A critical re‐examination and analysis of cognitive ability tests using the Thorndike model of fairness

Abstract

The literature investigating the bias of cognitive ability tests (CATs) is often conflated with the controversy surrounding which method for determining test bias is superior. The general acceptance of the Cleary (1968) model of test bias in industrial/organizational psychology has served to deter evaluations of tests against other models of test bias because acceptance of the Cleary model as ‘superior’ implies the limited relevance of investigations of tests against other models of bias. Although these other models are not considered to be models of predictive bias in the psychometric sense, they nonetheless have significant implications for workplace diversity. Most notably, the existing literature lacks the precision and depth necessary to extrapolate the actual false‐rejection rate in selection decisions that burden visible minority groups when CATs are used. The current study identifies these gaps in the literature in addition to evaluating CATs against the Thorndike (1971) model of test bias. Results indicate that a one standard deviation (SD) difference in Black‐White CAT scores is associated with a Black‐White difference in job performance of approximately 1/3 SD. The Black‐White difference in job performance is reduced to approximately 1/10 SD when objective, rather than subjective, job performance criteria are used. We therefore conclude that CATs are biased against Blacks when evaluated using the Thorndike model. The implications for use of CATs in personnel selection are discussed.