Modelos no lineales de ecuaciones estructuralesla influencia de las características de los modelos de medida en la precisión de las estimaciones

Abstract

Current doctoral dissertation assessed the impact of measurement model characteristics in the accuracy of nonlinear latent variable model estimates. In order to fulfill this goal, four Monte Carlo studies were conducted. The first study assessed the performance of the extended unconstrained approach (EXUC) and the latent moderated structural equation modeling (LMS) method, in situations where structural equation models (SEM) are comprised by linear, quadratic and interaction terms tested simultaneously, and we investigated the limitations of both procedures with regard to using parallel and congeneric indicators, a relatively large number of indicators and relatively low factor loadings. The second study analyzed the relevance of assessing interaction and quadratic effects as a base-line model when using the LMS method to fit a nonlinear SEM model and assessed the empirical consequences of structural model specification (i.e., correct specification, incorrect specification, underspecification and overespecification) in situations where sample size and composite factor reliability varied across conditions. The third study assessed the consequences for nonlinear model parameter estimates and Type I error rates of using items –instead of continuous indicators– comprised of five-response alternatives treated as continuous variables and the consequences of using item-parcels as continuous indicators of the measurement models used to fit nonlinear SEM models by using the LMS method. The fourth study aimed to explain how a new version of the LMS method adapted to categorical items (LMS-Cat) operates and how it can be implemented in Mplus; in addition, we assessed Type I error rates and power when using LMS-Cat to estimate models with a single interaction effect and models with interaction and quadratic effects tested simultaneously...