Multitrait-Multimethod Comparisons Across Populations: A Confirmatory Factor Analytic Approach

The advantages of multitrait-multimethod methodology and the power of maximum likelihood confirmatory factor analysis are combined in an ordered framework for the comparison of covariance structures and true means across populations.

Abstract

The advantages of multitrait-multimethod (MTMM) methodology and the power of maximum likelihood confirmatory factor analysis are combined in an ordered framework for the comparison of covariance structures and true means across populations. First, a sequence of tests of and between hierarchically nested confirmatory factor analytic models is described for the analysis of measurement equivalence and construct validity across populations. Second, a similar sequence of model comparisons is proposed for the detection of true score-observed score regression intercept differences and true mean differences between populations. The proposed procedure is contrasted with MANOVA comparisons of group means: (1) use of MANOVA assumes test equivalence and validity across populations, whereas the present procedure permits statistical analysis of these assumptions; (2) MANOVA bases discriminant function coefficients partially upon observed differences between groups, whereas the current procedure weights each variate according to its correlation with an underlying construct. The possibility of spurious results from MANOVA and verdical results from the proposed methodology is demonstrated via application of both procedures to an artificial data set.