4.1.1 Theoretical framework

To understand some of the theoretical questions and empirical challenges faced in this literature, it is instructive to first reconsider Lazear’s (2000) original analysis of the Safelite Glass Corporation, a large auto-glass firm in which the primary task of worker’s at the bottom-tier of the firm’s hierarchy is to install automobile windshields. Lazear used non-experimental methods to estimate the productivity effects of the firm moving from a compensation scheme in which workers were paid an hourly wage scheme, to one in which they were paid a piece rate for each windshield installed, with a minimum guarantee. This pioneering work brings to the fore many of the issues that have influenced all the subsequent literature, and allows us to highlight the specific issues that field experiments help address.

The model is as follows. Worker’s utility depends on income and effort , with .⁴⁴ Worker’s output depends on effort and her ability, , so with output assumed to be observable and . For any given output , there is a unique effort level that achieves this, denoted . It is then straightforward to see that so that higher ability workers need exert less effort to achieve a given output. If workers choose not to work at any firm, their outside option from leisure is denoted . Hence the lowest ability worker that would accept employment at a firm with a required output level and wage , is denoted and is such that,

     (27)

All workers of higher ability earn rents from employment over leisure. Similarly, suppose a worker of a given ability could take up employment at another firm offering a wage-minimum effort pair (). Hence with inter-firm competition there might exist an upper cutoff in ability, , such that,

     (28)

where workers of ability higher than prefer to take the alternative employment contract.

This framework makes clear that incentive structures will affect two types of behavior, an idea developed in more detail in Lazear (2005). First, there will be change in effort exerted by individual workers in response to monetary incentives. This is referred to as the “incentive effect”. Second, the compensation scheme will induce a differential composition of workers within the firm over time. Some workers will prefer to join this firm from other employers. These changes in workforce composition can be thought of as the “selection effect” of monetary incentives.

The incentive effect can be easily understood graphically. Figure 9 shows the relationship between output, , and compensation for the two schemes relevant for Lazear’s study: (i) a fixed hourly wage subject to a minimum output requirement , resulting in total compensation ; (ii) a linear piece rate scheme with a minimum guarantee of . As Fig. 9 shows, for output levels between and the worker receives under both compensation schemes, and for output higher than earns strictly more under the piece rate scheme.⁴⁵

Figure 9 Compensation before and after at Safelite.

On the incentive effect, the model makes clear that moving from the fixed hourly wage scheme to the piece rate scheme does not cause the output of any individual to fall, and causes average output to rise. Low ability workers, indicated with a solid indifference curve in Fig. 9, remain indifferent between the two schemes and would produce output at point A under both. Higher ability workers, indicated with a dashed indifference curve in Fig. 9, would prefer to increase their effort and move to point B. This is because the piece rate scheme allows higher ability workers to raise their utility through increased compensation that more than offsets any increase in their effort. As a result, the dispersion in worker effort and output rises as long as there is at least one worker that chooses to produce more than .⁴⁶

On the selection effect, under plausible conditions, the average ability of workers rises with the move to piece rates. This is because low ability individuals remain indifferent between working for this firm under either incentive scheme. If they were willing to work for the firm under the fixed wage scheme, they should remain willing to do so under piece rates all else equal. On the other hand, high ability workers might be attracted to this form from other firms that for example, have higher minimum output standards or pay piece rates but at a lower rate . In short, theory predicts that there should be no change in the number of low ability workers who are willing to work at the firm, but that piece rates attract high ability workers so the right tail of the ability distribution in the firm should thicken.

As described in more detail below, existing field experiments have focused on identifying the incentive effects, and the research designs used have been less amenable to pin down these types of selection effect. Yet it is important to emphasize the need for future research to provide credible research designs to uncover both effects.⁴⁷

In Lazear’s study, he documents the total effect of the change in monetary incentives to workers was around a 44% increase in worker productivity, defined to be the number of windshields installed by the worker per eight hour day. Around half the increase was due to incentive effects, namely a change in effort of the same worker as he moved from a fixed hourly wage to a piece rate scheme. However, the other half was entirely due to the selection effect, namely productivity changes due to endogenous changes in the composition of workers in response to change in monetary incentives. A legacy of Lazear’s study is to show that both motives likely underlie why firms choose to alter their output based incentive structures in the first place. A carefully crafted field experiment that begins to measure whether and how the compensation policies of a given firm have such spillover effects on other firms that compete for similar workers, would open up a rich research agenda tieing together the study of within-firm compensation policies on equilibrium wage-setting behavior in labor markets.⁴⁸

Linking to the design of field experiments

The model provides a series of implications that have impinged on the first generation of field experiments over the last decade. First, given worker heterogeneity, changes in compensation scheme will nearly always affect average effort and output, as well as the dispersion of effort and output. Given the linkage between performance and pay, this inevitably results in changes in the distribution of earnings across workers at the same tier of the firm hierarchy. Hence linking pay to performance might have unintended negative consequences on worker and firm performance as a result of such increased earnings inequality. This might manifest itself in the form of workers reducing cooperation with co-workers (Fehr and Schmidt, 1999; Charness and Rabin, 2002). Field experiments are particularly adept at detecting and quantifying such unintended consequences because researchers are engaged in primary data collection, and the firm would have no incentive to collect such information ex ante as part of its personnel files, especially if pay for performance type compensation schemes have not been previously implemented. Some of the field experiments described below have collected qualitative evidence from workers to explore these channels, in additional to using personnel files to measure the direct productivity effects of incentives.

