Social worker researchers have recognized the importance to control for selection bias and other threats to internal validity, and have started to employ PSM in program evaluation and analysis of survey data. For instance, Littell found that the outcomes of family preservation services (FPS) varied by the level of service participation, which partially explains why results regarding the effectiveness of FPS are inconsistent. Because participation levels of FPS are largely self-selected, even when treatment is mandated, active participants are likely to be different from passive or resistant recipients on outcomes. Thus, it is important to model the heterogeneity of service participation, and develop a modeling process that explicitly controls for participation levels. This strategy can complement randomized clinical trials in the pursuit of a rigorous program evaluation across the entire range of human services.

Moving from agreement that propensity score matching (PSM) is a desirable strategy to implementing PSM has been more difficult. Researchers often find that running propensity score analysis is troublesome, because the approach remains relatively new, and none of the commercial software packages offer a formal procedure to facilitate PSM analysis.

Last year at SSWR we presented an introduction to PSM using a SAS-macro that allowed one-to-one (caliper) matching. We have continued our pursuit of information about PSM options and have concluded that PSMATCH2 (developed by Edwin Leuven and Barbara Sianesi, as a user-supplied procedure in STATA) is the most comprehensive package that allows users to fulfill almost all kinds of tasks for PSM, and the routine is being continuously improved and updated.

Contents: Using illustrating examples from social work research, this workshop will demonstrate how to run propensity score matching using STATA/PSMATCH2. It will focus on the following topics:

1. Introduction to PSM by reviewing the counterfactual framework, the major steps involving in PSM, popular matching algorithms, and Heckman’s difference-in-differences method; 2. Running one-to-one or one-to-many matching using nearest neighbor(s) with caliper; 3. Running Mahalanobis-metric matching with or without propensity score as a matching variable; 4. Running Heckman’s difference-in-differences using the Gaussian kernel, Epanechnikov kernel, and local linear regression methods; 5. PSM analysis in conjunction with logistic regression, survival modeling, growth curve, and structural equation modeling.

Pedagogical Techniques: Teaching methods include lecture, PowerPoint presentation, and computer demonstration.

References

Littell, J.H. (2001). Client participation and outcomes of intensive family preservation services. Social Work Research, 25(2): 103-113.