Mistaking Labor Market Effects for Unequal Treatment in Regression Analysis

Abstract

Unequal treatment is defined as requiring a more severe test for one person to get a position or reward than for another. It is shown how a changing labor market can generate differences between the average amount of salary paid to men and women in the absence of unequal treatment. It is then demonstrated by generated models how using imperfectly measured independent variables commonly will not take account of the labor market effect in an application of regression analysis. While methodologists appreciate this limitation, it is ignored under particular circumstances. A method is then presented for showing unequal treatment without ambiguity. It is noted that when unequal treatment is not demonstrated by this method, additional procedures may be used to examine alternative ways in which unequal treatment can occur. Data are presented from the City University of New York to illustrate an application of the method.