Natural Experiment

mental-model

A natural experiment exploits an accidental real-world change that separates one suspected cause from competing explanations, making causal inference more credible than correlation alone.

COVID unexpectedly turned beauty into something researchers could partially switch off. The same professor taught comparable students online and in person, creating a rare test of whether appearance itself—not merely confidence, intelligence, or teaching style—affected grades.

E1

When the world supplies the comparison group

Ordinary correlations leave confounding variables tangled together: attractive students may differ from others in many ways, and polluted places may also differ in income, healthcare, or infrastructure. A natural experiment begins when an outside event changes the suspected cause for one group or period while leaving much of the surrounding system comparable. Outcomes on either side of that change can then reveal the effect of the altered variable more cleanly.

The crucial ingredient is not that the event is natural, but that its assignment is plausibly independent of the outcome being studied. The comparison still needs a credible causal mechanism and evidence that other consequential conditions did not change at the same time.

E1 E2

Where it shows up

Beauty with the visual channel removed

Online and in-person versions of the same class reduced some of the usual differences between classrooms. Comparing grades across those settings helped isolate how much seeing a student’s appearance contributed to the beauty premium.

E1

Toll booths, exhaust, and births

New Jersey’s E‑ZPass rollout reduced congestion and nearby pollution. Birth outcomes before and after the rollout provided a comparison for estimating pollution’s effects without deliberately exposing anyone.

E2

Accidental does not mean controlled

Real-world shocks rarely change only one thing. Online teaching may alter participation as well as visibility, while electronic tolling may change traffic patterns beyond pollution. A natural experiment becomes weak when those accompanying changes offer equally plausible explanations for the outcome.

E1 E2

Look for the discontinuity

When you encounter a causal claim, identify a policy rollout, cutoff, outage, relocation, or sudden format change that affected otherwise comparable people differently. Then list what else changed at that boundary before treating the outcome gap as the effect.

E1 E2

Episodes that teach this