Sample Size Requirements For Structural Equation Model Selection

Abstract

typical application of structural equation modeling in social and behavioral studies is to evaluate a theoretical hypothesis by selecting the best-fit model among several candidate models using a selection criterion. Determining the sample size required for identifying the true model from alternative ones is challenging since sample size requirements change as a function of variable type, model properties, and choice of estimation method. Although several rules-of-thumb exist for advising applied researchers, they are not model-specific and may lead to incorrect model selection. This study uses Monte Carlo simulation to estimate the sample size requirement for selecting the true model from alternative models with different degrees of misspecification. The effect of the number of latent and observed variables, the size of factor loadings and path coefficients, and the pattern of missing values is investigated systematically. We will also examine the effect of the estimation method used and the types of variables in the model. The empirical relationships between sample size requirements and a range of choices for parameter values will be explored to provide practitioners more specific guidance about what sample size is required for a specific model.