Abstract

Although structural equation models with observed variables are used in some psychological disciplines, they are not always recognized as a special case of models with latent variables. When these models are recognized as belonging to the family of structural equation models, the authors tend to report the same type of fit indices. In this paper, we evaluated whether these indices are appropriate when latent variables are not considered in the model. We studied the behavior of some well-known indices with a Monte Carlo experiment using four types of models and by manipulating the error. The analysis revealed that selected fit indices are generally independent of the error variance in the model, which in turn may render them not particularly informative. The only sensitive index is the Total Coefficient of Determination (CD), which does not seem to be correlated to the performance of other considered indices. Therefore, in structural equation models without latent variables, we propose to use the CD along with other indices.