Unbiased Estimation of Multilevel Structural Equation Models in Small to Moderate Sample Sizes: A Limited Information Estimator

• In Event: 34.094 - Fairmont Royal York Roundtable Session Six
  In Roundtable Session: 34.094-5 - Partially Clustered Data and Sample Size Investigations With Various Types of Multilevel Models

Abstract

Full information estimation of structural equation models (SEMs) is the preferred approach to operationalize sophisticated theories involving latent variables. Although full information estimators have desirable properties, a major limitation is that they often incur significant bias when implemented in small-to-moderate samples (e.g.,<200). Within the multilevel contexts of education, this critical limitation scales undesirably to the number of clusters (e.g.,schools) under multilevel SEM (MLSEM).
We develop a new class of limited information estimators that can be an attractive alternative for MLSEM with small-to-moderate samples (e.g.,<100 clusters). The proposed bias-corrected limited information method allows researchers to operationalize and test sophisticated theories involving latent variables at multiple levels while reliably accounting for the detrimental effects of measurement error even in small-to-moderate sample sizes.

Authors

• Benjamin Kelcey, University of Cincinnati

• Kyle T. Cox, University of Cincinnati

• Nianbo Dong, University of North Carolina at Chapel Hill