Regression Discontinuity Designs
A 3-Day Remote Seminar Taught by
Rocío Titiunik, Ph.D.
To see a sample of the course materials, click here.
This seminar focuses on methods for the analysis and interpretation of Regression Discontinuity (RD) designs. It will cover both introductory concepts and recent methodological developments.
The RD design is a non-experimental method that has high internal validity for estimating treatment effects. The design can be used when individuals are assigned to some treatment based entirely on a score—in education, this score is usually referred to as a “pretest score”. This could be any quantitative measure, such as an exam grade, income, age, or cholesterol level. All individuals whose score exceeds a predetermined cutoff are offered the treatment, while all individuals below the cutoff are not offered the treatment. For example, if a scholarship is given only to students who score 90 or more points in an exam, the effect of the scholarship could be analyzed with a RD design.
After treatment, an outcome is measured for all individuals–the “posttest score”–which could either be the same variable as the pretest score or a different measure. The analysis focuses on detecting possible discontinuities in the observed relationship between the pretest score and the outcome of interest at the cutoff, under appropriate continuity or local randomization assumptions.
Starting January 28, we are offering this seminar as a 3-day synchronous*, remote workshop for the first time. Each day will consist of a 4-hour live lecture held via the free video-conferencing software Zoom. You are encouraged to join the lecture live, but will have the opportunity to view the recorded session later in the day if you are unable to attend at the scheduled time.
Each lecture session will conclude with a hands-on exercise reviewing the content covered, to be completed on your own. An additional session will be held Thursday and Friday afternoons as an “office hour”, where you can review the exercise results with the instructor and ask any questions.
*We understand that scheduling is difficult during this unpredictable time. If you prefer, you may take all or part of the course asynchronously. The video recordings will be made available within 24 hours of each session and will be accessible for two weeks after the seminar, meaning that you will get all of the class content and discussions even if you cannot participate synchronously.
MORE DETAILS ABOUT THE COURSE CONTENT
RD designs appear naturally in cases where a policy is given to those who are most deserving or in greatest need. For example, a RD design might be implemented by ordering individuals from poorest to richest, and giving financial aid to the poorest individuals first until the budget has been exhausted. Although this would result in the treatment group being much poorer than the control group, the RD design relies on the assumption that near the cutoff, treated and control individuals would have had very similar outcomes in the absence of the treatment. Thus, one appealing feature of the RD design is that, if the required assumptions are met, it allows researchers to make internally valid causal inferences for those at the margin of switching from control to treatment assignment.
The course will discuss different assumptions under which the change in treatment status at the cutoff can be used to study causal treatment effects on outcomes of interest. The focus will be on methodology and empirical practice, not on theoretical results. The statistical and econometric theory underlying the results will be discussed at a conceptual, non-technical level.
The course will be based, in part, on the following two practical guides:
1. Cattaneo, Matias D., Nicolas Idrobo, and Rocío Titiunik. A Practical
Introduction to Regression Discontinuity Designs: Foundations. Forthcoming,
Cambridge University Press.
Click here for pre-publication draft.
2. Cattaneo, Matias D., Nicolas Idrobo, and Rocío Titiunik. A Practical
Introduction to Regression Discontinuity Designs: Extensions. In preparation,
Cambridge University Press.
Click here for preliminary draft.
This remote seminar is held via Zoom, a free video conferencing application. Instructions for joining a session via Zoom are available here. Before the seminar begins, you will receive an email with the meeting code and password you must use to join.
There will be several empirical illustrations, using Stata for the analysis. In addition, all functions and packages are also available in R, a free and open-source statistical software environment. The following Stata/R modules/commands will be used:
- rdrobust: RD analysis employing local polynomial and partitioning methods.
- rddensity: Manipulation testing for RD designs.
- rdlocrand: RD analysis employing randomization inference methods.
- rdmulti: Estimation and inference for RD designs with multiple cutoffs and multiple scores.
- rdpower: Power and sample size calculations using robust bias-corrected local polynomial inference methods.
All packages and associated references and help files can be found here.
All replication files for the empirical examples are available here.
If you’d like to take this course but are concerned that you don’t know enough R, there are excellent on-line resources for learning the basics. Here are our recommendations.
WHO SHOULD Register?
This seminar assumes that you have elementary working knowledge of statistics, econometrics and policy evaluation. It will be useful, but not required, if you are familiar with basic results from the literature on program evaluation and treatment effects at the level of Wooldridge (2010, Econometric Analysis of Cross Section and Panel Data, MIT Press). Nonetheless, the course is meant to be self-contained and most underlying statistics/econometrics concepts and results are introduced and explained in class.
1. Introduction to causal inference and program evaluation.
2. Introduction to RD designs; graphical illustration via RD plots.
3. Standard local polynomial methods and bandwidth selection for RD
4. Robust local polynomial methods for RD analysis; fuzzy RD designs.
5. Local randomization methods for RD analysis.
6. Falsification of RD designs.
7. Advanced RD topics: RD designs with discrete running variables and RD
analysis with additional covariates. If time permits: RD designs with multiple
cutoffs and geographic RD designs; RD extrapolation; power calculations
for RD analysis.
“I really enjoyed the RDD course. I feel like I can confidently go into a RD project using these materials. Fantastic lecturer! I really recommend this course to my colleagues.”
Yoshihito Goto, Kyoto University
“For anyone interested in non- or quasi-experimental framework for causal inference for their research, I highly encourage you to take this class.”
“This is an incredibly useful course for any social scientist. It offers a great overview of the RD framework with a variety of examples that help you understand how to apply and interpret this methodology. I strongly recommend.”
Gino Cattani, New York University