An Empirical Comparison of Metric and Nonmetric Analyses of Correlation Coefficients

Abstract

Factor analyses and nonmetric multidimensional scaling analyses were carried out on the same matrices. For error free data, the solutions are identical under choice of a similarity transformation. For real data the solutions were similar with the similarity usually maximal when both solutions had the same number of dimensions. The effect of sample size was looked at using Monte Carlo methods. As sample size increased the sample solutions on the average more closely resembled the population solution. It was found that achieving monotonicity between sample and population correlations might require large samples.