Item selection counts: a comparison of empirical key and rational scale validities in theory-based and...

Little explicit attention has been given to the impact of item pools on the validities and cross-validities of different background data scoring approaches. This study tests the idea that pools of items theoretically related to the performance of interest will outperform pools of items with no hypothesized relationship with the criterion. Validities and cross-validities of rational scales and empirical keys created from theory- and non-theory-based item pools were compared for 3 criteria. When size of the item pools was held constant, theory-based empirical keys (correlational and vertical percent) and rational scales showed larger validities and cross-validities than non-theory-based empirical keys (correlational and vertical percent) and showed minimal shrinkage in cross-validities. Even when item pool for the non-theory-based keys was expanded to include all items in the instrument, the theory-based keys showed comparable or slightly better validities and cross-validities for 2 of the 3 criteria, including college GPA, which was separated from the predictors by 4 years.