Use regression discontinuity design when you have a clear cutoff point that determines who receives an intervention. It allows you to estimate the causal effect of a policy by comparing people just above and just below a threshold, assuming those near the border are otherwise similar to one another.
Understanding Threshold-Based Impact
Many policies or business rules operate on a sharp threshold. Whether it is a marketing incentive triggered by a spending limit or a financial aid program based on income, these rules create a natural experiment. Regression discontinuity design targets these borders to isolate the impact of the policy itself from other confounding factors.
By focusing on the area immediately around the cutoff, the analysis filters out the influence of wider variations in the data. This provides a clear picture of how the treatment changes outcomes at the exact point where eligibility changes.
How the Analysis Works
Lattice fits a local linear model to the data points within a specific window around your threshold. It calculates the Local Average Treatment Effect (LATE), which represents the jump in the outcome variable exactly at the point where the treatment status switches from 'not eligible' to 'eligible.'
The platform automatically handles bandwidth selection using heuristic methods to determine how much data near the cutoff to include. This ensures that the estimate remains local and relevant while maintaining enough data points for a statistically meaningful calculation.
Verifying Data Integrity
A critical part of using regression discontinuity design is ensuring the 'running variable'—the value used to set the cutoff—has not been tampered with. If individuals can influence their own position relative to the threshold, the assumption of a natural experiment breaks down.
Lattice includes an automated McCrary density check to flag these potential issues. If the platform detects a significant discontinuity at the threshold, it will issue a warning that the results may not be credible. This built-in check ensures you are aware of potential biases before drawing conclusions about your results.
1 · Intent → method
An LLM picks causal_rdd from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
Why is the effect only calculated for people near the cutoff?
Regression discontinuity design focuses on the 'local' effect at the threshold because individuals on either side of a strict cutoff are most comparable. Expanding the analysis too far from the cutoff would risk comparing groups that are fundamentally different.
What is the McCrary test and why does it matter?
The McCrary test checks if the number of individuals just above the cutoff is suspiciously different from those just below. If there is a jump in density, it suggests people may have manipulated their data to get over the threshold, which makes the results of this regression discontinuity design unreliable.
Tool input schema
Schema for causal_rdd not exported yet (run pnpm export:registry).