423 – Contributed Oral Poster Presentations: Social Statistics Section
Propensity Score Analysis with Nested Data: Comparing Single and Multilevel Model Estimates
Aarti P. Bellara
University of South Florida
Kathryn M. Borman
University of South Florida
Patricia Rodriguez De Gil
University of South Florida
Eun Sook Kim
University of South Florida
Jeffrey Kromrey
University of South Florida
Rheta E. Lanehart
University of South Florida
Reginald Lee
University of South Florida
Propensity score (PS) methods provide viable strategies for reducing selection bias in nonexperimental (observational) studies. Most research on PS methods model the treatment assignment so that the estimated probability of receiving treatment allows for the identification of comparable individuals based on their individual characteristics. However, in nested data structures selection bias might result not only from differences in the characteristics of the individuals but also from differences in group membership. This study investigated differences in PS results from single-level and multi-level models. Data from an NSF funded project included school transcripts, demographics, enrollment, and achievement data. The impact of special educational programs on advanced mathematics course enrollment was investigated. Data were analyzed by comparing PS distributions, estimating the correlations between the two sets of propensity scores, and comparing the estimates of treatment effects. Results suggest a strong correlation between the PS obtained from single-level and multi-level models and only modest differences in resulting score distributions and estimates of treatment effects.