Evaluating Methods for Analyzing Subpopulation Data With Single-Level and Multilevel Pseudo Maximum Likelihood

• In Event: 42.034 - Estimation and Model Advances for Causal Inference and Treatment Effects

Abstract

The investigation of subpopulations helps contextualize our understanding of the education environment, and is thus critical to our pursuit of equal educational opportunities for all youth. In the context of complex probability sampling, there are multiple approaches for analyzing subpopulations. Whereas asymptotic theory provides some insight into the relative performance of these approaches under ideal sample conditions, their relative performance under more realistic conditions is unknown. Using simulation methods and an empirical example, this study examines the impact of subpopulation analysis method (multiple-group, zero-weight, subset) and estimation method (single-level vs. multilevel pseudo maximum likelihood) on the validity of subpopulation inferences. Results should be used to inform statistical practice with the ultimate goal of achieving more valid inferences in education research.

Authors

• Natalie Anne Koziol, University of Nebraska - Lincoln

• Jayden Nord, University of Nebraska - Lincoln

• Houston Lester, University of Nebraska - Lincoln