Supervisors and Performance Management Systems
NBER Working Paper No. 23351
We study how heterogeneity in performance evaluations across supervisors affects employee and supervisor careers and firm outcomes using data on the performance system of a Scandinavian service sector firm. We show that supervisors vary widely in how they rate subordinates of similar quality. To understand the nature of this heterogeneity, we propose a principal-agent model according to which supervisors can differ in their ability to elicit output from subordinates or in their taste for leniency when rating subordinates. The model also allows for variation in how informed firms are about this heterogeneity. Within the context of this model, we can discern the nature of the heterogeneity across supervisors and how informed firms are about this heterogeneity by relating estimated supervisor heterogeneity in ratings to worker, supervisor, and firm outcomes. We find that subordinates matched to a high-rating supervisor are paid significantly more and their pay is more closely aligned with performance. We also find that higher raters themselves are paid more and that the teams managed by higher raters perform better on objective performance measures. This evidence suggests that supervisor heterogeneity stems, at least in part, from real differences in managerial ability and that firms are at least partially informed about these differences. We conclude by quantifying how important heterogeneity in supervisor type is for workers' careers. For a typical worker, matching to a high rater (90th percentile) relative to a low rater (10th percentile) for just one year results in an increase in the present discounted value of earnings equivalent to 6-12 percent of an annual salary.
Document Object Identifier (DOI): 10.3386/w23351
Published: Anders Frederiksen & Lisa B. Kahn & Fabian Lange, 2020. "Supervisors and Performance Management Systems," Journal of Political Economy, vol 128(6), pages 2123-2187.
Users who downloaded this paper also downloaded* these: