An analytics concept defining statistical methods used to estimate relationships, evaluate interventions, and generate forecasts. It specifies data requirements, estimation procedures, and uncertainty measures used to support decision-making. It does not prove causation without an appropriate identification strategy, data quality checks, and sensitivity analysis. It supports performance management by translating data into estimates, predictions, and quantified uncertainty. The concept is generally stable, though tooling and best practices for measurement evolve over time.

Principle

If treatment assignment jumps discontinuously at a known cutoff while other determinants of the outcome evolve smoothly, the discontinuity in the outcome at the cutoff can be attributed to the treatment; identification is local to the threshold.

Demonstration

Estimate the effect of a scholarship that is awarded to students scoring above a test cutoff by comparing outcomes (e.g., university attendance) for students with scores narrowly above versus narrowly below the cutoff, controlling for score-polynomial and bandwidth choices.

Misapplication

Ignoring manipulation of the running variable near the cutoff (sorting), choosing an inappropriate bandwidth or polynomial order, or extrapolating RD estimates far from the cutoff, which invalidates the local causal interpretation.

Consequence

When the continuity assumption holds and manipulation is absent, RD yields credible local causal estimates at the cutoff; reporting density tests, bandwidth sensitivity, and local polynomial specifications is essential for validity.

Reversal

A setting with randomized assignment across the entire support of a variable, where causal effects can be estimated globally rather than locally, or a regression that assumes a parametric functional form without exploiting a cutoff.

Boundary

RD identifies treatment effects locally at the threshold and requires a clearly defined assignment variable and cutoff; it does not identify average treatment effects away from the cutoff and differs between sharp and fuzzy RD designs.

Semantic Tension

Overlaps with instrumental variable logic (a discontinuous instrument) and local randomization approaches; tension arises when choosing RD bandwidths versus IV averages or when generalizing local RD estimates to broader populations.

Synthesis

Regression discontinuity leverages a discontinuous assignment rule at a known cutoff to recover a local causal effect under continuity and no-manipulation assumptions; its strength is transparent local identification paired with diagnostic checks for credibility.