Do social interactions in the workplace lead to productivity spillover among co-workers?

Challenges in studying peer effects in co-worker productivity

Estimating peer effects in co-worker productivity poses several challenges. First, it requires measures of worker productivity and the possibility to identify the relevant co-workers with whom a worker interacts in the workplace. This may be more easily available when analyzing data from specific settings, such as a single occupation in a specific firm. Occupations in which relatively standardized tasks lead directly to an observable output (e.g. supermarket cashiers and fruit pickers), and settings in which workers interact in clearly defined shifts or teams (e.g. call centers and conveyer belt production lines) are especially suited to studying productivity spillover among co-workers. Such settings pertain in particular to low-skilled tasks or occupations, from which most of the current evidence on peer effects is derived.

Second, it is methodologically challenging to establish cause and effect in the behavior of peers. A positive observed correlation between the performances of two workers does not necessarily mean that both workers influence each other’s productivity positively. It is also consistent with, say, worker A influencing worker B strongly positively, but worker B not influencing worker A at all (or worker B even influencing worker A slightly negatively). This impossibility of inferring the direction and sign of the causal influence between peers from simple correlations of their behavior is known as the “reflection problem.” Some studies find ways to deal with this problem, others avoid it by analyzing the effect of a co-worker’s predetermined characteristic (e.g. his/her education) on a given worker’s performance, because then causality is likely to run in one direction only (namely from worker A’s education to worker B’s performance, but not from worker B’s performance to worker A’s education).

Third, even when estimating the effect of a predetermined co-worker characteristic (such as education) on individual worker performance, unobserved “confounding factors” pose a problem. Sometimes, the researcher does not observe important details of the work environment, such as the organization of work, hierarchies within the firm, or the compensation scheme. If such unobserved characteristics affect both workers’ wages and the quality of their peers, then this might erroneously be taken for a peer effect. For example, a firm may restructure its work processes, introducing a new performance evaluation system that increases a given worker’s performance; at the same time, the firm may hire better-educated co-workers, thereby increasing his/her co-workers’ educational characteristics. Such “confounding factors” might lead to wrong conclusions about peer effects, but the analysis can adjust for such factors when they are observed or when they can be controlled by the researcher. Therefore, compelling existing evidence on peer effects in co-worker productivity is often based on either laboratory experiments (in which the environment can be controlled) or on real-world data from single firms or occupations (in which detailed information on the environment is observed).

While studies based on specific settings provide compelling and clean evidence for the existence (or absence) of peer effects in those settings, an interesting question is to what extent such findings apply to the labor market in general. Studies based on data that are representative of more general workplace settings are currently emerging. These studies face the challenge of finding comparable productivity measures across occupations and of controlling for confounding factors. When examining the existing evidence, it is useful to distinguish between settings with low- and high-skilled tasks, where the process of productivity spillover may be quite different. In settings with low-skilled tasks, the more likely channel is social pressure (and the empirical literature has in general found positive productivity spillover effects), while in settings with high-skilled tasks the more likely channel is knowledge spillover (and less evidence of productivity spillover has been found).

Productivity spillover in low-skilled tasks

Partly for the reasons discussed above, much of the current evidence on peer effects is derived from settings characterized by low-skilled tasks or occupations. A well-known study analyzes the productivity of supermarket cashiers in a large supermarket chain and finds that cashiers tend to work faster (scan items faster at the till) when they are assigned to a shift with faster co-workers, especially if they can be observed by the faster co-workers [2]. This latter fact suggests that the peer effect in co-worker productivity is likely due to social pressure. Another study looks at soft-fruit pickers, taking the friendship structure between the workers into account, and finds that workers are more productive when working on the same field with a more able friend, but less productive if working only with less able friends [3]. This finding points toward a social desire among friends to adjust their productivity when working alongside one another. Further research based on firm data from a multinational telephone company shows that the general pattern of peer effects within co-worker networks is consistent with conformist behavior; that is, driven by social norms or peer pressure, co-workers tend to “imitate” the productivity of their co-workers [4].

