Statistical Discrimination in Labor Markets: An Experimental Analysis

This article reports results from controlled laboratory experiments designed to study second-moment (that is, risk-based) statistical discrimination in a labor market setting. Since decision makers may not view risk in the same way as economists or statisticians (that is, risk = variance of distribution), we also examine alternative measures of risk: the support of the distribution and the probability of earning less than the expected (maximum) profits for the employer. Our results indicate that employers made statistically discriminatory wage offers consistent with loss aversion in our full sample (though the result is driven by the male employer subsample). If one can transfer these results outside of the laboratory, they indicate that discrimination estimates based only on first-moment (mean-based) discrimination are biased. The public policy implication is that efforts and legislation aimed at reducing discrimination of various sorts face an additional challenge in trying to identify and limit relatively hidden, but significant, forms of statistical discrimination.