Regression Discontinuity Design

Regression Discontinuity Design Training

The Regression Discontinuity Design certificate program delivers a rigorous, hands-on introduction to one of the most credible quasi-experimental methods in modern causal inference. This course is designed for graduate students, policy analysts, data scientists, and applied researchers who need to estimate treatment effects when randomized experiments are infeasible. Through twelve focused lessons, participants move from the potential outcomes framework and selection bias to sharp and fuzzy RD designs, local polynomial estimation, kernel weighting, and optimal bandwidth selection. The main practical outcome is the ability to design, estimate, visualize, and defend a complete RD analysis using real-world data in R, Stata, or Python.

The program is structured as a beginner-friendly progression that balances theoretical foundations with applied implementation, ensuring that even newcomers to causal inference can follow along while experienced analysts deepen their toolkit. Core skill areas include causal identification at the cutoff, graphical diagnostics and binning, bandwidth selection and robustness checks, and coding implementations across three major statistical environments. The curriculum also covers advanced extensions such as multiple cutoffs, geographic and time-based RD designs, and current research directions, making it both comprehensive and current. Given the growing demand for credible causal evidence in policy evaluation and academic publishing, this training is a timely investment for anyone seeking to produce defensible research in 2025 and beyond.

What is Regression Discontinuity Design?

Regression Discontinuity Design (RD) is a quasi-experimental method that identifies causal effects by exploiting a cutoff or threshold that determines treatment assignment. The core idea is that units just above and just below the cutoff are nearly identical in all relevant aspects, so any observed difference in outcomes can be attributed to the treatment. RD designs come in two main forms: sharp RD, where treatment is deterministically assigned by the cutoff, and fuzzy RD, where the cutoff influences the probability of treatment but does not guarantee it. The method relies on a continuous running variable, a clearly defined cutoff, and the assumption that potential outcomes are smooth at the threshold.

RD has become a cornerstone of modern empirical research in economics, political science, epidemiology, and education policy, largely because it requires fewer assumptions than other observational methods and produces highly credible estimates. It is widely used to evaluate the effects of scholarship eligibility, minimum wage changes, medical treatment guidelines, and electoral outcomes, among many other applications. Recent methodological advances have focused on robust bias-corrected inference, optimal bandwidth selection, and formal manipulation checks, making RD designs more reliable and transparent than ever before. With the rise of administrative data and open-source statistical software, RD is now more accessible and more rigorously scrutinized than at any point in its history.

Mastering RD builds a distinctive skill stack that combines causal reasoning, statistical programming, data visualization, and critical evaluation of research design. These competencies are directly applicable in academic research, government agencies, think tanks, and industry settings where policy decisions and program evaluations demand rigorous evidence. The ability to implement RD correctly — from choosing bandwidths to running manipulation checks — distinguishes a researcher who merely runs regressions from one who produces credible, publishable causal estimates. For professionals working with observational data, this expertise opens doors to more sophisticated analyses and more influential findings.