0,1} is an index function that takes value 0 if individual i is selfish and 1 if individual i exhibits ORPs. Individual i’s ORP toward individual j≠i are a linear combination of two components, a reciprocity component, and an inequity aversion component; θi is the relative weight of the reciprocity component; and 1−θi of the inequity aversion. The reciprocal attitude toward individual j is described by the parameter ρi,j that takes value 1 if individual j has been kind with individual i and 0 otherwise, and the parameter γ∈(0,1), the weight of individual js monetary payoff in agent is utility, which is assumed to be the same for every R-ORP individual. Inequity aversion is modeled according to the standard Fehr and Schmidt (1999) model with, β≤α and 0≤β<1. Individuals who exhibit ORP are heterogeneous with respect to the weight assigned to the reciprocity component. For sake of simplicity, we assume that θi∈{0,1}; we then classify individuals with respect to their preferences in: selfish (SP), inequity averse (I-ORP), and reciprocity concerned (R-ORP). We assume that workers’ preferences are not correlated with their ability and that every player is SP, I-ORP, and R-ORP, respectively, with positive probability. We consider two different games. In both games the principal plays an action sP∈{H, L} that we interpret as the suggestion given to the agent on which worker to hire. In the baseline game the suggestion is binding, the agent is therefore a fictitious player who has no actions to play, and the hired worker t∈{H, L} chooses how much efforts x and y to exert. In the selection game the suggestion is not binding, the agent chooses which worker t∈{H, L} to hire, and the hired worked chooses the pair (x, y) of efforts to exert. Players’ reciprocity depend on their beliefs about the other players’ actions. R-ORP worker i∈{H, L} shows reciprocity concerns toward either the principal or the agent, depending on his beliefs s^iP about which type of worker the principal has suggested to hire. Specifically, we assume that ρi,P=s^iP. Similarly, an R-ORP worker believes that the agent was kind with him if he was hired against the principal’s suggestion, that is, ρi,A=1−s^iP. An R-ORP agent does not feel any reciprocity toward the principal, while she is grateful to the type of worker who will provide her with the largest monetary payoff. Finally, principal’s reciprocity toward the worker depends on his beliefs about the amount of effort x^t that a worker of ability t∈{H, L} will exert. We assume that ρP,H=1 if x^H>x^L and zero otherwise, and ρP,L=1 if x^L>x^H and zero otherwise, that is the principal believes that a worker of ability t is kind with him if he exerts strictly more effort than the other worker τ∈{H, L}, with τ≠t.7 We solve both games by backward induction. Consider first the baseline game. Starting from the last stage, selfish workers exert zero effort both in x and y. Whatever worker i∈{H, L} is hired, ρi,P=1, since the principal’s suggestion is binding. Given that, by design, the principal is the player with lowest monetary payoff (at least when x=y=0), in equilibrium both I-ORP and R-ORP workers exert a positive amount of effort x and zero effort in y, and the amount of effort exerted by H and L workers, on average, is the same. Principals, irrespective of their preferences, hire H workers because they exert the same effort than L workers but they have higher ability. The following proposition summarizes these findings. Proposition 1. In the baseline game, selfish workers exert zero effort in x and y. Workers with ORP exert a positive amount of effort x and zero effort y. On average, workers of different abilities exert the same amount of efforts. Principals hire H workers. Proof. See the Appendix.  Let xtBSL ( xtSEL) and ytBSL ( ytSEL) denote the average effort exerted in equilibrium of the baseline (selection) game by a worker of ability t∈{H, L}. From Proposition 1, we derive the following theoretical prediction about workers’ behavior: Prediction 1. In the baseline game, on average, H and L workers behave similarly, exerting a positive level of effort in x and zero in y: xHBSL=xLBSL>0, yHBSL=yLBSL=0. Consider now the selection game. This game shows a multiplicity of equilibria depending on the principal’s suggestion. We focus on the equilibrium in which principals suggest to hire H workers, as in the baseline game, also because it is the unique equilibrium that exists for any possible probability distribution over workers’ preferences (in fact, it is easy to check that this is the only equilibrium if all workers are selfish). Starting from the last stage, selfish workers exert zero effort. Consider R-ORP workers. In equilibrium beliefs are correct so ρH,P=1 and ρH,A=0, while ρL,P=0 and ρL,A=1. Therefore, H workers with R-ORP exert effort x while L workers with R-ORP exert effort y. Consider now I-ORP workers. Since the worker has always a lower monetary payoff than the agent (even when x=y=0) and the principal has a lower monetary payoff than workers if they choose x=y=0, it follows that in equilibrium  I-ORP workers, irrespective of their type t∈{H, L}, exert zero effort y and a positive amount of effort x. The following proposition summarizes these findings. Proposition 2. There exists an equilibrium of the selection game in which principals suggest to hire H workers. In this equilibrium, selfish workers exert zero effort. Inequity-averse workers exert a positive amount of effort x and zero effort y, irrespective of their ability. Reciprocity-concerned workers of ability H exert positive effort x and zero y, while reciprocity-concerned workers of ability L exert zero effort x and a positive effort y. Selfish and reciprocity-concerned agents hire L workers, while inequity-averse agents hire H workers if α≥c′(γ)−1c′(β1−β)−1+c′(γ)−1, and L workers otherwise. Proof. See the Appendix. From Proposition 2, we derive the following theoretical prediction about workers’ behavior: Prediction 2. In the selection game, on average, H workers behave as H (and also L workers) in the baseline game, exerting a positive level of effort in x and zero in y: xHBSL=xHSEL=xLBSL>0, yHBSL=yLBSL=yHSEL=0; L workers exert on average a lower level of effort x and a higher level of effort y than H workers: yLSEL>0 and xLSEL<xLBSL. Finally, the last theoretical prediction refers to the hiring of workers. Prediction 3. The fraction of low ability workers hired in the selection game is larger than the fraction of low ability workers hired in the baseline game. Up to now we have assumed, as is customary in economics, that preferences are exogenously determined by nature. According to our model, agents with ORP preferences are either inequity averse or reciprocal. However, which component of the ORP preferences, reciprocity or inequity aversion, turns out to be more prominent for an individual could also be influenced by the environment. To investigate, in an anonymous laboratory setting, whether the features of the working environment affect which one of these two potentially conflicting preferences prevails, we consider a modification of the workers’ payment. In particular, we introduce a tiny profit sharing compensation scheme that assigns an additional small fraction of the principal’s profits to the worker. Let ηw denote the fraction of the principal’s profits distributed to the worker. To avoid the monetary incentives could directly influence the worker’s effort x, we assume that ηw<c′(0). Nevertheless, we conjecture that ORPs could be influenced by the type of contract offered to the worker. Namely, a profit sharing contract could render the principal’s payoff salient to the workers, and therefore could induce all workers with ORP preference to be inequity averse. As a consequence, any differences between the two games, baseline and selection, should disappear when a profit sharing contract is offered to workers. Conjecture 1 . If a profit sharing contract renders the principal’s payoff salient to the workers, then reciprocity-concerned workers of ability H and L exert positive effort x and zero y both in the (profit sharing) baseline and selection games. In particular, workers with ability L should behave the same in the baseline and in the selection game in the presence of a profit sharing contract. Before concluding this section it is worth emphasizing that our simple model predicts that every worker never simultaneously exerts both efforts, x and y. It is easy to generalize the model to admit that some workers could choose to exert positive effort x and y. Suppose that θ has a continuous probability distribution with support [0,1] and Ui=πi+I(∑j≠iθiρi,jγπj+(1−θi)β(|πj−πi|)α), with 0<γ≤1, β<1, and α>1 (to avoid both components to be linear). It is easy to check that in the baseline game all ORP workers exert positive effort x and zero y, because both reciprocity and inequity aversion are addressed toward the principal’s payoff. In the selection game (focusing on the equilibrium in which principals suggest to hire H workers) H workers, irrespective of their θ, still exert zero effort y and positive x, while a L worker i with θi∈(0,1) will exert a positive amount of effort x, due to his inequity concern toward the principal, but also a positive amount of effort y, to reciprocate the agent’s kindness. Still, the qualitative results of the simple version of the model with θ∈{0,1} remain unchanged, and, importantly, the above predictions still hold. 4. Experimental Design and Parameters In our experiment we implement a between subjects analysis in a 2× 2 design, where we vary either the payoff structure of the game and whether the principal or the agent is making the hiring decision, mirroring the games discussed in the theoretical model, with few differences. In the laboratory, we render an organization composed by a principal (P), an agent (A), and a worker, either of low ability (L worker) or high ability (H worker). The principal hires an informed agent and pays her to select one of the two candidates. The principal gives instructions to the agent about which type of worker t∈{H, L} to hire, either H or L worker, and the agent selects the worker. The selected worker receives a fixed compensation by the principal and chooses how much effort to exert in project X and in activity Y. After the worker’s decision, payoffs are determined and the game ends. The non-selected candidate receives an unemployment benefit. In the baseline treatment (BSL), participants play the game just described and the principal gives binding instructions to the agent about which worker to hire: either L or H worker. In the selection treatment (SEL), players’ payoffs are the same as in the BSL treatment, but the agent (i) chooses whether to hire worker L or worker H, since principal’s suggestion is not binding anymore and (ii) may suggest to the hired worker a desired effort level to exert in X and Y. The baseline profit sharing (BSLPS) and selection profit sharing (SELPS) treatments are identical to the BSL and SEL treatments, respectively, with the only exception that we introduce a tiny profit sharing compensation for the workers. Panel A of Table 1 summarizes the values assigned to the parameters in each treatment, with 10 ECUs = 1 Euro implemented as exchange rate. Table 1. Experimental Parameters, Cost Function, and Material Payoff Functions Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Notes: In Panels B and C, the values are obtained considering a realization of σ=E(σ)=1. Panels B and C report 7.5% (15%) of the output, that is, the value produced in project X. Table 1. Experimental Parameters, Cost Function, and Material Payoff Functions Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Notes: In Panels B and C, the values are obtained considering a realization of σ=E(σ)=1. Panels B and C report 7.5% (15%) of the output, that is, the value produced in project X. In the experiment, we choose ability φH=0.5>φL=08; the levels of effort exerted in X and Y are integer numbers between {1,2,…,10} and {0,1,…,5}, respectively, and their sum has to be greater than one and less than or equal to 10. Effort in project X generates an output X=(x+φt)σ which is affected by a random variable, denoted by σ, uniformly distributed on the interval [0.8,1.2] with mean equal to 1. The random variable together with the small difference in ability makes it harder for the principal to infer the ability of the worker hired by the agent, since he only receives information about the value produced in X, which is a noisy signal of both the worker’s effort and ability. Effort in activity Y generates an output Y=δy, with δ=0.4, so that the workers’ ability does not affect their performance in activity Y. Panel B of Table 1 reports the worker i’s payoff function: πi=mW−c(e). mW is the fixed compensation that he receives from the principal when being hired, irrespective from his type t∈{H, L} and c(e) is the cost of the total effort exerted by the hired worker (i.e., e=x+y), which is taken from Fehr et al. (1998a), with the cost of the minimum level of enforceable effort (i.e., x = 1 in project X) being equal to 0. The cost of exerting effort, c(e), is an increasing function of the overall effort and does not depend on whether it is exerted in X and/or in Y. If not hired, the worker receives a fixed unemployment benefit of 10 ECUs. In the BSLPS and SELPS treatments, a tiny profit sharing compensation scheme for the worker, ηW=0.075, is introduced. Panel C of Table 1 reports the monetary payoffs for the agent and the principal. The agent’s monetary payoff is increasing both in the effort exerted by the hired worker in project X and in activity Y: πA=ηAX+δy+mA, where mA is a fixed compensation paid by the principal. The fraction of the output X assigned to the agent depends on the treatment: while in the BSL and SEL treatments ηA=0.15, in the BSLPS and SELPS treatments it is reduced to ηA=0.075, in order not to change the cost of the incentive schemes for the principal. Selfish and R-ORP agents, who believe that all workers are payoff maximizers, should choose to select H workers if ηA is positive. Most importantly, the agent gets the entire amount of the output produced by the worker’s effort in activity Y=δy and, since δ>ηA≥0, the agent’s monetary payoff is higher when a given level of effort is exerted in activity Y rather than in project X. This creates a potential conflict between the principal’s interests (who only benefits from effort in project X) and the agent’s ones. The agent has also the option to ex ante refuse the value produced in activity Y, thus devoting it to the organization following the rules adopted to distribute the value produced in X. Note that agents motivated by inequality aversion (I-ORP) might prefer to choose L workers if they believe that, due to the “entitlement effect,” L workers will exert more effort than H workers. The agent’s possibility to send a suggestion and to refuse the value produced in Y is the two features of our experimental design that allow us to understand the agent’s motivation in selecting a worker of ability t∈{H, L}. The principal’s payoff function is: πP=ηPX+B−mA−mW, where ηP≤1 is the fraction of X kept by the principal, with ηA+ηP=1; B is a monetary endowment while mA and mW are the fixed compensations paid to the agent and the worker, respectively. The experiment consists of two parts. In the first part, subjects participate in one of the treatments described above, playing the game as one shot. In the second part, they play the same game for 15 periods, maintaining the same role as in part 1 but under a stranger random matching protocol (see the experimental protocol in the Online Appendix B).9 In part 2, feedbacks are provided to every participant at the end of each of the 15 periods. The principal receives information regarding the value produced in X while the agent is informed about the effort exerted in both X and Y in their organization. Finally, all participants are informed about: (i) how many H and L workers have been hired in other organizations within the previous period of the session and (ii) the average effort exerted in X and Y by H and L workers in other organizations. Previous research has suggested (Keizer et al. 2008; Gino et al. 2009; Diekmann et al. 2015) that others’ norms violation affects individual choices: we thus expect that when a “collusive” norm begins to spread among players, that is, the proportion of agents choosing L worker and/or the proportion of workers exerting effort in Y increases, then the propensity for subjects to pursue their personal interest at the expense of the organization increases, generating a snowball effect (Chang and Lai 2002). 4.1 Experimental Procedures The experiment was programed using zTree (Fischbacher 2007). In all treatments, participants were undergraduate students from the Friedrich Schiller University in Jena, recruited via the ORSEE software (Greiner 2004).10 We conducted 21 sessions of the BSL, SEL, and SELPS treatments at the experimental laboratory of the Max Planck Institute of Economics (Jena, Germany), from November 2012 to February 2013. Respectively, 216, 212, and 216 subjects participated in the BSL, SEL, and SELPS treatments. A total of 140 subjects participated in the BSLPS treatment, at the laboratory of the Friedrich Schiller University in Jena in January 2017.11 The experimental procedures, length, and average earnings were the same in all treatments.12 More details on the experimental procedures can be found in the Online Appendix B. The duration of each session was about 110 min and the average payment was 17 Euros, including a show up and participation fee of 4 Euros. 