Model Fit in Confirmatory Factor Analysis: Differences Across Software

• In Event: Poster Session 11
  In Poster Session: Common Latent Variable Modeling Issues and Potential Answers

Abstract

Monte Carlo simulation was used to examine potential differences in model fit statistics (χ2, CFI, NNFI, RMSEA, and SRMR) across various structural equation modeling (SEM) software packages (LISREL, Mplus, EQS, R Lavaan) when testing confirmatory factor analysis models. Manipulated variables included sample size, model size, data type (continuous; 5-point ordinal scale), specification (correct, minimally misspecified, and moderately misspecified), data distribution, and estimator (ML, robust ML, DWLS, WLSMV). Model fit statistics were only consistent across software packages and versions with correctly specified models and were highly discrepant when models were misspecified. The implications of this study are far reaching, including software shopping to force results, consequences in theory testing, and journal reviewers’ decisions to accept or reject a manuscript.

Authors

• Susan R. Hutchinson, University of Northern Colorado

• Eric D. Teman, University of Wyoming