Trends in predictive validity coefficients across time or repeated performance assessments show significant trends. Commonly used predictors show decreasing validities for the prediction of temporally more remote performance assessments. Within-study corrections for statistical artifacts across performance assessments increased the negative slopes of the regressions of predictive validities on ordinal position of performance assessment. The average corrected validity decrement from initial to final performance assessment was -.60; the trimmed average decrement was -.45; the median decrement was -.29. The average within-study correlation between predictive validity and time of performance assessment was -.80. An analysis of stability coefficients of Time Period X Time Period performance assessments revealed similar findings. Several theoretical explanations are discussed as reasons for predictive validity decrements.Studies of the predictions of future performance from current abilities have typically ignored the time facet in prediction equations. With the exception of Humphreys (1968;Humphreys & Taber, 1973), Hulin (1972), andFleishman (1960), the goals of most investigators have been to establish generalizability across populations of individuals, abilities (used as predictors), tasks, and situations. Most analyses of the generalizability of predictive relationships have examined whether variance in predictive validities across elements of these four facets or populations can be attributed to statistical artifacts or to real differences in predictive relationships (Hunter, Schmidt, & Jackson, 1982;Schmidt & Hunter, 1977;Schmidt, Hunter, & Caplan, 1981).A narrative review by Henry and Hulin (1987) of the literature relevant to the stability of predictive validities across time suggested that most empirical predictive validities were less stable than has commonly been assumed and that the instability was general across content areas. Henry and Hulin reported that temporally decreasing predictive validities have been found in most areas of skilled performance. Psychomotor skills such as discriminant reaction time (Fleishman & Hempel, 1954, 1955, two-dimensional tracking (Dunham, 1974), rotary pursuit (Fleishman, 1960), two-handed coordination (Fleishman & Rich, 1963), and student pilot performance during training (Alvares & Hulin, 1973) were typically found to have decreasing predictive validities or decreasing intertrial correlations. Stud-This study was supported in part by Contract #F33615-87-C-0014 from Brooks Air Force Base. Our views do not necessarily represent those of the Air Force.We thank James Austin, Kathy Hanisch, Lloyd Humphreys, and Mary Roznowski for comments on earlier drafts of this article. Their comments improved and strengthened the analyses and interpretation of the results.
