Robustness of a Limited Information Estimator to Misspecified Multilevel Structural Equation Models and Small Samples

• In Event: Innovative Methods in Structural Equation Modeling

Abstract

To provide accurate and reliable estimates of multilevel structural equation models (MLSEMs), full information maximum likelihood (FIML) requires sample sizes that are often unavailable in planned educational research. A bias-corrected limited information estimator (BCLI) has shown promise in addressing this gap but its performance relative to other estimation approaches has yet to be fully explored. Using three simulation studies, we evaluated BCLI estimation under model misspecification and with small sample sizes. We found the new estimator outperformed the prevalent current approaches in terms of bias, was robust to model misspecifications, and avoided convergence issues even at small sample sizes. Altogether, these results imply BCLI estimation to be an effective alternative to FIML estimation of MLSEMs.

Authors

• Kyle T. Cox, University of North Carolina - Charlotte

• Benjamin Kelcey, University of Cincinnati