«IZA DP No. 3901 A Behavioral Account of the Labor Market: The Role of Fairness Concerns Ernst Fehr Lorenz Goette Christian Zehnder December 2008 ...»
Gneezy & List (2006) hire workers to enter books into a library information system. The workers are made aware that this is a one-time employment with no scope for further work later on. The workers are either paid a low wage ($12) or a high wage ($20) per hour, with no particular reason given for the pay. Overall, output is approximately 10 percent higher when workers are paid the high wage, but because of large variations across individual output levels and a small sample size (about 10 participants per condition), this difference is not statistically significant.9 Al-Ubaydli et al (2006) conduct a similar study with larger sample sizes (about 30 participants per condition). They use a temporary work agency to recruit workers for stuffing envelopes. Their rich setup consists of two treatments which allow a clean comparison regarding fairness. All workers in these two conditions were told upon hiring that their hourly wage will be somewhere between $8 and $16. Subsequently, the workers in one condition were paid $8 per hour, while those in the other condition were paid $16 per hour. The number of finished envelopes is about 22 percent lower when workers are paid $8 compared to when they are paid $16. This difference is highly significant, but it is unclear whether the effect was mainly caused by workers feeling that they were treated unfairly after they received $8, or whether the highly paid workers put in extra effort.
The positive impact of wage increases fades over time in the studies reported in Gneezy and List (2006).
However, other studies (mentioned below) did not replicate this finding. It is also surprising that Gneezy and List did not find a learning effect in their task (average output is constant or even decreasing over time), while other studies implementing the same task (Kube et al. 2006 and 2008) find a steep rise in output over time in all treatment conditions. Likewise, the experiment of Cohn et al (2007) – which lasted for several weeks and is described in more detail below – did not replicate the finding of Gneezy and List.
However, when workers received the low wage of 10 Euros, the reduction in effort relative to the baseline of 15 Euros is so large (27 percent) that it is significant despite the small sample.
These results confirm that another well established result from the laboratory is also present in the field: while the positive effects of fair treatment on behavior are usually small, the negative impact of unfair behavior is often rather large (Offerman 2002).
The studies discussed so far did not examine the specific mechanisms underlying the response (or absence of a response) to wages. One possibility would be to vary the nature of the treatment, e.g., if the principal signals thoughtfulness and caring about one's workers. Kube et al. (2008) hire students for a data entry task and announce earnings of €36 for the three-hour work episode. They compare their baseline treatment where they pay the €36 to two different gift treatments. In the monetary gift treatment, the wage is raised to €43, i.e., a wage increase of €7, when the subjects show up for work. In the material gift treatment, the workers receive a gift in the form of a Nalgene bottle worth 7 Euros. The wage increase of 7 Euros induces the subjects to enter approximately 6 percent more data than in the baseline treatment. As usual, the effect is not large enough to be significant, given the small samples.10 However, when the subjects receive the Nalgene bottle, there is a highly significant increase of about 30 percent in entered data. This effect also prevails in a treatment in which the price of the Nalgene bottle was saliently visible for the subjects. Moreover, in a subsequent binary choice experiment, the authors gave (other) subjects the choice between the bottle and €$7; the vast majority of the subjects preferred the money over the bottle. This suggests that the positive effect of the material gift on effort is not primarily due to the economic value of the gift but that the signaling value of the gift is also important. A plausible interpretation is that the workers appreciate it if their employer cares for their welfare and this appreciation may lead to higher effort levels. The results of Kube et al (2008) provide a rationale for why job rents may contain non-monetary components.
Notice, however, that the effect of raising the wage above the baseline is positive in every study we have discussed so far. Taking all five studies into account, the probability of this occurring is p = 0.55 0.05 under the null hypothesis of no effect. Thus, the overall evidence rejects the hypothesis that wage increases do not affect effort.