Other specific settings for which productivity spillover among co-workers has been identified include data-entry workers in an Indian company and cosmetics salespeople in a Chinese department store. In most of these settings, social-behavioral effects, such as peer pressure, are the most likely channels. An exception is the finding that a training program among call center workers increased not only the productivity of the participants themselves, but also the productivity of their untreated teammates, suggesting that this effect may at least be partly driven by knowledge spillover [4], [5]. Thus, while peer pressure seems to be the more common source of productivity spillover in low-skilled occupations, knowledge spillover is likely to be relevant in specific situations, for example in the interaction between newly trained and untrained workers.

Besides such field studies, there exist a number of experimental lab studies based on simple tasks. Some of them show that the mere presence of another individual carrying out the same task can increase productivity (most likely due to social pressure). While lab studies in general confirm the existence of positive productivity spillover, one study found that very low and very high levels of peer pressure decreased productivity, serving as a reminder that excessive peer pressure can be detrimental for worker productivity.

Productivity spillover in high-skilled tasks

Considerably fewer studies have analyzed high-skilled occupations, where knowledge spillover is more often the suspected channel of peer effects in productivity, and the picture that emerges is more mixed. While there is evidence for learning from co-workers among teachers in the same schools [6], knowledge spillover has not been documented among scientists in the same university departments [7]. However, the quality of academic staff in university departments affects doctoral students positively [8]. Similarly, very high quality co-authors and collaborators influence individual research output in medical sciences and mathematics positively [9], [10]. Conversely, at a wider geographical level, the large influx of Soviet mathematicians to the US in the 1990s strongly reduced the output of American mathematicians working in similar fields as the Soviet mathematicians [10], and there seems to have been no positive knowledge spillover of German Jewish scientists who emigrated to the US in the 1930s on individual US inventors working in similar fields [11].

The following important distinction may help explain the divergence in empirical results on knowledge spillover [10]. Among individuals who are collaborators (such as co-authors) or who are in a superior-subordinate relationship (such as PhD advisors and PhD students), a positive effect of peer quality on a worker’s output can be expected simply because in a collaborative team, the productivity of a worker directly depends (positively) on the productivity of his co-workers. On the other hand, among peers who do not actively collaborate but who share the same geographical space (same university department) or idea space (same academic field), there is no reason to expect that their productivities directly depend positively on one another. On the contrary, there may even be a negative effect due to competition (i.e. the more successful are one’s peers who work in the same field, the harder it is to get one’s own research published). In line with this distinction between peers who are actively collaborating and those who are not, it has been found that the premature death of an inventor reduces the patenting and earnings of the co-inventors in their team, but not of other inventors in the same firm [12].

The bigger picture

The literature provides compelling and clean evidence for the existence (or absence) of peer effects in specific settings and conveys the notion that peer effects in co-worker productivity exist in a range of low-skilled occupations, but that they are less frequent in high-skilled occupations, where they seem primarily to be related to collaborators (team members who work on the same task, including supervisor-subordinate pairs). An interesting question thus becomes: to what extent do these findings based on specific firms or occupations, or on experimental evidence, apply to the labor market in general?

A first step in answering this question is to systematically analyze the existing studies; this has recently been done in a meta-analysis of 34 individual studies, from which an estimate of a peer effect in co-worker productivity can be derived [13]. The results are organized in Figure 1. The average coefficient 0.12 implies that the effect of a, for example, 10% change in co-worker productivity on a worker’s own productivity is about 1.2% on average across the included studies. Further interesting results emerge from this meta-analysis. First, the difference in the peer effect estimated in field studies and lab experiments is small and not statistically significant, which prompts the authors to conclude that the results on peer effects in co-worker productivity from lab experiments do generalize to the field. Second, in line with the more mixed evidence for high-skilled occupations, the peer effect is smaller in samples that include more complex tasks. A further noteworthy result of the meta-analysis is that peer effects in co-worker productivity are larger under group piece-rate or fixed-wage conditions, that is, in the absence of individual piece rate pay. With team-based production, this is exactly the situation in which workers have incentives to free-ride on their co-workers (i.e. workers reduce their own effort because co-workers pick up the slack); peer pressure and workers’ desire to conform to social norms can help prevent workers from taking too much advantage of their co-workers and are therefore thought to help overcome free-rider problems, so it makes sense that peer effects are found to be higher in this situation.

