- Email: [email protected]
In good times and bad – Reciprocal behavior at the workplace in times of economic crises
Journal of Economic Behavior & Organization 134 (2017) 228–239
Contents lists available at ScienceDirect
Journal of Economic Behavior & Organization journal homepage: www.elsevier.com/locate/jebo
In good times and bad – Reciprocal behavior at the workplace in times of economic crises夽 Leonie Gerhards a,∗ , Matthias Heinz b a b
University of Hamburg, Von-Melle-Park 5, 20146 Hamburg, Germany University of Cologne, Albert-Magnus Platz 1, 50923 Cologne, Germany
a r t i c l e
i n f o
Article history: Received 11 December 2015 Received in revised form 13 December 2016 Accepted 22 December 2016 Available online 29 December 2016 JEL classification: C9 J3 M5
a b s t r a c t This laboratory experiment analyzes employer–employee relations in the face of an exogenous shock. We implement a two-period gift-exchange game in which the employer can be hit by an adverse shock in the second period. We find that the mere possibility of the shock encourages employers to pay significantly higher wages in the first period. These higher first period wages translate into increased effort levels in the second period, independent of the actual occurrence of the shock. Our results suggest that the mere possibility of an exogenous shock strengthens the cooperation between individuals, which can ultimately mitigate its negative impact on total profits. © 2016 Elsevier B.V. All rights reserved.
Keywords: Gift-exchange Wage cuts Exogenous shock Reciprocity
“When written in Chinese the word crisis is composed of two characters. One represents danger, and the other represents opportunity.” John F. Kennedy (1959)1
1. Introduction In the presence of incomplete contracts, work morale, that is, the employees’ voluntary cooperation with their employers, crucially determines the success of employment relationships. In particular during economic downturns, this mechanism
夽 This work was financially supported by Goethe University Frankfurt. We would like to thank Johannes Abeler, Arno Apffelstaedt, Iwan Barankay, Leonie Baumann, Gary Charness, Ruben Durante, Florian Englmaier, Armin Falk, Guido Friebel, Simon Gächter, Lorenz Götte, Georg Kirchsteiger, Alexander Koch, Joep Konings, Michael Kosfeld, Julia Nafziger, Gerd Mühlheußer, Daniel Schunk, Heiner Schumacher, Ferdinand von Siemens, Joël van der Weele, and seminar participants at the MPI for Research on Collective Goods in Bonn as well as the conference audiences at the European Economic Association Meeting in Oslo 2011, at the Workshop on Behavioral Personnel Economics in Mannheim 2011 and at the Colloquium of Personnel Economics in Vienna 2015. Finally, we would like to thank the referees and editors whose comments allowed us to greatly improve the manuscript. ∗ Corresponding author. E-mail addresses: [email protected] (L. Gerhards), [email protected] (M. Heinz). 1 Quote taken from Shapiro (2006). http://dx.doi.org/10.1016/j.jebo.2016.12.021 0167-2681/© 2016 Elsevier B.V. All rights reserved.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
229
seems to be of utmost importance. In this paper we study whether reciprocal gift-exchange between firms and workers can help to overcome the negative effects of adverse economic shocks by sustaining cooperation in the face of rough times. We consider shocks that hit firms through no fault of their own. Examples are increasing input prices for raw materials, demand drops caused by political conflicts or terroristic attacks or decreases in selling prices due to over-capacities on the global market. Earlier interview studies, conducted with both managers and employees suggest that workers are willing to help their firms by accepting wage cuts in these kind of situations (see Agell and Lundborg, 1995; Bewley, 1999, 2005; Blinder and Choi, 1990; Campbell and Kamlani, 1997; Charness and Levine, 2002; Kahneman et al., 1986). The interview studies, however, also report that firms can only count on support from their workers after a wage cut if they are in severe trouble. This, in turn, disciplines their employers not to reduce wages in more stable economic times, leading to the aggregate phenomenon of wage rigidity. In our laboratory experiment, we implement the adverse shock as a clearly exogenous event that has a substantial impact on employers’ profits. While previous studies considered reciprocal gift-exchange as a means to overcome shocks, we study if and how (the possibility of) exogenous shocks affect gift-exchange. In particular, we analyze how the mere possibility of a shock that can hit the employer influences employment relations. We set out to answer the following questions: How do employers adjust their wages if they are facing the possibility of an adverse shock? How do their employees respond? And lastly, does reciprocal gift-exchange have the potential to mitigate the negative payoff consequences of a shock? In our two-period gift-exchange game, in each period, the employer first sets a wage and the employee subsequently chooses an effort level. Both wages and effort levels are costly and not contractible. Across treatments we vary the probability with which the employer is hit by an exogenous shock in the second period. Our first benchmark treatment, the Prob0 treatment, is a twice repeated standard gift-exchange game. Employers and employees simply take their respective wage and effort choices twice, in two subsequent periods. In our second benchmark treatment, the Prob100 treatments, the payoff functions in the first period are the same as in the Prob0 treatment. In the second period, conversely, the employer is hit by an exogenous shock which implies a reduction of the worker’s marginal productivity in the employer’s payoff function. In the main treatment, the Prob50 treatment, the exogenous shock occurs in the second period with 50% probability. If the shock hits the employer, the payoff functions are the same as in the Prob100 treatment. Otherwise, they are the same as in the Prob0 treatment. In all treatments both parties have full information about the probabilities and the actual realization of the shock. Our main findings can be summarized as follows: The mere possibility of an adverse shock (in the Prob50 and Prob100 treatment) encourages the employers to pay on average higher wages in the first period – irrespective of whether the shock later occurs with 50% probability or for certain. When the employers actually experience the shock in the second period, they pay comparably lower wages. Their employees nevertheless keep up reciprocity. Interestingly, we generally find that in all treatment variants, in which the employer faced the exogenous shock (with their treatment-specific probabilities) in the first period, their employees behave on average significantly more reciprocal in the second period Taken as a whole, our results suggest that the possibility of an exogenous shock can strengthen reciprocal gift-exchange between firms and workers. Ultimately, our findings can provide guidance for firms on how to establish a cooperative relationship with their workers that they can reap in times of economic distress. This seems particularly important for firms that operate in volatile markets. We discuss several policy implications of our findings in the conclusion of this paper. This paper contributes to various fields of research. First, our paper relates to the literature on employees’ reactions to wage cuts. Several field experiments and field studies show that employees decrease their performance as a response to a wage reduction (see for instance Cohn et al., 2014; Kube et al., 2013; Krüger and Friebel, 2016). However, this literature mainly studies employees’ reactions to wage cuts where the employer either does not explain the reasons for the wage reduction or where the reasons are arguably perceived as unfair. Our experiment, on the other hand, focuses on employees’ responses to lowered wages that can be attributed to worsened economic conditions. Consistent with previous interview studies (Agell and Lundborg, 1995; Bewley, 1999, 2005; Blinder and Choi, 1990; Campbell and Kamlani, 1997; Charness and Levine, 2002; Kahneman et al., 1986), we find that this type of justifiable wage cuts seems to trigger less negative reactions from the employees’ side. Furthermore, our study corroborates and extends findings from the literature on reciprocity based on Akerlof’s (1982) gift-exchange model showing that reciprocity is an effective contract enforcement device in the presence of incomplete contracts (see among others Fehr et al., 1993, 1997). The main result from these studies is that workers’ efforts are positively related to the positive wages provided. For employers who anticipate that a non-negligible proportion of the population is endowed with reciprocal preferences, it hence pays off to choose generous wages. Recent experiments have extended this workhorse model by introducing economic crises in the lab. Kocher and Strasser (2011), for instance, consider a 15-periods labor market and model economic cycles by varying the employees’ productivity over periods. They show that, despite the fact that firms cut wages in times of economic distress, workers reduce effort levels in these periods only slightly. In Linardi and Camerer’s (2012) setup, firms and workers interact on a 30-periods labor market. In each of the periods the firms can be hit by a publicly observable stochastic shock, which prevents hiring for the coming three periods. Similar to Kocher and Strasser (2011), they find that the implemented stochastic shocks do not have significant negative effects on wages and effort levels. The employment relations are found to be rather long-lasting and to be characterized by reciprocal gift-exchange of high efforts, high wages and job security in times of economic distress.
230
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
Table 1 Payoff functions across treatments. Period 1
Period 2 No shock
Prob0
1 = 10 − w + 5e 2 = 10 − e + 5w
1 = 10 − w + 5e 2 = 10 − e + 5w
Prob50
1 = 10 − w + 5e 2 = 10 − e + 5w
1 = 10 − w + 5e 2 = 10 − e + 5w
Prob100
1 = 10 − w + 5e 2 = 10 − e + 5w
Shock realized
1 = 10 − w + 2.5e 2 = 10 − e + 5w 1 = 10 − w + 2.5e 2 = 10 − e + 5w
Employer’s payoff: 1 , employee’s payoff: 2 ; Effort e {0, 1, . . ., 10}, wage w {0, 1, . . ., 10}.
In line with Linardi and Camerer (2012) and Kocher and Strasser (2011), also employees in our experiment do not respond with low effort choices if employers lower their wages in case they are hit by an exogenous shock. This seems noteworthy since, different from their experiments, in our study a shock can occur only once and workers cannot become unemployed. While previous experiments have either focused on gift-exchange in the absence of exogenous shocks or have analyzed gift-exchange in times of adverse shocks, we are, to the best of our knowledge, the first to study how the mere possibility of being hit by an exogenous shock affects behavior in an otherwise standard gift-exchange game.2 2. Experimental design and procedures 2.1. Experimental design We implement a two-period gift-exchange game in which we apply Brandts and Charness’s (2004) symmetric and linear payoff functions. In the beginning of the first period, both parties receive an endowment of 10. First, the employer chooses a wage w from the integer set {0, 1, . . ., 10} and specifies a desired, non-binding effort level, a feature commonly used in gift-exchange experiments (see for instance Fehr et al., 1997). The chosen wage is multiplied by 5 and transferred to the employee. Subsequently, the employee decides which effort level e from the integer set {0, 1, . . ., 10} she wants to provide. This effort level is then analogously multiplied by 5 and sent to the employer. The first period is identical in all treatments. In the second period the payoff functions differ across treatments. In our first benchmark treatment (Prob0) the endowments, procedures and payoff functions are the same as in the first period. In our second benchmark treatment (Prob100), on the other hand, an exogenous shock hits the employer with certainty. This shock implies a reduction of the employee’s marginal productivity from 5 to 2.5. Apart from that the payoff functions do not change. Both parties still receive an endowment of 10 at the beginning of the second period, and the employer’s wage choice is still multiplied by 5 before it is transferred to the employee. In the Prob50 treatment, finally, the exogenous shock is realized with 50% probability, that is, exactly half of the employer-employee relationships are randomly selected to be hit by the shock. If the shock occurs, the second period payoff functions are equal to those from the Prob100 treatment. If the shock does not occur, they are the same as in the Prob0 treatment. Table 1 summarizes the employers’ and employees’ payoff functions for all treatments. Subjects have full information about the procedures of the experiment. In particular, the fact that the employer is going to (or can potentially) be hit by an exogenous shock in the second period in the Prob100 (Prob50) treatment and the resulting payoff functions are explained to both parties before the start of the experiment. Moreover, at the beginning of period 2 the employer as well as the employee are informed of whether the shock actually occurs or not. 2.2. Distinctive features of our experimental design The adverse shocks that we have in mind for our study are any kind of exogenous economic shocks that can potentially hit a firm and have a severe impact on its profits. In our experiment the shock implies a substantial reduction of the employee’s marginal productivity. Our design could thus reflect situations in which firms have to deal with decreasing output prices (for instance, due to a new competitor entering the market or over-capacities on the global market) or conditions in which firms face increasing input prices for raw materials (for example, as a result of political conflicts). The common and decisive feature of these situations is that exogenous factors lead to severe economic problems which make the firm heavily dependent on support from its employees.