A second approach towards a better understanding of the mechanisms behind gift exchanges is to collect information from outside the experiment to examine how the response to a wage increase is modulated. Cohn et al (2007) implement a wage increase during a newspaper promotion that took place when a publisher launched a new newspaper. It was clear from the outset that the promotion would last at most a few weeks. Therefore, the workers who were hired to distribute the newspaper at train stations and other public places did not have the prospect of long-term employment. The workers were given a CHF 5 increase over their regular hourly pay of CHF 22 and asked to approach the passers-by as actively as possible in return for the higher pay. Ten weeks after the experimental wage increase, Cohn et al. (2007) conducted a follow-up survey among the workers in which they measured the wage the workers thought would be appropriate for their work. In addition, they conducted a laboratory experiment in order to elicit workers’ propensity for reciprocal fairness. They find that only those workers who felt treated unfairly at the base wage and who displayed positive reciprocity in the laboratory experiment showed a significantly positive effort response to the wage increase. Individuals who already felt treated fairly at CHF 22 did not respond to the wage increase. Likewise, those who behaved selfishly in the lab experiment also did not respond to the wage increase – regardless of whether they felt underpaid or not. This finding shows that lab experiments can be very useful for understanding what is going on in field experiments. In addition, the finding is in line with the fair wage-effort hypothesis (Akerlof & Yellen 1990) which predicts no effort response for fairly paid or overpaid employees, while underpaid workers are predicted to respond to wage variations. These results also show the importance of explicit controls for workers' fairness preferences and fairness perceptions; and they indicate the relevance of individual heterogeneity and the current base wage for an aggregate gift exchange effects: If workers already receive a high current wage, the percentage of individuals likely to judge this wage as unfair is relatively low; only few individuals will therefore respond to a wage increase with higher effort, and the average effect is likely to be small. In contrast, if the current base wage is relatively low, many individuals will feel underpaid and finding a positive average gift exchange effect is thus more likely.11 This argument may be relevant for the interpretation of Gneezy and List (2006). According to personal communication from Uri Gneezy, the going market wage for the library task in this study was $7 per hour while To summarize, field experiments confirm the lab finding that paying higher wages leads to an increase in effort, although – depending on the circumstances – the effects may be small. The wage elasticity of workers outputs ranges from roughly 0.15 in Gneezy and List (2006) to
0.30 in Kube et al. (2006) and Kube et al. (2008), and to 0.44 in Al Ubaydly (2006). In the case of unambiguous wage cuts, the elasticity is 0.82 in Kube et al. (2006), and it rises to roughly 1.45 if a material gift is provided (Kube et al 2008). This evidence is consistent with the idea that effort is more responsive to wage cuts than to wage increases and that the psychological properties of the gift (money versus a non-monetary gift) matter. One study (Cohn et al 2007) also supports the notion that the fairness perceptions mediate effort responses to wage increases, thus providing more direct support for the gift exchange hypothesis (Akerlof and Yellen 1990). These results also contradict the proposition that laboratory findings rarely carry over to field setting because – so the claim – the field is fundamentally different (List and Levitt 2007). Like in the laboratory, the field experiments show that the impact of fairness in one-shot interactions is likely to be small. Likewise, lab and field experiments suggest that the response to unfairness (i.e., a wage cut) is stronger than the response to kind behavior (i.e., a same-sized wage increase). Finally, as Cohn et al. (2007) show, there is not only consistency between lab responses and field responses to kind acts but the lab experiment also helps us to better understand the effort responses observed in the field.
4.2 Fairness in ongoing Relations In the studies described above, the authors deliberately implemented conditions that rule out incentives arising from long-term labor relations because they were interested in the forces governing work effort when reputational incentives are absent. However, employment relationships are seldom spot market transactions where trading partners interact only once.
Therefore, we summarize in the following labor market studies from settings that are the base wage in the neutral treatment was $12 per hour. Thus, by paying such a high wage in the neutral treatment it seems likely that workers perceived this wage as already quite fair. Therefore, a further wage increase to $20, even though it is considerable, should have little effect.
characterized by the possibility of forming long-term relations. With regard to the experimental evidence, we will focus on papers that implemented opportunities for repeated interactions in a setting with a publicly known finite number of periods. This design feature implies that if the selfishness of all participants is common knowledge, the only equilibrium in the finitely repeated game is identical to the equilibrium in the one-shot game. We can thus conclude that deviations in behavior in the repeated interaction treatments with regard to the one-shot treatments arise from the interactions between fairness and reputation effects.
An early paper that investigates the effect of repeated interactions in a gift-exchange setup is Falk and Gächter (2002). They set up a laboratory experiment with two treatments. In the baseline treatment, each participant plays a sequence of ten one-shot gift-exchange games with 10 different partners. In the main treatment, subjects play the same gift-exchange game ten times with the same partner. Each pair of subjects in the second treatment thus has a common history, and both participants can always condition their actions on their past experience with their partner. The results of this study reveal that reputational incentives in finitely repeated interactions amplify the positive impact of fairness on performance. The wage-effort relationship is steeper, and average efforts are significantly higher in the treatment with repeated interactions than in the one-shot treatment. The effort level, which can be chosen between 0.1 and 1, stabilizes at about 0.55 in the repeated treatment after the first few periods, while the effort level in the one-shot treatment evens out around 0.35. The reason for this difference is that in the repeated condition many subjects in the role of an employer offer high wages only if their worker always provided high effort in the past. As a consequence, selfish workers have a strong incentive to hide their type and imitate the behavior of fair workers. By providing high effort in response to high wage offers, selfish workers can build up a reputation of being fair. Due to the conditional offering strategy of employers, such a reputation is of value, as it gives the workers access to attractive future offers from which they would be excluded if their true type were revealed. In the final period, when reputational concerns no longer matter, the effort level in the treatment with repeated interactions drops approximately to the level of the one-shot treatment. This end-game effect shows that fairness concerns genuinely motivate roughly the same fraction of subjects in both treatments. However, the long-term nature of the repeated treatment disciplines many selfish individuals who would – in the absence of repeated interaction – play uncooperatively.
Brown et al. (2004) allow long-term employment relationships to arise endogenously in a market environment with an excess supply of labor. In this experiment, which lasts for 15 periods, employers may address their wage offers to specific workers. They can therefore endogenously build up a long-term relationship with a worker by renewing offers to the same worker in consecutive periods. The comparison of this treatment to a treatment in which longterm relations are excluded provides the basis for measuring the impact of reputation incentives and the interaction of such incentives with fairness concerns. The possibility of contract renewals has strong positive effects on performance: It increases average effort from
3.3 in the treatment with one-shot interactions to 6.9 in the “relations treatment”. In fact, the modal effort choice is at the minimum in the one-shot condition, while the maximum effort is the modal choice in the relations treatment. There is an end-game effect as in the Falk and Gächter (2002) paper because the selfish players do no longer put forward non-minimal effort levels in the final period.