Regression Discontinuity Design (RDD) Introduction
What you'll learn: How to estimate causal effects when treatment is assigned based on crossing a specific threshold.
The Core Idea
Regression Discontinuity Design (RDD) is a causal inference method that exploits a sharp cutoff rule in treatment assignment. Imagine a scholarship program that awards funding to all students who score 70% or higher on an entrance exam. Students scoring 69.9% get nothing; students scoring 70.1% receive the scholarship. This discontinuous jump creates a natural experiment.
The key insight: students just below and just above the 70% threshold are likely very similar in all characteristics—except one group received treatment and the other didn't. By comparing outcomes for people near this boundary, we can estimate the causal effect of the treatment.
Why RDD Works Near the Threshold
Far from the cutoff, treated and untreated groups may differ systematically (confounders!). But right at the threshold, assignment is effectively "as good as random." A student scoring 69.8% versus 70.2% probably differs only due to minor luck or measurement error, not fundamentally different abilities or backgrounds.
RDD focuses on this local region around the cutoff, where we can credibly compare outcomes and attribute differences to the treatment itself, not to pre-existing differences between groups.
The Visual Intuition
Picture a graph where the x-axis is the exam score and the y-axis is future earnings. If the scholarship has a causal effect, you'd see a discontinuous jump in the earnings curve right at the 70% cutoff—the line would suddenly jump upward as you cross from 69.9% to 70.1%.
Key Takeaway: RDD estimates causal effects by comparing individuals just above and just below a treatment assignment threshold, leveraging the fact that these groups are nearly identical except for treatment status.