UTILITIES AND THE ISSUE OF FAIRNESS IN A DECISION THEORETIC MODEL FOR SELECTION

Abstract

Selection procedures in education and employment have come under increasing scrutiny in the last decade due to the rising concern for civil rights and equal opportunity. Such scrutiny has come primarily from minority racial-ethnic groups and from women concerned that traditional selection processes discriminate against them. Minority groups have challenged employee selection practices through the courts (e.g., the notable Griggs v. Duke Power) and in several instances current hiring procedures have been ruled illegal. While many such cases continue in litigation, a more recent trend is the charge by white males that affirmative action and compensatory efforts for disadvantaged groups create reverse discrimination by hindering their own fair opportunity for education and employment. A notable example of such a charge was DeFunis v. State of Washington. The Supreme Court ruled the DeFunis issue moot leaving the question of reverse discrimination (like that of discrimination against minorities and women) largely unresolved. The political and legal unresolve about what fairness is, and to whom concern should be directed, has paralleled considerable unresolve about the same issues when debated on technical rather than policy grounds. From the technical standpoint, the traditional statistically-prescribed selection procedure has called for the use of quantified assessments such as test scores in regression equations, as predictors of some performance or criterion variable. Selection is then based on the predicted criterion. Until recently, such a procedure was accepted without question as fair and appropriate. In the past few years, however, questions about this widely accepted regression approach have been raised (Cole, 1973; Darlington, 1971; Einhorn & Bass, 1971; Thorndike, 1971). The questions focus on what it means to be fair and whose concerns for fairness deserve greatest weight. Different authors have proposed different definitions of fairness, and have produced objective statistical selection techniques to satisfy their own definitions. The result of the statistical debate of the past five years is a growing recognition that the issues of fair selection are not primarily technical ones but are matters of value judgments about the proper definition of fairness. We do not see much profit in seeking a single universally-accepted definition of fairness. There is, however, a candidate for such a definition in the concept of giving equal treatment to equal individuals and reserving preferred treatment for those in-