Second, given the selection effect of monetary incentives, there is inevitably a change in workers’ peer group over time. The composition of peers, or their social ties with each other, play no role in the standard neoclassical model in which worker preferences are only defined over their own income and effort. There are good reasons to probe this assumption in the field. First, if peer effects determine workplace behavior because they alter the marginal return to worker’s effort, then understanding how workers respond to changes in monetary incentives requires an understanding of the mechanisms underlying such peer effects. Second, extending the neoclassical model to take into account such peer effects or social concerns, as has been done in Kandel and Lazear (1992), Lazear (1989), Rotemberg (1994) and Fershtman et al. (2003), has implications for many aspects of firm behavior including the optimal design of incentives. Such concerns are a recurring theme in the series of natural field experiments conducted by Bandiera et al. that are described below.

Third, the model highlights that under fixed hourly wage schemes, there should not be much heterogeneity in workers output or effort, despite workers being heterogeneous in ability. This does not fit the evidence very well. For example, in Lazear’s study there was considerable dispersion in worker productivity even under the fixed hourly wage scheme. To better explain behavior in fixed wage settings, theory suggests workers might respond to other forms of non-monetary or implicit incentives, such as gift-exchange motives where workers exert more effort in response to employers paying higher than market clearing wages, the ability to shirk (Shapiro and Stiglitz, 1984; Macleod and Malcomson, 1989), or promotion prospects and career concerns (Dewatripont et al., 1999). Such mechanisms might also predict how workers sort across firms (Stiglitz, 1975). Some of these aspects are highlighted by the field experiments discussed below on the employment relationship.

Fourth, the model implies that low ability workers should not leave the firm with the move to piece rates. If they were willing to work for the firm under the fixed wage scheme, they should remain willing to do so under piece rates all else equal. On the other hand, given output differences among workers under price rates, management can more easily identify low ability workers when pay is tied to their performance. Over a longer time period, this might lead to them being fired. This type of selection effect caused by employer learning the true ability of workers, has not been studied by field experiments.⁴⁹

Finally, given worker heterogeneity, the first best for the firm would be to set a worker specific piece rate, . This would be chosen to equate the marginal benefit of effort to its marginal cost. However, we generally observe firms being constrained to offer bottom-tier workers the same compensation scheme. This may be because of legal, technological or informational constraints (Lazear, 1989; Bewley, 1999; Encinosa et al., 1997; Fehr et al., 2004). To overcome this, one hypothesis is that firms can get closer to this first best by linking managers’ pay to the firm’s performance. Managers then have greater incentives to target their effort to specific workers, and from the worker’s point of view it is then as if they face an individual specific incentive scheme. This idea is developed in the natural field experiment by Bandiera et al. (2007) described below, which then links managerial incentives to pay inequality among workers.

4.1.2 Evidence from the field

In the decade following the wave of studies using personnel data and insider econometrics to understand responses to monetary incentives (Ichniowski and Shaw, forthcoming), field experiments have begun to exploiting exogenous and randomly timed variation in compensation pay to both reinforce these existing results using non-experimental methods, as well as providing new insights.

Among the first of these studies was Shearer’s (2004) natural field experiment, designed to estimate the productivity gains moving from a piece rate to a fixed wage scheme for tree planters in British Columbia, Canada. In contrast to Lazear (2000), this setting is one in which tree planters are usually paid a piece rate with no guaranteed minimum, and fixed wages are rarely used ex ante. Workers were randomly assigned at the start of a work day, to plant under one of the incentive schemes. Hence a within worker comparison can be exploited to estimate the incentive effects. To reduce the likelihood of this comparison confounding other effects unrelated to the compensation schemes in place, work took place on fields of similar conditions, and there was a constant length of the work day. The incentive effect estimate is then based on a total of 120 observations on daily worker productivity, 60 under each scheme. The relatively small sample size—nine male workers were randomly selected from the firm—reflects the difficulty researchers initially faced in real world settings in convincing firms to randomly assign individuals to alternative compensation schemes. As more field experiments are conducted, some of these constraints are being eased. For example, some of the field experiments described below are based on data on hundreds of workers.⁵⁰

Shearer’s field experiment reveals the incentive effect of having a piece rate rather than a fixed wage compensation schemes to be a 20% productivity increase. The magnitude of this is comparable to Lazear’s (2000) findings—moving from a fixed wage scheme—based on non-experimental data, although this is wholly by chance There is no reason a priori to expect the behavioral response of workers to these two incentive schemes to be of the same magnitude given the very different types of worker involved, nature of the production function, and that the piece rate was not the same across settings. In line with theory, Shearer finds the standard deviation of output across workers was higher under piece rates. Overall, it was found that unit costs under piece rates were around 13% lower than under fixed wages.

To shed more light on how workers would have responded in a slightly different environment and to alternative compensation schemes, Shearer then develops and estimates a structural model. In terms of altering the economic environment, the structural model is used to shed light on what would have been the productivity gains if management was imperfectly informed about planting conditions. This yields comparable estimates of the productivity gains of moving to piece rates. To shed light on alternative compensation schemes, Shearer explores how workers would have responded to an efficiency wage scheme. The efficiency wage that would induce effort levels equal to those observed under fixed wages in the field experiment, is calculated and its implied unit costs are compared with those achieved under piece rates. This exercise suggests that fixed wages would lead to a 2.7% increase in unit costs relative to piece rates.

Shearer (2004) uses a close to best practice methodology in combining estimates from a field experiment with structural modelling within the same setting. This combination, following from discussions in Heckman and Smith (1995) and Keane and Wolpin (1997), first identifies the existence and magnitude of important causal effects using reduced form evidence. To then move away from such black-box findings, the researcher then uses structural modelling to posit an underlying behavioral mechanism behind the effects, assess the sensitivity of the estimates to slight alterations in the economic environment, and to make headway in understanding the optimal compensation structure. Of course, the validity of the structural model can itself be tested by exploring whether it predicts the responses observed to the exogenous variation engineered by the field experiment.

