IN SEARCH OF FAIR SELECTION PROCEDURES

Prior to 1971 the primary notions of what constitutes unfair test use in selection (commonly labeled test bias) were reasonably consistent, at least among specialists in measurement. As summarized by Bray and Moses, "The key question turned out to be whether test scores meant the same thing in terms of their predictions of job performance for those who were disadvantaged and those who were not" (1972, p. 548). For other purposes "job" might be replaced by "educational" and "disadvantaged" by "members of a minority group" but otherwise the Bray and Moses statement could apply to selection in educational as well as employment settings. The most commonly used approach for evaluating whether test scores meant the same thing in terms of prediction for members of different groups was based on what Petersen and Novick (1976) refer to as the Regression Model. This model, which has sometimes been referred to as the Cleary Model (1968), enjoyed fairly general acceptance with some minor deviations. Assuming a linear regression model, the prediction systems for different subgroups