The purpose of this project is to develop user-friendly software and clear guidelines for using propensity score methods with complex survey data. While the randomized control trial remains the strongest design for rendering causal inferences, random assignment is often not feasible for a variety of practical reasons. One such situation occurs with national datasets, which provide a wealth of data but which typically feature no manipulation of an independent variable. Among the quasi-experimental approaches available to researchers is propensity score matching, a technique that has been extensively researched and developed but which has not been studied much in conjunction with complex, large-scale survey data.
The research team will use Monte Carlo simulations to investigate various approaches to propensity score estimation and propensity score usage that take into account complexities of survey data. They will then apply the techniques to two real-data studies of the Early Childhood Longitudinal Study – Kindergarten Cohort (ECLS-K) data. The research team will utilize a website, seminars, and peer-reviewed publications to describe the propensity score techniques that were successful and to provide software for conducting those techniques. Researchers will also provide guidelines for determining which approach to use for various applied research designs.
