| Sumario: | Using both the Bayesian validity generalization procedure presented in Schmidt, Hunter, Pearlman, and Shane and a second, recently developed Bayesian procedure, this study shows that most of the between-study variance in four distributions of observed validity coefficients of the Programmer Aptitude Test for measures of proficiency on the job is artifactual in nature. The average percentage of variance accounted for by artifacts was 69 using the Schmidt et al. procedure (without the Fisher's z transformation) and 65 using the new procedure. These figures were lower (39% and 41%, respectively) for the single distribution of training criterion validities. Results for both procedures indicated that validities were generalizable to new settings in four of five cases. Further analyses indicated that corrections for sampling error alone were sufficient to support the conclusions of validity generalizability. These findings taken together indicate that validity generalization is a robust phenomenon.
|
|---|