Confirmatory Factor Analyses of Multitrait-Multimethod Data: A Comparison of Alternative Models

Publisher Website
Google Scholar
Abstract
Alternative models for confirmatory factor analysis of multitrait-multimethod (MTMM) data were evaluated by varying the number of traits and methods and sample size for 255 MTMM matrices constructed from real data (Study 1), and for 180 MTMM matrices constructed from simulated data (Study 2). The correlated uniqueness model converged to proper solutions for 99% (Study 1) and 96% (Study 2) of the MTMM matrices, whereas the general model typically used converged to proper solutions for only 24% (Study 1) and 22% (Study 2) of the MTMM matrices. The general model was usually ill-defined (100% in Study 1, 90% in Study 2) for small MTMM matrices with small Ns, but performed better when the size of the MTMM matrix and N were larger. Even when both models converged to proper solutions, however, parameter estimates for the correlated uniqueness model were more accurate and precise in relation to known population parameters in Study 2.