Abstract: When are the wage distributions for two groups consistent with a general reduced form model of statistical discrimination? In our model, each group's productivities are drawn from different distributions with common means. Productivities are unobserved but inferred from noisy signals. Wages are determined by a strictly increasing (but otherwise unrestricted) function of the posterior expectation of the productivities (computed from the signal). We show that a pair of wage distributions are consistent with this model of statistical discrimination if, and only if, neither wage distribution first-order stochastically dominates the other. A rejection of this condition thus provides evidence of bias.
Keywords: discrimination, nonparametric testing, inequality
JEL Classification: D02;D04;D31;J31;J70