Validity Generalization: A Critical Review

Since Schmidt and Hunter's seminal publication in 1977 (Schmidt & Hunter, 1977) on situational specificity, the assessment of the generalizability of organizational interventions has received a great deal of attention among researchers and practitioners, especially among industrial/organizational (I/O) psychologists concerning the generalizability of the validity of predictors across organizations. The art and science of validity generalization (VG) revolves around the estimation of the mean and variance of population validities. That is, given a set of k validity coefficients (correlations between the same or similar predictors and criteria in the same or similar jobs) obtained from samples drawn from k different populations, one is interested in obtaining an estimate of the mean and variance of population validities in order to establish whether the validity in question is (a) significant and substantial and (b) generalizable across populations. An estimate of the mean of population validities is used to answer the first question and an estimate of the variance of population validities is used to answer the second question. The estimation of the mean and variance of population validities and the resulting substantive interpretations is the crux of VG research and practice.

There are currently several VG models and procedures for estimating the mean and variance of population validities. Some models and procedures are designed for use with observed correlations

-263-