5. Experimental Results In this section, we present our experimental findings. First, we analyze the hiring decisions (Section 5.1). Then, we focus on the effort exerted by the workers in project X and in activity Y (Sections 5.2) and on the agents’ intention (Section 5.3). Throughout the analysis we will focus on the 15 periods of part 2, when information about others’ behavior is spread among participants.13 Given that spreading information may render participants’ choices in part 2 dependent upon previous periods in the same session, we perform both (i) two-sample Mann–Whitney tests and (ii) Somers’ D median difference tests (Newson 2002), reporting in the main text only (i) unless the two tests give different results.14 5.1 Hiring Our first result shows how the hiring decisions vary across treatments. Result 1. The proportion of L workers hired is higher in the SEL and SELPS treatments than in the BSL treatment. When hiring L workers, the majority of agents do not comply with the instructions received by the principals. Support for Result 1 can be found in Figure 1, which reports the proportion of principals giving instruction to hire L worker and the proportion of agents hiring L worker in part 2. In the BSL treatment 28.30% of the principals hire L workers while this percentage increases to 47.04% when the agents are making the hiring decision in the SEL treatment. Figure 1. View largeDownload slide Proportion of L Workers Suggested by the Principals and Hired by the Agents across Treatments. Figure 1. View largeDownload slide Proportion of L Workers Suggested by the Principals and Hired by the Agents across Treatments. The decision to hire L workers in the BSL treatment may be motivated by the expectation that, feeling less entitled for the position, they will be more grateful than H workers and therefore will exert a higher effort in project X; as a consequence, the productivity loss due to the difference in ability will be overcompensated, as shown by Montinari et al. (2016).15 It is important to note that in the SEL treatment, in the majority of the cases, L workers are hired against the principal’s suggestion. Specifically, this happens in 58.56% (N = 219/374) of the cases. In the BSLPS treatment, principals increase the proportion of L workers hired compared with the BSL treatment (MW test: z = 1.91, p = 0.06)16: the profit sharing compensation scheme appears to render more salient for them the possibility that an “entitlement” effect could induce L workers to feel more grateful for being hired. Comparing the SEL and SELPS treatments, when agents are making the hiring decision, we find that the tendency to hire more L workers than the ones suggested by the principal is basically unchanged (WSR tests: L suggested versus L hired, SEL z = 1.86, p = 0.06, SELPSz = 2.29, p = 0.02). As already noted for the SEL treatment, also in the SELPS treatment if we focus on those agents who hired L workers, we find that the majority of them did not follow the principal’s suggestion (59.95%, N = 220/367). We interpret these results as a distortion in the hiring process, which is not surprising if we consider that in the SELPS treatment the agents only get ηA=0.075 of the value produced in project X, that is, 50% less than what they got in the SEL treatment, and therefore they have less incentive to hire H workers. When focusing on the individual hiring decisions, we find that in the BSL and BSLPS treatments, on average, 52.27% of the principals decides to hire a L worker between zero and four times while only 25.23% of the agents do so in the SEL and SELPS treatments; on average 42.99% of agents hires L workers eight or more times. This suggests that the reported differences between treatments are not driven by the behavior of a minority of agents. The dynamics of the hiring decision reveals that in the SEL and SELPS treatments the distortion in hiring emerges from the first period and persists until the last one: the proportion of L workers hired increases across periods starting from an average of 41.12% in periods 1−5; increasing to 47.48% in periods 6−10 and reaching 49.91% in the last five periods.17 5.2 Workers’ Effort In the following we concentrate on the effort exerted by the hired workers. Results 2–4 state our main findings, statistical support is provided afterward. Result 2. In the BSL and BSLPS treatment, the average level of effort exerted in project X and in activity Y is the same for H and L workers. Result 2 shows that, on average, workers’ reciprocity does not vary depending on the worker’s ability: in the BSL and BSLPS treatments, H and L workers behave similarly (MW tests H versus L workers: BSL, for x: z = 0.06, p = 0.95; for y: z = 0.45, p = 0.65. BSLSEL for x: z = 0.75, p = 0.45; for y: z = 0.57, p = 0.56). Workers exert an average effort in project X equals to 2.68 and 3.71 in the BSL and BSLPS treatments, respectively, which are significantly higher than the minimum level of effort of 1 (Wilcoxon signed-rank test, BSL: z = 2.37, p = 0.02; BSLPS: z = 7.67, p = 0.00 WSR test, henceforth).18 In activity Y both workers exert an effort significantly higher than zero and equals to 0.89 and 0.71 in BSL and BSLPS treatments, respectively (WSR test, BSL: z = 2.37, p = 0.02; BSLPS: z = 7.67, p = 0.00).19 Result 3. In the SEL treatment, workers exert greater effort in activity Y and less effort in project X compared with the BSL treatment. This difference is driven by the behavior of L workers, who on average increase their effort in activity Y and reduce their effort in project X. When comparing the BSL treatment to the SEL treatment, in line with our theoretical predictions, L workers reciprocate the agents’ hiring choice. In the SEL treatment workers exert more effort in activity Y and reduce their effort in project X. In particular, L workers reduce their effort in X of about 17% passing from an average of 2.67–2.21 and increase their effort in Y of about 81% passing from an average of 0.95 to 1.72 (MW one-sided tests: BSL vs SEL, for x: z = 1.34, p = 0.09; for y: z = 2.75, p = 0.00). Note that, according to our model, a positive amount of effort in Y identifies the behavior of R-ORP workers; I-ORP workers indeed would never exert effort in Y since it advantages the agent, who is the member of the organization with highest monetary payoff.20H workers, instead, do not significantly modify their behavior: they exert an average effort of 2.33 in project X and of 0.99 in activity Y (MW test BSL vs SEL, for x: z = 0.96, p = 0.34; for y: z = 0.32, p = 0.75). Result 4. Profit sharing is effective in limiting backscratching by L workers: differences in the effort exerted in activity Y are not significant when comparing the BSLPS and SELPS treatments. The profit sharing compensation scheme has quite a strong effect on the effort exerted in project X, despite workers only getting a small fraction ηW=0.075 of the value generated in X. We find that, overall, workers increases their effort in X of about 40% in the BSLPS treatment compared with the BSL treatment, and of about 26% in the SELPS treatment compared with the SEL treatment. This increase is significant when considering L and H workers separately, according to a set of MW tests, one-sided: BSLPS versus BSL, xH, z = 2.17, p = 0.03; xL, z = 1.53, p = 0.06. SELPS versus SEL, yH, z = 1.45, p = 0.07; yL, z = 2.11, p = 0.02. Moreover, we observe that the profit sharing induces a reduction of the effort exerted in Y. When comparing the SELPS and SEL treatments, we find that both H and L workers significantly reduces their effort in Y (MW tests, one-sided: yH, z = 1.60, p = 0.06; yL, z = 2.75, p = 0.01). Further support for Results 2–4 can be found in Tables 2 and 3 and in Figure 2. Figure 2 plots, separately for H and L workers, the average effort exerted in X and Y in each treatment. Table 2. Average Total Effort (x+y) Exerted by the Workers, Standard Deviation in Parentheses MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 Note: The tests are conducted considering each session as an independent observation, that is, 7(9) sessions for BSL, SEL, SELPS (BSLPS). Table 2. Average Total Effort (x+y) Exerted by the Workers, Standard Deviation in Parentheses MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 Note: The tests are conducted considering each session as an independent observation, that is, 7(9) sessions for BSL, SEL, SELPS (BSLPS). Table 3. The Effect of Profit Sharing: Comparison of BSL–SEL vs. BSLPS–SELPS Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Notes: * p<0.10, ** p<0.05, *** p<0.01. In all models standard errors are bootstrapped at the level of 30 sessions. N identifies the number of participants who took at least one decision in the second part of the experiment. Over the 15 periods of play, both L and H workers have the chance to be selected in each period. Table 3. The Effect of Profit Sharing: Comparison of BSL–SEL vs. BSLPS–SELPS Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Notes: * p<0.10, ** p<0.05, *** p<0.01. In all models standard errors are bootstrapped at the level of 30 sessions. N identifies the number of participants who took at least one decision in the second part of the experiment. Over the 15 periods of play, both L and H workers have the chance to be selected in each period. Figure 2. View largeDownload slide Average Effort Exerted in Project X and Activity Y in Each Treatment Depending on Workers’ Ability. Figure 2. View largeDownload slide Average Effort Exerted in Project X and Activity Y in Each Treatment Depending on Workers’ Ability. Consider now Table 2: it compares the total effort (e=x+y) exerted by the hired workers across the four treatments. The average total effort exerted by the hired workers (irrespective of their ability) is equal to 3.53 and 3.63 in the BSL and SEL treatments and to 4.43 and 3.65 in the BSLPS and SELPS treatments, respectively. Pairwise comparisons across treatments are not statistically significant except for the BSL and BSLPS treatments (z = 1.79, p = 0.07). Our results show that, overall, the treatments affect how workers allocate their effort between project X and activity Y, but not the total effort exerted per se. In the SEL treatment, therefore, we interpret the effort exerted in activity Y as an effort distortion while in the BSLPS and SELPS treatments the profit sharing payment scheme acts by shifting effort from activity Y to project X. Additional analyses focusing on the frequency of individual effort choices reveal that the observed backscratching is not due to the behavior of a small minority of workers, but it is rather a widespread phenomena, see the Online Appendix A.2.1 for more details. The dynamics of the effort exertion in X and Y reveals that in the SEL backscratching emerges from the first period and persists until the last one, detailed results are reported in the Online Appendix A.2.1. Table 3 allows us to evaluate the impact of the profit sharing incentive scheme on the emergence of backscratching, by performing a difference in differences analysis that compares the difference in effort exertion in project X and in activity Y between the BSL and SEL treatments to the difference between the BSLPS and SELPS treatments. The estimations reported are obtained by means of from a set of Zellner’s seemingly unrelated regressions, which simultaneously estimates two equations (where the dependent variables are the effort exerted by workers in X and Y, respectively) allowing for errors to be correlated. This estimation method allows us to account for the fact that workers simultaneously choose an effort level in X and Y. Standard errors are bootstrapped at the level of session.21 In all models we use as independent variables a set of dummies to identify the Selection treatments (i.e., Selection takes value 1 for the SEL and SELPS treatments and 0 otherwise) and the profit sharing incentive scheme (i.e., profit sharing takes value 1 for the BSLPS and SELPS treatments and 0 otherwise). The variable L worker takes value 1 if the worker has low ability and 0 otherwise while hired in t−1 identifies whether the worker was hired in the previous period or not. Results of the Breusch–Pagan test of independence confirm that residuals from the two equations are not independent (p < 0.000 in both cases), with a coefficient of correlation of the residuals of around 14%. Consider models 1a and 1b: the coefficient associated to the variable Selection shows the difference between the BSL and SEL treatments, as the BSL treatment is the omitted category. In model 1a, the coefficient has a negative and significant effect, while it has a positive and significant effect in model 1b, confirming our previous findings: when passing from the BSL to the SEL treatment, we observe a reduction in the effort exerted in X and an increase in the effort exerted in Y, signaling the emergence of backscratching. The coefficient associated to the variable Profit sharing represents the difference between the BSL and the BSLPS treatments, which is positive for the effort exerted in X and negative for the effort exerted in Y. In models 2a and 2b, we introduce the interaction between the Selection and the Profit sharing treatments. The coefficient of the interaction term identifies the difference in differences, that is the difference in effort exertion between the BSL and SEL treatments versus the difference in effort exertion between the BSLPS and SELPS treatments. We observe that this coefficient has a negative and significant effect on the effort exerted both in X and Y, while other effects remain unchanged. Finally, in models 3a and 3b, we introduce the interaction terms between: (i) the Selection treatment and L workers; (ii) the Profit sharing treatment and L workers, and (iii) the Selection treatment, the Profit sharing treatment, and L workers. In model 3a, the coefficients of Selection and of the interaction between Selection and Profit sharing loose significance, the first interaction displays a negative and significant effect, while the second a positive and significant effect. No significant effect is found for the third triple interaction, meaning that there is no additional effect on the exertion of effort in project x that can be explained when considering the behavior of L workers. Other results remain unchanged with respect to model 2a. When considering model 3b, we find that our main results from model 2b remain unchanged. In addition, the interaction between the Selection treatment and L workers, and the interaction between the Selection treatment, the Profit sharing treatment, and L workers display a negative and significant effect: when considering the exertion of effort in Y, we observe a difference in differences between the BSL versus SEL treatment and the BSLPS versus SELPS treatment, plus an additional effect related to the behavior of L workers. Our results based on the difference in differences analysis confirm that the profit sharing incentive is effective in limiting the emergence of backscratching by L workers (linear combination of coefficients: –0.419, p = 0.070): despite the agents still hiring a high fraction of L workers in the attempt to engage in a backscratching relationship, the selected L workers significantly reduce their reciprocal response toward them, shifting their effort exertion from activity Y to project X. Most importantly, these results indicate that reward systems do not only provide monetary incentives but they may also affect workers’ attitudes toward the organization: the small incentive provided to workers is not enough to change their behavior because of a merely pecuniary motive. 5.3 The Agents’ Intentions In this section we investigate the agents’ intentions by looking (i) at the effort suggestions to the hired worker and (ii) at their decision to refuse the effort exerted in activity Y. Moreover, we report the results of a robustness treatment, identified as control – Selection (cSEL), aimed at testing the role of the agent’s suggestion in shaping backscratching reciprocity. Result 5 summarizes our findings. Result 5. On average agents suggest exerting more effort in activity Y and less effort in project X to L workers compared with H workers. Agents who hire L workers following the instructions of the principal suggest on average higher effort in X compared with agents who hire L workers against the principal’s instructions. Agents who select L workers are more likely to accept what was produced in Y in the SEL treatment than in the BSL treatment. Support for Result 5 can be found in Tables 4 and 5. Table 4 reports, for the SEL and SELPS treatments, the average effort suggested by the agents in project X and in activity Y, depending on the ability of the selected worker. It can be noted that in the different treatments agents try to engage in a backscratching relationship in different ways: in the SEL treatment they suggest to L workers to exert more effort in Y compared with H workers; in the SELPS treatment, they suggest to L workers to exert less effort in X compared with H workers. Table 4. Effort Suggested by the Agents, Standard Deviation in Parenthesis SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 Notes: The p-values reported refer to the one-sided tests. The tests are conducted considering each session as an independent observation, that is, seven sessions in each treatment. Table 4. Effort Suggested by the Agents, Standard Deviation in Parenthesis SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 Notes: The p-values reported refer to the one-sided tests. The tests are conducted considering each session as an independent observation, that is, seven sessions in each treatment. Table 5. Effort Suggested by the Agents Depending on Principal’s Instructions (Standard Error in Parenthesis) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Notes: The complete estimation of the regression is reported in Models 3 and 4 of Table A.7 in Section A.3 of the Online Appendix A. Table 5. Effort Suggested by the Agents Depending on Principal’s Instructions (Standard Error in Parenthesis) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Notes: The complete estimation of the regression is reported in Models 3 and 4 of Table A.7 in Section A.3 of the Online Appendix A. Table 5 reports the coefficients of a Zellner’s seemingly unrelated regression analysis focusing on the agents’ suggestions and on the principals’ instruction. The dependent variables are represented by the effort suggested in X and Y. We find that the effort suggested in X is significantly higher when a worker is hired following the principal’s instruction than when he is not, and this is true also when restricting the analysis to the agents who hire L workers.22 When considering those agents who ex ante decide not to accept the value eventually produced in Y, we do not find any significant difference in the percentage of L workers hired in the SEL and in the SELPS treatments with respect to the BSL one (MW tests one-sided: z = 1.05, p = 0.145, and z = 0.657, p = 0.256, respectively). On the contrary, when considering those agents who are willing to accept what produced in Y, we find that they are more likely to select L workers in the SEL and in the SELPS treatments with respect to the BSL one (MW tests one-sided: z = 2.00, p = 0.023, respectively). Our results thus show that both the effort suggestion to workers and the possibility to ex ante refuse the value produced in Y are important features of our design in making the agents’ intentions clear: agents who select L workers act in this way to maximize their own monetary payoff. However, compared with the BSL treatment, in the SEL treatment two features of the experimental design changed: (1) the possibility for the agent not to follow the principal’s suggestion and (2) the possibility to suggest to the hired worker a desired effort level. In our robustness treatment cSEL we replicated the SEL treatment with the only exception that the agent cannot make any suggestion to the hired worker about how much effort to exert in X and Y. When considering both the agents’ hiring behavior and the workers’ effort exertion, in the cSEL treatment we do find a similar pattern of behavior with respect to the SEL treatment, even if the effects have a lower magnitude. Compared with the BSL treatment, L workers increase their effort in activity Y while reducing their effort in project X. This allows us to exclude that backscratching is merely explained by the agents’ suggestion; nonetheless, the possibility of communication magnifies it, by making agents’ intentions more salient to the L workers. See the Online Appendix A.4 for a detailed analysis. 