Validity Generalization of Navy Selector Composites

Abstract

Abstract : In recent years, Schmidt and Hunter (1977) and their colleagues have presented strong evidence that most of the variability in validity coefficients among various studies may be artifactual in nature. That is, the variability in observed validity coefficients across studies may be due to a number of factors, chief of which are sampling error, criterion unreliability, test unreliability, and restriction in range of ability of sample members. The validities of interest for validity generalization therefore are estimates of true validities. The mean and standard deviation of the true validities are obtained by correcting the mean and standard deviation of the observed validity distribution for criterion unreliability and restriction in range. The resulting distribution of estimated true validities is considered the Bayesian prior distribution whose mean, standard deviation, and credibility value (CV) represent the degree of validity generation represented by that particular distribution of test composite scores. Rejection of the hypothesis of validity generalization for a distribution of validities requires that the 90% CV include zero.