Employer Learning, Statistical Discrimination and Occupational Attainment

Abstract

Statistical discrimination plays a prominent role in theoretical discussions of the return to education as well as race and gender differences. Building on some earlier work by Henry Farber and Robert Gibbons (1996) (hereafter FG) and others on employer learning in the labor market, Altonji and Pierret (2001) (hereafter AP) provide a test for statistical discrimination in an environment in which firms are learning about characteristics of workers over time. However, that model assumes that the rate at which employers learn about a given worker is independent of the type of job that he or she is in. Consequently, the AP framework cannot readily be used to study statistical discrimination in hiring or how statistical discrimination and employer learning influence the skill level of the initial occupation and progression over a career. In this paper, I extend the AP analysis in three key respects. First, as in Robert Gibbons et al. (2004), the sensitivity of output to worker skill is assumed to depend on the skill level of the job. Second, the rate at which employers learn about the worker’s skill is assumed to depend on the skill level of the job. Finally, the probability of being hired into a given job depends on expected productivity relative to the wage. I show that statistical discrimination influences initial employment rates, wage levels, and job type and that employers’ initial estimate of productivity influences wage growth even in an environment in which access to training is not an issue. The implication is that the market may be slow to learn that a worker is highly skilled if the worker’s best early job opportunity given the information available to employers is a low-skill-level job that reveals little about the worker’s talent. I close with a brief discussion of directions for empirical work suggested by the analysis.