5.4 Conflicts of Interest and Earnings In this section, we focus on how the distortions on workers’ effort and agents’ selection affect the monetary payoffs of each member of the organization and the total earnings achieved by the organization. Our main findings are summarized in Result 6. In the Online Appendix A.5, we provide further analysis. Result 6. A profit sharing compensation scheme increases the earnings of the principals in both the BSLPS and the SELPS treatments compared with the BSL and SEL treatments, respectively. It decreases the earnings of the agents in the SELPS compared with the SEL treatment, displaying its effectiveness in limiting backscratching. As regards the total earnings, the maximum amount is reached in the BSLPS treatment, while the minimum amount is obtained in the SEL treatment. Support for Result 6 is provided in Figure 3. With respect to the BSL treatment, the distortions in the hiring process and in the effort exertion in the SEL treatment result in a significant increase of the agent’s earnings and a reduction of the principal’s earnings (MW tests BSL versus SEL, one-sided: principal, z = 1.98, p = 0.03; agent, z = 2.24, p = 0.02). When comparing the SEL and the SELPS treatments we observe that, once the profit sharing is introduced, the principal’s earnings increase while the earnings of the agents decrease (MW test SEL versus SELPS: principal, z = 1.98, p = 0.03; agent z = 3.13, p = 0.00). In particular, in the SEL treatment, hiring a L worker rather than a H worker is beneficial for the agents, providing them with higher earnings, while this is not the case in the SELPS treatment (MW test one-sided H versus L: SEL z = 1.60, p = 0.06; SELPSz = 0.45, p = 0.33). The opposite effect is observed when considering the principal’s earnings (MW test one-sided H versus L: SEL z = 1.47, p = 0.07; SELPSz = 1.09, p = 0.14). Figure 3. View largeDownload slide Average Earnings of the Principal, Agent, and Worker in Each Treatment. Figure 3. View largeDownload slide Average Earnings of the Principal, Agent, and Worker in Each Treatment. When considering the total earnings of the organization, obtained as the sum of the monetary payoff of all its members, they decrease by about 2% in the SEL treatment with respect to the BSL treatment. In the SELPS treatment, total earnings increase since the profit sharing affects L workers’ behavior: they substantially increase their effort in X and reduce their effort in Y, compared with the SEL treatment. In the BSLPS treatment, all workers exert more effort than in the BSL treatment in project X, which is beneficial for the organization: the largest amount of total earning is obtained in absence of backscratching reciprocity. Note that, in our design, exerting effort in activity Y would increase total earnings compared with not exerting any effort at all. However, this does not seem to occur in our experiment where, as shown in Table 2, the treatment manipulations affected the effort allocation between X and Y rather than the total effort exerted. To get an idea of the distortion associated to backscratching, we can calculate the hypothetical earnings across all parties, that is, the total earnings obtained if all the effort would have been exerted in X. The hypothetical earnings would have been equal to 135.48 and 136.26 in the BSL and SEL treatments and to 144.96 and 137.95 in the BSLPS and SELPS treatments, respectively. By comparing the hypothetical and actual earnings across all parties we can get a measure of the loss in the total earnings of the organization, which is equal to 5.1, 8.17, 6.93, and 6.56 ECU’s in the BSL, SEL, BSLPS, and SELPS treatment, respectively. It can be noted how the losses in the total earnings are the highest in the absence of profit sharing and, in particular, in the SEL treatment, providing additional evidence of the distortion induced by backscratching. 6. Conclusion In this paper, we provide evidence that workers’ reciprocity concerns may exacerbate, rather than alleviate, conflicts of interest within a hierarchical organization. We find that agents are more likely to hire low ability workers than high ability ones, since they are more likely to exert effort in an activity benefiting the agent rather than the principal. The organizational performance is negatively affected by the emergence of hidden gaming, because of the hiring and effort distortions observed in our main treatment. We show that a tiny profit sharing compensation scheme, which may foster workers’ identification with the organization, is effective in reducing workers’ inefficient effort exertion in favor of their foremen. This result is in line with the social psychology literature (Haslam 2004) and more recent studies in economics (Akerlof and Kranton 2005, 2008) suggesting that workers’ effort also depends on how they view themselves in relation to the organization. Interestingly, sharing part of the profits with the agents is not enough to modify their selfish behavior, while extending this compensation scheme to workers significantly reduces their distorted reciprocity toward agents: other-regarding preferences seem to be more susceptible than self-regarding ones to external stimuli induced by different institutional and organizational frameworks. In our experiment, workers’ subjective entitlements are based on their different abilities: low ability workers, who do not expect to be hired, are more grateful toward their agents compared with the high ability ones. However, other individuals’ observable characteristics could possibly be used to induce the emergence of backscratching reciprocity. For example, the decision to hire a candidate identifiable as a member of a discriminated group might be used to make him/her feel (more) indebted toward the agent. Alternatively, agents may favor members of their own social group expecting that the reduced social distance strengthens their reciprocity, as in Bramoullé and Goyal (2016). Compared with the gift exchange literature, we extend the hierarchical structure of the organization by adding an intermediate level: we focus our attention on how workers’ reciprocity toward the agent affects the organization. However, in the present study, we do not allow the principals to act kindly toward the agent or toward the worker, leaving room for future research to study the effects of multiple and simultaneous exchanges of gifts among the different levels of the hierarchy. Supplementary Material Supplementary material is available at Journal of Law, Economics, & Organization online. Conflict of interest statement. None declared. We acknowledge the Max Planck Institute of Economics and the University of Manchester for funding the experiment. V.M.’s contribution to this research was also performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of Universitè de Lyon, within the program Investissements d’Avenir (ANR-11-IDEX-007) operated by the French National Research Agency (ANR). Natalia Montinari’s contribution was performed while she was a researcher at the Department of Economics of the Lund University, Sweden. We thank Maria Bigoni, David Hugh-Jones, Marco Piovesan, Pedro Rey-Biel, Marie Claire Villeval for useful comments. We are indebted with Oliver Kirchkamp for letting us use the Experimental Laboratory of the Friedrich Schiller University in Jena. We also thank participants at the 3rd Annual Behavioral and Experimental Economics Workshop in Toulouse, the workshop on Understanding employee dishonesty behaviors in the workplace in Dijon, the 8th Alhambra Experimental Workshop in Rome, the 63rd Annual Meeting of the French Economic Association (AFSE) in Lyon, the 2014 ASSET meeting in Aix-en-Provence, the workshop on Industrial Organization: Theory, Empirics and Experiments in Alberobello, the workshop Norms Actions Games in London, the workshop Incentives and Unethical Behavior Conference in San Diego for useful comments. We also thank two anonymous referees for thoughtful comments that significantly improved the paper. Appendix Proof of Proposition 1. It is straightforward to note that selfish workers exert zero effort. Consider any RORP worker i. For both t∈{H, L}, worker i is grateful to the principal, then ρi,Pt=1 and ρi,At=0 and therefore yi∗=0 and xi∗=c′(γ)−1>0. Consider any IORP worker i. Since the principal has zero payoff when x=y=0, and the agent has higher payoff then the worker for any pair (x,y), then yi∗=0 and xi∗=c′(β1−β)−1. Workers exert the same level of effort irrespective of their ability. By assumption φH>φL, hence principals, irrespective of their preferences, select workers with ability H. Proof of Proposition 2. Suppose that principals suggest to hire a H worker in equilibrium, and notice immediately that if any principal deviates and play s~P=L, the deviation is not observed by any worker and therefore has no consequence. Beliefs are correct in equilibrium and therefore H workers behave as in the baseline game: selfish H workers exert zero effort; RORP H worker i  chooses yi∗=0 and c′(γ)−1=xi∗; a IORP H worker i chooses yi∗=0 and xi∗=c′(β1−β)−1. Consider now any L worker i; if he is selfish, then he exerts zero effort; if he has RORP then ρi,AL=1 and therefore yi∗=c′(γ)−1 and xi∗=0; finally, if he has IORP preferences then he choses yi∗=0 and xi∗=c′(β1−β)−1. Finally, selfish and RORP agents choose L workers, because only low ability workers exert effort with positive probability ;  IORP agents choose to hire L workers only if c′(γ)−1−α(mA+c′(γ)−1)≤−α(mA−c′(β1−β)−1), (A.1) that is, α≥c′(γ)−1c′(β1−β)−1+c′(γ)−1. (A.2) Footnotes 1. In what follows, we use the female pronoun for the agent and the male pronoun for the principal and for the workers. 2. Some concrete examples of this behavior are provided by The Conflict of Interest Board (COIB), New York. In COIB v. Fischetti [COIB Case No. 2010-035 (2010)], the Senior Deputy Director for Infrastructure Technology in the Information Technology Division at the New York City Housing Authority (NYCHA) was fined for his multiple violations of the City’s conflicts of interest law. In particular, “he used his NYCHA subordinate, a Data Technician, to perform work on a regular basis at the restaurant without compensation. He further admitted that he caused his subordinate to use his NYCHA computer, e-mail account, and Blackberry to perform work related to the restaurant, at times the subordinate was required to be working for the City.” For further evidence look at the COIB web site at http://www.nyc.gov/html/conflicts/html/topic/position.shtml. 3. An exception is provided by the field experiment with children by Belot and Van de Ven (2011). 4. Another reason for managers to promote low-quality workers is found in the fact that incompetent managers would feel threatened by competent subordinates, and inevitably drive away competent employees (Bedeian and Armenakism 1998). In line with this reasoning, Prendergast (1993) notes that “yes men” tends to be concentrated among less able workers and among workers with less able managers. 5. Gneezy et al. (2013) have provided experimental evidence that greed, rather than reciprocity, may explain the emergence of corruption. 6. We take as reference the model by Charness and Rabin (2002). 7. Principal’s reciprocity toward the agent is not relevant because the principal cannot affect the agent’s payoff. Nonetheless, it is natural to assume that the principal is grateful to the agent who hires the worker who exerts the highest effort x. 8. Fixing a small difference between φH and φL provides a more robust test for our hypotheses: everything equals, we would expect a bigger difference in ability to produce a stronger backscratching in the SEL treatment. 9. Find the Online Appendix B at the following web site: https://sites.google.com/site/vmaggian/research. 10. Each subject participated only in one experimental session. We excluded those who previously participated in experiments featuring a gift exchange game or trust game and also those who participated in the sessions where data for Montinari et al. (2016) were collected since their experimental design was similar to our own. 11. We thank an anonymous referee for suggesting us to run this additional treatment. 12. The experimental laboratories of the Max Planck Institute of Economics and of the Friedrich Schiller University in Jena share the same recruiting system and, therefore, the same subject pool. Subjects’ invitations for the BSLPS treatment were sent to exactly the same subject pool used in the other treatments, also maintaining the same criteria for the eligibility. 13. Result from Part 1 is qualitatively identical to those of Part 2 and they are reported in the Online Appendix A, Section 6. Since 5/54, 8/53, 9/34, and 10/54 L workers are hired in the BSL, SEL, BSLPS, and SELPS treatments, respectively, the analysis on effort has very little statistical power and would not lead to any additional benefit to the reader. Find the Online Appendix A at the following web site: https://sites.google.com/site/vmaggian/research. 14. When performing the Mann–Whitney U-test we average the data within a session and treat each session as a single observation. While conservative, this approach makes the power of all tests low and not robust to outliers. To account for these limitations we use the rank-order statistics Somers’ D (provided by the “somersd” package in Stata) that looks at the individuals’ choices accounting for the presence of clusters at the level of sessions in the data. 15. In Montinari et al. (2016) about 30% of employers hire low ability workers. 16. All statistical tests reported are two samples and two-sided unless differently specified. When we have clear, explicit, and theory-driven predictions about agents’ and worker’ behavior in the different treatments we used one-sided tests. 17. See the Online Appendix A.1.2. for additional analyses on the dynamics of part 2. 18. In both BSL and BSLPS treatments, H workers exert less effort than L workers in project X (2.54 versus 2.67, in the BSL and 3.46 versus 3.98 in the BSLPS), but this difference fails to reach significance. 19. We fixed a minimum effort for project X equal to 1 and this may have created an anchoring effect, inducing some workers to exert a positive effort in Y too. 20. A worker aimed at equalizing payoffs of all members of the organization should exert xt = 5 and yt = 0. In our data, a one-sample WSR test rejects the null hypothesis that the average effort exerted by the hired worker in project X is equal to five in each treatment, p < 0.02 in all cases. 21. For the estimations, we used the “sureg” command in Stata. For the standard errors, we used the “bootstrap” command in Stata, which executes a nonparametric bootstrap of the statistics in the list of the explanatory variables by resampling observations (with replacement) from the data in memory 1000 times. Our results do not change if we cluster the standard errors at the subjects level. 22. The whole regression and a more complete analysis of the determinants of the effort suggestion are provided in the Online Appendix A.3. References Abbink Klaus. 2004 . “ Staff Rotation as an Anti-corruption Policy: An Experimental Study,” 20 European Journal of Political Economy 887 – 906 . Google Scholar CrossRef Search ADS Abbink Klaus , Irlenbusch Bernd , Renner Elke . 2002 . “ An Experimental Bribery Game ,” 18 Journal of Law, Economics, and Organization 428 – 54 . Google Scholar CrossRef Search ADS Akerlof George A. , Kranton Rachel E. . 2005 . “ Identity and the Economics of Organizations ,” 19 Journal of economic perspectives 9 – 32 . Google Scholar CrossRef Search ADS Akerlof George A. , Kranton Rachel E. 2008 . “ Identity, Supervision, and Work Groups ,” 98 American Economic Review 212 – 7 . Google Scholar CrossRef Search ADS Bac Mehmet. 1996 . “ Corruption and Supervision Costs in Hierarchies ,” 22 Journal of Comparative Economics 99 – 118 . Google Scholar CrossRef Search ADS Bandiera Oriana , Barankay Iwan , Rasul Imran . 