2 Relatedly, Rubin and Sheremeta (2016), Davis et al. (2016) consider more extended versions of the gift-exchange game in which, first, the principal sets a wage, then, the agent chooses an effort level that can subsequently be affected by an (unobservable) shock and, finally, the principal chooses a wage adjustment. These experimental studies consistently find that principals reward their agents both based on realized output and effort (in treatments in which it is observable) in the last stage of the game. However, while Rubin and Sheremeta (2016) find that the observability of the shock does not affect choices, Davis et al. (2016) find significant behavioral differences across the treatments in which principals observe either only effort or both effort and the occurrence of the shock.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
231
Another noteworthy feature in this context are the probabilities with which the shock occurs in our treatments (100% and 50%). Arguably, a shock that occurs for certain and a shock that occurs with 50% capture different problems of an employer. We mainly chose these probabilities for methodological reasons. The Prob100 treatment first and foremost serves as a benchmark treatment for the shock case. The Prob50 treatment, on the other hand, provides us with an easy way to study the impact of the threat of a shock. In the second period it splits up into two equally sized treatment variants – one variant with and one without shock realization (for simplicity, called Prob50shock and Prob50noshock in the following). Comparing behavior from each of these treatment variants to that from the respective benchmark treatment (Prob0 or Prob100) allows us to disentangle the influence of the ex-ante varying shock probabilities in the main and benchmark treatments from further effects resulting from the differing payoff functions. Moreover, by comparing employers’ and employees’ decisions of the Prob50shock and Prob50noshock treatment variants in period 2, we are able to study the influence of the shock on wages and efforts, holding constant the history from period 1. In order to study the impact of the shock in isolation, we minimized the complexity of the decision situation to the greatest possible extent. We opted for a fixed and exogenous partner matching and excluded the opportunity to lay off workers. This reduces the impact of confounding factors resulting from market interactions and the threat of being dismissed. Moreover, we inform both employers and employees about the probabilities and actual occurrence of the shock, which allows us to rule out potential tensions between contracting parties emerging from information asymmetries.3 The payoff functions that we borrowed from Brandts and Charness (2004) come with several additional advantages. First, due to their linearity, the marginal effect of effort on the employer’s profit is independent of the wage. Second, given the symmetry in all periods in which the employer is not hit by the shock, reciprocal employees can easily see that equality of wage and effort equalizes payoffs. And third, the fact that the shock only affects the employee’s marginal productivity in the employer’s payoff function makes it evident that the employer can choose from the same set of wages as in period 1. We simplified the design further by implementing a decision setting where a shock can occur only once. Although we acknowledge that a design with repeated shocks would allow subjects to learn from previous rounds and might hence increase their understanding of the situation, we are confident that our experimental game is the method of choice when studying the effects of announcing the possibility of an economic shock in isolation. The simplicity of the game already reduces the likelihood that observed treatment effects are mainly driven by subjects’ confusion and erroneous behavior. Hence, potential additional learning resulting from repeated interaction seem negligible. On the contrary, our setup allows us to study behavior in the face of an adverse shock undisturbed from any feedback or reputation building. 2.3. Procedures At the beginning of the session all subjects read the general instructions to the experiment in which we apply a natural framing using expressions such as “employer”, “employee” and “economic crisis” to ensure a common understanding of the experimental game.4 To clarify the payoff calculations, two examples are given in the instructions. Additionally, subjects have to answer several control questions concerning the sequence of actions during the experiment and the monetary consequences of their decisions. Once all subjects have answered all control questions correctly, the computer randomly assigns subjects to their roles (employer or employee) and matches them on a one-to-one basis. In the end of the session, after subjects have taken their decisions in the experimental game we administer a short survey, in which we elicit, among other things, subjects’ general willingness to take risk (“How willing are you to take risks, in general?”). Answers are given on a 10 point scale. We hence elicit risk attitudes similar to Dohmen et al. (2011), who provide evidence that individuals’ self-reported risk preferences are correlated with actual risk-taking behavior in the field. The experiment was conducted at the FLEX laboratory of Goethe University Frankfurt. All subjects were undergraduate students from different disciplines at the university and were recruited via ORSEE (Greiner, 2004). In total 129 employeremployee pairs participated (258 subjects in total), split into 35 employer–employee pairs in the Prob0, 34 pairs in the Prob100 and 60 pairs in the Prob50 (30 pairs with shock realization (Prob50shock) and 30 without (Prob50noshock)) treatment. A session lasted approximately 60 minutes. All decisions were made on a computer screen using z-Tree (Fischbacher, 2007). During the sessions, a fictitious experimental currency called “Taler” was used. The exchange rate of 1 Taler = 0.15 Euro was announced in advance. Talers were converted into Euros at the end of the experiment when we paid the subjects in private. On average, the subjects earned 14.81 Euro (s.d. 3.15 Euro, min: 6.20 Euro, max: 21.65 Euro) including a show-up fee of 5 Euro. 3. Hypotheses Before presenting the results of our experiment, this section provides intuitions and predictions on how employers and employees might behave in our treatments. The predictions under standard assumptions of payoff maximizing individuals are straightforward. As the game is finitely repeated and wage and effort choices above the minimum level are costly, neither employers nor employees should choose positive wage or effort levels in any treatment. The resulting payoffs are
3 4
See for instance Li and Matouschek (2013), who study the impact of asymmetric information in relational contracts theoretically. The translated instructions can be found in the Appendix.
232
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
Table 2 Average wages and effort levels across treatments. Period 1
Period 2
Wage
Effort
Reciprocity
Wage
Effort
Reciprocity
Prob0
6.286 (2.270)
6.486 (2.853)
0.521
0.200 (2.530)
5.714 (3.130)
3.629 (3.565)
0.639
−2.086 (2.726)
Prob50
6.817 (2.228)
0.669
−0.467 (1.761)
6.833 (2.379) 5.333 (2.796)
6.100 (3.055) 5.133 (3.126)
0.630
7.283 (1.896)
−0.733 (2.532) −0.200 (2.722)
7.059 (2.558)
6.235 (3.153)
0.638
5.618 (2.913)
4.676 (3.591)
0.648
Prob100
−0.824 (2.263)
Prob50noshock Prob50shock
0.537
−0.941 (2.806)
The table displays mean values of chosen wages, efforts and reciprocity, defined as effortt − waget , t = 1, 2. Corresponding standard deviations are given in parentheses. Spearman’s rank correlation coefficients for pairwise correlations between wages and efforts are presented in the columns titled “”. All these correlations are highly significant with p < 0.01.
Pareto-dominated by more cooperative play and there is considerable scope for cooperation to improve joint payoffs. Numerous experiments have shown that many individuals do not act in a purely selfish manner, but reveal social preferences (for an overview see Fehr and Schmidt, 2006). Particularly interesting for our study are experimental findings based on Akerlof ’s (1982) gift-exchange model showing that the norm of reciprocity is an effective contract enforcement device in the presence of incomplete contracts (see among others Fehr et al., 1993, 1997). The main result in these studies is that workers’ efforts are positively related to the wage level provided. For employers who anticipate that a non-negligible proportion of the population is endowed with reciprocal preferences,5 it pays off to choose generous wages, in order to appeal to their employees’ reciprocity. To formulate our hypotheses we refer to a simple model of reciprocity (presented in the Appendix), which is based on the assumption that at least some employees exhibit inequity aversion in the sense of Fehr and Schmidt (1999). This model does not differentiate between the potential reasons for the employees’ inequity aversion and we do not want to claim that inequity aversion is the sole driver behind reciprocal behavior. The model is, however, easily applicable to our context and captures the most important aspects of reciprocal behavior. It is based on the assumption that inequity averse employees reciprocate high wages with high efforts in order to reduce their utility loss from gaining a larger profit than their employer that would result if they did not behave this way. Inequity averse agents will therefore always choose positive efforts in response to positive wages – in both periods of the game. In the pooling equilibrium that we derive, also selfish agents choose positive efforts in the first period. They imitate the behavior of their inequity averse counterparts in order to receive high wages in the second period (which they respond to with zero effort choices then). Based on this simple benchmark model, we expect that, on average, wages and efforts are positively correlated. Since we do not expect the fraction of inequity averse agents to change with the possibility or the actual occurrence of the crisis, the model does not predict differences in employees’ reciprocal behavior across treatments. We therefore state our first hypothesis in a general manner: Hypothesis 1.
Wages and efforts are on average positively correlated in all treatments and in both periods of the game.
Moreover, due to the employees’ comparatively lower marginal productivity in case of shock realization the benchmark model predicts that employers are more reluctant to pay high wages in case they are hit by the adverse shock in period 2. From this we derive our second hypothesis: Hypothesis 2.
Employers pay on average lower wages in case they are hit by the adverse shock in period 2.
In the following section we will test these hypotheses and present and discuss the corresponding results as well as further findings in light of existing studies. 4. Results Table 2 presents summary statistics of chosen wages and efforts in both periods of the game. As becomes evident, both employers and employees choose on average positive wage and effort levels in both rounds. To quantify the employees’ degree of reciprocity, we consider two measures. First, we define the employment relation’s reciprocityt as the difference effortt − waget , for each period t = 1, b. Second, we calculate the Spearman’s rank correlation coefficients (Spearman’s ) between chosen wages and efforts. Both measures suggest a significant positive relationship between wage and effort choices in all periods and all treatment variants (in fact, all Spearman’s rank correlations are highly significant with p < 0.01). This finding is in line with previous gift-exchange experiments and confirms our Hypothesis 1.
5
Fischbacher et al. (2001), for instance, find that approximately half of their experimental population are conditional cooperators.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
233
Fig. 1. Wages and corresponding effort choices in period 2, by treatment variant.
In the following, we will analyze the wage and effort decisions in more detail. We begin with a discussion of employers’ and employees’ second period behavior in Section 4.1 and study their first period behavior in Section 4.2. 4.1. Gift-exchange in the second period To start with, we compare behavior of employers and employees in the two variants of the Prob50 treatment. They have – in terms of expectations – the same history from period 1 and offer a clean way to study the direct behavioral effects of the adverse shock. This enables us to cleanly test Hypothesis 2. Second, we contrast second period behavior in the Prob50shock treatment variant and the Prob100 treatment. These treatments have ex-post identical payoff structures, which allows us to study whether a shock that occurs with certainty yields different effects compared to a shock that was expected to occur with 50% probability. Third and finally, we study whether there is any effect of having been threatened, but ultimately not being hit by the exogenous shock. For this, we compare second period behavior in the Prob50noshock treatment variant and the Prob0 treatment. Since the ex-post realized payoff functions are identical across the respective treatment variants, neither the standard model of payoff maximizing individuals nor our benchmark model of inequity aversion predict treatment differences for the latter two comparisons. It is an empirical question if the difference in announced shock probabilities and whether being threatened by a shock in the first place, triggers different sets of behaviors. 4.1.1. Prob50shock vs. Prob50noshock In line with Hypothesis 2, Table 2 as well as a comparison of the two lower histograms in Fig. 1(a) reveals that employers pay on average significantly lower wages when they are hit by the shock (Mann–Whitney rank-sum test result: p = 0.0456 ). Nonetheless, their employees do not reduce effort in case of a shock. Mann–Whitney rank-sum tests reveal neither for the employees’ effort choices, nor for their degree of reciprocity significant treatment differences (Mann–Whitney rank-sum test results: p = 0.237 and p = 0.246, respectively; see also the descriptive statistics in Table 2 and the two lower histograms of Fig. 1(b). We summarize these findings as follows: Summary 1. Comparing behavior in the two Prob50 treatment variants, we find that employers pay on average lower wages if they are hit by the adverse shock in the second period. Employees’ reciprocity, on the other hand, is not significantly affected by the realization of the adverse shock. The employees’ effort choices suggest preferences for equity that could potentially be captured in a more general model of inequity aversion. Their behavior corroborates observations from several interview studies which indicate that employees are willing to help their firms in times of severe economic distress by keeping up their work morale despite wage cuts (Bewley, 1999, 2005; Charness and Levine, 2002; Kahneman et al., 1986). The observed effort choices are also in line with findings from lab experiments on first movers’ intentions (Charness, 2004; Falk and Fischbacher, 2006) and field experiments on wage reductions (e.g. Chen and Horton, 2009) suggesting that cutting wages does not necessarily lead to reduced effort provision, as long as the employer can credibly claim that she was forced to take this step for exogenous reasons.
