On choosing the appropriate measure of association when analyzing rating scale data

This paper reviewed the current state of the literature on the appropriateness of various correlation measures when data are ordinal. Emphasis was on the use of these measures as input to multivariate analysis. Based on the review, four correlation indexes were further studied via a simulation design. The simulation results indicated that the polychoric correlation outperformed product-moment, Spearman’s rho and Kendall’s tau-b measures on the basis of bias and squared error criteria. The present study suggests that the chapter investigating pairwise correlations may be closed and attention may now be directed towards non-normal data and actual multivariate analysis with correlation matrices.