An Evaluation of James et al.'s (1992) VG Estimation Procedure when Artifacts and True Validity are Correlated*

Abstract

James, Demaree, Muliak, and Ladd (1992) outlined a procedure for estimating the mean (M) and variance (V) of true validities (ρ's). This procedure was designed to take into account the potential nonzero intercorrelations among the three artifacts (predictor reliability, criterion reliability, and range restriction) and ρ. The accuracy of this new validity generalization procedure was compared with the accuracy of the Model 2‐based (in which correlations are individually corrected for artifacts) procedure because this latter procedure does not require the assumption of uncorrelated artifacts. The current study included two different ρ distributions, three sample sizes, and six different levels of intercorrelations among the three artifacts and ρ. Both procedures yielded relatively accurate estimates of M and V even when the intercorrelations among the three artifacts and ρ were nonzero. The Model 2‐based estimates were slightly more accurate than the James et al. estimates, and the accuracy of the Model 2‐based estimates was much more stable across sample sizes and different levels of intercorrelatedness. Sample‐based artifact data were used in this investigation.