6
Note that all reported p-values are two-sided unless clearly stated otherwise.
234
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
4.1.2. Prob50shock vs. Prob100 In a next step we study whether a shock that occurs with certainty yields different effects than a shock that was predicted to occur with 50% probability. When comparing the Prob100 and the Prob50shock treatment variant, we find, however, neither for employers’ wages nor for employees’ reciprocity a significant treatment effect (Mann–Whitney rank-sum tests results: p = 0.645 and p = 0.334, respectively). Employers’ and employees’ behavior in period 2 seems to be mainly driven by the experience of the shock and less by the previously announced probability of its occurrence. 4.1.3. Prob50noshock vs. Prob0 Lastly, we consider whether there is any behavioral effect of having been threatened, but finally not being hit by the exogenous shock. An inspection of Table 2 and the two left histograms in Fig. 1(a) shows that wages are not significantly differently distributed in the respective Prob50noshock and the Prob0 treatment (Mann–Whitney rank-sum test result: p = 0.170). Interestingly however, employees nevertheless react significantly different to these wages. In the Prob0 treatment, we observe the usual sharp drop in reciprocal gift-exchange in the final period of the game, which is in line with our simple benchmark model. Only inequity averse employees provide high efforts throughout the game, their selfish counterparts lower their efforts in the second period. In the Prob50noshock treatment variant, conversely, the employees’ average degree of reciprocity is more centered around 0, indicating that employees choose on average the same effort level as they have received as wage. Potentially, this behavior could be explained by a more general model of inequity aversion. The difference in the distributions of employees’ reciprocity is significant according to a Mann–Whitney rank-sum test (p = 0.037).7 Interestingly, overall, Fig. 1(b) reveals that in all treatments, except for the Prob0 treatment, the distribution of the reciprocity measure is virtually centered around 0. In the Prob0 treatment, on the contrary, the mass of the distribution lies below 0.8 Summary 2. In all treatment variants, in which the employer faced the exogenous shock (with their treatment-specific probabilities) in the first period, their employees behave on average significantly more reciprocal in the second period, for given wages. In Section 4.3 we study to what extent the information about the (possibility of an) exogenous shock shapes period 2 effort decisions and whether the employers’ wage decisions play an additional role. 4.2. Gift-exchange in the first period To study how the mere possibility of an exogenous shock affects reciprocal relationships, we compare first-period wages and effort levels from the Prob0, Prob50 and Prob100 treatments. In terms of first-period payoff functions, all treatments are the same. Therefore, neither the standard model of payoff maximizing individuals nor our benchmark model of inequity aversion predict behavioral differences across treatments. The treatments do, however, differ in the fact that employers in the latter two treatments are potentially hit by a shock in the second period with their respective probabilities. One could hence assume that employers in the Prob50 and Prob100 treatments pay higher first period wages since they are (or at least perceive to be) more reliant on the generous behavior of their employees in the second period and want to set the ground for a well-functioning gift-exchange relation. Higher first period wages in the Prob50 and Prob100 treatments would also be in line with other experimental studies showing that investments into gift-exchange relations increase in the potential for future benefits of cooperation. Gächter et al. (2008), for instance, find significantly higher early cooperation rates in a 50-round public good game than in a 10-round version of the game. In a similar vein, Andreoni and Samuelson (2006) observe that in case the payoffs in a twice-played prisoners’ dilemma are increased in the second round as compared to the first round, subjects start with significantly higher cooperation rates in the first round of the game. It is an empirical question whether employers in our experiment entertain similar considerations and pay comparatively high period 1 wages in the Prob50 and Prob100 treatments compared to the Prob0 treatment to lay a foundation for a well-functioning gift-exchange relationship. An inspection of Fig. 2(a) reveals that the distribution of wages in the Prob50 treatment is significantly shifted to the right compared to the Prob0 treatment (Mann–Whitney rank-sum test result: p = 0.027), while the distributions are virtually identical in the Prob50 and Prob100 treatments (Mann–Whitney rank-sum test result for the latter comparison: p = 0.920, see Table 2 for further descriptive statistics). We thus find that employers from both the Prob50 and Prob100 treatment pay higher wages in the first period, suggesting that they try to appeal to their employees’ social preferences in order to increase their cooperativeness in the next period. As we have seen in Section 4.1 above, the employers expectations about their employees’ reciprocity are on average correct, such that their relatively high first period wage choices actually pay off. Summary 3. Employers, who face a positive probability of being hit by the shock in the second period, pay significantly higher wages in period 1. There is no significant difference with regard to whether the shock occurs for certain or only with 50% probability.
7 A further Mann–Whitney rank-sum test confirms that also the difference in effort levels across the Prob0 and the Prob50nochock treatment variant is highly significant (p = 0.005). 8 Additional Mann–Whitney-rank-sum test results from treatment comparisons: p = 0.047 for Prob0 vs. Prob100, p = 0.004 for Prob0 vs. Prob50shock.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
235
Fig. 2. Wages and reciprocity in period 1, by treatment variant.
Table 3 Intertemporal link. Dep. variable: Effort in period 2 No shock
Shock realized
Wage in period 1
0.572*** (0.196)
0.505*** (0.181)
Prob50 Treatment
1.968** (0.806)
0.285 (0.811)
Constant
0.036 (1.199)
1.115 (1.274)
Observations R2
65 0.234
64 0.122
OLS regressions with Huber White robust standard errors given in parantheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Model 1 considers only observations from the Prob0 and from the Prob50 treatment, which are not hit by the shock. Model 2 considers only observations from the Prob100 and Prob50 treatment, which are hit by the shock.
It remains to be determined whether employees react to the relatively high wages in the Prob50 and Prob100 treatments already in period 1 or only later, i.e. in period 2. Fig. 2(b) depicts the distributions of our reciprocity measure in the first period. As becomes evident, there is no significant difference in employees’ reciprocity in the first period across treatments (Mann–Whitney rank-sum test results: p = 0.343 for Prob0 vs. Prob50, p = 0.344 for Prob50 vs. Prob100). This indicates that employees in the Prob50 and Prob100 treatments do not help their employers beforehand by choosing excessively high efforts, but – if at all – only do so in the second period of the game when the shock actually occurs.
4.3. Inter-temporal link between first and second period behavior and the role of risk preferences The question remains whether it is the more generous period 1 wages or rather the fact that the employers faced (the possibility of) an adverse shock, that explains the considerably more reciprocal behavior of employees in the second period of the Prob50 and Prob100 treatments compared to the Prob0 treatment. To provide an answer, we regress second period effort choices on period 1 wages and treatment dummies. A significant coefficient for the first-period wage in this regression would suggest that high initial wages trigger more generous effort choices in period 2 – even when controlling for any additional treatment effect. Similarly, a significant coefficient for the treatment dummies would imply that information about the probability of a future shock has an independent effect on reciprocal behavior in period 2 – controlling for any additional wage effects. In the first specification presented in Table 3 we consider only employment relations from the Prob0 treatment and the Prob50noshock treatment variant. These treatments have ex-post identical payoff structures and differ only in the fact that employers in the Prob50noshock treatment variant have ex-ante been threatened to be hit by the shock. The estimated
236
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
coefficients indicate that both higher period 1 wages as well as the threat of a potential shock have independent, positive effects on employees’ second period effort choices.9 In the regression presented in the second column of Table 3, we consider only employer–employee pairs from the Prob100 treatment and the Prob50shock treatment variant. Ex-post, these treatments have identical payoff structures. However, while the occurrence of the shock was anticipated with certainty in the Prob100 treatment it was expected with only 50% probability in the Prob50shock treatment variant. Again we find a significantly positive influence of first period wages on second period effort levels. The differing probabilities of the shock occurrence in the two treatments do not additionally significantly affect employees’ effort choices. In further regressions, presented in Table A.1 in the Appendix, we modify and extend the regressions from Table 3 by including period 2 wages. Most importantly, the coefficient of the treatment dummy remains significant when considering those employment relations that are not hit by the shock in period 2. This emphasizes the independent, positive effect of previously having been threatened by the shock on last period effort choices. The coefficients of period 1 wages, on the other hand, turns insignificant once we control for period 2 wages. This suggests that the observed intertemporal wage effect works indirectly through the high correlation of wages: High period 1 wages are followed by high period 2 wages, which in turn positively stimulate the employees’ effort decision in period 2. A further interesting question is in how far the observed behavior is driven by the subjects’ risk preferences. In this regard, our experiment adds to the literature on the impact of risk attitudes in social dilemma games. So far, the evidence in the literature is rather mixed. Eckel and Wilson (2004), for instance, provide an overview of the existing studies and find only a weak relationship between self-reported risk preferences and trust in their own trust game experiment. We study the impact of employers’ risk preferences on their first and second period wages in two separate OLS regressions. In the first specification, we regress first period wages on employers’ self-reported willingness to take risk, two treatment dummies and interaction terms between the treatment dummies and the employers’ risk preferences. As presented in column 1 in Table A.2, employers’ wages significantly increase in their willingness to take risks. This correlation does not differ significantly across treatments. However, when we run a similar regression, using wages in period 2 as dependent variable, we do not find a significant effect of employers’ risk preferences any longer. This suggests that risk preferences shape wage decisions only if the employers have not yet been able to gather information about their matched employee’s type. As soon as they have observed their employees’ effort decisions in period 1, employers’ risk preferences play less of a role in their subsequent wage decision in period 2. We expect that employees similarly deduce information about their employer’s type from the wage they receive, which ultimately diminishes the impact of their own risk attitudes on their effort choices. Therefore, we refrained from replicating the above regression analysis for the employees’ effort decisions. Furthermore, one could argue that the employees’ choices are not so much affected by their general willingness to take risks (for themselves), but rather by their willingness to take risks for others – a phenomenon more closely inspected by, for instance, Charness and Jackson (2009) and Pahlke et al. (2015). These studies consistently show that individuals who are responsible for another players’ (positive) payoffs take more risk averse decisions. Since we did not elicit the employees’ risk attitudes when taking risky decisions on behalf of others, we can only assume that similar forces are also at work in our experiment. 4.4. The impact on profits As a last step, we investigate how the introduction of a (potential) exogenous shock affects the realized profits. For this purpose, we compare the sum of employer’s and employee’s total payoffs from both periods across treatments. The mean total profits amount to 128.457 (s.d. 38.681) in the Prob0, 136.783 (s.d. 29.986) in the Prob50 and 122.662 (s.d. 34.842) in the Prob100 treatment.10 Interestingly, according to Mann–Whitney rank-sum tests the distributions of total profits do not differ significantly across treatments. Only the difference between the Prob50 and Prob100 turns out to be borderline significant.11 This simple analysis of the total payoffs has, however, to be taken with a grain of salt since the productivity factors in the employers’ payoff function differ between treatments. In order to account for this, we introduce a more sound total profit measure. It quantifies how close employment relations get to the treatment-specific efficiency frontier that depends on the ex-post realized payoff functions. This frontier is reached if both parties invest their full endowment in both periods. We call this variable realized efficiency and define it as follows:12 realized efficiency =
sum of actual employer s and employee s total earnings
sum of maximum feasible employer s and employee s total earnings
.
