Optimal sampling strategies for validation studies

    loading  Checking for direct PDF access through Ovid

Abstract

In selection instrument validation studies the situation occasionally arises in which there are a large number of observations on the predictor but criterion data are very expensive or difficult to obtain, thus making it necessary to sample values of the predictor. Three strategies (random, rectangular, and extreme groups) for sampling predictor values were compared with respect to accuracy and statistical power in estimating the total group validity. Comparisons were made on samples drawn from 6 large N (approximately 10,000) bivariate test score distributions known to contain some departures from linearity and homoscedasticity. It was shown that in this situation selecting values of the predictor that form a rectangular distribution gave, in all instances studied, at least equal accuracy and greater statistical power in estimating the total group validity compared with random sampling. When the predictor-criterion relationship was generally linear with only modest departures from linearity, selecting values from the extremes of the predictor distribution was optimal in terms of accuracy and statistical power and clearly superior to rectangular sampling. (14 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved)

Related Topics

    loading  Loading Related Articles