An Empirical Comparison of the Accuracy of Univariate and Multivariate Corrections for Range Restriction

Abstract

Abstract : A study was conducted to determine whether uncorrected validity coefficients, those corrected for univariate range restriction, or those corrected for multivariate range restriction would be the most accurate in estimating the unrestricted population validity coefficients, and hence the most appropriate for use in Navy personnel section research. Correlations were computed between scores on Armed Services Vocational Aptitude Battery (ASVAB) predictors--ASVAB tests and school selector composites--and student performance scores for seven Navy technical schools with numbers of students ranging from 880 to 2598. These correlations were regarded as 'true' validity coefficients and each school regarded as a population. Restricted samples were selected from each 'population'. Validity coefficients were computed for each restricted; these values were corrected using the multivariate correction procedure. Validity coefficients that were corrected using a univariate and multivariate procedure were generally more accurate than those corrected using univariate procedure, and univariate corrected validity coefficients were more accurate than uncorrected validity coefficients. Keywords include: Corrections for range restriction; Test validity; validity coefficients; and Multivariate selection.