Evaluation of Propensity Score Strategies With Multilevel Data When Treatment Assignment Mechanism Varies Between Clusters

• In Event: Advances in Statistical Methodology

Abstract

Secondary data analyses are becoming popular due to the increased availability of national studies. One issue with secondary data analyses is that data is collected before particular study so it is impossible to randomly assign to treatment. Hierarchical structure is common for large scale datasets. Researchers should take additional step to deal with non-random treatment assignment and hierarchical nature of the data. However, it is hard to fit parametric complex multilevel models to model the relationship. We will conduct a Monte Carlo simulation study to understand consequences of fitting random intercepts and generalized boosted regression models to estimate propensity scores when a complex model is required. Objective is to compare parsimonious models to data mining methods in multilevel observational studies.

Authors

• Sungur Gurel, University of Florida

• Walter L. Leite, University of Florida