Syllabus

Propensity Score Matching: Theories and Applications

Sponsor: American Evaluation Association

Workshop Information

Course Description

Propensity score matching is becoming an increasingly popular technique for non-experimental evaluation research in the fields of public health, developmental economics, education, social work, and criminology.  In this workshop, we will discuss the rationale and importance of controlling for observed confounders in non-experimental research, and the role that propensity scores play in that regard.  We will demonstrate the common procedures of fitting a propensity score model, forming matched pairs, and using weighting to adjust for confounders, and estimating the effect of the treatment after the matching. Demonstration and hands-on activities will be conducted using add-on packages for R.

By the end of the course, you should have demonstrated the ability to discuss the theoretical basis of propensity score matching and related models, and know how to:

• import, export, and process data files between Excel and R,
• assess covariate balance before and after matching,
• fit a propensity score model to the data, and
• estimate the program effect after the matching.

Course Prerequisites

No prior knowledge of R or PSM is required, although a basic understanding of OLS regression might be helpful. You are encouraged to bring your own laptops.

Class Schedule and Activities

8:45am—9:00: Sign-in

9:00—10:30am: Introduction and Foundations

• Icebreaker
• Observational studies and causal Inference frameworks
• Principles of PSM and its limitations
• Commonly-used PSM methods
• Current issues and new developments

10:30—10:45am: Break (drinks in the hall)

10:45am—12:00pm: Introduction to R and related PSM programs

• Installing R software and PSM packages
• Importing external data
• Assessing covariate balance of unmatched data
• Fit stepwise propensity score model through logistic regression
• Form matched pairs
• Apply covariance adjustment
• Assessing post-match covariance balance
• Participants with laptops can follow the demonstration

12:00—1:00pm: Lunch Break (on your own)

1:00—2:30pm: Demo of PSM with more examples

• Navigating the process of optimal and full matching
• Impact analysis after matching using Zelig
• Advanced topics: Sensitivity analysis, instrumental variables, hierarchical PSM (depending on amount of time available)
• Participants are welcome to follow the demonstration to replicate the greedy, optimal, and full matching approaches and compare the results

2:30—2:45pm: Break (drinks in the hall)

2:45—3:45pm: Hands-on Group Activities

• Participants will work in teams to import assigned data files and assess initial covariate balance
• Participants will fit a PSM model and assess the improvement of covariate balance as a result of the matching
• The instructors walk around to help answer questions and provide technical guidance
• A representative from each team will present the results of their analysis to the class

3:45—4:00pm: Wrap-up and Course Evaluation

• Instructors will summarize the group activities
• Collecting course evaluation forms