2009 . “ Social Connections and Incentives in the Workplace: Evidence from Personnel Data ,” 77 Econometrica 1047 – 94 . Google Scholar CrossRef Search ADS Baron James N. 2013 . “ Empathy Wages? Gratitude and Gift Exchange in Employment Relationships ,” 33 Research in Organizational Behavior 113 – 34 . Google Scholar CrossRef Search ADS Barr Abigail , Serra Danila . 2009 . “ The Effects of Externalities and Framing on Bribery in a Petty Corruption Experiment,” 12 Experimental Economics 488 – 503 . Google Scholar CrossRef Search ADS Bedeian Arthur G. , Armenakism Achilles A. . 1998 . “ The Cesspool Syndrome: How Dreck Floats to the Top of Declining Organizations ,” 12 Academy of Management Executive 58 – 67 . Bellemare Charles , Shearer Bruce . 2009 . “ Gift Giving and Worker Productivity: Evidence from a Firm-Level Experiment,” 67 Games and Economic Behavior 233 – 44 . Google Scholar CrossRef Search ADS Belot Michle , Van de Ven Jeroen . 2011 . “ Friendships and Favouritism on the Schoolground—A Framed Field Experiment ,” 121 Economic Journal 1228 – 51 . Google Scholar CrossRef Search ADS Bramoullé Yann , Goyal Sanjeev . 2016 . “ Favoritism ,” 122 Journal of Development Economics 16 – 27 . Google Scholar CrossRef Search ADS Chang Juin Jen , Lai Ching Chong . 2002 . “ Is the Efficiency Wage Efficient? The Social Norm and Organizational Corruption ,” 104 Scandinavian Journal of Economics 27 – 47 . Google Scholar CrossRef Search ADS Charness Gary. 2004 . “ Attribution and Reciprocity in an Experimental Labor Market,” 22 Journal of Labor Economics 665 – 88 . Google Scholar CrossRef Search ADS Charness Gary , Rabin Matthew . 2002 . “ Understanding Social Preferences with Simple Tests,” 117 Quarterly Journal of Economics 817 – 69 . Google Scholar CrossRef Search ADS Cialdini Robert. 1996 . “Social Infuence and the Triple Tumor Structure of Organizational Dishonesty,” in Messick D.M. , Tenbrunsel A.E. eds., Codes of Conduct: Behavioral Research into Business Ethics, 44 – 58 . New York, NY : Russell Sage . Coyle-Shapiro Jacqueline A.-M. , Morrow Paula C. , Richardson Ray , Dunn Stephen R. . 2002 . “ Using Profit Sharing to Enhance Employee Attitudes: A Longitudinal Examination of the Effects on Trust and Commitment ,” 41 Human Resource Management 423 – 39 . Google Scholar CrossRef Search ADS Diekmann Andreas , Przepiorka Wojtek , Rauhut Heiko . 2015 . “ Lifting the Veil of Ignorance: An Experiment on the Contagiousness of Norm Violations ,” 27 Rationality and Society 309 – 33 . Google Scholar CrossRef Search ADS Edwards Richard. 1979 . Contested Terrain: The Transformation of the Workplace in the Twentieth Century . New York, NY : Basic Books . Englmaier Florian , Leider Steve . 2012 . “Managerial Payoff and Gift Exchange in the Field.” Cesifo Working Paper Series No. 3707. Eriksson Tor , Villeval Marie Claire . 2012 . “ Respect and Relational Contracts ,” 81 Journal of Economic Behavior and Organization 286 – 98 . Google Scholar CrossRef Search ADS Fehr Ernst , Falk Armin . 1999 . “ Wage Rigidity in a Competitive Incomplete Contract Market,” 107 Journal of Political Economy 106 – 34 . Google Scholar CrossRef Search ADS Fehr Ernst , Schmidt Klaus M. . 1999 . “ A Theory of Fairness, Competition, and Cooperation ,” 114 Quarterly Journal of Economics 817 – 68 . Fehr Ernst , Kirchsteiger Georg , Riedl Arno . 1993 . “ Does Fairness Prevent Market Clearing? An Experimental Investigation ,” 108 Quarterly Journal of Economics 437 – 59 . Google Scholar CrossRef Search ADS Fehr Ernst , Gachter Simon , Kirchsteiger Georg . 1997 . “ Reciprocity as a Contract Enforcement Device: Experimental Evidence ,” 65 Econometrica 833 – 60 . Google Scholar CrossRef Search ADS Fehr Ernst , Kirchsteiger Georg , Riedl Arno . 1998a . “ Gift Exchange and Reciprocity in Competitive Experimental Markets ,” 42 European Economic Review 1 – 34 . Google Scholar CrossRef Search ADS Fehr Ernst , Kirchler Erich , Weichbold Andreas , Simon Gachter . 1998b . “ When Social Norms Overpower Competition: Gift Exchange in Experimental Labor Markets ,” 16 Journal of Labor Economics 324 – 51 . Google Scholar CrossRef Search ADS Fischbacher Urs. 2007 . “ z-Tree: Zurich Toolbox for Ready-Made Economic Experiments ,” 10 Experimental Economics 171 – 8 . Google Scholar CrossRef Search ADS Gino Francesca , Shahar Ayal , Ariely Dan . 2009 . “ Contagion and Differentiation in Unethical Behavior: The Effect of One Bad Apple on the Barrel,” 20 Psychological Science 393 – 8 . Google Scholar CrossRef Search ADS Gneezy Uri , List John A. . 2006 . “ Putting Behavioral Economics to Work: Testing for Gift Exchange in Labor Markets Using Field Experiments ,” 74 Econometrica 1365 – 84 . Google Scholar CrossRef Search ADS Gneezy Uri , Saccardo Silvia , van Veldhuizen Roel . 2013 . “Bribery: Greed versus Reciprocity.” U. C. San Diego Working Papers. Greiner Ben. 2004 . “ An Online Recruitment System for Economic Experiments ,” 63 Forschung und Wissenschaftliches Rechnen 2003. Gwdg Bericht 79 – 93 . Haslam S. Alexander 2004 . Psychology in Organizations . Thousand Oaks, CA : Sage Publications . Hennig-Schmidt Heike , Sadrieh Abdolkarim , Rockenbach Bettina . 2010 . “ In Search of Workers’ Real Effort Reciprocity—A Field and a Laboratory Experiment,” 8 Journal of the European Economic Association 817 – 37 . Heywood John S. , Jirjahn Uwe , Tsertsvadze Georgi . 2005 . “ Getting along with Colleagues—Does Profit Sharing Help or Hurt?,” 58 Kyklos 557 – 73 . Google Scholar CrossRef Search ADS Keizer Kees , Lindenberg Siegwart , Steg Linda 2008 . “ The Spreading of Disorder,” 322 Science 1681 – 5 . Google Scholar CrossRef Search ADS Kerr Jeffrey , Slocum John W. . 1987 . “ Managing Corporate Culture through Reward Systems ,” 1 Academy of Management Executive 99 – 107 . Khatri Naresh , Tsang Eric W. K. . 2003 . “ Antecedents and Consequences of Cronyism in Organizations ,” 43 Journal of Business Ethics 289 – 303 . Google Scholar CrossRef Search ADS Knez Marc , Simester Duncan . 2001 . “ Firm Wide Incentives and Mutual Monitoring at Continental Airlines ,” 19 Journal of Labor Economics 743 – 72 . Google Scholar CrossRef Search ADS Kolm Serge-Christophe , ed. 2006 . “Reciprocity: Its Scope, Rationales, and Consequences,” in Handbook of the Economics of Giving, Altruism and Reciprocity, vol. 1, 371 – 541 . North Holland : Imprint . Kramarz Francis , Thesmar David . 2013 . “ Social Networks in the Boardroom,” 11 Journal of the European Economic Association 780 – 807 . Google Scholar CrossRef Search ADS Kube Sebastian , Maréchal Michel Andr , Puppe Clemens . 2013 . “ Do Wage Cuts Damage Work Morale? Evidence from a Natural Field Experiment ,” 11 Journal of the European Economic Association 853 – 70 . Google Scholar CrossRef Search ADS Laffont Jean-Jacques. 1988 . “ Hidden Gaming in Hierarchies: Facts and Model ,” 64 Economic Record 295 – 306 . Google Scholar CrossRef Search ADS Laffont Jean-Jacques. 1990 . “ Analysis of Hidden Gaming in a Three-Level Hierarchy,” 6 Journal of Law, Economics and Organization 301 – 24 . Levine David K. , Weinschelbaum Federico , Zurita Felipe . 2010 . “ The Brother-in-Law Effect ,” 51 International Economic Review 497 – 507 . Google Scholar CrossRef Search ADS Malmendier Ulrike , Klaus Schmidt . 2017 . “ You Owe Me ,” 107 American Economic Review 493 – 526 . Google Scholar CrossRef Search ADS Montinari Natalia , Nicolò Antonio , Oexl Regine . 2016 . “ The Gift of Being Chosen,” 19 Experimental Economics 460 – 79 . Google Scholar CrossRef Search ADS Newson Roger. 2002 . “ Parameters behind ‘Nonparametric’ Statistics: Kendall’s Tau, Somers’ d and Median Differences ,” 2 Stata Journal 45 – 64 . Osterman Paul. 1994 . “ How Common Is Workplace-Transformation and Who Adopts It ,” 47 Industrial and Labor Relations Review 173 – 88 . Google Scholar CrossRef Search ADS Pearce Jone L. , Branyiczki Imre , Bigley Gregory A. . 2000 . “ Insufficient Bureaucracy: Trust and Commitment in Particularistic Organizations,” 11 Organization Science 148 – 62 . Google Scholar CrossRef Search ADS Prendergast Canice. 1993 . “ A Theory of ‘Yes Men’,” 83 American Economic Review 757 – 70 . Prendergast Canice , Topel Robert H. . 1993 . “Favoritism in Organizations.” National Bureau of Economic Research No. w4427. Robin Stphane , Rusinowska Agnieszka , Villeval Marie Claire . 2014 . “ Ingratiation: Experimental Evidence ,” 66 European Economic Review 16 – 38 . Google Scholar CrossRef Search ADS Schlicht Ekkehart. 1998 . On Custom in the Economy . Oxford : Clarendon Press . Google Scholar CrossRef Search ADS Thiele Veikko. 2013 . “ Subjective Performance Evaluations, Collusion, and Organizational Design ,” 29 Journal of Law, Economics, and Organization 35 – 59 . Google Scholar CrossRef Search ADS Tirole Jean. 1986 . “ Hierarchies and Bureaucracies: On the Role of Collusion in Organizations,” 2 Journal of Law, Economics, and Organization 181 – 214 © The Author(s) 2018. Published by Oxford University Press on behalf of Yale University. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Journal of Law, Economics, and Organization Oxford University Press

Backscratching in Hierarchical Organizations

Loading next page...
 
/lp/ou_press/backscratching-in-hierarchical-organizations-iUcxAF0OuI
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of Yale University. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
8756-6222
eISSN
1465-7341
D.O.I.
10.1093/jleo/ewy006
Publisher site
See Article on Publisher Site

Abstract

Abstract We experimentally investigate the role of reciprocity in sustaining the emergence of implicit collusive agreements in hierarchical organizations. We show that when an agent hires, on behalf of the principal, one worker out of two candidates: (i) low ability workers, being less entitled to be selected, are more likely to exert effort in a task that is exclusively beneficial to the agent; (ii) as a consequence, agents distort the hiring process in favor of low ability workers; and (iii) sharing a small part of the organization’s profits with the workers alleviates their effort distortion 1. Introduction The possibility that managers and bureaucrats buy the cooperation and the loyalty of their subordinates at the expense of the organization characterizes many hierarchical organizations (Edwards 1979). Employees who are promoted or hired based on their loyalty toward their supervisors, and not on their merit, are more likely to invest their time and attention to maintaining good relations with their boss, rather than improving their job performance (Pearce et al. 2000). Tirole (1986) highlights how, in a three-tier principal–supervisor–agent model, collusion between the supervisor and the agent can be sustained by a norm of reciprocity. Laffont (1988, 1990) generalizes the notion of moral hazard to include hidden gaming, defined as the “ability that some players may have to design and play games with other members of the hierarchy by which they benefit from others while they are not observable by the principal” (Laffont 1990: 302). In this paper, we provide the first experimental evidence that reciprocity sustains hidden gaming in hierarchical organizations. First, we show the emergence of backscratching between members at the lowest and at the intermediate level of a three-tier organization, at the expense of the principal. Agents, who select workers for a job on behalf of the principal, are more likely to hire candidates with lower ability. These candidates feel less entitled for the position and thus they are more prone to devote their effort in favor of the agent who selected them. Second, we design a compensation scheme which, without increasing the cost of the principal, limits the detrimental effect of backscratching reciprocity. In our experiment, we render a three-tier organization formed by one principal, one agent, and one worker. The worker has to be chosen from a pool of two candidates who differ in their ability.1 The principal, in contrast to the agent, is not able to distinguish the candidates’ ability and, therefore, delegates the hiring decision to the latter. Once employed, the worker receives a fixed wage and chooses a level of non-verifiable effort that can be exerted both in project X, which is beneficial for the principal and the agent, and in activity Y, that provides a private benefit solely to the agent. The joint payoff of the principal and the agent is maximized when all the effort of the worker is devoted to project X; however, the agent’s payoff is maximized when the worker’s effort is exerted in activity Y. Candidates differ in their abilities: for each level of effort exerted in project X the high ability candidate is more productive than the low ability one. When exerting effort in activity Y, the two candidates are equally productive. While it is public information that candidates have different abilities, only the agent is able to distinguish among them and this is precisely the reason why she is hired by the principal. Our aim is to capture a situation where the agent can exploit her position in order to induce subordinates to do certain activities that go beyond their formal job descriptions but give to her a personal benefit.2 In the Baseline treatment, the principal privately gives instructions to the agent about which candidate to hire: either the high or the low ability one. The agent selects one candidate and the hired worker chooses how much effort to exert in project X and in activity Y. The agent has to follow the principal’s instructions and thus cannot take any decision. In the Selection treatment, the agent is not forced to follow the instructions received by the principal. Moreover, the agent has the opportunity to privately suggest to the hired worker a level of effort to exert both in X and Y. Communication renders the agent’s intentions clear: the agent’s decision to select one of the two candidates may be aimed either at increasing the profit of the organization as a whole or at getting personal benefits from the employment relationship. Compared with the Baseline treatment, we observe a significant increase in the number of low ability candidates hired in the Selection treatment. Moreover, we provide evidence that such a hiring distortion is due to the fact that low ability workers exert more effort in activity Y and less effort in project X than the high ability ones. Agents do strategically exploit the reciprocal concerns of low ability workers who feel less entitled to get the job and thus are more grateful to the agent for being selected. We then examine whether distributing a small fraction of the profits to the workers is effective in limiting the emergence of backscratching reciprocity. The Baseline – Profit sharing and Selection – Profit sharing treatments replicate the design implemented in the corresponding Baseline and Selection treatments with the single difference that the worker receives a small fraction of the value generated in project X. In the Selection – Profit sharing treatment, we find that distributing a small share of the value of project X to the workers is effective in reducing the low ability workers’ effort distortion in favor of agents with respect to the Selection treatment. Moreover, when comparing the Baseline – Profit sharing to the Baseline treatment we find that workers, irrespectively from their ability, exert significantly higher effort in project X. Given that the share of the value of project X distributed to the worker is very small, we suggest that such a result is not simply driven by the monetary motive associated with the profit sharing payoff structure. It seems rather due to a positive effect on the worker’s attitude toward the firm: even the presence of a tiny link between payment and effort is shown to be relevant in shaping the way workers perceive the organization. 