| Walter Leite
University of Florida
Combining latent growth modeling with propensity score matching to estimate the time-varying effect of student mobility
Propensity score matching has been used as a bias reduction strategy in the estimation of treatment effects in observational studies (Dehejia & Wahba, 2002; Rubin, 1973). Essentially, matching involves the use of a proximity function to pair treated individuals with similar, comparable non-treated individuals, based on the available resource of observable covariates. The problem of estimating time-varying treatment effects is more complex, because individuals that received the treatment at a later occasion are not comparable to individuals that received the treatment at earlier occasions (Raudenbush, 2001). Propensity score matching can be combined with latent growth models (within a structural equation modeling framework) (Bollen, 1989) to obtain an appropriate comparison group at each measurement occasion. Because the treatment varies with time, the membership in the comparison group will also vary with time, providing adequate counterfactual outcomes to be compared with the treatment outcomes.
The objective of this study is to present and evaluate a new method that combines propensity score matching with latent growth modeling to estimate the effect of a time-varying treatment. This new method has potential for extensive use with the longitudinal datasets collected by NCES. A demonstration of the new method will be performed with data from the Early Childhood Longitudinal Study, Kindergarten Class of 1998-1999. The objective of the application will be to estimate the difference in academic achievement between children who changed schools during the elementary school years and children who were at very high risk of changing schools but did not do it (i.e., the control group). Previous research has shown that students who change schools during elementary grades are more likely to have lower academic performances, repeat grades, drop out of school and exhibit symptoms of behavioral and emotional problems or stress-induced illnesses on account of the adjustments to new schools, teachers and classmates. A sensitivity analysis will also be performed to evaluate how selection bias would affect the estimates.
Back to Funded Research Grants Page |