Evaluating the Effects of Different Informative Priors on Bayesian SEM Model Fit for Personality Measures

• In Event: AERA Roundtable Session Friday 8:00 am Four Seasons Ballroom 4
  In Roundtable Session: Advancing Psychometric Modeling: Innovative Techniques for Educational Assessment (Table 8)

Abstract

As the selection of the prior distribution is tricky, evaluating the impacts of the informative prior distribution on model fit is critical in Bayesian Structural Equation Modeling. The purpose of this study is to evaluate and compare the posterior predictive P-value and recently developed approximate fit statistics (BCFIs, BTLIs, and BRMSEAs) of BSEM bifactor models with different sets of informative priors using IPIP-NEO-120 data. BSEM with informative cross-loading and residual covariance priors produced the greatest average BCFIs and the lowest average BRMSEAs. The present study aims to provide researchers with useful information and practical guidelines to apply BSEM techniques to designing personality inventories and other self-report measures and inform the strengths and weaknesses of using different informative priors in BSEM.

Authors

• Hyeri Hong, California State University - Fresno

• Kahee Han, University of Kansas

• Alfonso J. Martinez, Fordham University