A Comparison of Population-Averaged and Cluster-Specific Approaches in the Context of Unequal Probabilities of Selection

• In Event: 29.036 - Marvelous Multilevel Modeling

Abstract

Survey sampling designs are typically complex, involving multiple design features across multiple stages of selection. Researchers must appropriately account for these design features in order to obtain unbiased point estimators and make valid inferences about population parameters. Weighted single-level (i.e., population-averaged) and weighted multilevel (i.e., cluster-specific) estimators provide researchers with two alternative methods for analyzing survey data when the sample design involves both cluster sampling and sampling with unequal probabilities of selection. The present study used Monte Carlo methods to determine the relative performance of weighted single-level and multilevel estimators across different conditions, including different distributions, degrees of misspecification, levels of design informativeness, cluster sizes, and ICCs. Implications for applied education researchers are discussed.

Authors

• Natalie Anne Koziol, University of Nebraska - Lincoln

• James A. Bovaird, University of Nebraska - Lincoln

• HyeSun Lee, University of Nebraska - Lincoln