Another example of how field experiments can and should learn from other methodologies is at the heart of the natural field experiment of Hossain and List (2009). They use theoretical insights on framing effects from behavioral economics that have previously found empirical support in laboratory experiments (Kahneman and Tversky, 1979; Thaler, 1980; Samuelson and Zeckhauser, 1988; Ellingsen and Johannesson, 2008), to see if, in the field, framing manipulations affect worker responses to bonus incentives. Their setting is a high tech Chinese firm producing consumer electronics, where workers are organized into both individual and team production. They find that bonuses framed as both “losses” and “gains” increase productivity, for both individuals and teams. Teams respond more to bonuses posed as losses than as comparable bonuses posed as gains. The comparable effects for individuals are of the same sign but are not statistically significantly different to each other. Team productivity is enhanced by 1% purely due to the framing manipulation. Neither the framing nor the incentive effect lose their importance over the six month study period. Nor are there any detrimental effects on the quality of work as measured by product defect rates.

On a practical note, the results highlight that conditional on bonuses being provided, framing matters, and as framing can be adjusted almost costlessly, there are simply ways in which firms can further enhance productivity responses to monetary incentives. Theoretically, these results from the field provide an example of the prevalence of loss aversion in a natural labor market setting. As such the results provide external validity to laboratory evidence, and should be seen to provide a strong argument in favor of field and laboratory experiments being complements, not substitutes.

4.2.1 Theoretical framework

Suppose worker ’s payoff depends on three components. First, she derives some benefit from exerting effort towards a productive task. This benefit, , reflects in part how her effort maps into income through the monetary compensation scheme. To cover a wide range of compensation schemes including absolute performance evaluation incentive schemes such as piece rates, relative performance evaluation schemes such as rank order tournaments, or team incentives, these benefits will in general also depend on co-workers’ effort, . Second, the worker faces a convex cost of effort, , where workers are of heterogeneous ability, . Finally, we assume worker places some weight on the utility of co-worker , . In turn, such social preferences might depend on the existence or strength of the social tie between individuals and . This third component of the worker’s payoff function generates social incentives.⁵⁷ Workers simultaneously choose their efforts to maximize their total payoff,

     (29)

The first order condition is⁵⁸,

     (30)

The monetary compensation scheme determines the marginal benefit of effort, . As the worker has social incentives, she takes account of the fact that on the margin, her effort also affects the benefits that accrue to others, . As mentioned above, the precise sign of this social interaction depends on the nature of peer effects between workers that are socially connected, and the monetary compensation scheme in place.

The theoretical predictions of such models generate a wide range of behavioral responses. For example, working alongside friends might make work more enjoyable, generate contagious enthusiasm among friends, provide positive role models, or generate incentives to compete to be the best in the network of friends. All such mechanisms, that effectively increase the net benefits of effort, imply workers exert more effort in the presence of their friends relative to themselves when they work in the absence of their friends. Alternatively, working with friends might create contagious malaise, or lead to low effort norms within friends or co-workers more generally. All such mechanisms, that effectively decrease the net benefits of effort, imply workers exert less effort in the presence of their friends. Finally, the presence of friends might have heterogeneous effects across workers in that some exert more effort in the presence of their friends relative to when they work solely with non-friends, and others exert less effort. For example, friends or co-workers may conform to a common norm (Bernheim, 1994), or workers might be averse to pay inequality within their network (Fehr and Schmidt, 1999; Charness and Rabin, 2002). In either case, relative to when they work only with non-friends— (i) low ability workers exert more effort in the presence of their friends, and; (ii) high ability workers exert less effort in the presence of their friends. These aspects are highlighted by the field experiments discussed below on non-monetary incentives.

The field experiments in Bandiera et al. are designed to engineer exogenous variation in the incentives faced by workers to identify , corresponding to a similar reduced form parameter as in Lazear (2000) and Shearer (2004). They then combine this variation with primary data collected on social networks and plausibly exogenous variation in the assignment of friends as co-workers over time, to identify social incentives as embodied in . In these experiments, the authors examine the effects of social incentives both within and across tiers of the firm hierarchy. Namely in some studies and are co-workers engaged in the same tasks, and in other studies the pair correspond to a manager and her subordinate. Moreover, they study cases in which: (i) individual effort hurts co-workers , as in the case of relative incentive schemes; (ii) where it benefits them , as in the case of team incentives; and (iii) where it has no effect , as in the case of a piece rate scheme. We now summarize the main insights from these natural field experiments.

4.2.2 Evidence from the field

Social incentives among bottom tier workers

The firm studied in Bandiera et al. is a leading UK producer of soft fruit. Managerial staff belongs to three classes. The first class consists of a single general manager whom we refer to as the Chief Operating Officer (COO), the second comprises ten field managers, and the bottom-tier of the firm hierarchy consists of workers whose main task is to pick fruit. Field managers are responsible for field logistics, most importantly to assign workers to rows of fruit within the field and to monitor workers. Managerial effort can therefore be targeted to individual workers and is complementary to worker’s effort. The main task of the COO is to decide which workers are selected to pick fruit each day, and which are assigned to non-picking tasks. The field experiments described below together provide insights on behavior at each tier of the firm’s hierarchy.

