Asymmetric Employer Learning and Statistical Discrimination

This paper develops a simple model of statistical discrimination in which firms learn about worker productivity over time and may use race to infer worker productivity. The framework we propose nests both symmetric and asymmetric employer learning by allowing outside employers to learn about workers productivity with noisier signals relative to current employers. We derive testable hypotheses on race-based statistical discrimination under different processes of employer learning. Testing the model with data from the NLSY79, we find that employers statistically discriminate against black workers at time of hiring in the non-college market where learning appears to be mostly asymmetric. For college graduates, employers directly observe most of the productivity of potential employees at hiring and learn very little over time. A series of sensitivity tests provide further support for our main findings