Discrimination: Fact or Fiction? An Examination Using an Alternative Approach

Women earn less than men. Blacks earn less than whites. These facts are blatantly observable in the economy. With the proliferation of civil rights legislation in the last two decades, social scientists have become increasingly concerned with the determination of the wage gap attributable to discrimination. Because workers have different productivities, statistical methods have been used to isolate the discrimination effect from wage differences justified by differences in productivity. The primary method estimates an equation with wages expressed as a function of productivity and sex or race. Since productivity cannot be directly observed, a number of proxies for productivity have been used, such as education, tenure, and experience. Statisticians [11] have shown that the necessity of using proxies results in statistical biases which may tend to overestimate discrimination. They have shown that mismeasuring any independent variable can impart a bias on each regression coefficient, thereby causing a severe misestimate of discrimination. These errors are proportional to the degree of mismeasurement (the relative degree of error variance in the independent variables) as well as to the amount of intercorrelation between each variable and race or