In each natural field experiment, the researchers worked closely with the CEO of the firm to engineer exogenously timed changes in monetary incentives to workers or managers. The same workers and managers are observed under both incentive schemes and therefore it is possible to control for time invariant sources of heterogeneity across workers, such as their ability, and across managers, such as their management style.⁵⁹ The most important remaining empirical concern is that the estimates of such changes might still reflect naturally occurring time trends in productivity. This is addressed using a battery of tests in each paper. In addition, the time span of study allows the authors to check in each case whether the behavioral response to incentives is long-lasting, or whether they reflect Hawthorne effects, as discussed earlier, whereby individuals respond in the short run to any change in their workplace environment. Being able to use field experiments to estimate short and long run responses to changes in management practice is a theme we will return to below when we present field experimental evidence on gift-exchange in firms, and contrast the evidence from the field and the laboratory.⁶⁰

In each natural field experiment, the authors collected primary data on the social networks of each individual worker. With such a precise mapping of the structure of friendship networks in the firm, personnel data providing workers productivity over time, and the field experiment on monetary incentives, the authors are able to shed light on the interplay between monetary and social incentives in this setting.

Finally, they have daily information on the pool of workers available to pick fruit. This allows them to precisely identify the effect of monetary incentives on the selection of workers from this pool. The entire pool of workers is observed in this context because individuals are hired seasonally from Eastern Europe, and they live on the farm for the duration of their stay. This margin of selection—driven by the COO’s demand for workers—from the firm’s internal labor market proves to be an important margin of response to some changes in incentives, particularly in relation to changes in managerial incentives. Still, these field experiments, like Shearer (2004), are silent on the selection effect highlighted by Lazear (2000) in relation to workers choice of which firm to supply their labor to.

Another obvious similarity between Shearer (2004) and Bandiera et al. is that they study agricultural environments in which worker productivity is easy to measure, comparable across workers at the same moment in time, and comparable within a worker over time. The fact that worker productivity is measured electronically with little measurement error, also makes analysis of the impact of the field experiment on the distribution of productivity, again as highlighted by Lazear (2000), particularly amenable to quantile regression methods for example. However, it remains true that settings in which worker’s output is hard to measure, verify or compare, which might represent the bulk of tasks in the modern service based economy, remain relatively unexplored in field experiments.

In Bandiera et al. (2005) the natural field experiment exogenously changes the monetary incentives to the bottom-tier workers whose primary task is to pick fruit. The study compares the behavior of these workers under a relative incentive scheme to a piece rate scheme. The comparison is revealing because under relative incentives individual effort imposes a negative externality on co-workers’ pay whereas under piece rates individual effort has no effect on others’ pay. The difference in workers’ performance under the two schemes, if any, then provides evidence on whether and to what extent workers internalize the externality they impose on their colleagues. To see this, the framework above is tailored to this specific field experiment as follows.

Consider a group of workers, each worker exerts effort which determines her productivity. The cost of effort is assumed to be . Under relative incentives the benefit from pay depends on the worker’s productivity relative to all her co-workers, , where . The relative scheme has the key characteristics that an increase in worker ’s effort—(i) increases her pay; (ii) increases average effort and hence imposes a negative externality by reducing the pay of co-workers. The effort choice under relative incentives then depends on whether workers have social incentives and therefore internalize this externality. Assuming worker places the same social weight on all co-workers, so , the equilibrium effort for worker solves,

     (31)

Assuming worker chooses her effort taking the effort of others as given, the Nash equilibrium effort for worker solves,

     (32)

Under piece rates, individual effort is paid at a fixed rate per unit and worker chooses her effort as follows,

     (33)

The equilibrium effort level solves the first order condition,

     (34)

As worker ’s effort does not affect her co-workers’ pay, her optimal choice of effort is independent of . To compare effort choices under the two schemes, evaluate (34) at so that for a given , the pay per unit of effort is the same under both incentive schemes. The first order condition under piece rates then is,

     (35)

so the difference between the first order conditions (32) and (35) can be ascribed to two sources. The first is the externality worker imposes on others under relative incentives, the magnitude of which depends on . When worker ’s productivity is lower under relative incentives compared to piece rates. Second, by exerting more effort, each worker lowers the pay she receives for each unit of effort under relative incentives. This effect, captured by the term, also reduces productivity under relative incentives but is negligible in large groups.

The main results from Bandiera et al. (2005) are then as follows. First, the reduced form estimates suggest that the exogenously timed switch from relative incentives to piece rates had a significant and permanent impact on worker productivity. For the average worker, productivity increased by at least 50% moving from relative incentives to piece rates. As in the earlier literature, both the mean and dispersion of productivity significantly increase with the move to piece rates. The productivity gains achieved under piece rates are not found to be at the expense of a lower quality of picking.

The authors then assess whether this productivity change is consistent with the standard assumption that workers ignore the externality they impose on others under the relative scheme (), or whether they fully internalize it (). To do this they use the structural model above, imposing a functional form assumption on and a production function linking effort to observed output, to calibrate the first order conditions of the workers’ maximization problem to compute an estimate of each worker’s cost parameter, , under each incentive scheme and behavioral assumption. Since worker’s ability is innate, they ought to find the same implied distributions of costs across workers under both incentive schemes if the underlying behavioral assumption is correct.

Calibration of the first order conditions for worker’s efforts reveals that the observed change in productivity is too large to be consistent with the assumption that workers ignore the negative externality they impose on others. At the same time, the observed change in productivity is also too small to be consistent with the assumption that workers maximize the welfare of the group and fully internalize the negative externality. The authors then uncover the distribution of social weights across workers that would explain the productivity increases. To do so they assume the true cost of effort of each worker is that derived under piece rates, and then substitute into the first order condition (32). They find the data is consistent with the average worker placing a weight of  on the benefits accruing to all other co-workers, assuming they place a weight of one on their own benefits.