2. Related Literature There is a vast experimental literature on gift exchange games, initiated by Fehr et al. (1993) and followed by many applications both in the laboratory (Fehr et al. 1997, 1998a, 1998b; Fehr and Falk 1999; Charness 2004; Eriksson and Villeval 2012) and in the field (Gneezy and List 2006; Bellemare and Shearer 2009; Hennig-Schmidt et al. 2010; Kube et al. 2013). In the field, the effectiveness of gift exchange seems to depend on the details of the environment and, in particular, it could be enhanced when targeted toward reciprocal employees (Englmaier and Leider 2012). These studies all focus on two-tier settings and test the positive effect of reciprocity in limiting opportunistic behaviors of workers. The novelty of our approach consists in testing the gift exchange hypothesis in three-tier organizations. By mean of an experiment we show that reciprocity may damage hierarchical organizations, when their members use it as an enforcement device to acquire personal illegitimate benefits. A second branch of literature related to our paper refers to the distortions in the hiring process due to favoritism within organizations. Several studies have pointed out that organizations’ performance is usually negatively affected when candidates’ evaluation is not based on their ability (Levine et al. 2010; Kramarz and Thesmar 2013). This can happen when candidates are hired or promoted on the basis of subjective rather than objective criteria.3 Managers may indeed favor workers according to their social connections (Bandiera et al. 2009) and personal preferences when objective evaluations of workers’ performance are not available (Prendergast and Topel 1993), or may favor those who engage in ingratiatory behavior regardless of their objective ability (Robin et al. 2014). With respect to these studies, we look to an additional motivation of the distortion of the hiring process: the attempt to induce reciprocal behavior in less entitled workers by favoring them in the selection, in the hope of receiving future benefits.4 The dark side of reciprocity has already been analyzed by scholars in other disciplines. Studies in Social Psychology underline how recipients of (unsolicited) gifts feel indebted toward the gift givers and are more likely to “return the favor,” once requested. These studies show how individuals can trigger reciprocity in order to gain an unfair advantage (Cialdini 1996). In Organizational Science backscratching is identified as (vertical) cronyism and indicates a favoritism of the superior toward subordinates (as, e.g., the assignment of promotion, bonus, pay rise, or better job) based on criteria different than merit in exchange for the latter’s personal loyalty (Khatri and Tsang 2003). Reciprocity has also been shown to be a key feature in sustaining corruption agreements, which cannot be enforced by third parties (Abbink et al. 2002; Abbink 2004; Barr and Serra 2009).5 We are not aware of any experimental study looking at the emergence of implicit collusion sustained by reciprocity in hierarchical organizations. In a recent study, Malmendier and Klaus (2017) analyze the emergence of a dark side of reciprocity in a different context, a client–producer relationship. They find that when a decision maker has to buy a product on behalf of a client, and two producers compete to sell the product, the possibility of one producer sending a small gift to the decision maker increases the probability that the recipient chooses the gift-giver’s product, even if favoring the gift-giver will damage the client. Compared with Malmendier and Klaus (2017), there are two main differences in our design. First, in our experiment workers hired by agents can reciprocate toward them without damaging their principal, and therefore there is not necessarily a tension between reciprocating a gift and fulfilling a duty. Second, our design allows us to investigate the role of subjective entitlement within organizations. Following Schlicht (1998: 24), we define entitlements as “subjectively perceived rights that go along with a motivational disposition to defend them”. In our setting, the low ability candidate is less entitled to be selected for the job compared with the high ability one. The relevance of subjective entitlement has been displayed by recent papers showing that gifts offered by employers to workers who belong to relatively disadvantaged groups and/or to the lower part of the performance distribution are likely to elicit more gratitude (Baron 2013; Montinari et al. 2016). Our findings are in line with Kolm (2006) evidencing how the beneficiaries of (unsolicited) gift tend to feel “morally indebted” toward the gift giver. Moreover, our contribution highlights the crucial role of subjective entitlement in activating this feeling. A third branch of literature related to our work analyzes how to reduce the negative effects of internal collusion. Bac (1996) studies how hierarchical structures affect the level of corruption in an organization; Thiele (2013) suggests decreasing the incentives for employees while increasing managers’ compensations. Chang and Lai (2002) investigated the impact of social norms on supervisors’ corrupt behavior showing that, when in the presence of corruption, paying supervisors more than workers limits workers’ slack. We show that reward systems do not only provide monetary incentives but also affect workers’ attitudes toward the corporate culture, ultimately affecting their productivity. Using a profit sharing compensation scheme to increase the organizational performance, via positive changes in employees’ attitude, has already been suggested (Osterman 1994; Knez and Simester 2001; Coyle-Shapiro et al. 2002; Heywood et al. 2005) and is a well-known phenomenon in the economics literature (Kerr and Slocum 1987). 3. The Model The model analyzes a three-tier organization, formed by a Principal (P), an Agent (A), and a worker. The worker is selected from a pool of two candidates, a low ability worker (L worker) and a high ability worker (H worker). The principal needs a worker to carry out a project but, being unable to distinguish between the ability of the two candidates, he hires an informed agent and pays her a fixed compensation to select one of the two candidates. Once selected, the worker receives a fixed compensation and chooses how much effort x≥0 to exert in the project valuable to the principal and how much effort y≥0 in a different activity that benefits the agent, with the overall effort resulting from the sum of the two effort components: e=x+y. The (monetary) cost of effort c(e) is a differentiable, strictly increasing and convex function in the overall amount of effort e exerted by the worker, with c(0)=0. The principal’s monetary payoff is equal to πP=(x+φt), where φt is the ability of worker t∈{H, L}, with φH>φL≥0. The (hired) worker’s monetary payoff is equal to πt=mW−c(e), where mW > 0 is the compensation paid to the worker which does not depend on his ability. The unemployed candidate receives a monetary payoff equal to zero. The agent’s monetary payoff is equal to πA=mA+y, with mA > 0. We assume that mA > mW which implies that if a worker with ability t∈{H, L} exerts a total amount of effort equal to zero, then πA>πH=πL>πP. We assume that individuals may exhibit other regarding preferences (ORPs) and therefore their utility may not coincide with their monetary payoff.6 Specifically, the utility function of individual i∈{P, A, H, L}, who belongs to the organization, is given by: Ui=πi+I(∑j≠i(θiρi,jγπj−(1−θi)(αmax {(πj−πi),0}+βmax {(πi−πj),0}))). (1)   I∈{0,1} is an index function that takes value 0 if individual i is selfish and 1 if individual i exhibits ORPs. Individual i’s ORP toward individual j≠i are a linear combination of two components, a reciprocity component, and an inequity aversion component; θi is the relative weight of the reciprocity component; and 1−θi of the inequity aversion. The reciprocal attitude toward individual j is described by the parameter ρi,j that takes value 1 if individual j has been kind with individual i and 0 otherwise, and the parameter γ∈(0,1), the weight of individual js monetary payoff in agent is utility, which is assumed to be the same for every R-ORP individual. Inequity aversion is modeled according to the standard Fehr and Schmidt (1999) model with, β≤α and 0≤β<1. Individuals who exhibit ORP are heterogeneous with respect to the weight assigned to the reciprocity component. For sake of simplicity, we assume that θi∈{0,1}; we then classify individuals with respect to their preferences in: selfish (SP), inequity averse (I-ORP), and reciprocity concerned (R-ORP). We assume that workers’ preferences are not correlated with their ability and that every player is SP, I-ORP, and R-ORP, respectively, with positive probability. We consider two different games. In both games the principal plays an action sP∈{H, L} that we interpret as the suggestion given to the agent on which worker to hire. In the baseline game the suggestion is binding, the agent is therefore a fictitious player who has no actions to play, and the hired worker t∈{H, L} chooses how much efforts x and y to exert. In the selection game the suggestion is not binding, the agent chooses which worker t∈{H, L} to hire, and the hired worked chooses the pair (x, y) of efforts to exert. Players’ reciprocity depend on their beliefs about the other players’ actions. R-ORP worker i∈{H, L} shows reciprocity concerns toward either the principal or the agent, depending on his beliefs s^iP about which type of worker the principal has suggested to hire. Specifically, we assume that ρi,P=s^iP. Similarly, an R-ORP worker believes that the agent was kind with him if he was hired against the principal’s suggestion, that is, ρi,A=1−s^iP. An R-ORP agent does not feel any reciprocity toward the principal, while she is grateful to the type of worker who will provide her with the largest monetary payoff. Finally, principal’s reciprocity toward the worker depends on his beliefs about the amount of effort x^t that a worker of ability t∈{H, L} will exert. We assume that ρP,H=1 if x^H>x^L and zero otherwise, and ρP,L=1 if x^L>x^H and zero otherwise, that is the principal believes that a worker of ability t is kind with him if he exerts strictly more effort than the other worker τ∈{H, L}, with τ≠t.7 We solve both games by backward induction. Consider first the baseline game. Starting from the last stage, selfish workers exert zero effort both in x and y. Whatever worker i∈{H, L} is hired, ρi,P=1, since the principal’s suggestion is binding. Given that, by design, the principal is the player with lowest monetary payoff (at least when x=y=0), in equilibrium both I-ORP and R-ORP workers exert a positive amount of effort x and zero effort in y, and the amount of effort exerted by H and L workers, on average, is the same. Principals, irrespective of their preferences, hire H workers because they exert the same effort than L workers but they have higher ability. The following proposition summarizes these findings. Proposition 1. In the baseline game, selfish workers exert zero effort in x and y. Workers with ORP exert a positive amount of effort x and zero effort y. On average, workers of different abilities exert the same amount of efforts. Principals hire H workers. Proof. See the Appendix.  Let xtBSL ( xtSEL) and ytBSL ( ytSEL) denote the average effort exerted in equilibrium of the baseline (selection) game by a worker of ability t∈{H, L}. From Proposition 1, we derive the following theoretical prediction about workers’ behavior: Prediction 1. In the baseline game, on average, H and L workers behave similarly, exerting a positive level of effort in x and zero in y: xHBSL=xLBSL>0, yHBSL=yLBSL=0. Consider now the selection game. This game shows a multiplicity of equilibria depending on the principal’s suggestion. We focus on the equilibrium in which principals suggest to hire H workers, as in the baseline game, also because it is the unique equilibrium that exists for any possible probability distribution over workers’ preferences (in fact, it is easy to check that this is the only equilibrium if all workers are selfish). Starting from the last stage, selfish workers exert zero effort. Consider R-ORP workers. In equilibrium beliefs are correct so ρH,P=1 and ρH,A=0, while ρL,P=0 and ρL,A=1. Therefore, H workers with R-ORP exert effort x while L workers with R-ORP exert effort y. Consider now I-ORP workers. Since the worker has always a lower monetary payoff than the agent (even when x=y=0) and the principal has a lower monetary payoff than workers if they choose x=y=0, it follows that in equilibrium  I-ORP workers, irrespective of their type t∈{H, L}, exert zero effort y and a positive amount of effort x. The following proposition summarizes these findings. Proposition 2. There exists an equilibrium of the selection game in which principals suggest to hire H workers. In this equilibrium, selfish workers exert zero effort. Inequity-averse workers exert a positive amount of effort x and zero effort y, irrespective of their ability. Reciprocity-concerned workers of ability H exert positive effort x and zero y, while reciprocity-concerned workers of ability L exert zero effort x and a positive effort y. Selfish and reciprocity-concerned agents hire L workers, while inequity-averse agents hire H workers if α≥c′(γ)−1c′(β1−β)−1+c′(γ)−1, and L workers otherwise. Proof. See the Appendix. From Proposition 2, we derive the following theoretical prediction about workers’ behavior: Prediction 2. In the selection game, on average, H workers behave as H (and also L workers) in the baseline game, exerting a positive level of effort in x and zero in y: xHBSL=xHSEL=xLBSL>0, yHBSL=yLBSL=yHSEL=0; L workers exert on average a lower level of effort x and a higher level of effort y than H workers: yLSEL>0 and xLSEL<xLBSL. Finally, the last theoretical prediction refers to the hiring of workers. Prediction 3. The fraction of low ability workers hired in the selection game is larger than the fraction of low ability workers hired in the baseline game. Up to now we have assumed, as is customary in economics, that preferences are exogenously determined by nature. According to our model, agents with ORP preferences are either inequity averse or reciprocal. However, which component of the ORP preferences, reciprocity or inequity aversion, turns out to be more prominent for an individual could also be influenced by the environment. To investigate, in an anonymous laboratory setting, whether the features of the working environment affect which one of these two potentially conflicting preferences prevails, we consider a modification of the workers’ payment. In particular, we introduce a tiny profit sharing compensation scheme that assigns an additional small fraction of the principal’s profits to the worker. Let ηw denote the fraction of the principal’s profits distributed to the worker. To avoid the monetary incentives could directly influence the worker’s effort x, we assume that ηw<c′(0). Nevertheless, we conjecture that ORPs could be influenced by the type of contract offered to the worker. Namely, a profit sharing contract could render the principal’s payoff salient to the workers, and therefore could induce all workers with ORP preference to be inequity averse. As a consequence, any differences between the two games, baseline and selection, should disappear when a profit sharing contract is offered to workers. Conjecture 1 . If a profit sharing contract renders the principal’s payoff salient to the workers, then reciprocity-concerned workers of ability H and L exert positive effort x and zero y both in the (profit sharing) baseline and selection games. In particular, workers with ability L should behave the same in the baseline and in the selection game in the presence of a profit sharing contract. Before concluding this section it is worth emphasizing that our simple model predicts that every worker never simultaneously exerts both efforts, x and y. It is easy to generalize the model to admit that some workers could choose to exert positive effort x and y. Suppose that θ has a continuous probability distribution with support [0,1] and Ui=πi+I(∑j≠iθiρi,jγπj+(1−θi)β(|πj−πi|)α), with 0<γ≤1, β<1, and α>1 (to avoid both components to be linear). It is easy to check that in the baseline game all ORP workers exert positive effort x and zero y, because both reciprocity and inequity aversion are addressed toward the principal’s payoff. In the selection game (focusing on the equilibrium in which principals suggest to hire H workers) H workers, irrespective of their θ, still exert zero effort y and positive x, while a L worker i with θi∈(0,1) will exert a positive amount of effort x, due to his inequity concern toward the principal, but also a positive amount of effort y, to reciprocate the agent’s kindness. Still, the qualitative results of the simple version of the model with θ∈{0,1} remain unchanged, and, importantly, the above predictions still hold. 4. Experimental Design and Parameters In our experiment we implement a between subjects analysis in a 2× 2 design, where we vary either the payoff structure of the game and whether the principal or the agent is making the hiring decision, mirroring the games discussed in the theoretical model, with few differences. In the laboratory, we render an organization composed by a principal (P), an agent (A), and a worker, either of low ability (L worker) or high ability (H worker). The principal hires an informed agent and pays her to select one of the two candidates. The principal gives instructions to the agent about which type of worker t∈{H, L} to hire, either H or L worker, and the agent selects the worker. The selected worker receives a fixed compensation by the principal and chooses how much effort to exert in project X and in activity Y. After the worker’s decision, payoffs are determined and the game ends. The non-selected candidate receives an unemployment benefit. In the baseline treatment (BSL), participants play the game just described and the principal gives binding instructions to the agent about which worker to hire: either L or H worker. In the selection treatment (SEL), players’ payoffs are the same as in the BSL treatment, but the agent (i) chooses whether to hire worker L or worker H, since principal’s suggestion is not binding anymore and (ii) may suggest to the hired worker a desired effort level to exert in X and Y. The baseline profit sharing (BSLPS) and selection profit sharing (SELPS) treatments are identical to the BSL and SEL treatments, respectively, with the only exception that we introduce a tiny profit sharing compensation for the workers. Panel A of Table 1 summarizes the values assigned to the parameters in each treatment, with 10 ECUs = 1 Euro implemented as exchange rate. Table 1. Experimental Parameters, Cost Function, and Material Payoff Functions Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Notes: In Panels B and C, the values are obtained considering a realization of σ=E(σ)=1. Panels B and C report 7.5% (15%) of the output, that is, the value produced in project X. Table 1. Experimental Parameters, Cost Function, and Material Payoff Functions Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel A: Experimental parameters Parameter Value Treatment x ∈{1,2…,10} All y ∈{0,1…,5} All x + y ∈{1,2,…,10} All σ ∼u[0.8,1.2] All φH 0.5 All φL 0 All δ 0.4 All mA=mW 50 All B 100 All ηP 0.85 All ηA 0.15 BSL, SEL 0.075 BSLPS, SELPS ηW 0 BSL, SEL 0.075 BSLPS, SELPS Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel B: Cost of total effort exerted in X and Y and material payoffs of the worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 All c(x+y) 0 1 2 4 6 8 10 12 15 18 All mW−c(x+y) 50 49 48 46 44 42 40 38 35 32 Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 H Worker BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 L Worker BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Panel C: Material payoffs of the principal and the agent depending on the effort choices of the hired worker Effort in x+y Treatment 1 2 3 4 5 6 7 8 9 10 Principal All ηP(x+φH)σ 12.8 21.3 29.8 38.3 46.8 55.3 63.8 72.3 80.8 89.3 All ηP(x+φL)σ 8.5 17.0 25.5 34.0 42.5 51.0 59.5 68.0 76.5 85.0 Agent All mA 50 50 50 50 50 50 50 50 50 50 BSL, SEL ηA(x+φH)σ 2.3 3.8 5.3 6.8 8.3 9.8 11.3 12.8 14.3 15.8 BSL, SEL ηA(x+φL)σ 1.5 3.0 4.5 6.0 7.5 9.0 10.5 12.0 13.5 15.0 BSLPS, SELPS ηA(x+φH)σ 1.1 1.9 2.6 