Correct Estimation Methods for RMSEA Under Missing Data. Issue 2 (4th March 2021)

• Record Type: Journal Article
• Authors: Lai, Keke
• Is Part Of: Structural equation modeling. Volume 28:Issue 2(2021)
• Page Start: 207
• Page End: 218
• Publication Date: 2021-03-04
• Subjects: RMSEA -- point estimation -- confidence interval -- missing data; Multivariate analysis -- Periodicals; Social sciences -- Statistical methods -- Periodicals; 519.535
• DOI: 10.1080/10705511.2020.1755864
• Languages: English
• ISSNs: 1070-5511

Abstract

ABSTRACT: The full information maximum likelihood (FIML) and the two-stage (TS) procedure are two popular likelihood-based approaches to estimating structural equation model (SEM) parameters with missing data. Although both FIML and TS can give consistent model parameter estimates and standard errors, the RMSEA point estimate and confidence interval (CI) based on these two procedures are problematic. In this paper, we propose a correct point estimator and a correct CI for the population RMSEA, given missing (completely) at random normal data. We first explain with theoretical derivations why the new methods are correct but the traditional methods fail. Then we conduct three simulation studies to compare the methods using various model types (CFA; SEM; growth), missing mechanisms (MAR; MCAR; mixture of MAR and MCAR), missing data amounts, and sample sizes that are typical in the social sciences. Results indicate that the new methods perform well whereas all the traditional methods break down.