A second step in assessing the generalizability of findings for specific settings is to analyze large-scale data for a representative set of workers, firms, and sectors, as recently provided by a study analyzing matched employer-employee data for a large local labor market in Germany [1]. The study addresses the challenge of finding a comparable output measure across a large range of occupations by focusing on peer effects in wages instead of physical productivity. As discussed above, unobserved confounding factors pose a challenge. As an example, bad management decisions may lead to a loss of market share and revenue, which may force a firm to raise wages at a slower rate than other firms, subsequently motivating the best workers to leave. This would induce a correlation between peer quality and wages for reasons unrelated to peer effects. The study controls for such unobserved factors by implementing a research design that ultimately identifies the peer effect by estimating to what extent an individual’s wage changes (relative to wage changes in other occupations in the same firm) in response to a given change in peer quality within their own occupation in the firm (relative to worker quality in other occupations in the same firm). By looking at changes over time for those who remain in the same peer group, unobserved factors at occupation-firm level are held constant, and by using variation relative to other occupations in the same firm, unobserved factors that affect the whole firm in a given year are netted out (such as the example of the bad management decision mentioned above).

Interestingly, after having thus controlled for a large set of confounding factors, the study finds almost no peer effects in wages for all occupations in the sample [1]. However, when distinguishing between occupations with a high and low incidence of repetitive and predefined tasks, the authors do find larger peer effects in occupations with more repetitive and predefined tasks, such as cashiers, agricultural workers, and other mostly low-skilled manual occupations. This is depicted in Figure 2, which shows that the peer effect in wages for the occupations with the most repetitive tasks is found to be about half as big in magnitude as the average peer effect found in the meta-analysis discussed above [13] and depicted in Figure 1. A smaller peer effect in wages is what one would expect if productivity differentials do not propagate fully into wages (i.e. if wages do not fully reflect actual differences between worker productivity).

The same study further addresses the question of what the channel of the peer effect in the sample of low-skilled occupations is, distinguishing between social pressure and knowledge spillover [1]. First, because these are occupations with relatively repetitive and predefined tasks, knowledge might be less important, and output might be reasonably well observable (a precondition for social pressure), so that the more likely channel is social pressure. However, learning on the job might still be relevant in these occupations, especially when a worker first enters the labor market, or when a worker starts a new job with a new employer. If the peer effects disappear after a few years of work experience or job tenure, then this would suggest knowledge spillover as a channel. However, the results show that even for older workers with more job tenure, part of the peer effect remains. Moreover, if knowledge spillover was the relevant channel, then the quality of past peers should have an effect because workers should retain knowledge learned from their peers, even after those peers leave the firm. But the results show that past peers do not matter. The authors thus conclude that social pressure rather than knowledge spillover is the main source of peer effects in the group of low-skilled occupations.

Perils of peer pressure

While peer pressure can increase worker effort, which in turn can help mitigate the well-known free-rider problem associated with teamwork, it does not necessarily make workers better off. Workers may dislike working under peer pressure, and if they are given the choice, they may well prefer working in an environment without it. Firms in which peer pressure is high may then face problems with worker retention, having to pay higher wages in order to retain workers. Theoretical and empirical results support this view [1]. Thus, while a moderate amount of peer pressure may be profit-enhancing from the point of view of the firm, excessive peer pressure may hurt both firm profits and worker well-being. Knowledge spillover, on the other hand, probably does not have any negative effect on worker well-being, though encouraging knowledge transfer among co-workers may still be associated with some cost to the firm, such as providing the space and time needed for workers to meet and communicate.