Using matched employer-employee data to study labor market discrimination

Wage gaps between individuals of difference races, sexes, and ethnicities have been documented and replicated extensively, and have generated a long history in labor economics research of empirical tests for labor market discrimination. The most widely-used approach to test for labor market discrimination is based on wage regressions estimated at the level of individual workers, with the estimate of discrimination inferred from the residual race, sex, or ethnic group differential in wages that remains unexplained after including a wide array of proxies for productivity. What is absent from the residual wage approach and in our view leaves the approach vulnerable to being regarded as uninformative regarding discrimination is any directly observable measure of productivity with which to adjust differentials in wages in trying to infer whether a particular group suffers from discrimination. The ideal solution would be individual-level productivity data that can be compared with wages. Any of the variables that differ across groups and are unobserved in the residual wage regression approach should affect wages and productivity equally, and hence not bias the test. However, such data are extremely rare, in large part because individual productivity is often unobservable and seldom measured. This chapter focuses on the use of matched employer-employee data sets to carry out a version of this ideal test, but at the establishment level. When these data sets permit the measurement of the demographic characteristics of establishments' workforces, as well as the estimation of production functions, they can be used to infer productivity differentials between workers in different groups. Comparisons of these productivity differentials with wage differentials then provide versions of the ideal test for discrimination at the establishment level. In addition to providing tests of discrimination, matched employer-employee data sets have proven useful in studying other questions that arise in the economics of discrimination, including measuring labor market segregation and assessing its consequences, and examining hypotheses or predictions that are central to economic models of discrimination.