3.4 4.1 4.9 5.6 6.4 7.1 7.9 BSLPS, SELPS ηA(x+φL)σ 0.8 1.5 2.3 3.0 3.8 4.5 5.3 6.0 6.8 7.5 Effort in y Treatment 0 1 2 3 4 5 Agent All δy 0 4.3 8.5 12.8 17.0 21.3 Notes: In Panels B and C, the values are obtained considering a realization of σ=E(σ)=1. Panels B and C report 7.5% (15%) of the output, that is, the value produced in project X. In the experiment, we choose ability φH=0.5>φL=08; the levels of effort exerted in X and Y are integer numbers between {1,2,…,10} and {0,1,…,5}, respectively, and their sum has to be greater than one and less than or equal to 10. Effort in project X generates an output X=(x+φt)σ which is affected by a random variable, denoted by σ, uniformly distributed on the interval [0.8,1.2] with mean equal to 1. The random variable together with the small difference in ability makes it harder for the principal to infer the ability of the worker hired by the agent, since he only receives information about the value produced in X, which is a noisy signal of both the worker’s effort and ability. Effort in activity Y generates an output Y=δy, with δ=0.4, so that the workers’ ability does not affect their performance in activity Y. Panel B of Table 1 reports the worker i’s payoff function: πi=mW−c(e). mW is the fixed compensation that he receives from the principal when being hired, irrespective from his type t∈{H, L} and c(e) is the cost of the total effort exerted by the hired worker (i.e., e=x+y), which is taken from Fehr et al. (1998a), with the cost of the minimum level of enforceable effort (i.e., x = 1 in project X) being equal to 0. The cost of exerting effort, c(e), is an increasing function of the overall effort and does not depend on whether it is exerted in X and/or in Y. If not hired, the worker receives a fixed unemployment benefit of 10 ECUs. In the BSLPS and SELPS treatments, a tiny profit sharing compensation scheme for the worker, ηW=0.075, is introduced. Panel C of Table 1 reports the monetary payoffs for the agent and the principal. The agent’s monetary payoff is increasing both in the effort exerted by the hired worker in project X and in activity Y: πA=ηAX+δy+mA, where mA is a fixed compensation paid by the principal. The fraction of the output X assigned to the agent depends on the treatment: while in the BSL and SEL treatments ηA=0.15, in the BSLPS and SELPS treatments it is reduced to ηA=0.075, in order not to change the cost of the incentive schemes for the principal. Selfish and R-ORP agents, who believe that all workers are payoff maximizers, should choose to select H workers if ηA is positive. Most importantly, the agent gets the entire amount of the output produced by the worker’s effort in activity Y=δy and, since δ>ηA≥0, the agent’s monetary payoff is higher when a given level of effort is exerted in activity Y rather than in project X. This creates a potential conflict between the principal’s interests (who only benefits from effort in project X) and the agent’s ones. The agent has also the option to ex ante refuse the value produced in activity Y, thus devoting it to the organization following the rules adopted to distribute the value produced in X. Note that agents motivated by inequality aversion (I-ORP) might prefer to choose L workers if they believe that, due to the “entitlement effect,” L workers will exert more effort than H workers. The agent’s possibility to send a suggestion and to refuse the value produced in Y is the two features of our experimental design that allow us to understand the agent’s motivation in selecting a worker of ability t∈{H, L}. The principal’s payoff function is: πP=ηPX+B−mA−mW, where ηP≤1 is the fraction of X kept by the principal, with ηA+ηP=1; B is a monetary endowment while mA and mW are the fixed compensations paid to the agent and the worker, respectively. The experiment consists of two parts. In the first part, subjects participate in one of the treatments described above, playing the game as one shot. In the second part, they play the same game for 15 periods, maintaining the same role as in part 1 but under a stranger random matching protocol (see the experimental protocol in the Online Appendix B).9 In part 2, feedbacks are provided to every participant at the end of each of the 15 periods. The principal receives information regarding the value produced in X while the agent is informed about the effort exerted in both X and Y in their organization. Finally, all participants are informed about: (i) how many H and L workers have been hired in other organizations within the previous period of the session and (ii) the average effort exerted in X and Y by H and L workers in other organizations. Previous research has suggested (Keizer et al. 2008; Gino et al. 2009; Diekmann et al. 2015) that others’ norms violation affects individual choices: we thus expect that when a “collusive” norm begins to spread among players, that is, the proportion of agents choosing L worker and/or the proportion of workers exerting effort in Y increases, then the propensity for subjects to pursue their personal interest at the expense of the organization increases, generating a snowball effect (Chang and Lai 2002). 4.1 Experimental Procedures The experiment was programed using zTree (Fischbacher 2007). In all treatments, participants were undergraduate students from the Friedrich Schiller University in Jena, recruited via the ORSEE software (Greiner 2004).10 We conducted 21 sessions of the BSL, SEL, and SELPS treatments at the experimental laboratory of the Max Planck Institute of Economics (Jena, Germany), from November 2012 to February 2013. Respectively, 216, 212, and 216 subjects participated in the BSL, SEL, and SELPS treatments. A total of 140 subjects participated in the BSLPS treatment, at the laboratory of the Friedrich Schiller University in Jena in January 2017.11 The experimental procedures, length, and average earnings were the same in all treatments.12 More details on the experimental procedures can be found in the Online Appendix B. The duration of each session was about 110 min and the average payment was 17 Euros, including a show up and participation fee of 4 Euros. 5. Experimental Results In this section, we present our experimental findings. First, we analyze the hiring decisions (Section 5.1). Then, we focus on the effort exerted by the workers in project X and in activity Y (Sections 5.2) and on the agents’ intention (Section 5.3). Throughout the analysis we will focus on the 15 periods of part 2, when information about others’ behavior is spread among participants.13 Given that spreading information may render participants’ choices in part 2 dependent upon previous periods in the same session, we perform both (i) two-sample Mann–Whitney tests and (ii) Somers’ D median difference tests (Newson 2002), reporting in the main text only (i) unless the two tests give different results.14 5.1 Hiring Our first result shows how the hiring decisions vary across treatments. Result 1. The proportion of L workers hired is higher in the SEL and SELPS treatments than in the BSL treatment. When hiring L workers, the majority of agents do not comply with the instructions received by the principals. Support for Result 1 can be found in Figure 1, which reports the proportion of principals giving instruction to hire L worker and the proportion of agents hiring L worker in part 2. In the BSL treatment 28.30% of the principals hire L workers while this percentage increases to 47.04% when the agents are making the hiring decision in the SEL treatment. Figure 1. View largeDownload slide Proportion of L Workers Suggested by the Principals and Hired by the Agents across Treatments. Figure 1. View largeDownload slide Proportion of L Workers Suggested by the Principals and Hired by the Agents across Treatments. The decision to hire L workers in the BSL treatment may be motivated by the expectation that, feeling less entitled for the position, they will be more grateful than H workers and therefore will exert a higher effort in project X; as a consequence, the productivity loss due to the difference in ability will be overcompensated, as shown by Montinari et al. (2016).15 It is important to note that in the SEL treatment, in the majority of the cases, L workers are hired against the principal’s suggestion. Specifically, this happens in 58.56% (N = 219/374) of the cases. In the BSLPS treatment, principals increase the proportion of L workers hired compared with the BSL treatment (MW test: z = 1.91, p = 0.06)16: the profit sharing compensation scheme appears to render more salient for them the possibility that an “entitlement” effect could induce L workers to feel more grateful for being hired. Comparing the SEL and SELPS treatments, when agents are making the hiring decision, we find that the tendency to hire more L workers than the ones suggested by the principal is basically unchanged (WSR tests: L suggested versus L hired, SEL z = 1.86, p = 0.06, SELPSz = 2.29, p = 0.02). As already noted for the SEL treatment, also in the SELPS treatment if we focus on those agents who hired L workers, we find that the majority of them did not follow the principal’s suggestion (59.95%, N = 220/367). We interpret these results as a distortion in the hiring process, which is not surprising if we consider that in the SELPS treatment the agents only get ηA=0.075 of the value produced in project X, that is, 50% less than what they got in the SEL treatment, and therefore they have less incentive to hire H workers. When focusing on the individual hiring decisions, we find that in the BSL and BSLPS treatments, on average, 52.27% of the principals decides to hire a L worker between zero and four times while only 25.23% of the agents do so in the SEL and SELPS treatments; on average 42.99% of agents hires L workers eight or more times. This suggests that the reported differences between treatments are not driven by the behavior of a minority of agents. The dynamics of the hiring decision reveals that in the SEL and SELPS treatments the distortion in hiring emerges from the first period and persists until the last one: the proportion of L workers hired increases across periods starting from an average of 41.12% in periods 1−5; increasing to 47.48% in periods 6−10 and reaching 49.91% in the last five periods.17 5.2 Workers’ Effort In the following we concentrate on the effort exerted by the hired workers. Results 2–4 state our main findings, statistical support is provided afterward. Result 2. In the BSL and BSLPS treatment, the average level of effort exerted in project X and in activity Y is the same for H and L workers. Result 2 shows that, on average, workers’ reciprocity does not vary depending on the worker’s ability: in the BSL and BSLPS treatments, H and L workers behave similarly (MW tests H versus L workers: BSL, for x: z = 0.06, p = 0.95; for y: z = 0.45, p = 0.65. BSLSEL for x: z = 0.75, p = 0.45; for y: z = 0.57, p = 0.56). Workers exert an average effort in project X equals to 2.68 and 3.71 in the BSL and BSLPS treatments, respectively, which are significantly higher than the minimum level of effort of 1 (Wilcoxon signed-rank test, BSL: z = 2.37, p = 0.02; BSLPS: z = 7.67, p = 0.00 WSR test, henceforth).18 In activity Y both workers exert an effort significantly higher than zero and equals to 0.89 and 0.71 in BSL and BSLPS treatments, respectively (WSR test, BSL: z = 2.37, p = 0.02; BSLPS: z = 7.67, p = 0.00).19 Result 3. In the SEL treatment, workers exert greater effort in activity Y and less effort in project X compared with the BSL treatment. This difference is driven by the behavior of L workers, who on average increase their effort in activity Y and reduce their effort in project X. When comparing the BSL treatment to the SEL treatment, in line with our theoretical predictions, L workers reciprocate the agents’ hiring choice. In the SEL treatment workers exert more effort in activity Y and reduce their effort in project X. In particular, L workers reduce their effort in X of about 17% passing from an average of 2.67–2.21 and increase their effort in Y of about 81% passing from an average of 0.95 to 1.72 (MW one-sided tests: BSL vs SEL, for x: z = 1.34, p = 0.09; for y: z = 2.75, p = 0.00). Note that, according to our model, a positive amount of effort in Y identifies the behavior of R-ORP workers; I-ORP workers indeed would never exert effort in Y since it advantages the agent, who is the member of the organization with highest monetary payoff.20H workers, instead, do not significantly modify their behavior: they exert an average effort of 2.33 in project X and of 0.99 in activity Y (MW test BSL vs SEL, for x: z = 0.96, p = 0.34; for y: z = 0.32, p = 0.75). Result 4. Profit sharing is effective in limiting backscratching by L workers: differences in the effort exerted in activity Y are not significant when comparing the BSLPS and SELPS treatments. The profit sharing compensation scheme has quite a strong effect on the effort exerted in project X, despite workers only getting a small fraction ηW=0.075 of the value generated in X. We find that, overall, workers increases their effort in X of about 40% in the BSLPS treatment compared with the BSL treatment, and of about 26% in the SELPS treatment compared with the SEL treatment. This increase is significant when considering L and H workers separately, according to a set of MW tests, one-sided: BSLPS versus BSL, xH, z = 2.17, p = 0.03; xL, z = 1.53, p = 0.06. SELPS versus SEL, yH, z = 1.45, p = 0.07; yL, z = 2.11, p = 0.02. Moreover, we observe that the profit sharing induces a reduction of the effort exerted in Y. When comparing the SELPS and SEL treatments, we find that both H and L workers significantly reduces their effort in Y (MW tests, one-sided: yH, z = 1.60, p = 0.06; yL, z = 2.75, p = 0.01). Further support for Results 2–4 can be found in Tables 2 and 3 and in Figure 2. Figure 2 plots, separately for H and L workers, the average effort exerted in X and Y in each treatment. Table 2. Average Total Effort (x+y) Exerted by the Workers, Standard Deviation in Parentheses MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 Note: The tests are conducted considering each session as an independent observation, that is, 7(9) sessions for BSL, SEL, SELPS (BSLPS). Table 2. Average Total Effort (x+y) Exerted by the Workers, Standard Deviation in Parentheses MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 MW test BSL SEL BSL versus SEL BSL versus BSLPS H workers 3.44 3.32 z = 0.06 z = 0.90 (0.51) (0.96) p = 0.95 p = 0.37 L workers 3.62 3.94 z = 0.70 z = 1.64 (0.89) (0.45) p = 0.48 p = 0.10 MW test BSLPS SELPS BSLPS versus SELPS SEL versus SELPS H workers 4.05 3.52 z = 0.58 z = 0.06 (1.07) (0.80) p = 0.56 p = 0.95 L workers 4.80 3.79 z = 1.54 z = 0.96 (1.73) (0.87) p = 0.12 p = 0.34 Note: The tests are conducted considering each session as an independent observation, that is, 7(9) sessions for BSL, SEL, SELPS (BSLPS). Table 3. The Effect of Profit Sharing: Comparison of BSL–SEL vs. BSLPS–SELPS Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Notes: * p<0.10, ** p<0.05, *** p<0.01. In all models standard errors are bootstrapped at the level of 30 sessions. N identifies the number of participants who took at least one decision in the second part of the experiment. Over the 15 periods of play, both L and H workers have the chance to be selected in each period. Table 3. The Effect of Profit Sharing: Comparison of BSL–SEL vs. BSLPS–SELPS Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Model 1 2 3 a b a b a b Estimation method Seemingly unrelated regressions Dependent variable Effort x Effort y Effort x Effort y Effort x Effort y Independent variables Selection −0.556*** 0.260*** −0.366*** 0.447*** −0.107 0.260*** (0.076) (0.045) (0.098) (0.063) (0.118) (0.077) Profit sharing 0.816*** −0.517*** 1.056*** −0.279*** 0.803*** −0.269** (0.074) (0.045) (0.128) (0.065) (0.160) (0.082) Selection×Profit sharing – – −0.425*** −0.422*** −0.285 −0.277** (0.156) (0.090) (0.209) (0.109) L Worker 0.251*** 0.405*** 0.232*** 0.386*** 0.429*** 0.211** (0.079) (0.046) (0.081) (0.048) (0.162) (0.100) Selection×L Worker – – – – −0.604*** 0.449*** (0.195) (0.140) Profit sharing× L Worker – – – – 0.510* −0.034 (0.278) (0.144) Selection×Profit sharing x L Worker – – – – 0.290 −0.353* (0.337) (0.196) Hired in t−1 0.036 0.038 0.035 0.037 0.062 0.025 (0.081) (0.049) (0.075) (0.047) (0.076) (0.049) Periods 6–10 −0.263*** 0.065 −0.262*** 0.066 −0.264*** 0.061 (0.097) (0.055) (0.096) (0.057) (0.098) (0.057) Periods 11–15 −0.396*** −0.141** −0.396*** −0.141** −0.387*** −0.149** (0.094) (0.057) (0.096) (0.055) (0.100) (0.054) Constant 2.900*** 0.890*** 2.811*** 0.805*** 2.733*** 0.871*** (0.109) (0.064) (0.108) (0.068) (0.118) (0.072) N 2730 2730 2730 2730 2730 2730 Subjects 200 200 200 200 200 200 Periods 2–15 2–15 2–15 2–15 2–15 2–15 Treatments BSL, BSLPS, SEL, SELPS R2 0.059*** 0.079*** 0.062*** 0.086*** 0.071*** 0.091*** Breusch Pagan test 14.788*** 13.173*** 19.853*** Notes: * p<0.10, ** p<0.05, *** p<0.01. In all models standard errors are bootstrapped at the level of 30 sessions. N identifies the number of participants who took at least one decision in the second part of the experiment. Over the 15 periods of play, both L and H workers have the chance to be selected in each period. Figure 2. View largeDownload slide Average Effort Exerted in Project X and Activity Y in Each Treatment Depending on Workers’ Ability. Figure 2. View largeDownload slide Average Effort Exerted in Project X and Activity Y in Each Treatment Depending on Workers’ Ability. Consider now Table 2: it compares the total effort (e=x+y) exerted by the hired workers across the four treatments. The average total effort exerted by the hired workers (irrespective of their ability) is equal to 3.53 and 3.63 in the BSL and SEL treatments and to 4.43 and 3.65 in the BSLPS and SELPS treatments, respectively. Pairwise comparisons across treatments are not statistically significant except for the BSL and BSLPS treatments (z = 1.79, p = 0.07). Our results show that, overall, the treatments affect how workers allocate their effort between project X and activity Y, but