Reliability Artifacts in Comparable Worth Procedures

Comparable worth principles suggest that regression line differences between male- and female-dominated jobs be examined where pay is regressed on job evaluation point values. When the lines differ, the typical recommendation is to raise the wages for female-dominated jobs using the male-regression line as a target wage line. We argue and demonstrate that unreliable job evaluation measures could artifactually produce regression line differences between male- and female-dominated jobs, even when no such bias exists. We subsequently examine reverse regression (Goldberger, 1984) and other procedures as alternative methods of determining bias. We show that lisrel procedures (Jöreskog and Sörbom, 1984) provide accurate and nonbiased estimates of wage discrimination in comparable worth analyses.