9 A similar positive effect of high initial wages has been made in Brown et al.’s (2004) 15-period gift-exchange experiment. The authors show that high effort levels in the last period of the game can be explained by a successful reciprocal employer–employee relationship in previous periods. 10 In all treatments the employees receive the major part of the total profits. The average employers’ share amounts to 44.962% (s.d. 9.458%) in the Prob0, 45.283% (s.d. 6.640%) in the Prob50 and 40.824% (s.d. 9.351%) in the Prob100 treatment. 11 Comparing the distribution of total profits in the Prob0 and the Prob50 treatment: Mann–Whitney rank-sum test result: p = 0.305; comparing Prob0 and Prob100: p = 0.653; comparing Prob50 and Prob100: p = 0.092. 12 To give an example: For the treatment variants without shock realization in the second period the maximum achievable total profit amounts to 200 points. If both parties choose wage and effort levels of 5 in both periods, the actual total profit equals 120 points. The realized efficiency is thus 120/200 = 60%.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
237
Fig. 3. Realized efficiency, by treatment.
The box and whiskers diagram in Fig. 3 illustrates the distribution of this measure across treatments. The upper and lower limits of the boxes indicate the 75th and 25th percentile, the horizontal lines inside the boxes denote the median. While the employment relations in the Prob0 treatment realize on average 64.229% (s.d. 19.341%) of the maximum feasible total profit, the corresponding numbers for the Prob50 and Prob100 treatments are 72.895% (s.d. 14.894%) and 70.092% (s.d. 19.910%). Comparing the Prob50 and the Prob0 treatment, we thus observe that the employer’s threat of being hit by the shock has a significantly positive impact on realized efficiency (Mann–Whitney rank-sum test result: p = 0.044). The realized efficiency in the Prob50 and Prob100 treatments, on the other hand, is statistically indistinguishable (Mann–Whitney rank-sum test: p = 0.708). We summarize our findings as follows: Summary 4. Comparing the Prob50 and Prob100 treatments to the Prob0 treatment, the announcement of (the possibility of) an exogenous shock has on average no negative effect on total profits. It even has a positive effect on realized efficiency when taking the treatment specific payoff functions into account.
5. Conclusion In our two period gift-exchange experiment, employers who face the possibility of being hit by a shock in the second period offer significantly higher wages in the first period. This finding holds irrespective of whether the probability of the later shock realization is 50% or whether the shock occurs with certainty. With these comparatively high wages, employers arguably make an investment into a cooperative relationship with their employees, who indeed reciprocate their decision. Employees in both, the Prob50 and the Prob100 treatment, reward the behavior of their employers by choosing an exceptionally high effort level in the final period of the game – even if their employers are not hit by the shock. At least in the framework of the experiment, the exogenous shock thus does not necessarily have a negative impact on total profits. In a broader perspective, our findings suggest that exogenous adverse shocks discipline firms in reality to provide “gifts” to their employees – be it in the form of adequate wages, fringe benefits or a good working environment. By doing so they can set the ground for a cooperative relation with their employees that they can draw on in times of economic well-being, but also, and more importantly so, in times of economic distress. Moreover, our findings are supportive of the view that firms should be transparent about the economic situation of the firm, for instance, by ‘opening the books’ to works councils, as suggested by Freeman and Lazear (1995), Englmaier and Segal (2011). When firms communicate potential upcoming economic shocks early on to their employees, their employees’ general cooperativeness should increase in relatively stable economic times and could be benefited from in times of economic distress. It remains to be empirically determined whether this effect is ultimately sufficiently strong to mitigate or even overcome potential negative side effects such as increasing employee turnover resulting from such candid communication strategies. Our simple experimental design allowed us to cleanly study how exogenous shocks affect reciprocal behavior in employment relations. Obviously, this abstract and controlled design comes at the cost of realism. One could, however, easily adapt and extend it in order to tackle various further issues. Extended versions of our design could, for instance, shed light on the question of whether the relatively high cooperation rates in the Prob50 and the Prob100 treatments persist in the long run. In our present experimental setup, economic shocks occur only once and their realizations are perfectly observed by both employers and employees. It would be interesting to see whether the reciprocity enhancing effects of warnings about economic shocks stay persistent over time in experimental designs with n > 2 periods (or in the field). Further extensions of our design could allow for dismissals, changing the information structure of the game by announcing the actual occurrence
238
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
of the shock only to the employers, or introducing an additional free-form communication (that is, a chat) stage into the design. We leave this to future research. Appendix A. Additional tables
Table A.1 Intertemporal link. Dep. variable: Effort in period 2 No shock
Shock realized
(1)
(2)
(3)
(4)
0.766*** (0.118) 1.615** (0.683) −0.747 (0.542)
−0.243 (0.192) 0.888*** (0.123) 1.692** (0.693) 0.081 (0.965)
0.731*** (0.099) 0.665 (0.671) 0.569 (0.679)
−0.163 (0.167) 0.818*** (0.130) 0.745 (0.699) 1.235 (0.907)
65 0.487
65 0.498
64 0.385
64 0.392
Wage in period 1 Wage in period 2 Prob50 Treatment Constant Observations R2
OLS regressions with Huber White robust standard errors given in parentheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Model 1 and 2 consider only observations from the Prob0 and from the Prob50 treatment, which are not hit by the shock. Model 3 and 4 consider only observations from the Prob100 and Prob50 treatment, which are hit by the shock.
Table A.2 The role of risk preferences. Dep. variable:
Wage in period 1
Wage in period 2
Risk preferences
0.229** (0.113) 0.769 (1.371) 0.711 (1.471) −0.304 (0.244) −0.155 (0.259) 5.910*** (0.675)
−0.025 (0.192) −0.031 (1.924) −1.323 (1.943) −0.068 (0.349) 0.145 (0.334) 6.236*** (1.115)
129 0.055
129 0.009
Prob0 Treatment Prob100 Treatment Prob 0 × Risk preferences Prob 100 × Risk preferences Constant Observations R2
OLS regressions with (Huber White) robust standard errors given in parantheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Dependent variables: Chosen wages. Risk preferences are measured on a 10 point scale, 1 representing risk averse, 10 representing risk loving. In both models we use the Prob50 treatment as reference treatment.
Appendix B. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.jebo.2016.12.021. References Agell, J., Lundborg, P., 1995. Theories of pay and unemployment: survey evidence from Swedish manufacturing firms. Scand. J. Econ. 97 (2), 295–307. Akerlof, G.A., 1982. Labor contracts as partial gift exchange. Q. J. Econ. 97 (4), 543–569. Andreoni, J., Samuelson, L., 2006. Building rational cooperation. J. Econ. Theory 127 (1), 117–154. Bewley, T.F., 1999. Why Wages don’t Fall During a Recession. Harvard Univ. Press, Cambridge, MA. Bewley, T.F., 2005. Fairness, Reciprocity, and Wage Rigidity. MIT Press, Cambridge, MA. Blinder, A.S., Choi, D.H., 1990. A shred of evidence on theories of wage stickiness. Q. J. Econ. 105 (4), 1003–1015. Brandts, J., Charness, G., 2004. Do labour market conditions affect gift exchange? Some experimental evidence. Econ. J. 114 (497), 684–708. Brown, M., Falk, A., Fehr, E., 2004. Relational contracts and the nature of market interactions. Econometrica 72 (3), 747–780. Campbell, C.M., Kamlani, K.S., 1997. The reasons for wage rigidity: evidence from a survey of firms. Q. J. Econ. 112 (3), 759–789. Charness, G., 2004. Attribution and reciprocity in an experimental labor market. J. Labor Econ. 22 (3), 665–688.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
239
Charness, G., Jackson, M.O., 2009. The role of responsibility in strategic risk-taking. J. Econ. Behav. Organ. 69 (3), 241–247. Charness, G., Levine, D.I., 2002. Changes in the employment contract? Evidence from a quasi-experiment. J. Econ. Behav. Organ. 47 (4), 391–405. Chen, D.L., Horton, J.J., 2009. The Wages of Pay Cuts: Evidence from a Field Experiment. Working paper. Harvard University. Cohn, A., Fehr, E., Herrmann, B., Schneider, F., 2014. Social comparison and effort provision: evidence from a field experiment. J. Eur. Econ. Assoc. 12 (4), 877–898. Davis, B., Kerschbamer, R., Oexl, R., 2016. Beliefs in a Gift Exchange Game with Random Shocks. MIMEO, University of Innsbruck. Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G.G., 2011. Individual risk attitudes: measurement, determinants, and behavioral consequences. J. Eur. Econ. Assoc. 9 (3), 522–550. Eckel, C.C., Wilson, R.K., 2004. Is trust a risky decision? J. Econ. Behav. Organ. 55 (December (4)), 447–465. Englmaier, F., Segal, C., 2011. Conflict and Cooperation in Organizations. Tech. rep., CEPR Working Paper. Falk, A., Fischbacher, U., 2006. A theory of reciprocity. Games Econ. Behav. 54 (2), 293–315. Fehr, E., Kirchsteiger, G., Riedl, A., 1993. Does fairness prevent market clearing? An experimental investigation. Q. J. Econ. 108 (2), 437–459. Fehr, E., Schmidt, K.M., 1999. A theory of fairness, competition, and cooperation. Q. J. Econ. 114 (3), 817–868. Fehr, E., Schmidt, K.M., 2006. The economics of fairness, reciprocity and altruism – experimental evidence and new theories. In: Serge-Christophe, K., Jean Mercier, Y. (Eds.), In: Handbook on the Economics of Giving, Reciprocity and Altruism: Foundations, 1. Elsevier, pp. 615–691. Fehr, E., Simon, G., Kirchsteiger, G., 1997. Reciprocity as a contract enforcement device: experimental evidence. Econometrica 65 (4), 833–860. Fischbacher, U., 2007. z-Tree: Zurich toolbox for ready-made economic experiments. Exp. Econ. 10 (2), 171–178. Fischbacher, U., Gächter, S., Fehr, E., 2001. Are people conditionally cooperative? Evidence from a public goods experiment. Econ. Lett. 71 (3), 397–404. Freeman, R., Lazear, E., 1995. An economic analysis of works councils. In: Works Councils: Consultation, Representation, and Cooperation in Industrial Relations. National Bureau of Economic Research Comparative Labor Markets Series. University of Chicago Press, Chicago and London, pp. 27–50 (Chapter). Gächter, S., Renner, E., Sefton, M., 2008. The long-run benefits of punishment. Science 322 (5907), 1510. Greiner, B., 2004. The Online Recruitment System ORSEE – A Guide for the Organization of Experiments in Economics. Papers on Strategic Interaction 2003–10. Max Planck Institute of Economics, Strategic Interaction Group. Kahneman, D., Knetsch, J.L., Thaler, R., 1986. Fairness as a constraint on profit seeking: entitlements in the market. Am. Econ. Rev. 76 (4), 728–741. Kocher, M.G., Strasser, S., 2011. The Fair Employment Hypothesis: Reciprocity in Unstable Environments. Working paper. Ludwig Maximilians University Munich. Krüger, M., Friebel, G., 2016. What are the Costs of a Pay Cut? Evidence from a Personnel Consulting Firm. Working Paper. Goethe-University Frankfurt. Kube, S., Marchal, M.A., Puppe, C., 2013. Do Wage Cuts Damage Work Morale? Evidence From a Natural Field Experiment. J. Eur. Econ. Assoc. 11 (4), 853–870. Li, J., Matouschek, N., 2013. Managing conflicts in relational contracts. Am. Econ. Rev. 103 (6), 2328–2351. Linardi, S., Camerer, C.T., 2012. Can Relational Contracts Survive Stochastic Interruptions? Experimental Evidence. Working Paper. University of Pittsburgh. Pahlke, J., Strasser, S., Vieider, F.M., 2015. Responsibility effects in decision making under risk. J. Risk Uncertain. 51 (2), 125–146. Rubin, J., Sheremeta, R., 2016. Principal-agent settings with random shocks. Manag. Sci. 62 (4), 985–999. Shapiro, F., 2006. The Yale Book of Quotations. Yale University Press.