not the total effort exerted per se. In the SEL treatment, therefore, we interpret the effort exerted in activity Y as an effort distortion while in the BSLPS and SELPS treatments the profit sharing payment scheme acts by shifting effort from activity Y to project X. Additional analyses focusing on the frequency of individual effort choices reveal that the observed backscratching is not due to the behavior of a small minority of workers, but it is rather a widespread phenomena, see the Online Appendix A.2.1 for more details. The dynamics of the effort exertion in X and Y reveals that in the SEL backscratching emerges from the first period and persists until the last one, detailed results are reported in the Online Appendix A.2.1. Table 3 allows us to evaluate the impact of the profit sharing incentive scheme on the emergence of backscratching, by performing a difference in differences analysis that compares the difference in effort exertion in project X and in activity Y between the BSL and SEL treatments to the difference between the BSLPS and SELPS treatments. The estimations reported are obtained by means of from a set of Zellner’s seemingly unrelated regressions, which simultaneously estimates two equations (where the dependent variables are the effort exerted by workers in X and Y, respectively) allowing for errors to be correlated. This estimation method allows us to account for the fact that workers simultaneously choose an effort level in X and Y. Standard errors are bootstrapped at the level of session.21 In all models we use as independent variables a set of dummies to identify the Selection treatments (i.e., Selection takes value 1 for the SEL and SELPS treatments and 0 otherwise) and the profit sharing incentive scheme (i.e., profit sharing takes value 1 for the BSLPS and SELPS treatments and 0 otherwise). The variable L worker takes value 1 if the worker has low ability and 0 otherwise while hired in t−1 identifies whether the worker was hired in the previous period or not. Results of the Breusch–Pagan test of independence confirm that residuals from the two equations are not independent (p < 0.000 in both cases), with a coefficient of correlation of the residuals of around 14%. Consider models 1a and 1b: the coefficient associated to the variable Selection shows the difference between the BSL and SEL treatments, as the BSL treatment is the omitted category. In model 1a, the coefficient has a negative and significant effect, while it has a positive and significant effect in model 1b, confirming our previous findings: when passing from the BSL to the SEL treatment, we observe a reduction in the effort exerted in X and an increase in the effort exerted in Y, signaling the emergence of backscratching. The coefficient associated to the variable Profit sharing represents the difference between the BSL and the BSLPS treatments, which is positive for the effort exerted in X and negative for the effort exerted in Y. In models 2a and 2b, we introduce the interaction between the Selection and the Profit sharing treatments. The coefficient of the interaction term identifies the difference in differences, that is the difference in effort exertion between the BSL and SEL treatments versus the difference in effort exertion between the BSLPS and SELPS treatments. We observe that this coefficient has a negative and significant effect on the effort exerted both in X and Y, while other effects remain unchanged. Finally, in models 3a and 3b, we introduce the interaction terms between: (i) the Selection treatment and L workers; (ii) the Profit sharing treatment and L workers, and (iii) the Selection treatment, the Profit sharing treatment, and L workers. In model 3a, the coefficients of Selection and of the interaction between Selection and Profit sharing loose significance, the first interaction displays a negative and significant effect, while the second a positive and significant effect. No significant effect is found for the third triple interaction, meaning that there is no additional effect on the exertion of effort in project x that can be explained when considering the behavior of L workers. Other results remain unchanged with respect to model 2a. When considering model 3b, we find that our main results from model 2b remain unchanged. In addition, the interaction between the Selection treatment and L workers, and the interaction between the Selection treatment, the Profit sharing treatment, and L workers display a negative and significant effect: when considering the exertion of effort in Y, we observe a difference in differences between the BSL versus SEL treatment and the BSLPS versus SELPS treatment, plus an additional effect related to the behavior of L workers. Our results based on the difference in differences analysis confirm that the profit sharing incentive is effective in limiting the emergence of backscratching by L workers (linear combination of coefficients: –0.419, p = 0.070): despite the agents still hiring a high fraction of L workers in the attempt to engage in a backscratching relationship, the selected L workers significantly reduce their reciprocal response toward them, shifting their effort exertion from activity Y to project X. Most importantly, these results indicate that reward systems do not only provide monetary incentives but they may also affect workers’ attitudes toward the organization: the small incentive provided to workers is not enough to change their behavior because of a merely pecuniary motive. 5.3 The Agents’ Intentions In this section we investigate the agents’ intentions by looking (i) at the effort suggestions to the hired worker and (ii) at their decision to refuse the effort exerted in activity Y. Moreover, we report the results of a robustness treatment, identified as control – Selection (cSEL), aimed at testing the role of the agent’s suggestion in shaping backscratching reciprocity. Result 5 summarizes our findings. Result 5. On average agents suggest exerting more effort in activity Y and less effort in project X to L workers compared with H workers. Agents who hire L workers following the instructions of the principal suggest on average higher effort in X compared with agents who hire L workers against the principal’s instructions. Agents who select L workers are more likely to accept what was produced in Y in the SEL treatment than in the BSL treatment. Support for Result 5 can be found in Tables 4 and 5. Table 4 reports, for the SEL and SELPS treatments, the average effort suggested by the agents in project X and in activity Y, depending on the ability of the selected worker. It can be noted that in the different treatments agents try to engage in a backscratching relationship in different ways: in the SEL treatment they suggest to L workers to exert more effort in Y compared with H workers; in the SELPS treatment, they suggest to L workers to exert less effort in X compared with H workers. Table 4. Effort Suggested by the Agents, Standard Deviation in Parenthesis SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 Notes: The p-values reported refer to the one-sided tests. The tests are conducted considering each session as an independent observation, that is, seven sessions in each treatment. Table 4. Effort Suggested by the Agents, Standard Deviation in Parenthesis SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 SEL SELPS MW test Effort suggested in SEL vs. SELPS x y x y x y H workers 4.07 2.59 4.90 2.62 z = 1.98 z = 0.45 (0.71) (0.55) (0.85) (0.84) p = 0.02 p = 0.33 L workers 3.43 3.29 4.00 2.72 z = 1.09 z = 0.96 (0.88) (0.33) (1.02) (0.93) p = 0.14 p = 0.17 MW test: z = 1.21 z = 2.24 z = 1.73 z = 0.32 H vs. L p = 0.11 p = 0.02 p = 0.04 p = 0.38 Notes: The p-values reported refer to the one-sided tests. The tests are conducted considering each session as an independent observation, that is, seven sessions in each treatment. Table 5. Effort Suggested by the Agents Depending on Principal’s Instructions (Standard Error in Parenthesis) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Notes: The complete estimation of the regression is reported in Models 3 and 4 of Table A.7 in Section A.3 of the Online Appendix A. Table 5. Effort Suggested by the Agents Depending on Principal’s Instructions (Standard Error in Parenthesis) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Dependent variable: Effort suggested in project X in activity Y Coefficient Coefficient Workers hired following 0.25** −0.08 P’s suggestion (0.12) (0.08) L Workers hired following 0.54** −0.00 P’s suggestion (0.25) (0.16) Notes: The complete estimation of the regression is reported in Models 3 and 4 of Table A.7 in Section A.3 of the Online Appendix A. Table 5 reports the coefficients of a Zellner’s seemingly unrelated regression analysis focusing on the agents’ suggestions and on the principals’ instruction. The dependent variables are represented by the effort suggested in X and Y. We find that the effort suggested in X is significantly higher when a worker is hired following the principal’s instruction than when he is not, and this is true also when restricting the analysis to the agents who hire L workers.22 When considering those agents who ex ante decide not to accept the value eventually produced in Y, we do not find any significant difference in the percentage of L workers hired in the SEL and in the SELPS treatments with respect to the BSL one (MW tests one-sided: z = 1.05, p = 0.145, and z = 0.657, p = 0.256, respectively). On the contrary, when considering those agents who are willing to accept what produced in Y, we find that they are more likely to select L workers in the SEL and in the SELPS treatments with respect to the BSL one (MW tests one-sided: z = 2.00, p = 0.023, respectively). Our results thus show that both the effort suggestion to workers and the possibility to ex ante refuse the value produced in Y are important features of our design in making the agents’ intentions clear: agents who select L workers act in this way to maximize their own monetary payoff. However, compared with the BSL treatment, in the SEL treatment two features of the experimental design changed: (1) the possibility for the agent not to follow the principal’s suggestion and (2) the possibility to suggest to the hired worker a desired effort level. In our robustness treatment cSEL we replicated the SEL treatment with the only exception that the agent cannot make any suggestion to the hired worker about how much effort to exert in X and Y. When considering both the agents’ hiring behavior and the workers’ effort exertion, in the cSEL treatment we do find a similar pattern of behavior with respect to the SEL treatment, even if the effects have a lower magnitude. Compared with the BSL treatment, L workers increase their effort in activity Y while reducing their effort in project X. This allows us to exclude that backscratching is merely explained by the agents’ suggestion; nonetheless, the possibility of communication magnifies it, by making agents’ intentions more salient to the L workers. See the Online Appendix A.4 for a detailed analysis. 5.4 Conflicts of Interest and Earnings In this section, we focus on how the distortions on workers’ effort and agents’ selection affect the monetary payoffs of each member of the organization and the total earnings achieved by the organization. Our main findings are summarized in Result 6. In the Online Appendix A.5, we provide further analysis. Result 6. A profit sharing compensation scheme increases the earnings of the principals in both the BSLPS and the SELPS treatments compared with the BSL and SEL treatments, respectively. It decreases the earnings of the agents in the SELPS compared with the SEL treatment, displaying its effectiveness in limiting backscratching. As regards the total earnings, the maximum amount is reached in the BSLPS treatment, while the minimum amount is obtained in the SEL treatment. Support for Result 6 is provided in Figure 3. With respect to the BSL treatment, the distortions in the hiring process and in the effort exertion in the SEL treatment result in a significant increase of the agent’s earnings and a reduction of the principal’s earnings (MW tests BSL versus SEL, one-sided: principal, z = 1.98, p = 0.03; agent, z = 2.24, p = 0.02). When comparing the SEL and the SELPS treatments we observe that, once the profit sharing is introduced, the principal’s earnings increase while the earnings of the agents decrease (MW test SEL versus SELPS: principal, z = 1.98, p = 0.03; agent z = 3.13, p = 0.00). In particular, in the SEL treatment, hiring a L worker rather than a H worker is beneficial for the agents, providing them with higher earnings, while this is not the case in the SELPS treatment (MW test one-sided H versus L: SEL z = 1.60, p = 0.06; SELPSz = 0.45, p = 0.33). The opposite effect is observed when considering the principal’s earnings (MW test one-sided H versus L: SEL z = 1.47, p = 0.07; SELPSz = 1.09, p = 0.14). Figure 3. View largeDownload slide Average Earnings of the Principal, Agent, and Worker in Each Treatment. Figure 3. View largeDownload slide Average Earnings of the Principal, Agent, and Worker in Each Treatment. When considering the total earnings of the organization, obtained as the sum of the monetary payoff of all its members, they decrease by about 2% in the SEL treatment with respect to the BSL treatment. In the SELPS treatment, total earnings increase since the profit sharing affects L workers’ behavior: they substantially increase their effort in X and reduce their effort in Y, compared with the SEL treatment. In the BSLPS treatment, all workers exert more effort than in the BSL treatment in project X, which is beneficial for the organization: the largest amount of total earning is obtained in absence of backscratching reciprocity. Note that, in our design, exerting effort in activity Y would increase total earnings compared with not exerting any effort at all. However, this does not seem to occur in our experiment where, as shown in Table 2, the treatment manipulations affected the effort allocation between X and Y rather than the total effort exerted. To get an idea of the distortion associated to backscratching, we can calculate the hypothetical earnings across all parties, that is, the total earnings obtained if all the effort would have been exerted in X. The hypothetical earnings would have been equal to 135.48 and 136.26 in the BSL and SEL treatments and to 144.96 and 137.95 in the BSLPS and SELPS treatments, respectively. By comparing the hypothetical and actual earnings across all parties we can get a measure of the loss in the total earnings of the organization, which is equal to 5.1, 8.17, 6.93, and 6.56 ECU’s in the BSL, SEL, BSLPS, and SELPS treatment, respectively. It can be noted how the losses in the total earnings are the highest in the absence of profit sharing and, in particular, in the SEL treatment, providing additional evidence of the distortion induced by backscratching. 6. Conclusion In this paper, we provide evidence that workers’ reciprocity concerns may exacerbate, rather than alleviate, conflicts of interest within a hierarchical organization. We find that agents are more likely to hire low ability workers than high ability ones, since they are more likely to exert effort in an activity benefiting the agent rather than the principal. The organizational performance is negatively affected by the emergence of hidden gaming, because of the hiring and effort distortions observed in our main treatment. We show that a tiny profit sharing compensation scheme, which may foster workers’ identification with the organization, is effective in reducing workers’ inefficient effort exertion in favor of their foremen. This result is in line with the social psychology literature (Haslam 2004) and more recent studies in economics (Akerlof and Kranton 2005, 2008) suggesting that workers’ effort also depends on how they view themselves in relation to the organization. Interestingly, sharing part of the profits with the agents is not enough to modify their selfish behavior, while extending this compensation scheme to workers significantly reduces their distorted reciprocity toward agents: other-regarding preferences seem to be more susceptible than self-regarding ones to external stimuli induced by different institutional and organizational frameworks. In our experiment, workers’ subjective entitlements are based on their different abilities: low ability workers, who do not expect to be hired, are more grateful toward their agents compared with the high ability ones. However, other individuals’ observable characteristics could possibly be used to induce the emergence of backscratching reciprocity. For example, the decision to hire a candidate identifiable as a member of a discriminated group might be used to make him/her feel (more) indebted toward the agent. Alternatively, agents may favor members of their own social group expecting that the reduced social distance strengthens their reciprocity, as in Bramoullé and Goyal (2016). Compared with the gift exchange literature, we extend the hierarchical structure of the organization by adding an intermediate level: we focus our attention on how workers’ reciprocity toward the agent affects the organization. However, in the present study, we do not allow the principals to act kindly toward the agent or toward the worker, leaving room for future research to study the effects of multiple and simultaneous exchanges of gifts among the different levels of the hierarchy. Supplementary Material Supplementary material is available at Journal of Law, Economics, & Organization online. Conflict of interest statement. None declared. We acknowledge the Max Planck Institute of Economics and the University of Manchester for funding the experiment. V.M.’s contribution to this research was also performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of Universitè de Lyon, within the program Investissements d’Avenir (ANR-11-IDEX-007) operated by the French National Research Agency (ANR). Natalia Montinari’s contribution was performed while she was a researcher at the Department of Economics of the Lund University, Sweden. We thank Maria Bigoni, David Hugh-Jones, Marco Piovesan, Pedro Rey-Biel, Marie Claire Villeval for useful comments. We are indebted with Oliver Kirchkamp for letting us use the Experimental Laboratory of the Friedrich Schiller University in Jena. We also thank participants at the 3rd Annual Behavioral and Experimental Economics Workshop in Toulouse, the workshop on Understanding employee dishonesty behaviors in the workplace in Dijon, the 8th Alhambra Experimental Workshop in Rome, the 63rd Annual Meeting of the French Economic Association (AFSE) in Lyon, the 2014 ASSET meeting in Aix-en-Provence, the workshop on Industrial Organization: Theory, Empirics and Experiments in Alberobello, the workshop Norms Actions Games in London, the workshop Incentives and Unethical Behavior Conference in San Diego for useful comments. We also thank two anonymous referees for thoughtful comments that significantly improved the paper. Appendix Proof of Proposition 1. It is straightforward to note that selfish workers exert zero effort. Consider any RORP worker i. For both t∈{H, L}, worker i is grateful to the principal, then ρi,Pt=1 and ρi,At=0 and therefore yi∗=0 and xi∗=c′(γ)−1>0. Consider any IORP worker i. Since the principal has zero payoff when x=y=0, and the agent has higher payoff then the worker for any pair (x,y), then yi∗=0 and xi∗=c′(β1−β)−1. Workers exert the same level of effort irrespective of their ability. By assumption φH>φL, hence principals, irrespective of their preferences, select workers with ability H. Proof of Proposition 2. Suppose that principals suggest to hire a H worker in equilibrium, and notice immediately that if any principal deviates and play s~P=L, the deviation is not observed by any worker and therefore has no consequence. Beliefs are correct in equilibrium and therefore H workers behave as in the baseline game: selfish H workers exert zero effort; RORP H worker i  chooses yi∗=0 and c′(γ)−1=xi∗; a IORP H worker i chooses yi∗=0 and xi∗=c′(β1−β)−1. Consider now any L worker i; if he is selfish, then he exerts zero effort; if he has RORP then ρi,AL=1 and therefore yi∗=c′(γ)−1 and xi∗=0; finally, if he has IORP preferences then he choses yi∗=0 and xi∗=c′(β1−β)−1. Finally, selfish and RORP agents choose L workers, because only low ability workers exert effort with positive probability ;  IORP agents choose to hire L workers only if c′(γ)−1−α(mA+c′(γ)−1)≤−α(mA−c′(β1−β)−1), (A.1) that is, α≥c′(γ)−1c′(β1−β)−1+c′(γ)−1. (A.2) Footnotes 1. In what follows, we use the female pronoun for the agent and the male pronoun for the principal and for the workers. 