Estimating Unattenuated Correlations With Limited Information About Selection Variables

Abstract

Correcting attenuated correlations from selected samples is a common goal in organizational settings. Hunter and Schmidt introduced a procedure, called Case IV, for correcting correlations when a researcher has no information on the variable(s) used by an organization to form a suitability judgment. In this article, we compare Case IV to two other comparable procedures: the first correction (the expectation maximization algorithm) requires raw data about the selection variables used to form a suitability judgment. The second, the Pearson-Lawley correction, requires the variance-covariance matrix of the selection variables. We show that even when the variables used for selection are unobserved or unavailable, it is still possible to estimate parameters without making the restrictive assumptions of Case IV. In addition, these two corrections almost always outperform Case IV, particularly when the critical assumption of Case IV is violated. We also provide R code illustrating the use of these correction procedures.