Evaluating Imputation-Based Fit Statistics in Structural Equation Modeling With Ordinal Data: The MI2S Approach

Abstract

The multiple imputation two-stage (MI2S) approach holds promise for evaluating the model fit of structural equation models for ordinal variables with multiply imputed data. However, previous studies only examined the performance of MI2S-based residual-based test statistics. This study extends previous research by examining the performance of two alternative test statistics: the mean-adjusted test statistic ( T M) and the mean- and variance-adjusted test statistic ( T MV). Our results showed that the MI2S-based T MV generally outperformed other test statistics examined in a wide range of conditions. The MI2S-based root mean square error of approximation also exhibited good performance. This article demonstrates the MI2S approach with an empirical data set and provides Mplus and R code for its implementation.