Parsimony-Based Fit Indices for Multiple-Indicator Models: Do They Work?

A frequently used type of model in applications of covariance structure analysis is one referred to as a multiple-indicator regression model. This study takes a simulation approach to investigate seven parsimony-based indices used to evaluate this type of model. Four representative theoretical models were examined, and the number of indicators used to represent latent variables was varied with two of the models. Both correctly and incorrectly specified models were fit to the data. The results show that the Akaike information criteria, the root mean square index, and the Tucker-Lewis index were the most effective indices. The implications of the findings for the model selection process are discussed.