Gender Discrimination Determination in Faculty Salary Patterns from Small Population Colleges

Abstract

Attempts were made to develop multiple linear regression models to represent salary patterns from two small population (N=91, N=44) colleges. Multiple discriminant, canonical, and set correlation analyses were used to confirm the presence or absence of "tainted" variables. Problems with multicollinearity were solved by removing variables. "Fixed" models were formulated after using variable selection techniques to determine statistical significance. Entry salary (which acted as a suppressor variable) did not have a linear relationship to salary and the models involving it violated the normality of error terms assumption. Average percent increase in salary was used instead. However, the presence of heteroscedasticity in models for both colleges could not be eliminated. For these colleges, using multiple linear regression to determine, statistically, the presence or absence of gender discrimination in salary patterns was not possible.