Further analysis combines the experimental variation induced by the change in incentive scheme, with non-experimental variation of the assignment of workers to work alongside their friends on some days but not on other days. The field experiment method allows the collection of primary data on social networks of each worker on the farm. This reveals that under relative incentives workers internalize the externality more when the share of their personal friends in the group is larger and this effect is stronger in smaller groups. In line with the interpretation that social preferences explain the difference in productivity across the two schemes, the relationship among workers does not affect productivity under piece rates. Finally, they find that productivity under relative incentives was significantly lower only when workers were able to monitor each other. Given that monitoring is necessary to enforce collusion while it does not affect altruism, they take this finding to support the hypothesis that workers are able to sustain implicit collusive agreements when relative incentives are in place. Hence, building on a large body of evidence from laboratory settings, this evidence from the field suggests workers behave as if they have social preferences but do not, in structural form, have social preferences that make them unconditionally altruistic towards others.

The results beg the question of why, given the large gains to productivity and profits, of the move to piece rates, were relative incentives ever employed in the first place. The farm management suggested the relative scheme was mainly adopted to difference out common shocks that are a key determinant of workers productivity in this setting. While this is in line with the predictions of incentive theory, the superiority of relative incentives relies on the assumption that workers ignore the externality their effort imposes on others.⁶¹ This assumption on worker behavior is not supported by this field experiment. Relative incentives led to lower productivity because workers internalized the negative externality to some extent. The results of this natural field experiment then speak directly to Lazear’s (1989) observation on how rarely workers are compensated according to rank-order tournaments, and point to new and interesting directions for theory to develop on the optimal provision of incentives under more robust assumptions on worker preferences.

Social incentives among managers

While the evidence from field experiments discussed thus far has focused on the monetary incentives provided to bottom-tier workers, Rosen’s (1982) magnification principle implies the incentives provided higher up in the firm hierarchy can have larger effects on firms’ performance. Bandiera et al. (2007) present evidence from a field experiment in the same setting as previously described to explore this issue.

They examine the effects of providing bonuses to managers based on the average productivity of their subordinates. They extend the framework above to highlight that, as in most firms, in their context managers can affect worker productivity through two channels—(i) they can take actions that affect the productivity of existing workers, and, (ii) they can affect the identity of the workers selected into employment. A simple theoretical framework indicates that, when workers are of heterogeneous ability and managers’ and workers’ effort are complements, the introduction of managerial performance pay makes managers target their effort towards the most able workers. This is labeled a “targeting effect” of managerial incentives. In addition, the introduction of managerial performance pay makes managers select the most able workers into employment. This is labeled as a “selection effect” of managerial incentives.

As in Lazear’s framework, such targeting and selection effects influence both the mean and the dispersion of workers’ productivity. Mean productivity unambiguously rises as managers target the most able workers and fire the least able. The effect on the dispersion is however ambiguous. On the one hand, targeting the most able workers exacerbates the natural differences in ability and leads to an increase in dispersion. On the other hand, if only more able and hence more similar workers are selected into employment in the first place, the dispersion of productivity may fall, depending on the underlying distribution of ability across workers.

They key findings from Bandiera et al. (2007) are as follows. First, the introduction of managerial performance pay increases both the average productivity and the dispersion of productivity among lower-tier workers. The average productivity increases by 21 percent and the coefficient of variation increases by 38 percent.

Second, the increase in the mean and dispersion of productivity is due to both targeting and selection effects. The analysis of individual productivity data reveals that the most able workers experience a significant increase in productivity while the productivity of other workers is not affected or even decreases. This suggests that the targeting effect is at play—after the introduction of performance pay, managers target their effort towards more able workers. The individual data also provides evidence of a selection effect. More able workers, namely those who had the highest productivity when managers were paid fixed wages, are more likely to be selected into the workforce when managers are paid performance bonuses. Least able workers are employed less often and workers at the bottom of the productivity distribution are fired.⁶²

Third, the selection and targeting effect reinforce each other, as workers who experience the highest increase in productivity are also more likely to be selected into employment. The introduction of managerial performance pay thus exacerbates earnings inequality due to underlying differences in ability both because the most able workers experience a larger increase in productivity and because they are selected into employment more often.

Finally, they evaluate the relative importance of the targeting and selection effects through a series of thought experiments. They find that at least half of the 21 percent increase in average productivity is driven by the selection of more productive workers. In contrast, the change in dispersion is nearly entirely due to managers targeting the most able workers after the introduction of performance pay. Namely, the dispersion of productivity would have increased by almost the same amount had the selection of workers remained unchanged. The reason is that the distribution of ability across workers is such that even when the least able workers are fired, the marginal worker selected to pick is still of relatively low ability. Hence there remains considerable heterogeneity in productivity among selected workers.

These findings shed some light on why firms provide performance related pay to managers in the first place. While such incentive schemes are obviously designed to increase unobservable managerial effort, these results suggest another more subtle reason for their use. This stems from the general observation that firms are typically constrained to offer bottom-tier workers the same compensation scheme. This may be because of legal, technological or informational constraints (Lazear, 1989; Bewley, 1999; Encinosa et al., 1997; Fehr et al., 2004). To the extent that bottom-tier workers are of heterogeneous ability, however, offering the same compensation scheme to all of them will be sub-optimal. When managers’ pay is linked to firm’s performance, their interests become more aligned with those of the firm and they have greater incentives to target their effort to specific workers in order to offset the inefficiency that arises because of the common compensation scheme. From the worker’s point of view it is then as if they face an individual specific incentive scheme. This opens a broad research agenda to examine whether firms are indeed more likely to offer managers performance pay in settings where lower tier workers are of heterogeneous ability, managers are able to target their effort towards specific workers, and workers are offered the same compensation scheme.

