Natural experiment
An accident of the world that mimics random assignment when a true experiment is impossible.
What it means
A natural experiment is a study that exploits a naturally occurring event, policy, or boundary which assigns people to different conditions in a way that approximates random assignment, used when deliberately randomizing would be impossible, unethical, or impractical. The credibility of the design hinges entirely on how plausibly 'as-if random' the assignment is — that is, whether the people on either side of the dividing line (a date, a border, an eligibility cutoff, a lottery) are comparable in everything except the treatment. Because the researcher does not control assignment, natural experiments are quasi-experimental and require careful argument and statistical techniques — such as difference-in-differences, regression discontinuity, or instrumental variables — to rule out the confounders that randomization would otherwise have balanced. Their reward is the ability to study real-world treatments at scale with strong causal leverage and high ecological validity. Their risk is that hidden differences between the groups, or selection into the 'natural' assignment, can masquerade as treatment effects. It matters because many of the most important questions in economics, policy, and public health cannot be randomized, and natural experiments are often the best available route to causal answers.
Examples
A policy change applied at an arbitrary calendar date or only on one side of a state border lets researchers compare otherwise-similar groups exposed and not exposed to it.
A school entry cutoff falling on a birthdate means children born days apart start a year apart, letting researchers estimate the effect of extra schooling nobody ever assigned.
A city allocates subsidised housing by lottery, so winners and losers are comparable at the draw and their health years later reveals what the housing itself did.
First described in Econometrics; quasi-experimental design.