Employer learning and statistical discrimination

We show that if firms statistically discriminate among young workers on the basis of easily observable characteristics such as education, then as firm s learn about productivity, the coefficients on the easily observed variabl es should fall, and the coefficients on hard-to-observe correlates of produ ctivity should rise. We find support for this proposition using NLSY79 data on education, the AFQT test, father's education, and wages for young men a nd their siblings. We find little evidence for statistical discrimination i n wages on the basis of race. Our analysis has a wide range of applications in the labor market and elsewhere.