Date of Award
Doctor of Philosophy in Psychology
Wayne F. Velicer
Researchers who employ multivariate techniques often wish to compare solutions obtained from two independent studies. When factor analysis or component analysis procedures are employed, pattern matrices are compared, In the present study, the matching performance of several component matching indices, coefficient of congruence(c), s – statistic (i), Pearson r (r), and kappa (k), were compared under a variety of experimental conditions. Independent variables manipulated were (1) component saturation or size of component loadings, (2) sample size, (3) the number of variables, and (4) the number of components. Five sample component patterns were computer generated and matched, employing each index, to their population component pattern. Matching performance was assessed both within indices and between indices. No index was free of the influence of the variables manipulated. Component saturation and sample size exerted the primary influence on index matching performance. An increase in either of these variables resulted in index values that more accurately reflected whether or not components matched. With the exception of r, little difference in matching performance between indices was observed. For all practical purposes, matching performance for c, s, and k was indistinguishable when both corresponding components and non-corresponding components were compared.
Guadagnoli, Edward, "A Comparison of Several Component Matching Indices: A Simulation Study" (1986). Open Access Dissertations. Paper 935.