Contents lists available at ScienceDirect
Journal of Economic Behavior & Organization journal homepage: www.elsevier.com/locate/jebo
In good times and bad – Reciprocal behavior at the workplace in times of economic crises夽 Leonie Gerhards a,∗ , Matthias Heinz b a b
University of Hamburg, Von-Melle-Park 5, 20146 Hamburg, Germany University of Cologne, Albert-Magnus Platz 1, 50923 Cologne, Germany
a r t i c l e
i n f o
Article history: Received 11 December 2015 Received in revised form 13 December 2016 Accepted 22 December 2016 Available online 29 December 2016 JEL classification: C9 J3 M5
a b s t r a c t This laboratory experiment analyzes employer–employee relations in the face of an exogenous shock. We implement a two-period gift-exchange game in which the employer can be hit by an adverse shock in the second period. We find that the mere possibility of the shock encourages employers to pay significantly higher wages in the first period. These higher first period wages translate into increased effort levels in the second period, independent of the actual occurrence of the shock. Our results suggest that the mere possibility of an exogenous shock strengthens the cooperation between individuals, which can ultimately mitigate its negative impact on total profits. © 2016 Elsevier B.V. All rights reserved.
Keywords: Gift-exchange Wage cuts Exogenous shock Reciprocity
“When written in Chinese the word crisis is composed of two characters. One represents danger, and the other represents opportunity.” John F. Kennedy (1959)1
1. Introduction In the presence of incomplete contracts, work morale, that is, the employees’ voluntary cooperation with their employers, crucially determines the success of employment relationships. In particular during economic downturns, this mechanism
夽 This work was financially supported by Goethe University Frankfurt. We would like to thank Johannes Abeler, Arno Apffelstaedt, Iwan Barankay, Leonie Baumann, Gary Charness, Ruben Durante, Florian Englmaier, Armin Falk, Guido Friebel, Simon Gächter, Lorenz Götte, Georg Kirchsteiger, Alexander Koch, Joep Konings, Michael Kosfeld, Julia Nafziger, Gerd Mühlheußer, Daniel Schunk, Heiner Schumacher, Ferdinand von Siemens, Joël van der Weele, and seminar participants at the MPI for Research on Collective Goods in Bonn as well as the conference audiences at the European Economic Association Meeting in Oslo 2011, at the Workshop on Behavioral Personnel Economics in Mannheim 2011 and at the Colloquium of Personnel Economics in Vienna 2015. Finally, we would like to thank the referees and editors whose comments allowed us to greatly improve the manuscript. ∗ Corresponding author. E-mail addresses: [email protected] (L. Gerhards), [email protected] (M. Heinz). 1 Quote taken from Shapiro (2006). http://dx.doi.org/10.1016/j.jebo.2016.12.021 0167-2681/© 2016 Elsevier B.V. All rights reserved.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
229
seems to be of utmost importance. In this paper we study whether reciprocal gift-exchange between firms and workers can help to overcome the negative effects of adverse economic shocks by sustaining cooperation in the face of rough times. We consider shocks that hit firms through no fault of their own. Examples are increasing input prices for raw materials, demand drops caused by political conflicts or terroristic attacks or decreases in selling prices due to over-capacities on the global market. Earlier interview studies, conducted with both managers and employees suggest that workers are willing to help their firms by accepting wage cuts in these kind of situations (see Agell and Lundborg, 1995; Bewley, 1999, 2005; Blinder and Choi, 1990; Campbell and Kamlani, 1997; Charness and Levine, 2002; Kahneman et al., 1986). The interview studies, however, also report that firms can only count on support from their workers after a wage cut if they are in severe trouble. This, in turn, disciplines their employers not to reduce wages in more stable economic times, leading to the aggregate phenomenon of wage rigidity. In our laboratory experiment, we implement the adverse shock as a clearly exogenous event that has a substantial impact on employers’ profits. While previous studies considered reciprocal gift-exchange as a means to overcome shocks, we study if and how (the possibility of) exogenous shocks affect gift-exchange. In particular, we analyze how the mere possibility of a shock that can hit the employer influences employment relations. We set out to answer the following questions: How do employers adjust their wages if they are facing the possibility of an adverse shock? How do their employees respond? And lastly, does reciprocal gift-exchange have the potential to mitigate the negative payoff consequences of a shock? In our two-period gift-exchange game, in each period, the employer first sets a wage and the employee subsequently chooses an effort level. Both wages and effort levels are costly and not contractible. Across treatments we vary the probability with which the employer is hit by an exogenous shock in the second period. Our first benchmark treatment, the Prob0 treatment, is a twice repeated standard gift-exchange game. Employers and employees simply take their respective wage and effort choices twice, in two subsequent periods. In our second benchmark treatment, the Prob100 treatments, the payoff functions in the first period are the same as in the Prob0 treatment. In the second period, conversely, the employer is hit by an exogenous shock which implies a reduction of the worker’s marginal productivity in the employer’s payoff function. In the main treatment, the Prob50 treatment, the exogenous shock occurs in the second period with 50% probability. If the shock hits the employer, the payoff functions are the same as in the Prob100 treatment. Otherwise, they are the same as in the Prob0 treatment. In all treatments both parties have full information about the probabilities and the actual realization of the shock. Our main findings can be summarized as follows: The mere possibility of an adverse shock (in the Prob50 and Prob100 treatment) encourages the employers to pay on average higher wages in the first period – irrespective of whether the shock later occurs with 50% probability or for certain. When the employers actually experience the shock in the second period, they pay comparably lower wages. Their employees nevertheless keep up reciprocity. Interestingly, we generally find that in all treatment variants, in which the employer faced the exogenous shock (with their treatment-specific probabilities) in the first period, their employees behave on average significantly more reciprocal in the second period Taken as a whole, our results suggest that the possibility of an exogenous shock can strengthen reciprocal gift-exchange between firms and workers. Ultimately, our findings can provide guidance for firms on how to establish a cooperative relationship with their workers that they can reap in times of economic distress. This seems particularly important for firms that operate in volatile markets. We discuss several policy implications of our findings in the conclusion of this paper. This paper contributes to various fields of research. First, our paper relates to the literature on employees’ reactions to wage cuts. Several field experiments and field studies show that employees decrease their performance as a response to a wage reduction (see for instance Cohn et al., 2014; Kube et al., 2013; Krüger and Friebel, 2016). However, this literature mainly studies employees’ reactions to wage cuts where the employer either does not explain the reasons for the wage reduction or where the reasons are arguably perceived as unfair. Our experiment, on the other hand, focuses on employees’ responses to lowered wages that can be attributed to worsened economic conditions. Consistent with previous interview studies (Agell and Lundborg, 1995; Bewley, 1999, 2005; Blinder and Choi, 1990; Campbell and Kamlani, 1997; Charness and Levine, 2002; Kahneman et al., 1986), we find that this type of justifiable wage cuts seems to trigger less negative reactions from the employees’ side. Furthermore, our study corroborates and extends findings from the literature on reciprocity based on Akerlof’s (1982) gift-exchange model showing that reciprocity is an effective contract enforcement device in the presence of incomplete contracts (see among others Fehr et al., 1993, 1997). The main result from these studies is that workers’ efforts are positively related to the positive wages provided. For employers who anticipate that a non-negligible proportion of the population is endowed with reciprocal preferences, it hence pays off to choose generous wages. Recent experiments have extended this workhorse model by introducing economic crises in the lab. Kocher and Strasser (2011), for instance, consider a 15-periods labor market and model economic cycles by varying the employees’ productivity over periods. They show that, despite the fact that firms cut wages in times of economic distress, workers reduce effort levels in these periods only slightly. In Linardi and Camerer’s (2012) setup, firms and workers interact on a 30-periods labor market. In each of the periods the firms can be hit by a publicly observable stochastic shock, which prevents hiring for the coming three periods. Similar to Kocher and Strasser (2011), they find that the implemented stochastic shocks do not have significant negative effects on wages and effort levels. The employment relations are found to be rather long-lasting and to be characterized by reciprocal gift-exchange of high efforts, high wages and job security in times of economic distress.
230
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
Table 1 Payoff functions across treatments. Period 1
Period 2 No shock
Prob0
1 = 10 − w + 5e 2 = 10 − e + 5w
1 = 10 − w + 5e 2 = 10 − e + 5w
Prob50
1 = 10 − w + 5e 2 = 10 − e + 5w
1 = 10 − w + 5e 2 = 10 − e + 5w
Prob100
1 = 10 − w + 5e 2 = 10 − e + 5w
Shock realized
1 = 10 − w + 2.5e 2 = 10 − e + 5w 1 = 10 − w + 2.5e 2 = 10 − e + 5w
Employer’s payoff: 1 , employee’s payoff: 2 ; Effort e {0, 1, . . ., 10}, wage w {0, 1, . . ., 10}.
In line with Linardi and Camerer (2012) and Kocher and Strasser (2011), also employees in our experiment do not respond with low effort choices if employers lower their wages in case they are hit by an exogenous shock. This seems noteworthy since, different from their experiments, in our study a shock can occur only once and workers cannot become unemployed. While previous experiments have either focused on gift-exchange in the absence of exogenous shocks or have analyzed gift-exchange in times of adverse shocks, we are, to the best of our knowledge, the first to study how the mere possibility of being hit by an exogenous shock affects behavior in an otherwise standard gift-exchange game.2 2. Experimental design and procedures 2.1. Experimental design We implement a two-period gift-exchange game in which we apply Brandts and Charness’s (2004) symmetric and linear payoff functions. In the beginning of the first period, both parties receive an endowment of 10. First, the employer chooses a wage w from the integer set {0, 1, . . ., 10} and specifies a desired, non-binding effort level, a feature commonly used in gift-exchange experiments (see for instance Fehr et al., 1997). The chosen wage is multiplied by 5 and transferred to the employee. Subsequently, the employee decides which effort level e from the integer set {0, 1, . . ., 10} she wants to provide. This effort level is then analogously multiplied by 5 and sent to the employer. The first period is identical in all treatments. In the second period the payoff functions differ across treatments. In our first benchmark treatment (Prob0) the endowments, procedures and payoff functions are the same as in the first period. In our second benchmark treatment (Prob100), on the other hand, an exogenous shock hits the employer with certainty. This shock implies a reduction of the employee’s marginal productivity from 5 to 2.5. Apart from that the payoff functions do not change. Both parties still receive an endowment of 10 at the beginning of the second period, and the employer’s wage choice is still multiplied by 5 before it is transferred to the employee. In the Prob50 treatment, finally, the exogenous shock is realized with 50% probability, that is, exactly half of the employer-employee relationships are randomly selected to be hit by the shock. If the shock occurs, the second period payoff functions are equal to those from the Prob100 treatment. If the shock does not occur, they are the same as in the Prob0 treatment. Table 1 summarizes the employers’ and employees’ payoff functions for all treatments. Subjects have full information about the procedures of the experiment. In particular, the fact that the employer is going to (or can potentially) be hit by an exogenous shock in the second period in the Prob100 (Prob50) treatment and the resulting payoff functions are explained to both parties before the start of the experiment. Moreover, at the beginning of period 2 the employer as well as the employee are informed of whether the shock actually occurs or not. 2.2. Distinctive features of our experimental design The adverse shocks that we have in mind for our study are any kind of exogenous economic shocks that can potentially hit a firm and have a severe impact on its profits. In our experiment the shock implies a substantial reduction of the employee’s marginal productivity. Our design could thus reflect situations in which firms have to deal with decreasing output prices (for instance, due to a new competitor entering the market or over-capacities on the global market) or conditions in which firms face increasing input prices for raw materials (for example, as a result of political conflicts). The common and decisive feature of these situations is that exogenous factors lead to severe economic problems which make the firm heavily dependent on support from its employees.
