Regression discontinuity designs
Why this level?
The design family and its defining features have not been specified for this hierarchy. NHMRC 2009 sourceThe design relies on an assignment rule rather than on adjustment alone, but the typed vocabulary has no regression discontinuity fact, so it is left unclassified rather than guessed. How strictly the threshold was applied, and whether the score could be influenced, decide how close the comparison near the cutoff comes to a randomized one.
What question does this help answer?
Does receiving an intervention assigned by a cutoff on a continuous score change the outcome for people near that cutoff?
When to use it
Use it when eligibility is decided by a rule on a measured continuous variable, such as an age limit, a laboratory value, a birth weight, a risk score or a date, and outcomes are available for people on both sides of the threshold.
When not to use it
Avoid it when patients or clinicians can adjust the score to cross the threshold, when the threshold also triggers other changes, when few people fall near the cutoff, or when the effect of interest is for people far from it.
How it works in plain language
People just either side of a threshold are usually similar in everything except whether the rule assigned them the intervention. The analysis models how the outcome changes with the score on each side and estimates the jump at the threshold. When the rule is followed exactly the design is called sharp. When crossing the threshold only changes the chance of receiving the intervention, it is called fuzzy; the estimate then applies to people whose treatment was changed by the rule and needs additional instrumental-variable assumptions.
What the jump at the threshold looks like
Key assumptions and design requirements
- The threshold and the score were set before, and independently of, the outcome.
- The score is not manipulated. People bunching just on the favourable side of the cutoff is a warning sign.
- Characteristics the intervention cannot affect show no jump at the threshold.
- Nothing else changes at the same threshold.
- The window of data around the cutoff and the shape of the model are prespecified, or the result is shown to be stable across reasonable choices. Fits that let distant data drive the curve can distort the estimate at the cutoff.
Clinical example
A hospital enrols newborns below a birth-weight threshold in a follow-up programme. Babies just under and just over the threshold are otherwise similar, so a jump in a later outcome at the threshold can be attributed to the programme for babies near that weight, provided weights were not recorded selectively and no other care changed at the same value. This example is illustrative.
What to watch out for
The estimate applies to people near the threshold, not to everyone the rule affects. It fails when people can influence which side of the cutoff they fall on, when other treatments change at the same threshold, or when the result depends on a window or model shape chosen after seeing the data.
How to report it
No reporting guideline dedicated to regression discontinuity was found in a PubMed search on 2026-09-16. STROBE is the guideline to use. Report what it does not ask for: the assignment rule and whether it was followed, the bandwidth and how it was chosen, the functional form fitted either side of the threshold, and the density check for manipulation around it. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.
Before using this in your workspace
Record the assignment rule and score, whether the design is sharp or fuzzy, the source and precision of the score, the planned checks for manipulation and for balance at the threshold, the planned window and model, and the population the local estimate describes.
- Regression discontinuity designs in healthcare research (opens in a new tab) · BMJ · doi 10.1136/bmj.i1216
- Regression discontinuity designs in epidemiology: causal inference without randomized trials (opens in a new tab) · Epidemiology · doi 10.1097/EDE.0000000000000138
- Regression discontinuity designs are underutilized in medicine, epidemiology, and public health: a review of current and best practice (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2014.06.021
- Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies (opens in a new tab) · BMJ · doi 10.1136/bmj.39335.541782.AD