The findings from this field experiment also highlight the interplay between the provision of managerial incentives and the earnings inequality among lower-tier workers. Such a linkage exists whenever managers can target their efforts towards some workers and away from others, and managers choose which individuals are selected into the workforce. Hence that there might be an important interplay between managerial incentives and earnings inequality among workers highlights a possible link between two important trends in labor markets over the past twenty years that have previously been unconnected in the economics literature—the rising use of managerial performance pay, and the rising earnings inequality among observationally similar workers.⁶³

In Bandiera et al. (2009a,b), the authors use the same introduction of managerial bonuses to understand whether managers favor workers they are socially connected to. In general, social connections between managers and workers can help or harm firm performance. On the one hand, social connections may be beneficial to firm performance if they allow managers to provide non-monetary incentives to workers, or help reduce informational asymmetries within the firm. On the other hand, managers may display favoritism towards workers they are socially connected with, to the detriment of other workers and overall firm performance.⁶⁴

In this experiment, as managerial compensation becomes more closely tied to firm performance, we would expect managers to utilize social connections to a greater extent if indeed, such connections are beneficial for firm performance. On the other hand, if social connection are bad for the firm, we might expect managers to reallocate their effort across workers in response to managerial incentives, towards high ability workers, and away from workers they are socially connected to. To be precise, if the managers’ behavior towards connected workers changes once their interests are more closely aligned with the firm’s, their previous behavior under fixed wages could have not been maximizing the firm’s average productivity.

To measure social connections the authors use a survey they designed to exploit three sources of similarity between managers and workers—whether they are of the same nationality, whether they live in close proximity to each other on the farm, and whether they arrived at a similar time on the farm. The underlying assumption is that individuals are more likely to befriend others if they are of the same nationality, if they are neighbors, or if they share early experiences in a new workplace.⁶⁵

The main findings are as follows. First, when managers are paid fixed wages, the productivity of a given worker is 9% higher when he is socially connected to his manager, relative to when he is not. As workers are paid piece rates, this translates into the same proportionate change in earnings. Second, when managers are paid performance bonuses that tie their pay to the average productivity of workers they manage, being socially connected to the manager has no effect on workers’ productivity.

Third, the introduction of managerial performance pay significantly decreases the productivity of low ability workers when they are connected to their manager relative to when they were connected to their manager and she was paid a fixed wage. The introduction of managerial performance pay increases the productivity of high ability workers, especially when they are not connected to their managers. These findings indicate that when managers face low powered incentives, they favor the workers they are socially connected to, regardless of the workers’ ability. In contrast, when they face high powered incentives, managers favor high ability workers regardless of the workers’ connection status.

Fourth, an increase in the level of social connections between managers and workers has a detrimental effect on the firms’ average productivity when managers are paid fixed wages and has no effect when managers are paid performance bonuses. In this setting, social connections are therefore detrimental for the firm because their existence distorts the allocation of managerial effort in favor of lower ability workers.

This natural field experiment paper contributes to the growing empirical evidence on the interplay between social networks and individual and firm performance. In particular, the design allows the authors to identify not only whether social connections matter within the firm, but also exploit the exogenous variation in incentives to understand whether they are to the benefit or detriment of the firm.

Feedback

In a final natural field experiment from this setting, Bandiera et al. (2010, forthcoming) present evidence to evaluate the effect of performance feedback and monetary prize tournaments, when the workforce is organized in teams. Hence in this set-up workers effort imposes a positive externality on their team members, . They compare the effects of these forms of non-monetary and monetary incentives relative to when teams are paid piece rates, and analyze their effect on two outcomes: how workers sort into teams and team productivity.

This field experiment provides important contributions to the literature along three margins. First, despite the pervasiveness of teams in the workplace, field evidence on team incentives is scarce.⁶⁶ The existing evidence from individual reward schemes provides limited guidance because the margins along which individuals and teams can respond to incentives differ. Specifically, in addition to changes in individual effort, changes in team incentives can lead to changes in team composition. To the extent that workers effort depends on the identity of their team members because of social incentives, changes in team composition can affect the productivity of the individual teams and of the firm as a whole. ⁶⁷

Second, tournaments are widely used to provide incentives across diverse organizations such as salespeople competing for bonuses, managers competing for promotions, and politicians competing for vote shares (Bull et al., 1987; Baker et al., 1988). While several studies have tested whether the response to variation in tournament structure is consistent with theoretical predictions, field evidence on the comparison of monetary prize tournaments against alternative monetary and non-monetary incentive mechanisms is scarce.⁶⁸

Third, whenever tournaments are in place, workers inevitably receive some information on their relative performance. This information might have direct effects on productivity if individuals have concerns for their relative position or status (Moldovanu et al., 2007; Besley and Ghatak, 2008), inequality aversion (Fehr and Schmidt, 1999; Charness and Rabin, 2002) or conformity (Bernheim, 1994). The field experiment allows the authors to de-couple the effect of feedback from the effect of monetary prize tournaments. As the provision of feedback is almost costless, measuring its contribution to the overall tournament effect can lead to considerable cost savings if most of the positive effect of tournaments on productivity is actually due to worker responses to feedback.⁶⁹

In the experiment, at the beginning of the season, teams were paid piece rates based on their aggregate productivity. Halfway through the season teams were additionally provided feedback by posting daily histograms of each team’s productivity. This feedback makes precise the absolute productivity of each team, and their ranking relative to all other teams. Halfway through the remaining part of the season a monetary prize for the most productive team each week was introduced, in addition to the provision of feedback, and conditional on teams being paid according to piece rates.

When workers first arrive at the farm they are assigned to a team by the general manager for their first week. Thereafter workers are free to choose their own team members at a team exchange that takes place every week. A team is formed only if all its members agree. Hence in this setting workers have two choice variables: how much effort to exert into picking, and team composition.