2 Relatedly, Rubin and Sheremeta (2016), Davis et al. (2016) consider more extended versions of the gift-exchange game in which, first, the principal sets a wage, then, the agent chooses an effort level that can subsequently be affected by an (unobservable) shock and, finally, the principal chooses a wage adjustment. These experimental studies consistently find that principals reward their agents both based on realized output and effort (in treatments in which it is observable) in the last stage of the game. However, while Rubin and Sheremeta (2016) find that the observability of the shock does not affect choices, Davis et al. (2016) find significant behavioral differences across the treatments in which principals observe either only effort or both effort and the occurrence of the shock.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
231
Another noteworthy feature in this context are the probabilities with which the shock occurs in our treatments (100% and 50%). Arguably, a shock that occurs for certain and a shock that occurs with 50% capture different problems of an employer. We mainly chose these probabilities for methodological reasons. The Prob100 treatment first and foremost serves as a benchmark treatment for the shock case. The Prob50 treatment, on the other hand, provides us with an easy way to study the impact of the threat of a shock. In the second period it splits up into two equally sized treatment variants – one variant with and one without shock realization (for simplicity, called Prob50shock and Prob50noshock in the following). Comparing behavior from each of these treatment variants to that from the respective benchmark treatment (Prob0 or Prob100) allows us to disentangle the influence of the ex-ante varying shock probabilities in the main and benchmark treatments from further effects resulting from the differing payoff functions. Moreover, by comparing employers’ and employees’ decisions of the Prob50shock and Prob50noshock treatment variants in period 2, we are able to study the influence of the shock on wages and efforts, holding constant the history from period 1. In order to study the impact of the shock in isolation, we minimized the complexity of the decision situation to the greatest possible extent. We opted for a fixed and exogenous partner matching and excluded the opportunity to lay off workers. This reduces the impact of confounding factors resulting from market interactions and the threat of being dismissed. Moreover, we inform both employers and employees about the probabilities and actual occurrence of the shock, which allows us to rule out potential tensions between contracting parties emerging from information asymmetries.3 The payoff functions that we borrowed from Brandts and Charness (2004) come with several additional advantages. First, due to their linearity, the marginal effect of effort on the employer’s profit is independent of the wage. Second, given the symmetry in all periods in which the employer is not hit by the shock, reciprocal employees can easily see that equality of wage and effort equalizes payoffs. And third, the fact that the shock only affects the employee’s marginal productivity in the employer’s payoff function makes it evident that the employer can choose from the same set of wages as in period 1. We simplified the design further by implementing a decision setting where a shock can occur only once. Although we acknowledge that a design with repeated shocks would allow subjects to learn from previous rounds and might hence increase their understanding of the situation, we are confident that our experimental game is the method of choice when studying the effects of announcing the possibility of an economic shock in isolation. The simplicity of the game already reduces the likelihood that observed treatment effects are mainly driven by subjects’ confusion and erroneous behavior. Hence, potential additional learning resulting from repeated interaction seem negligible. On the contrary, our setup allows us to study behavior in the face of an adverse shock undisturbed from any feedback or reputation building. 2.3. Procedures At the beginning of the session all subjects read the general instructions to the experiment in which we apply a natural framing using expressions such as “employer”, “employee” and “economic crisis” to ensure a common understanding of the experimental game.4 To clarify the payoff calculations, two examples are given in the instructions. Additionally, subjects have to answer several control questions concerning the sequence of actions during the experiment and the monetary consequences of their decisions. Once all subjects have answered all control questions correctly, the computer randomly assigns subjects to their roles (employer or employee) and matches them on a one-to-one basis. In the end of the session, after subjects have taken their decisions in the experimental game we administer a short survey, in which we elicit, among other things, subjects’ general willingness to take risk (“How willing are you to take risks, in general?”). Answers are given on a 10 point scale. We hence elicit risk attitudes similar to Dohmen et al. (2011), who provide evidence that individuals’ self-reported risk preferences are correlated with actual risk-taking behavior in the field. The experiment was conducted at the FLEX laboratory of Goethe University Frankfurt. All subjects were undergraduate students from different disciplines at the university and were recruited via ORSEE (Greiner, 2004). In total 129 employeremployee pairs participated (258 subjects in total), split into 35 employer–employee pairs in the Prob0, 34 pairs in the Prob100 and 60 pairs in the Prob50 (30 pairs with shock realization (Prob50shock) and 30 without (Prob50noshock)) treatment. A session lasted approximately 60 minutes. All decisions were made on a computer screen using z-Tree (Fischbacher, 2007). During the sessions, a fictitious experimental currency called “Taler” was used. The exchange rate of 1 Taler = 0.15 Euro was announced in advance. Talers were converted into Euros at the end of the experiment when we paid the subjects in private. On average, the subjects earned 14.81 Euro (s.d. 3.15 Euro, min: 6.20 Euro, max: 21.65 Euro) including a show-up fee of 5 Euro. 3. Hypotheses Before presenting the results of our experiment, this section provides intuitions and predictions on how employers and employees might behave in our treatments. The predictions under standard assumptions of payoff maximizing individuals are straightforward. As the game is finitely repeated and wage and effort choices above the minimum level are costly, neither employers nor employees should choose positive wage or effort levels in any treatment. The resulting payoffs are
3 4
See for instance Li and Matouschek (2013), who study the impact of asymmetric information in relational contracts theoretically. The translated instructions can be found in the Appendix.
232
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
Table 2 Average wages and effort levels across treatments. Period 1
Period 2
Wage
Effort
Reciprocity
Wage
Effort
Reciprocity
Prob0
6.286 (2.270)
6.486 (2.853)
0.521
0.200 (2.530)
5.714 (3.130)
3.629 (3.565)
0.639
−2.086 (2.726)
Prob50
6.817 (2.228)
0.669
−0.467 (1.761)
6.833 (2.379) 5.333 (2.796)
6.100 (3.055) 5.133 (3.126)
0.630
7.283 (1.896)
−0.733 (2.532) −0.200 (2.722)
7.059 (2.558)
6.235 (3.153)
0.638
5.618 (2.913)
4.676 (3.591)
0.648
Prob100
−0.824 (2.263)
Prob50noshock Prob50shock
0.537
−0.941 (2.806)
The table displays mean values of chosen wages, efforts and reciprocity, defined as effortt − waget , t = 1, 2. Corresponding standard deviations are given in parentheses. Spearman’s rank correlation coefficients for pairwise correlations between wages and efforts are presented in the columns titled “”. All these correlations are highly significant with p < 0.01.
Pareto-dominated by more cooperative play and there is considerable scope for cooperation to improve joint payoffs. Numerous experiments have shown that many individuals do not act in a purely selfish manner, but reveal social preferences (for an overview see Fehr and Schmidt, 2006). Particularly interesting for our study are experimental findings based on Akerlof ’s (1982) gift-exchange model showing that the norm of reciprocity is an effective contract enforcement device in the presence of incomplete contracts (see among others Fehr et al., 1993, 1997). The main result in these studies is that workers’ efforts are positively related to the wage level provided. For employers who anticipate that a non-negligible proportion of the population is endowed with reciprocal preferences,5 it pays off to choose generous wages, in order to appeal to their employees’ reciprocity. To formulate our hypotheses we refer to a simple model of reciprocity (presented in the Appendix), which is based on the assumption that at least some employees exhibit inequity aversion in the sense of Fehr and Schmidt (1999). This model does not differentiate between the potential reasons for the employees’ inequity aversion and we do not want to claim that inequity aversion is the sole driver behind reciprocal behavior. The model is, however, easily applicable to our context and captures the most important aspects of reciprocal behavior. It is based on the assumption that inequity averse employees reciprocate high wages with high efforts in order to reduce their utility loss from gaining a larger profit than their employer that would result if they did not behave this way. Inequity averse agents will therefore always choose positive efforts in response to positive wages – in both periods of the game. In the pooling equilibrium that we derive, also selfish agents choose positive efforts in the first period. They imitate the behavior of their inequity averse counterparts in order to receive high wages in the second period (which they respond to with zero effort choices then). Based on this simple benchmark model, we expect that, on average, wages and efforts are positively correlated. Since we do not expect the fraction of inequity averse agents to change with the possibility or the actual occurrence of the crisis, the model does not predict differences in employees’ reciprocal behavior across treatments. We therefore state our first hypothesis in a general manner: Hypothesis 1.
Wages and efforts are on average positively correlated in all treatments and in both periods of the game.
Moreover, due to the employees’ comparatively lower marginal productivity in case of shock realization the benchmark model predicts that employers are more reluctant to pay high wages in case they are hit by the adverse shock in period 2. From this we derive our second hypothesis: Hypothesis 2.
Employers pay on average lower wages in case they are hit by the adverse shock in period 2.
In the following section we will test these hypotheses and present and discuss the corresponding results as well as further findings in light of existing studies. 4. Results Table 2 presents summary statistics of chosen wages and efforts in both periods of the game. As becomes evident, both employers and employees choose on average positive wage and effort levels in both rounds. To quantify the employees’ degree of reciprocity, we consider two measures. First, we define the employment relation’s reciprocityt as the difference effortt − waget , for each period t = 1, b. Second, we calculate the Spearman’s rank correlation coefficients (Spearman’s ) between chosen wages and efforts. Both measures suggest a significant positive relationship between wage and effort choices in all periods and all treatment variants (in fact, all Spearman’s rank correlations are highly significant with p < 0.01). This finding is in line with previous gift-exchange experiments and confirms our Hypothesis 1.
5
Fischbacher et al. (2001), for instance, find that approximately half of their experimental population are conditional cooperators.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
233
Fig. 1. Wages and corresponding effort choices in period 2, by treatment variant.
In the following, we will analyze the wage and effort decisions in more detail. We begin with a discussion of employers’ and employees’ second period behavior in Section 4.1 and study their first period behavior in Section 4.2. 4.1. Gift-exchange in the second period To start with, we compare behavior of employers and employees in the two variants of the Prob50 treatment. They have – in terms of expectations – the same history from period 1 and offer a clean way to study the direct behavioral effects of the adverse shock. This enables us to cleanly test Hypothesis 2. Second, we contrast second period behavior in the Prob50shock treatment variant and the Prob100 treatment. These treatments have ex-post identical payoff structures, which allows us to study whether a shock that occurs with certainty yields different effects compared to a shock that was expected to occur with 50% probability. Third and finally, we study whether there is any effect of having been threatened, but ultimately not being hit by the exogenous shock. For this, we compare second period behavior in the Prob50noshock treatment variant and the Prob0 treatment. Since the ex-post realized payoff functions are identical across the respective treatment variants, neither the standard model of payoff maximizing individuals nor our benchmark model of inequity aversion predict treatment differences for the latter two comparisons. It is an empirical question if the difference in announced shock probabilities and whether being threatened by a shock in the first place, triggers different sets of behaviors. 4.1.1. Prob50shock vs. Prob50noshock In line with Hypothesis 2, Table 2 as well as a comparison of the two lower histograms in Fig. 1(a) reveals that employers pay on average significantly lower wages when they are hit by the shock (Mann–Whitney rank-sum test result: p = 0.0456 ). Nonetheless, their employees do not reduce effort in case of a shock. Mann–Whitney rank-sum tests reveal neither for the employees’ effort choices, nor for their degree of reciprocity significant treatment differences (Mann–Whitney rank-sum test results: p = 0.237 and p = 0.246, respectively; see also the descriptive statistics in Table 2 and the two lower histograms of Fig. 1(b). We summarize these findings as follows: Summary 1. Comparing behavior in the two Prob50 treatment variants, we find that employers pay on average lower wages if they are hit by the adverse shock in the second period. Employees’ reciprocity, on the other hand, is not significantly affected by the realization of the adverse shock. The employees’ effort choices suggest preferences for equity that could potentially be captured in a more general model of inequity aversion. Their behavior corroborates observations from several interview studies which indicate that employees are willing to help their firms in times of severe economic distress by keeping up their work morale despite wage cuts (Bewley, 1999, 2005; Charness and Levine, 2002; Kahneman et al., 1986). The observed effort choices are also in line with findings from lab experiments on first movers’ intentions (Charness, 2004; Falk and Fischbacher, 2006) and field experiments on wage reductions (e.g. Chen and Horton, 2009) suggesting that cutting wages does not necessarily lead to reduced effort provision, as long as the employer can credibly claim that she was forced to take this step for exogenous reasons.