2. Some concrete examples of this behavior are provided by The Conflict of Interest Board (COIB), New York. In COIB v. Fischetti [COIB Case No. 2010-035 (2010)], the Senior Deputy Director for Infrastructure Technology in the Information Technology Division at the New York City Housing Authority (NYCHA) was fined for his multiple violations of the City’s conflicts of interest law. In particular, “he used his NYCHA subordinate, a Data Technician, to perform work on a regular basis at the restaurant without compensation. He further admitted that he caused his subordinate to use his NYCHA computer, e-mail account, and Blackberry to perform work related to the restaurant, at times the subordinate was required to be working for the City.” For further evidence look at the COIB web site at http://www.nyc.gov/html/conflicts/html/topic/position.shtml. 3. An exception is provided by the field experiment with children by Belot and Van de Ven (2011). 4. Another reason for managers to promote low-quality workers is found in the fact that incompetent managers would feel threatened by competent subordinates, and inevitably drive away competent employees (Bedeian and Armenakism 1998). In line with this reasoning, Prendergast (1993) notes that “yes men” tends to be concentrated among less able workers and among workers with less able managers. 5. Gneezy et al. (2013) have provided experimental evidence that greed, rather than reciprocity, may explain the emergence of corruption. 6. We take as reference the model by Charness and Rabin (2002). 7. Principal’s reciprocity toward the agent is not relevant because the principal cannot affect the agent’s payoff. Nonetheless, it is natural to assume that the principal is grateful to the agent who hires the worker who exerts the highest effort x. 8. Fixing a small difference between φH and φL provides a more robust test for our hypotheses: everything equals, we would expect a bigger difference in ability to produce a stronger backscratching in the SEL treatment. 9. Find the Online Appendix B at the following web site: https://sites.google.com/site/vmaggian/research. 10. Each subject participated only in one experimental session. We excluded those who previously participated in experiments featuring a gift exchange game or trust game and also those who participated in the sessions where data for Montinari et al. (2016) were collected since their experimental design was similar to our own. 11. We thank an anonymous referee for suggesting us to run this additional treatment. 12. The experimental laboratories of the Max Planck Institute of Economics and of the Friedrich Schiller University in Jena share the same recruiting system and, therefore, the same subject pool. Subjects’ invitations for the BSLPS treatment were sent to exactly the same subject pool used in the other treatments, also maintaining the same criteria for the eligibility. 13. Result from Part 1 is qualitatively identical to those of Part 2 and they are reported in the Online Appendix A, Section 6. Since 5/54, 8/53, 9/34, and 10/54 L workers are hired in the BSL, SEL, BSLPS, and SELPS treatments, respectively, the analysis on effort has very little statistical power and would not lead to any additional benefit to the reader. Find the Online Appendix A at the following web site: https://sites.google.com/site/vmaggian/research. 14. When performing the Mann–Whitney U-test we average the data within a session and treat each session as a single observation. While conservative, this approach makes the power of all tests low and not robust to outliers. To account for these limitations we use the rank-order statistics Somers’ D (provided by the “somersd” package in Stata) that looks at the individuals’ choices accounting for the presence of clusters at the level of sessions in the data. 15. In Montinari et al. (2016) about 30% of employers hire low ability workers. 16. All statistical tests reported are two samples and two-sided unless differently specified. When we have clear, explicit, and theory-driven predictions about agents’ and worker’ behavior in the different treatments we used one-sided tests. 17. See the Online Appendix A.1.2. for additional analyses on the dynamics of part 2. 18. In both BSL and BSLPS treatments, H workers exert less effort than L workers in project X (2.54 versus 2.67, in the BSL and 3.46 versus 3.98 in the BSLPS), but this difference fails to reach significance. 19. We fixed a minimum effort for project X equal to 1 and this may have created an anchoring effect, inducing some workers to exert a positive effort in Y too. 20. A worker aimed at equalizing payoffs of all members of the organization should exert xt = 5 and yt = 0. In our data, a one-sample WSR test rejects the null hypothesis that the average effort exerted by the hired worker in project X is equal to five in each treatment, p < 0.02 in all cases. 21. For the estimations, we used the “sureg” command in Stata. For the standard errors, we used the “bootstrap” command in Stata, which executes a nonparametric bootstrap of the statistics in the list of the explanatory variables by resampling observations (with replacement) from the data in memory 1000 times. Our results do not change if we cluster the standard errors at the subjects level. 22. The whole regression and a more complete analysis of the determinants of the effort suggestion are provided in the Online Appendix A.3. References Abbink Klaus. 2004 . “ Staff Rotation as an Anti-corruption Policy: An Experimental Study,” 20 European Journal of Political Economy 887 – 906 . Google Scholar CrossRef Search ADS Abbink Klaus , Irlenbusch Bernd , Renner Elke . 2002 . “ An Experimental Bribery Game ,” 18 Journal of Law, Economics, and Organization 428 – 54 . Google Scholar CrossRef Search ADS Akerlof George A. , Kranton Rachel E. . 2005 . “ Identity and the Economics of Organizations ,” 19 Journal of economic perspectives 9 – 32 . Google Scholar CrossRef Search ADS Akerlof George A. , Kranton Rachel E. 2008 . “ Identity, Supervision, and Work Groups ,” 98 American Economic Review 212 – 7 . Google Scholar CrossRef Search ADS Bac Mehmet. 1996 . “ Corruption and Supervision Costs in Hierarchies ,” 22 Journal of Comparative Economics 99 – 118 . Google Scholar CrossRef Search ADS Bandiera Oriana , Barankay Iwan , Rasul Imran . 2009 . “ Social Connections and Incentives in the Workplace: Evidence from Personnel Data ,” 77 Econometrica 1047 – 94 . Google Scholar CrossRef Search ADS Baron James N. 2013 . “ Empathy Wages? Gratitude and Gift Exchange in Employment Relationships ,” 33 Research in Organizational Behavior 113 – 34 . Google Scholar CrossRef Search ADS Barr Abigail , Serra Danila . 2009 . “ The Effects of Externalities and Framing on Bribery in a Petty Corruption Experiment,” 12 Experimental Economics 488 – 503 . Google Scholar CrossRef Search ADS Bedeian Arthur G. , Armenakism Achilles A. . 1998 . “ The Cesspool Syndrome: How Dreck Floats to the Top of Declining Organizations ,” 12 Academy of Management Executive 58 – 67 . Bellemare Charles , Shearer Bruce . 2009 . “ Gift Giving and Worker Productivity: Evidence from a Firm-Level Experiment,” 67 Games and Economic Behavior 233 – 44 . Google Scholar CrossRef Search ADS Belot Michle , Van de Ven Jeroen . 2011 . “ Friendships and Favouritism on the Schoolground—A Framed Field Experiment ,” 121 Economic Journal 1228 – 51 . Google Scholar CrossRef Search ADS Bramoullé Yann , Goyal Sanjeev . 2016 . “ Favoritism ,” 122 Journal of Development Economics 16 – 27 . Google Scholar CrossRef Search ADS Chang Juin Jen , Lai Ching Chong . 2002 . “ Is the Efficiency Wage Efficient? The Social Norm and Organizational Corruption ,” 104 Scandinavian Journal of Economics 27 – 47 . Google Scholar CrossRef Search ADS Charness Gary. 2004 . “ Attribution and Reciprocity in an Experimental Labor Market,” 22 Journal of Labor Economics 665 – 88 . Google Scholar CrossRef Search ADS Charness Gary , Rabin Matthew . 2002 . “ Understanding Social Preferences with Simple Tests,” 117 Quarterly Journal of Economics 817 – 69 . Google Scholar CrossRef Search ADS Cialdini Robert. 1996 . “Social Infuence and the Triple Tumor Structure of Organizational Dishonesty,” in Messick D.M. , Tenbrunsel A.E. eds., Codes of Conduct: Behavioral Research into Business Ethics, 44 – 58 . New York, NY : Russell Sage . Coyle-Shapiro Jacqueline A.-M. , Morrow Paula C. , Richardson Ray , Dunn Stephen R. . 2002 . “ Using Profit Sharing to Enhance Employee Attitudes: A Longitudinal Examination of the Effects on Trust and Commitment ,” 41 Human Resource Management 423 – 39 . Google Scholar CrossRef Search ADS Diekmann Andreas , Przepiorka Wojtek , Rauhut Heiko . 2015 . “ Lifting the Veil of Ignorance: An Experiment on the Contagiousness of Norm Violations ,” 27 Rationality and Society 309 – 33 . Google Scholar CrossRef Search ADS Edwards Richard. 1979 . Contested Terrain: The Transformation of the Workplace in the Twentieth Century . New York, NY : Basic Books . Englmaier Florian , Leider Steve . 2012 . “Managerial Payoff and Gift Exchange in the Field.” Cesifo Working Paper Series No. 3707. Eriksson Tor , Villeval Marie Claire . 2012 . “ Respect and Relational Contracts ,” 81 Journal of Economic Behavior and Organization 286 – 98 . Google Scholar CrossRef Search ADS Fehr Ernst , Falk Armin . 1999 . “ Wage Rigidity in a Competitive Incomplete Contract Market,” 107 Journal of Political Economy 106 – 34 . Google Scholar CrossRef Search ADS Fehr Ernst , Schmidt Klaus M. . 1999 . “ A Theory of Fairness, Competition, and Cooperation ,” 114 Quarterly Journal of Economics 817 – 68 . Fehr Ernst , Kirchsteiger Georg , Riedl Arno . 1993 . “ Does Fairness Prevent Market Clearing? An Experimental Investigation ,” 108 Quarterly Journal of Economics 437 – 59 . Google Scholar CrossRef Search ADS Fehr Ernst , Gachter Simon , Kirchsteiger Georg . 1997 . “ Reciprocity as a Contract Enforcement Device: Experimental Evidence ,” 65 Econometrica 833 – 60 . Google Scholar CrossRef Search ADS Fehr Ernst , Kirchsteiger Georg , Riedl Arno . 1998a . “ Gift Exchange and Reciprocity in Competitive Experimental Markets ,” 42 European Economic Review 1 – 34 . Google Scholar CrossRef Search ADS Fehr Ernst , Kirchler Erich , Weichbold Andreas , Simon Gachter . 1998b . “ When Social Norms Overpower Competition: Gift Exchange in Experimental Labor Markets ,” 16 Journal of Labor Economics 324 – 51 . Google Scholar CrossRef Search ADS Fischbacher Urs. 2007 . “ z-Tree: Zurich Toolbox for Ready-Made Economic Experiments ,” 10 Experimental Economics 171 – 8 . Google Scholar CrossRef Search ADS Gino Francesca , Shahar Ayal , Ariely Dan . 2009 . “ Contagion and Differentiation in Unethical Behavior: The Effect of One Bad Apple on the Barrel,” 20 Psychological Science 393 – 8 . Google Scholar CrossRef Search ADS Gneezy Uri , List John A. . 2006 . “ Putting Behavioral Economics to Work: Testing for Gift Exchange in Labor Markets Using Field Experiments ,” 74 Econometrica 1365 – 84 . Google Scholar CrossRef Search ADS Gneezy Uri , Saccardo Silvia , van Veldhuizen Roel . 2013 . “Bribery: Greed versus Reciprocity.” U. C. San Diego Working Papers. Greiner Ben. 2004 . “ An Online Recruitment System for Economic Experiments ,” 63 Forschung und Wissenschaftliches Rechnen 2003. Gwdg Bericht 79 – 93 . Haslam S. Alexander 2004 . Psychology in Organizations . Thousand Oaks, CA : Sage Publications . Hennig-Schmidt Heike , Sadrieh Abdolkarim , Rockenbach Bettina . 2010 . “ In Search of Workers’ Real Effort Reciprocity—A Field and a Laboratory Experiment,” 8 Journal of the European Economic Association 817 – 37 . Heywood John S. , Jirjahn Uwe , Tsertsvadze Georgi . 2005 . “ Getting along with Colleagues—Does Profit Sharing Help or Hurt?,” 58 Kyklos 557 – 73 . Google Scholar CrossRef Search ADS Keizer Kees , Lindenberg Siegwart , Steg Linda 2008 . “ The Spreading of Disorder,” 322 Science 1681 – 5 . Google Scholar CrossRef Search ADS Kerr Jeffrey , Slocum John W. . 1987 . “ Managing Corporate Culture through Reward Systems ,” 1 Academy of Management Executive 99 – 107 . Khatri Naresh , Tsang Eric W. K. . 2003 . “ Antecedents and Consequences of Cronyism in Organizations ,” 43 Journal of Business Ethics 289 – 303 . Google Scholar CrossRef Search ADS Knez Marc , Simester Duncan . 2001 . “ Firm Wide Incentives and Mutual Monitoring at Continental Airlines ,” 19 Journal of Labor Economics 743 – 72 . Google Scholar CrossRef Search ADS Kolm Serge-Christophe , ed. 2006 . “Reciprocity: Its Scope, Rationales, and Consequences,” in Handbook of the Economics of Giving, Altruism and Reciprocity, vol. 1, 371 – 541 . North Holland : Imprint . Kramarz Francis , Thesmar David . 2013 . “ Social Networks in the Boardroom,” 11 Journal of the European Economic Association 780 – 807 . Google Scholar CrossRef Search ADS Kube Sebastian , Maréchal Michel Andr , Puppe Clemens . 2013 . “ Do Wage Cuts Damage Work Morale? Evidence from a Natural Field Experiment ,” 11 Journal of the European Economic Association 853 – 70 . Google Scholar CrossRef Search ADS Laffont Jean-Jacques. 1988 . “ Hidden Gaming in Hierarchies: Facts and Model ,” 64 Economic Record 295 – 306 . Google Scholar CrossRef Search ADS Laffont Jean-Jacques. 1990 . “ Analysis of Hidden Gaming in a Three-Level Hierarchy,” 6 Journal of Law, Economics and Organization 301 – 24 . Levine David K. , Weinschelbaum Federico , Zurita Felipe . 2010 . “ The Brother-in-Law Effect ,” 51 International Economic Review 497 – 507 . Google Scholar CrossRef Search ADS Malmendier Ulrike , Klaus Schmidt . 2017 . “ You Owe Me ,” 107 American Economic Review 493 – 526 . Google Scholar CrossRef Search ADS Montinari Natalia , Nicolò Antonio , Oexl Regine . 2016 . “ The Gift of Being Chosen,” 19 Experimental Economics 460 – 79 . Google Scholar CrossRef Search ADS Newson Roger. 2002 . “ Parameters behind ‘Nonparametric’ Statistics: Kendall’s Tau, Somers’ d and Median Differences ,” 2 Stata Journal 45 – 64 . Osterman Paul. 1994 . “ How Common Is Workplace-Transformation and Who Adopts It ,” 47 Industrial and Labor Relations Review 173 – 88 . Google Scholar CrossRef Search ADS Pearce Jone L. , Branyiczki Imre , Bigley Gregory A. . 2000 . “ Insufficient Bureaucracy: Trust and Commitment in Particularistic Organizations,” 11 Organization Science 148 – 62 . Google Scholar CrossRef Search ADS Prendergast Canice. 1993 . “ A Theory of ‘Yes Men’,” 83 American Economic Review 757 – 70 . Prendergast Canice , Topel Robert H. . 1993 . “Favoritism in Organizations.” National Bureau of Economic Research No. w4427. Robin Stphane , Rusinowska Agnieszka , Villeval Marie Claire . 2014 . “ Ingratiation: Experimental Evidence ,” 66 European Economic Review 16 – 38 . Google Scholar CrossRef Search ADS Schlicht Ekkehart. 1998 . On Custom in the Economy . Oxford : Clarendon Press . Google Scholar CrossRef Search ADS Thiele Veikko. 2013 . “ Subjective Performance Evaluations, Collusion, and Organizational Design ,” 29 Journal of Law, Economics, and Organization 35 – 59 . Google Scholar CrossRef Search ADS Tirole Jean. 1986 . “ Hierarchies and Bureaucracies: On the Role of Collusion in Organizations,” 2 Journal of Law, Economics, and Organization 181 – 214 © The Author(s) 2018. Published by Oxford University Press on behalf of Yale University. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

The Journal of Law, Economics, and OrganizationOxford University Press

Published: Feb 27, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off