Defining a "fair" or "unbiased" selection model: A question of utilities

Abstract

Describes a general selection model for selecting minority and majority applicants. The procedure consists of assigning utilities or losses to the "consequences" of selection decisions and deriving acceptance scores which maximize (minimize) the expected utility (loss). The question of "unfair" selection arises when applicants do not agree with the institution's utility (loss) assignment. 5 recently proposed models of fair selection are viewed in terms of an expected loss model and are shown to be procedures which implicitly assign loss values. It is concluded that utility (loss) values are always present in a selection problem, whether assigned explicitly or implicitly. Further, since applicants will probably never agree on a utility structure there cannot be a universal, fair, selection procedure. (PsycInfo Database Record (c) 2025 APA, all rights reserved)