6
Note that all reported p-values are two-sided unless clearly stated otherwise.
234
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
4.1.2. Prob50shock vs. Prob100 In a next step we study whether a shock that occurs with certainty yields different effects than a shock that was predicted to occur with 50% probability. When comparing the Prob100 and the Prob50shock treatment variant, we find, however, neither for employers’ wages nor for employees’ reciprocity a significant treatment effect (Mann–Whitney rank-sum tests results: p = 0.645 and p = 0.334, respectively). Employers’ and employees’ behavior in period 2 seems to be mainly driven by the experience of the shock and less by the previously announced probability of its occurrence. 4.1.3. Prob50noshock vs. Prob0 Lastly, we consider whether there is any behavioral effect of having been threatened, but finally not being hit by the exogenous shock. An inspection of Table 2 and the two left histograms in Fig. 1(a) shows that wages are not significantly differently distributed in the respective Prob50noshock and the Prob0 treatment (Mann–Whitney rank-sum test result: p = 0.170). Interestingly however, employees nevertheless react significantly different to these wages. In the Prob0 treatment, we observe the usual sharp drop in reciprocal gift-exchange in the final period of the game, which is in line with our simple benchmark model. Only inequity averse employees provide high efforts throughout the game, their selfish counterparts lower their efforts in the second period. In the Prob50noshock treatment variant, conversely, the employees’ average degree of reciprocity is more centered around 0, indicating that employees choose on average the same effort level as they have received as wage. Potentially, this behavior could be explained by a more general model of inequity aversion. The difference in the distributions of employees’ reciprocity is significant according to a Mann–Whitney rank-sum test (p = 0.037).7 Interestingly, overall, Fig. 1(b) reveals that in all treatments, except for the Prob0 treatment, the distribution of the reciprocity measure is virtually centered around 0. In the Prob0 treatment, on the contrary, the mass of the distribution lies below 0.8 Summary 2. In all treatment variants, in which the employer faced the exogenous shock (with their treatment-specific probabilities) in the first period, their employees behave on average significantly more reciprocal in the second period, for given wages. In Section 4.3 we study to what extent the information about the (possibility of an) exogenous shock shapes period 2 effort decisions and whether the employers’ wage decisions play an additional role. 4.2. Gift-exchange in the first period To study how the mere possibility of an exogenous shock affects reciprocal relationships, we compare first-period wages and effort levels from the Prob0, Prob50 and Prob100 treatments. In terms of first-period payoff functions, all treatments are the same. Therefore, neither the standard model of payoff maximizing individuals nor our benchmark model of inequity aversion predict behavioral differences across treatments. The treatments do, however, differ in the fact that employers in the latter two treatments are potentially hit by a shock in the second period with their respective probabilities. One could hence assume that employers in the Prob50 and Prob100 treatments pay higher first period wages since they are (or at least perceive to be) more reliant on the generous behavior of their employees in the second period and want to set the ground for a well-functioning gift-exchange relation. Higher first period wages in the Prob50 and Prob100 treatments would also be in line with other experimental studies showing that investments into gift-exchange relations increase in the potential for future benefits of cooperation. Gächter et al. (2008), for instance, find significantly higher early cooperation rates in a 50-round public good game than in a 10-round version of the game. In a similar vein, Andreoni and Samuelson (2006) observe that in case the payoffs in a twice-played prisoners’ dilemma are increased in the second round as compared to the first round, subjects start with significantly higher cooperation rates in the first round of the game. It is an empirical question whether employers in our experiment entertain similar considerations and pay comparatively high period 1 wages in the Prob50 and Prob100 treatments compared to the Prob0 treatment to lay a foundation for a well-functioning gift-exchange relationship. An inspection of Fig. 2(a) reveals that the distribution of wages in the Prob50 treatment is significantly shifted to the right compared to the Prob0 treatment (Mann–Whitney rank-sum test result: p = 0.027), while the distributions are virtually identical in the Prob50 and Prob100 treatments (Mann–Whitney rank-sum test result for the latter comparison: p = 0.920, see Table 2 for further descriptive statistics). We thus find that employers from both the Prob50 and Prob100 treatment pay higher wages in the first period, suggesting that they try to appeal to their employees’ social preferences in order to increase their cooperativeness in the next period. As we have seen in Section 4.1 above, the employers expectations about their employees’ reciprocity are on average correct, such that their relatively high first period wage choices actually pay off. Summary 3. Employers, who face a positive probability of being hit by the shock in the second period, pay significantly higher wages in period 1. There is no significant difference with regard to whether the shock occurs for certain or only with 50% probability.
7 A further Mann–Whitney rank-sum test confirms that also the difference in effort levels across the Prob0 and the Prob50nochock treatment variant is highly significant (p = 0.005). 8 Additional Mann–Whitney-rank-sum test results from treatment comparisons: p = 0.047 for Prob0 vs. Prob100, p = 0.004 for Prob0 vs. Prob50shock.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
235
Fig. 2. Wages and reciprocity in period 1, by treatment variant.
Table 3 Intertemporal link. Dep. variable: Effort in period 2 No shock
Shock realized
Wage in period 1
0.572*** (0.196)
0.505*** (0.181)
Prob50 Treatment
1.968** (0.806)
0.285 (0.811)
Constant
0.036 (1.199)
1.115 (1.274)
Observations R2
65 0.234
64 0.122
OLS regressions with Huber White robust standard errors given in parantheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Model 1 considers only observations from the Prob0 and from the Prob50 treatment, which are not hit by the shock. Model 2 considers only observations from the Prob100 and Prob50 treatment, which are hit by the shock.
It remains to be determined whether employees react to the relatively high wages in the Prob50 and Prob100 treatments already in period 1 or only later, i.e. in period 2. Fig. 2(b) depicts the distributions of our reciprocity measure in the first period. As becomes evident, there is no significant difference in employees’ reciprocity in the first period across treatments (Mann–Whitney rank-sum test results: p = 0.343 for Prob0 vs. Prob50, p = 0.344 for Prob50 vs. Prob100). This indicates that employees in the Prob50 and Prob100 treatments do not help their employers beforehand by choosing excessively high efforts, but – if at all – only do so in the second period of the game when the shock actually occurs.
4.3. Inter-temporal link between first and second period behavior and the role of risk preferences The question remains whether it is the more generous period 1 wages or rather the fact that the employers faced (the possibility of) an adverse shock, that explains the considerably more reciprocal behavior of employees in the second period of the Prob50 and Prob100 treatments compared to the Prob0 treatment. To provide an answer, we regress second period effort choices on period 1 wages and treatment dummies. A significant coefficient for the first-period wage in this regression would suggest that high initial wages trigger more generous effort choices in period 2 – even when controlling for any additional treatment effect. Similarly, a significant coefficient for the treatment dummies would imply that information about the probability of a future shock has an independent effect on reciprocal behavior in period 2 – controlling for any additional wage effects. In the first specification presented in Table 3 we consider only employment relations from the Prob0 treatment and the Prob50noshock treatment variant. These treatments have ex-post identical payoff structures and differ only in the fact that employers in the Prob50noshock treatment variant have ex-ante been threatened to be hit by the shock. The estimated
236
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
coefficients indicate that both higher period 1 wages as well as the threat of a potential shock have independent, positive effects on employees’ second period effort choices.9 In the regression presented in the second column of Table 3, we consider only employer–employee pairs from the Prob100 treatment and the Prob50shock treatment variant. Ex-post, these treatments have identical payoff structures. However, while the occurrence of the shock was anticipated with certainty in the Prob100 treatment it was expected with only 50% probability in the Prob50shock treatment variant. Again we find a significantly positive influence of first period wages on second period effort levels. The differing probabilities of the shock occurrence in the two treatments do not additionally significantly affect employees’ effort choices. In further regressions, presented in Table A.1 in the Appendix, we modify and extend the regressions from Table 3 by including period 2 wages. Most importantly, the coefficient of the treatment dummy remains significant when considering those employment relations that are not hit by the shock in period 2. This emphasizes the independent, positive effect of previously having been threatened by the shock on last period effort choices. The coefficients of period 1 wages, on the other hand, turns insignificant once we control for period 2 wages. This suggests that the observed intertemporal wage effect works indirectly through the high correlation of wages: High period 1 wages are followed by high period 2 wages, which in turn positively stimulate the employees’ effort decision in period 2. A further interesting question is in how far the observed behavior is driven by the subjects’ risk preferences. In this regard, our experiment adds to the literature on the impact of risk attitudes in social dilemma games. So far, the evidence in the literature is rather mixed. Eckel and Wilson (2004), for instance, provide an overview of the existing studies and find only a weak relationship between self-reported risk preferences and trust in their own trust game experiment. We study the impact of employers’ risk preferences on their first and second period wages in two separate OLS regressions. In the first specification, we regress first period wages on employers’ self-reported willingness to take risk, two treatment dummies and interaction terms between the treatment dummies and the employers’ risk preferences. As presented in column 1 in Table A.2, employers’ wages significantly increase in their willingness to take risks. This correlation does not differ significantly across treatments. However, when we run a similar regression, using wages in period 2 as dependent variable, we do not find a significant effect of employers’ risk preferences any longer. This suggests that risk preferences shape wage decisions only if the employers have not yet been able to gather information about their matched employee’s type. As soon as they have observed their employees’ effort decisions in period 1, employers’ risk preferences play less of a role in their subsequent wage decision in period 2. We expect that employees similarly deduce information about their employer’s type from the wage they receive, which ultimately diminishes the impact of their own risk attitudes on their effort choices. Therefore, we refrained from replicating the above regression analysis for the employees’ effort decisions. Furthermore, one could argue that the employees’ choices are not so much affected by their general willingness to take risks (for themselves), but rather by their willingness to take risks for others – a phenomenon more closely inspected by, for instance, Charness and Jackson (2009) and Pahlke et al. (2015). These studies consistently show that individuals who are responsible for another players’ (positive) payoffs take more risk averse decisions. Since we did not elicit the employees’ risk attitudes when taking risky decisions on behalf of others, we can only assume that similar forces are also at work in our experiment. 4.4. The impact on profits As a last step, we investigate how the introduction of a (potential) exogenous shock affects the realized profits. For this purpose, we compare the sum of employer’s and employee’s total payoffs from both periods across treatments. The mean total profits amount to 128.457 (s.d. 38.681) in the Prob0, 136.783 (s.d. 29.986) in the Prob50 and 122.662 (s.d. 34.842) in the Prob100 treatment.10 Interestingly, according to Mann–Whitney rank-sum tests the distributions of total profits do not differ significantly across treatments. Only the difference between the Prob50 and Prob100 turns out to be borderline significant.11 This simple analysis of the total payoffs has, however, to be taken with a grain of salt since the productivity factors in the employers’ payoff function differ between treatments. In order to account for this, we introduce a more sound total profit measure. It quantifies how close employment relations get to the treatment-specific efficiency frontier that depends on the ex-post realized payoff functions. This frontier is reached if both parties invest their full endowment in both periods. We call this variable realized efficiency and define it as follows:12 realized efficiency =
sum of actual employer s and employee s total earnings
sum of maximum feasible employer s and employee s total earnings
.