The field experiment is again closely tied to an underlying model. This makes precise two key forces that drive team formation: workers’ ability and social connections. As individual earnings are increasing in the ability of team members, workers have incentives to assortatively match by ability. On the other hand, workers might prefer to form teams with friends because this might limit free-riding within teams (Rosen (1986); Hamilton et al. (2003).⁷⁰ To the extent that workers are not socially connected to colleagues of similar ability, a trade-off emerges. The theoretical framework then makes precise how the introduction of feedback and prizes affect this trade-off.

The key empirical results from the field experiment are as follows. First, the introduction of feedback and of monetary prizes leads to significant changes in team composition. Relative to the piece rate regime, the share of team members connected by social ties is lower and team members’ ability levels are more similar under the feedback and tournament regimes.

Second, the feedback and tournament schemes have opposite effects on average productivity. Relative to the piece rate regime, the introduction of feedback significantly reduces average team productivity by 14%. The further introduction of a monetary prize tournament, conditional on the provision of feedback, significantly increases productivity by 24%. As made precise in the theoretical framework, the reduction in average productivity when feedback is provided is consistent with workers being better off sorting into teams on the basis of ability rather than friendship as feedback increases the strength of incentives faced, and the firm being worse off because it no longer harnesses the ability of socially connected workers to ameliorate free-riding within the team. Hence the endogenous formation of teams under feedback reduces the firm’s productivity overall. In contrast, the tournament incentives are sufficiently high-powered so the increase in worker’s effort more than offsets any increase in free-riding within teams. Hence the firm’s overall productivity rises.

Third, the dispersion of productivity increases under both regimes because both effects are heterogeneous as indicated by the theoretical framework. Quantile regression results show that the introduction of feedback reduces the productivity of teams at the bottom of the conditional productivity distribution compared to piece rates, while it has no effect on teams above the 40th percentile. In contrast, the introduction of prizes increases the productivity of teams at the top of the conditional productivity distribution compared to piece rates, while it has no effect on teams below the 30th percentile.

Fourth, focusing on the teams that remain intact after each change in incentives, the authors evaluate the effect of feedback and prizes on effort, holding constant team composition. They find that while the effect of feedback on team productivity is positive the magnitude appears small. This emphasizes that the documented negative effect of feedback is primarily due to the endogenous changes in team composition caused by the provision of feedback, rather than changes in behavior of the same team. In contrast the additional introduction of monetary prizes increases team productivity by 25% for teams that choose to remain intact. Hence the provision of monetary prizes affects firm performance through both the endogenous changes in team composition and changes in behavior within the same team.

Finally, the authors present qualitative evidence from a worker survey they conducted. As highlighted at the start of this section, this type of primary data collection that is inherent in field experiments, allows the authors to shed light on other margins of behavior between workers that might be affected by the monetary and non-monetary incentives provided, but that the firm does not collect data on ex ante. This survey data reveals that relative to the piece rate regime, during the tournament regime significantly fewer workers report pushing their team members to work hard or giving team members instructions. This is consistent with workers being better matched by ability and having fewer social connections with their team members under the tournament regime, so that peer pressure within the team becomes less effective.

By exploring changes in behavior on a range of dimensions, this evidence from the field highlights new directions for research in understanding how agents react to monetary and non-monetary incentives in workplaces characterized by team production where teams form endogenously.

4.3.1 Gift exchange

The standard labor market model assumes in equilibrium firms pay market clearing wages and workers provide minimum effort. This prediction does not receive uniform support empirically. There are numerous cases where employers are observed paying above the market equilibrium wage Akerlof (1982), and where workers exert more than the minimum effort level, as we have already discussed in relation to Lazear (2000) and in many other studies on employee performance under fixed wages. This has led to the development of the gift-exchange model which is based on the assumption of their being a positive association between wages and worker effort (Akerlof, 1982; Akerlof and Yellen, 1988, 1990). In this class of model, employers offer higher than market clearing wages, and workers are viewed to positively reciprocate by providing higher than the minimum required effort.

Clearly such theories are hard to test using non-experimental data: there might be a host of unobservable factors that create a correlation between wages and worker effort. Hence, there has been a large body of evidence established in laboratory settings on gift-exchange in firm settings, which began with Fehr et al. (1993). In this original study, they constructed a labor market equilibrium with excess labor supply so that the equilibrium wage was low. Employees also had no pecuniary incentive to raise the quality of their work above the minimum required level, so the best response of employers was to pay the low equilibrium wage. Contrary to the prediction, the majority of employers attempted to induce employees to invest greater effort by offering them higher than market-clearing wages. On average, this high wage was reciprocated by greater employee effort. Overall, it was profitable for employers to offer high wage contracts.

Gneezy and List (2006) use a natural field experiment then look for evidence of gift-exchange in similar real world environments in which equilibrium wages are low and workers earnings are not tied to their performance. In moving from the lab to the field, one important comparative static to evaluate is how behavior changes with the duration of the task. In other words, are the types of positive reciprocity observed by workers in the lab, a long run phenomena. The psychology literature provides two reasons why the duration of tasks might matter. First, there is the distinction between hot and cold decision making (Loewenstein and Schkade, 1999; Loewenstein, 2005). Second, there can be adaptation of behavior over time (Gilbert et al., 1998).

Two subject pools were utilized for the field experiments. In each a between subject design was used. The first field experiment recruited undergraduate students to participate in an effort to computerize the holdings of a small library at the university. The task was to enter data regarding the books into a computer database. In the no-Gift treatment, individuals were offered a flat wage of $12 per hour. In the Gift treatment, once the task was explained to participants, they were surprisingly paid $20 per hour rather then $12 per hour as advertised. In total 19 workers were hired for six hours each; 10 were randomly assigned to the no-Gift treatment. The second field experiment was part of a door to door fundraising drive to support a university research center. Fundraising solicitors were recruited. All solicitors were told they would be paid $10 per hour, and those in the Gift treatment were surprisingly told they would actually receive $20 per hour. In total 23 solicitors were employed over two days, with 10 being randomly assigned to the no-Gift treatment.⁷¹

The main results are as follows. First, in line with earlier evidence from laboratory settings, there are signs of significant gift exchange in the first few hours of the task, as measured both by effort in the library task and money raised in the fundraising task. For example, in the library task, effort is around 25% higher for those in the Gift treatment.

