Methods for Handling Missing Non-Normal Data in Structural Equation Modeling

Abstract

Methods for dealing with non-normal data have been broadly discussed in the structural equation modeling (SEM) literature. The issue of how to properly handle normal missing data has also received enough attention. However, much less research has been done to deal with the situation where non-normality and missingness coexist. Generally speaking, there are three classes of methods for dealing with missing non-normal data (continuous and ordinal) in SEM: a) robust procedures, b) Bayesian analysis, and c) multiple imputation. None of these methods, except for robust full information maximum likelihood (robust FIML), have been systematically evaluated in the SEM context with incomplete non-normal data. In this dissertation, I investigated and compared the performance of the three classes of methods under a broad range of conditions for the two types of missing non-normal data.