9 A similar positive effect of high initial wages has been made in Brown et al.’s (2004) 15-period gift-exchange experiment. The authors show that high effort levels in the last period of the game can be explained by a successful reciprocal employer–employee relationship in previous periods. 10 In all treatments the employees receive the major part of the total profits. The average employers’ share amounts to 44.962% (s.d. 9.458%) in the Prob0, 45.283% (s.d. 6.640%) in the Prob50 and 40.824% (s.d. 9.351%) in the Prob100 treatment. 11 Comparing the distribution of total profits in the Prob0 and the Prob50 treatment: Mann–Whitney rank-sum test result: p = 0.305; comparing Prob0 and Prob100: p = 0.653; comparing Prob50 and Prob100: p = 0.092. 12 To give an example: For the treatment variants without shock realization in the second period the maximum achievable total profit amounts to 200 points. If both parties choose wage and effort levels of 5 in both periods, the actual total profit equals 120 points. The realized efficiency is thus 120/200 = 60%.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
237
Fig. 3. Realized efficiency, by treatment.
The box and whiskers diagram in Fig. 3 illustrates the distribution of this measure across treatments. The upper and lower limits of the boxes indicate the 75th and 25th percentile, the horizontal lines inside the boxes denote the median. While the employment relations in the Prob0 treatment realize on average 64.229% (s.d. 19.341%) of the maximum feasible total profit, the corresponding numbers for the Prob50 and Prob100 treatments are 72.895% (s.d. 14.894%) and 70.092% (s.d. 19.910%). Comparing the Prob50 and the Prob0 treatment, we thus observe that the employer’s threat of being hit by the shock has a significantly positive impact on realized efficiency (Mann–Whitney rank-sum test result: p = 0.044). The realized efficiency in the Prob50 and Prob100 treatments, on the other hand, is statistically indistinguishable (Mann–Whitney rank-sum test: p = 0.708). We summarize our findings as follows: Summary 4. Comparing the Prob50 and Prob100 treatments to the Prob0 treatment, the announcement of (the possibility of) an exogenous shock has on average no negative effect on total profits. It even has a positive effect on realized efficiency when taking the treatment specific payoff functions into account.
5. Conclusion In our two period gift-exchange experiment, employers who face the possibility of being hit by a shock in the second period offer significantly higher wages in the first period. This finding holds irrespective of whether the probability of the later shock realization is 50% or whether the shock occurs with certainty. With these comparatively high wages, employers arguably make an investment into a cooperative relationship with their employees, who indeed reciprocate their decision. Employees in both, the Prob50 and the Prob100 treatment, reward the behavior of their employers by choosing an exceptionally high effort level in the final period of the game – even if their employers are not hit by the shock. At least in the framework of the experiment, the exogenous shock thus does not necessarily have a negative impact on total profits. In a broader perspective, our findings suggest that exogenous adverse shocks discipline firms in reality to provide “gifts” to their employees – be it in the form of adequate wages, fringe benefits or a good working environment. By doing so they can set the ground for a cooperative relation with their employees that they can draw on in times of economic well-being, but also, and more importantly so, in times of economic distress. Moreover, our findings are supportive of the view that firms should be transparent about the economic situation of the firm, for instance, by ‘opening the books’ to works councils, as suggested by Freeman and Lazear (1995), Englmaier and Segal (2011). When firms communicate potential upcoming economic shocks early on to their employees, their employees’ general cooperativeness should increase in relatively stable economic times and could be benefited from in times of economic distress. It remains to be empirically determined whether this effect is ultimately sufficiently strong to mitigate or even overcome potential negative side effects such as increasing employee turnover resulting from such candid communication strategies. Our simple experimental design allowed us to cleanly study how exogenous shocks affect reciprocal behavior in employment relations. Obviously, this abstract and controlled design comes at the cost of realism. One could, however, easily adapt and extend it in order to tackle various further issues. Extended versions of our design could, for instance, shed light on the question of whether the relatively high cooperation rates in the Prob50 and the Prob100 treatments persist in the long run. In our present experimental setup, economic shocks occur only once and their realizations are perfectly observed by both employers and employees. It would be interesting to see whether the reciprocity enhancing effects of warnings about economic shocks stay persistent over time in experimental designs with n > 2 periods (or in the field). Further extensions of our design could allow for dismissals, changing the information structure of the game by announcing the actual occurrence
238
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
of the shock only to the employers, or introducing an additional free-form communication (that is, a chat) stage into the design. We leave this to future research. Appendix A. Additional tables
Table A.1 Intertemporal link. Dep. variable: Effort in period 2 No shock
Shock realized
(1)
(2)
(3)
(4)
0.766*** (0.118) 1.615** (0.683) −0.747 (0.542)
−0.243 (0.192) 0.888*** (0.123) 1.692** (0.693) 0.081 (0.965)
0.731*** (0.099) 0.665 (0.671) 0.569 (0.679)
−0.163 (0.167) 0.818*** (0.130) 0.745 (0.699) 1.235 (0.907)
65 0.487
65 0.498
64 0.385
64 0.392
Wage in period 1 Wage in period 2 Prob50 Treatment Constant Observations R2
OLS regressions with Huber White robust standard errors given in parentheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Model 1 and 2 consider only observations from the Prob0 and from the Prob50 treatment, which are not hit by the shock. Model 3 and 4 consider only observations from the Prob100 and Prob50 treatment, which are hit by the shock.
Table A.2 The role of risk preferences. Dep. variable:
Wage in period 1
Wage in period 2
Risk preferences
0.229** (0.113) 0.769 (1.371) 0.711 (1.471) −0.304 (0.244) −0.155 (0.259) 5.910*** (0.675)
−0.025 (0.192) −0.031 (1.924) −1.323 (1.943) −0.068 (0.349) 0.145 (0.334) 6.236*** (1.115)
129 0.055
129 0.009
Prob0 Treatment Prob100 Treatment Prob 0 × Risk preferences Prob 100 × Risk preferences Constant Observations R2
OLS regressions with (Huber White) robust standard errors given in parantheses: * p < 0.10, ** p < 0.05, *** p < 0.01. Dependent variables: Chosen wages. Risk preferences are measured on a 10 point scale, 1 representing risk averse, 10 representing risk loving. In both models we use the Prob50 treatment as reference treatment.
Appendix B. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.jebo.2016.12.021. References Agell, J., Lundborg, P., 1995. Theories of pay and unemployment: survey evidence from Swedish manufacturing firms. Scand. J. Econ. 97 (2), 295–307. Akerlof, G.A., 1982. Labor contracts as partial gift exchange. Q. J. Econ. 97 (4), 543–569. Andreoni, J., Samuelson, L., 2006. Building rational cooperation. J. Econ. Theory 127 (1), 117–154. Bewley, T.F., 1999. Why Wages don’t Fall During a Recession. Harvard Univ. Press, Cambridge, MA. Bewley, T.F., 2005. Fairness, Reciprocity, and Wage Rigidity. MIT Press, Cambridge, MA. Blinder, A.S., Choi, D.H., 1990. A shred of evidence on theories of wage stickiness. Q. J. Econ. 105 (4), 1003–1015. Brandts, J., Charness, G., 2004. Do labour market conditions affect gift exchange? Some experimental evidence. Econ. J. 114 (497), 684–708. Brown, M., Falk, A., Fehr, E., 2004. Relational contracts and the nature of market interactions. Econometrica 72 (3), 747–780. Campbell, C.M., Kamlani, K.S., 1997. The reasons for wage rigidity: evidence from a survey of firms. Q. J. Econ. 112 (3), 759–789. Charness, G., 2004. Attribution and reciprocity in an experimental labor market. J. Labor Econ. 22 (3), 665–688.
L. Gerhards, M. Heinz / Journal of Economic Behavior & Organization 134 (2017) 228–239
239
Charness, G., Jackson, M.O., 2009. The role of responsibility in strategic risk-taking. J. Econ. Behav. Organ. 69 (3), 241–247. Charness, G., Levine, D.I., 2002. Changes in the employment contract? Evidence from a quasi-experiment. J. Econ. Behav. Organ. 47 (4), 391–405. Chen, D.L., Horton, J.J., 2009. The Wages of Pay Cuts: Evidence from a Field Experiment. Working paper. Harvard University. Cohn, A., Fehr, E., Herrmann, B., Schneider, F., 2014. Social comparison and effort provision: evidence from a field experiment. J. Eur. Econ. Assoc. 12 (4), 877–898. Davis, B., Kerschbamer, R., Oexl, R., 2016. Beliefs in a Gift Exchange Game with Random Shocks. MIMEO, University of Innsbruck. Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G.G., 2011. Individual risk attitudes: measurement, determinants, and behavioral consequences. J. Eur. Econ. Assoc. 9 (3), 522–550. Eckel, C.C., Wilson, R.K., 2004. Is trust a risky decision? J. Econ. Behav. Organ. 55 (December (4)), 447–465. Englmaier, F., Segal, C., 2011. Conflict and Cooperation in Organizations. Tech. rep., CEPR Working Paper. Falk, A., Fischbacher, U., 2006. A theory of reciprocity. Games Econ. Behav. 54 (2), 293–315. Fehr, E., Kirchsteiger, G., Riedl, A., 1993. Does fairness prevent market clearing? An experimental investigation. Q. J. Econ. 108 (2), 437–459. Fehr, E., Schmidt, K.M., 1999. A theory of fairness, competition, and cooperation. Q. J. Econ. 114 (3), 817–868. Fehr, E., Schmidt, K.M., 2006. The economics of fairness, reciprocity and altruism – experimental evidence and new theories. In: Serge-Christophe, K., Jean Mercier, Y. (Eds.), In: Handbook on the Economics of Giving, Reciprocity and Altruism: Foundations, 1. Elsevier, pp. 615–691. Fehr, E., Simon, G., Kirchsteiger, G., 1997. Reciprocity as a contract enforcement device: experimental evidence. Econometrica 65 (4), 833–860. Fischbacher, U., 2007. z-Tree: Zurich toolbox for ready-made economic experiments. Exp. Econ. 10 (2), 171–178. Fischbacher, U., Gächter, S., Fehr, E., 2001. Are people conditionally cooperative? Evidence from a public goods experiment. Econ. Lett. 71 (3), 397–404. Freeman, R., Lazear, E., 1995. An economic analysis of works councils. In: Works Councils: Consultation, Representation, and Cooperation in Industrial Relations. National Bureau of Economic Research Comparative Labor Markets Series. University of Chicago Press, Chicago and London, pp. 27–50 (Chapter). Gächter, S., Renner, E., Sefton, M., 2008. The long-run benefits of punishment. Science 322 (5907), 1510. Greiner, B., 2004. The Online Recruitment System ORSEE – A Guide for the Organization of Experiments in Economics. Papers on Strategic Interaction 2003–10. Max Planck Institute of Economics, Strategic Interaction Group. Kahneman, D., Knetsch, J.L., Thaler, R., 1986. Fairness as a constraint on profit seeking: entitlements in the market. Am. Econ. Rev. 76 (4), 728–741. Kocher, M.G., Strasser, S., 2011. The Fair Employment Hypothesis: Reciprocity in Unstable Environments. Working paper. Ludwig Maximilians University Munich. Krüger, M., Friebel, G., 2016. What are the Costs of a Pay Cut? Evidence from a Personnel Consulting Firm. Working Paper. Goethe-University Frankfurt. Kube, S., Marchal, M.A., Puppe, C., 2013. Do Wage Cuts Damage Work Morale? Evidence From a Natural Field Experiment. J. Eur. Econ. Assoc. 11 (4), 853–870. Li, J., Matouschek, N., 2013. Managing conflicts in relational contracts. Am. Econ. Rev. 103 (6), 2328–2351. Linardi, S., Camerer, C.T., 2012. Can Relational Contracts Survive Stochastic Interruptions? Experimental Evidence. Working Paper. University of Pittsburgh. Pahlke, J., Strasser, S., Vieider, F.M., 2015. Responsibility effects in decision making under risk. J. Risk Uncertain. 51 (2), 125–146. Rubin, J., Sheremeta, R., 2016. Principal-agent settings with random shocks. Manag. Sci. 62 (4), 985–999. Shapiro, F., 2006. The Yale Book of Quotations. Yale University Press.