Natural experiment

A study that uses an external event or policy to create exposure groups in a way that resembles random assignment. Its credibility depends on whether the exposure is plausibly independent of participants’ other outcome-related characteristics.

Causal inference: The process of estimating whether and how one factor causes changes in another. Natural experiments use observational variation to estimate causal effects.
Randomized controlled trial: An experiment that assigns participants to treatment or control groups by random allocation. It provides the benchmark that natural experiments approximate without researcher-controlled assignment.
Angrist's Vietnam draft lottery study: Joshua Angrist's study using lottery-based draft eligibility to estimate the effects of military service on earnings. Lottery numbers generated plausibly exogenous variation in service for the study.
Quasi-experiment: A study design that estimates causal effects without random assignment by exploiting structured variation in exposure. Natural experiments are commonly treated as a subset of quasi-experimental designs.
Instrumental variable: A variable used to estimate a causal effect when it influences treatment but is otherwise independent of the outcome. Some natural experiments use an external event as an instrument for exposure.
Confounding: Bias arising when a factor associated with both exposure and outcome distorts their estimated relationship. External assignment is valuable when it reduces confounding between exposure and outcome.
Card and Krueger minimum-wage study: A study comparing fast-food employment in New Jersey and Pennsylvania after New Jersey raised its minimum wage in 1992. The state policy change served as a natural experiment for employment effects.
Field experiment: An experiment that manipulates treatment under real-world conditions, often with randomized assignment. Researchers assign exposure in a field experiment; natural experiments rely on external assignment.
Regression discontinuity design: A quasi-experimental method that compares units near a threshold determining treatment assignment. A policy cutoff can create locally comparable groups without random assignment.
Selection bias: Systematic error caused by how units enter a study or its comparison groups. An apparently external event can still produce selected groups that undermine comparison.

Linked from 18 pages