A meta-analysis of the validity of personality/test scores in the prediction of employee turnover

The literature would suggest that the relationship between personality and turnover is weak and contradictory--leading numerous researchers to downplay the importance of personality constructs/ variables. The Hunter, Schmidt, and Jackson (1982) Bayesian validity generalization procedure was employed to determine whether this state of affairs might be explained by statistical and/or methodological artifacts. Rather than relying on published data as in most meta-analytic research, data in this study were collected on 24 personality test scores (utilizing both the Minnesota Multiphasic Personality Inventory and the Guilford-Zimmerman Temperament Survey) from 532 commercial real estate sales representatives in seven geographical regions. Utilizing traditional interpretational procedures led to the conclusion of situational specificity. However, when two sources of artifacts were removed statistically (range restriction and sampling error) and three held constant (criterion contamination and deficiency, predictor unreliability, and slight differences in factor structures between measures of similar constructs), nonsignificant overall variation in validities remained--thereby leaving little room for the operation of situational moderation. Support for generalizability (transportability) was strong in 12 test distributions, which had validities between -.02 and -.49. Results also indicated that a battery composed of the individual scales: Schizophrenia, Ego-Strength, L, K, and Anxiety, predicted 19% of the criterion variance. Moreover, the latter two scales acted as suppressor variables. Although limitations with the Hunter et al. procedure might exist when applied to personality-turnover relationships in a field setting, the procedure helps to explain the conflicting results across studies and shows the importance of not totally ignoring personality variables.