Comparative study among Mplus, AMOS and LISREL in a model with categorical data and small sample

Abstract

Structural equation models can be solved using several specific software. The precepts underlying the asymptotic theory – large samples, continuous or numerical variables, and reflected indicators – constitute the statistical and mathematical foundation for identifying the model's parameters. However, empirical studies cannot always be carried out in this ideal scenario. Small samples and categorical data are commonly found in quantitative work in the social sciences. The Mplus app has enormous applicability in quantitative research that makes the assumptions of asymptotic theory more flexible. From a structural equation model that analyzes the effectiveness of a public policy, the Mplus, AMOS, and LISREL software – which are widely used in the resolution of structural models – with their respective discrepancy functions will be used to verify the convergence of parameters in a survey with a small sample and categorical data. In these situations, according to the specialized literature, Mplus is the most appropriate application. It is verified that the processing of data in AMOS with the bootstrapping technique presents results similar to those of Mplus. Therefore, in the absence of Mplus, the use of AMOS with bootstrapping can result in the estimation of reliable parameters for non-numeric data.