Two Approaches to Estimate Multilevel Confirmatory Factor Analysis Models With Small Macro-Level Sample Sizes

• In Event: Advanced Topics in Structural Equation Modeling

Abstract

Multilevel measurement invariance is necessary to evaluate macro level structural paths (e.g., contextual effects). Unfortunately, multilevel measurement invariance testing is often impractical due to insufficient macro level observations (e.g., classrooms).This study explores two alternatives to traditional item-indicator estimation of a multilevel confirmatory factor analysis model (MCFA). Three research objectives are investigated: the extent to which (1) parceling and Bayesian estimation allow reliable parameter estimation with smaller macro level sample sizes and (2) permit reliable estimation with smaller micro level sample sizes; and (3) the extent to which parceling and Bayesian estimation benefits are impacted by construct reliability. Preliminary results reveal that parceling is a viable alternative that allows MCFA to be estimated with fewer macro units than traditional item-indicator MCFA.

Authors

• Houston Lester, University of Nebraska - Lincoln

• James A. Bovaird, University of Nebraska - Lincoln