Comparing Two Small-Sample Correction Methods in Three-Level Models

• In Event: AERA Roundtable Session Wednesday 3:45 pm - Gold 3
  In Roundtable Session: Recent Developments in Multilevel Modeling: Model Estimation and Evaluation

Abstract

Multilevel models are widely used in education research but often encounter small-sample constraints at one or more levels. This study uses a full factorial Monte Carlo simulation to evaluate two small-sample correction methods—REML with Kenward–Roger adjustment and CR2 with Satterthwaite degrees of freedom—in three-level models with treatment assigned at either the top or middle level. Simulations reflect realistic educational designs and systematically vary sample sizes, treatment allocation, covariate presence, ICCs, and cluster size balance. By comparing performance across 640 conditions, this study will provide timely and practical guidance for researchers analyzing school- and classroom-based interventions when conventional multilevel assumptions are violated due to small-cluster designs.

Authors

• Huibin Zhang, University of Tennessee

• Zuchao Shen, University of Georgia