Second, there are significant falls in effort over time. After a few hours, there are no longer any significant differences in effort between the no-Gift and Gift treatments in either task. Figure 10 illustrates clearly how any positive reciprocity by workers is a short run phenomena in these two settings. Overall, the results suggest that with the same budget, the employer would have been better off just paying the market clearing wage as in the no-Gift treatments.

Figure 10 Gift exchange and the duration of tasks.

While the results go against the standard gift-exchange explanation of the positive association of wages and effort, they are in line with survey evidence on wage rigidity.⁷² For example Bewley (1999) considers why wages are downwardly rigid during a recession. He reports that managers are worried that wage cuts might result in decreases in morale that would subsequently result in poor worker performance when the economy recovered, if not immediately. This highlights the importance of fairness considerations in cases of negative reciprocity. With respect to positive reciprocity, as in this field setting, Bewley’s evidence is less conclusive. He argues that morale is less important when considering wage increases, but finds that one main consideration when determining raises is the effect on employee turnover once the recession ends. Bewley’s work suggests that there appears to be little connection between increasing pay and productivity, except to the extent that higher wages make it possible to attract, and retain, higher quality workers. This ties back to the earlier discussion of Lazear (2000) and the subsequent work on monetary incentives in firms, where it is thought there are qualitatively large selection effects of incentives driven by changes in workforce composition. Again, more evidence on these channels related to employee turnover are required to test more precisely a fuller set of theoretical predictions.

4.3.2 Shirking

Field experiments had also provided insights on research questions related to employee shirking behaviors. The standard economic framework emphasizes employees are rational shirkers: they will slack when the marginal benefits of doing so outweigh the marginal costs. Firms respond to such behaviors by choosing compensation and monitoring policies to reduce shirking. As emphasized above, this view that workers will behave so opportunistically when the marginal returns on their effort are low, is often contradicted by empirical evidence and predictions on behavior in such settings from the psychological and sociological literatures (Pfeffer, 1996; Kreps, 1997; Baron and Kreps, 1999). While we have earlier studied the role of compensation schemes and wage setting behavior to raise employee effort, we now focus on the effect employer monitoring has on worker behavior.

If employees are rational cheats then, conditional on a given incentive pay arrangement, a reduction in monitoring will lead to an increase in shirking. The most powerful sanction available to employers is typically dismissal. Thus, an increase in shirking resulting from reduced monitoring should be greatest among individuals for whom the ongoing employment relationship is least valuable.

As with most of the research questions posed in this chapter, establishing credible empirical evidence to support the theory is not straightforward. In this case there are two concerns that have plagued the non-experimental literature. First, shirking behavior is by its nature hard to detect. Moreover, the ability of the econometrician to detect shirkers might itself be endogenously related to the employer’s monitoring practices. Second, there might be unobserved factors, such as other hiring policies, that cause there to be a correlation between monitoring and shirking.

A carefully designed natural field experiment of Nagin et al. (2002) addresses both challenges. The setting was a telephone solicitation company, with employees dispersed across 16 call centers. At each call center, telephone solicitors were paid according to the same piece rate incentive scheme, one in which salary increased with the number of successful solicitations. This piece rate, together with imperfect information on the outcome of pledges, created incentives for employees to falsely claim that they had solicited a donation.

To curb opportunistic behavior, the employer monitored for false donations by calling back a fraction of those who had responded positively to a solicitation. Employees were informed when hired that their activities would be checked by “callbacks” made by management. The results of each week’s callbacks were communicated to both employees and their immediate supervisors, and the bad calls were deducted from each individual’s weekly incentive pay. Stronger sanctions for bad calls were not generally imposed on employees because the number of bad calls was understood to be a noisy indicator of cheating. For example, donors might sometimes change their mind after agreeing to pledge money. ⁷³

To see if the costs of implementing this monitoring system could be reduced, the company conducted a controlled field experiment. This experiment was “double blind” in the sense that neither the employees nor their immediate supervisors were aware of departures from “business as usual.” In the experiment, the employer varied the fraction of bad calls that were reported back to employees and supervisors at each call center. To more precisely estimate the true rate of fictitious pledges, the firm simultaneously increasing the true callback rate from 10% to 25% of pledges. By working closely with the firm, the researchers were able to collect primary survey data on employee attitudes toward the job, their expected job tenure, and the perceived difficulty of finding another, comparable job. These relate closely to the underlying theory of rational shirking behavior. This information is used to test whether those for whom the job was most valuable were also the employees least likely to engage in opportunistic behavior.

The main results are as follows. First, a significant fraction of employees behave according to the predictions of the rational cheater model. In particular, employees respond to a reduction in the perceived cost of opportunistic behavior by increasing the rate at which they shirk. Using the survey data collected, the authors find the employees who responded to reductions in monitoring tended to be those who perceived the employer as being unfair and uncaring. On the other hand, there is no evidence that individuals with good outside options increased shirking by more than other workers when the rate of monitoring declined. Second, a substantial proportion of employees do not appear to respond at all to manipulations in the monitoring rate. As with responses to monetary and non-monetary incentives documented above, there is considerable heterogeneity in how workers respond to employer monitoring. This underlying heterogeneity highlights the need to balance the need to reduce the shirking behavior of some workers inclined to rationally cheat, against those that are unlikely to do so under normal circumstances.