The Effects of Differentially Computed Factor Scores on Statistical Decisions

Abstract

Factor scores generated via five different computational procedures were used, independently, as dependent variables in 117 univariate analyses of variance, 18 multivariate analyses of variance, 18 multiple discriminant analyses, and 14 canonical correlation analyses. Significance probabilities and their associated F ratios and chi-square. values were examined across the scoring methods to determine which of the differentially computed scores produced the smallest probability in each statistical test. Obtained frequencies (of smallest p values) were submitted to an appropriate chi-square test to contrast the effectiveness of the scoring procedures in the various statistical analyses. The results indicated no significant differences (at alpha =.05) in the observed versus expected frequencies across the five factor score algorithms in the univariate or multivariate tests. However a recurrent trend shown in the data suggests that the regressions technique may be the most effective way to compute subjects’ factor scores when they are to be subsequently used as criterion variables in statistical analyses.