The Robustness of 2SLS Estimation of a Non-normally Distributed Confirmatory Factor Analysis Model

Publisher Website
Google Scholar
Abstract
A Monte Carlo study was conducted to assess the robustness of the limited information two-stage least squares (2SLS) estimation procedure on a confirmatory factor analysis model with non-normal distributions. Full information maximum likelihood (FIML) methods were also used for a comparison. Both procedures were used to estimate model parameters, contingent upon a design utilizing two levels of model loadings (small versus large) over four different distribution conditions (normally distributed, small asymmetry, medium asymmetry, and large asymmetry, with kurtosis held constant). One hundred model replications were conducted to generate the study data. For the model and parameters studied, results indicated both estimators were robust with respect to moderate deviations from normality. Both estimators produced similar estimates across all conditions studied, thus providing a measure of high relative efficiency for the 2SLS procedure.