Fun with numbers: Alternative models for predicting salary levels

The increasing awareness and concern with equity issues in higher education, along with the escalating litigation, has prompted institutions to undertake salary prediction studies. Four prediction models (built on a males only and total sample) were compared: (1) entering all variables, (2) excluding rank and tenure, (3) using predicted rank and tenure, and (4) using only objective variables. Models were tested using all permanent full-time faculty at a large midwestern university. Using predicted rank and tenure was the most suitable for equity studies. Including all variables yielded the best results for explaining/predicting reward systems. The other two models did not appear appropriate for either purpose. The males only sample consistently produced the largest bias effects. Institutions considering a salary prediction study should find these outcomes helpful in determining appropriate analytical strategies.