One-Step Ordinal Structural Equation Modeling Using Latent Trait Scores: A Bayesian Perspective

• In Event: Structural Equation Modeling With Ordinal/Categorical Data

Abstract

The present study proposes and demonstrates a one-step Bayesian estimation of ordinal structural equation models (SEM) that use latent trait scores from graded response models (GRM). The approach is effective even in small samples and non-normal data commonly encountered in behavioral research. Using a small scale simulation, this study addresses issues related to incorporating standard errors at the appropriate levels. Credibility intervals from the simulation illustrate the penalty in disregarding the errors. The posterior means and standard errors of path coefficients are underestimated in the two-step approach that ignores the errors in parameter estimates from the first step. The final paper will include an illustration of the technique using empirical data.

Author

• Prathiba Natesan, University of North Texas