Statistical Methods for Analyzing Claims of Employment Discrimination

Abstract

This paper shows that in several recent EEO cases, lawyers and the courts have not used as powerful a statistical test of discrimination as they could have. The author describes two methods for combining into a single measure the results of statistical analysis of each of several data sets common in EEO cases, such as the hiring or promotion rates of minorities and majorities in each of several occupations in a company. He argues that these methods—combining one-sample binomial tests and the Mantel-Haenszel procedure—are more appropriate and usually more powerful than other tests of significance, such as Fisher's, that have been used in many EEO cases. He illustrates his argument with data from several of those cases.