Testing the Impact of Omitted School Variables in Hierarchical Linear Models and Obtaining Robust Statistical Estimators

Jee-Seon Kim

Despite decades of effort, results concerning the effects of school variables on student learning remain mixed. Recently, it has been argued that these inconsistent findings may be due in part to the inappropriateness of the models utilized in the statistical analysis. For example, if school and teacher characteristics are not included in a model, the tests for the effects of variables included in the model are invalid unless it is assumed that school and teacher characteristics are uncorrelated with the variables in the model. This restrictive assumption is often violated in reality, leading to what is referred to as omitted variable bias. Even though researchers are aware of the danger of this bias, it is often infeasible to collect the requisite data in studies that are strictly observational.This project develops and investigates a battery of statistical tests for assessing the impact of omitted school variables in the analysis of complex educational data sets. Using a general multilevel modeling framework, this project aims to integrate a number of existing approaches into a unified methodology, and it provides new options for handling omitted variables in the types of hierarchical and/or longitudinal data sets that are ubiquitous in educational research.

About Jee-Seon Kim

Jee-Seon Kim is an Assistant Professor of Quantitative Methods in the Department of Educational Psychology at the University of Wisconsin-Madison. She was born and raised in Seoul, Korea before coming to the US for her doctoral study at the University of Illinois at Urbana-Champaign. Following her graduate study, she received the 2002 Mary Catherine Ellwein Outstanding Dissertation Award for Quantitative Methods from the American Education Research Association. Her research interests are concerned with the development and application of quantitative methods in education and the social sciences, especially focusing on multilevel models and other latent variable models, methods for modeling change, learning, and human development using longitudinal data, categorical data analysis, issues related to omitted variables, and school effectiveness.