Test Construction and Targeted Factor Solutions Derived by Multiple Group and Procrustes Methods

An hypothesized multidimensional structure for items comprising a psychological measure generally requires empirical verification. Least-squares confirmatory factor strategies have involved (a) multiple group factor analysis and (b) targeted (Procrustean) rotations of arbitrary factor solutions. Alternatively, inferences of structure have been based simply on (c) product-moment correlations of item responses with total scale scores. All three methods were used in this study to verify empirically the assumed structures of data from four published psychological test batteries. Confirmatory maximum likelihood factor analysis was similarly evaluated in supplementary analyses. Because, among other reasons, solutions derived by multiple group analysis and item-total correlation analysis were generally most interpretable from a psychological perspective, it was concluded that their application to test construction is to be preferred over Procrustean or confirmatory maximum likelihood approaches.