Understanding Regression Discontinuity
Uploaded by
planerpopUnderstanding Regression Discontinuity
Uploaded by
planerpopCommon questions
Powered by AIRegression Discontinuity is suitable in scenarios where the intervention is not randomly assigned, but eligibility is determined based on a continuous variable with an arbitrary threshold. This method is appropriate when the other conditions such as the continuity and no sorting assumptions are met, especially in contexts with a clear eligibility criterion that can create close comparable groups above and below the threshold .
Sharp Regression Discontinuity Design occurs when the assignment to treatment is perfectly determined by whether the running variable is above or below the threshold. In contrast, Fuzzy Regression Discontinuity occurs when the probability of treatment changes discontinuously but not perfectly at the threshold, implying partial compliance. Sharp RD estimates the treatment effect directly at the cutoff, while Fuzzy RD involves estimating a local treatment effect using instrumental variable approaches to handle the imperfect compliance .
Ensuring that all observable variables are continuous at the cutoff is significant because it supports the assumption that the only factor causing a discontinuity in outcomes is the intervention itself. This continuity ensures that there are no abrupt changes in other confounding variables that could bias the identification of the causal effect .
The arbitrariness of eligibility criteria is crucial as it ensures that the criteria do not systematically differ between near-threshold treatment and control groups, thereby allowing RD to reasonably mimic randomization. Arbitrary thresholds make it credible that any observed differences in outcomes at the cutoffs are attributable to the intervention rather than pre-existing differences between units .
The "no sorting across the cutoff" assumption is crucial because it ensures that individuals do not manipulate their treatment status by strategically positioning themselves relative to the threshold, which would compromise the comparability of groups just above and below the cutoff. Violation of this assumption would result in biased estimates of the treatment effect as the presumption of random assignment near the cutoff would be invalidated .
The eligibility or selection variable is pivotal as it determines the threshold that differentiates who receives the treatment. This variable must be continuous and its threshold can be arbitrary, aiding the design in mimicking random assignments when assessing causal impacts. The selection variable facilitates the clear delineation of near-threshold units into treatment and control groups, essential for estimating the treatment effect accurately .
The validity of a Regression Discontinuity Design relies on two key assumptions: 1) Continuity Assumption - All other observable variables must be continuous at the cutoff, ensuring that there is no jump in the underlying variables except for the intervention; 2) No Sorting Across the Cutoff - Units cannot perfectly manipulate their assignment or self-select around the cutoff to become eligible or ineligible, preserving the integrity of the comparability across the threshold .
Researchers can visually diagnose the suitability of using Regression Discontinuity by producing a scatter plot of the outcome variable against the eligibility variable and looking for a visible discontinuity at the eligibility cutoff. If a clear discontinuity is observed, RD may be a suitable approach; if not, the results may not be strong without further statistical support .
Regression Discontinuity mimics randomization by leveraging arbitrary eligibility criteria set over a continuous variable to create a comparison between treatment and control groups. Units that are just above or below the eligibility threshold are assumed to be similar in all respects except for their access to the intervention, thereby mimicking the conditions of a randomized controlled trial .
Regression Discontinuity might be applied in evaluating the impact of financial aid on student performance, where aid is given to students below a certain income threshold. Challenges involved include ensuring the threshold genuinely creates comparability and that students or their families cannot manipulate the income reporting to qualify for aid, thus maintaining the no sorting assumption .