Adding a dimension: Time as a factor in the generalizability of predictive relationships.

Abstract : An analysis of trends in predictive validity coefficients across time and repeated performance assessments shows highly significant and consistent trends in validities as a function of time and/or interpolated practice. Commonly used ability measures show decreasing predictive validities for the prediction of temporally more remote performance assessments. Within study corrections for differential restrictions of range and attenuation due to unreliability across the different performance assessments increased the negative slopes of the regressions of predictive validity on time or ordinal position of performance assessment. The median validity decrement from initial to final performance assessment, corrected for differential range restriction, attenuation, and within study sampling fluctuations was -.29. The mean of the trimmed distribution of corrected validity decrements, after eliminating the two most extreme cases, was -.45. The average within study correlation between predictive validity and time or ordinal position of performance assessment was 0.80. A similar analysis of stability coefficients of time period-by-time period or trial-by-trial performance assessment correlations revealed very similar albeit slightly more consistent findings. Theoretical explanations stressing the dynamic nature of human abilities, the changing nature of abilities required for task performance, and social competition factors are discussed as reasons for the predictive validity decrements.