The Experiment Nobody Assigned

On October 11, 2021, the Royal Swedish Academy of Sciences recognised three economists who made that problem more tractable. Half of the Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel went to David Card for empirical contributions to labour economics. The other half went jointly to Joshua D. Angrist and Guido W. Imbens for methodological contributions to analysing causal relationships.

Their shared territory was the natural experiment: a situation in which policy changes, institutional rules or chance create differences that researchers can use to investigate causes. Nobody needs to assign an entire population its life choices. The world supplies a contrast; the difficult work is deciding what that contrast can tell us.

The missing world next door

Suppose people who spend longer in education earn more. That association alone cannot tell us how much their extra schooling caused their higher earnings. People who stay in school may differ from those who leave in ways that also affect income. A tidy graph can contain a very untidy explanation.

The missing quantity is what would have happened without the intervention. We cannot watch the same person live both histories. Natural experiments offer a way to construct a credible comparison—not a portal to an alternate universe, which would presumably have its own funding problems.

The Academy’s background account describes a famous example from Card and Alan Krueger’s work. New Jersey raised its minimum hourly wage from $4.25 to $5.05, while neighbouring Pennsylvania did not. The researchers compared employment in fast-food restaurants in New Jersey and eastern Pennsylvania.

Simply counting New Jersey jobs before and after the increase would not isolate the wage policy: other economic changes could also affect employment. The neighbouring comparison supplied a way to distinguish a shared trend from a change associated with the higher wage. Its credibility depended on the expectation that, without the increase, employment would have evolved similarly in the two areas.

The Academy summarises the study as finding no effect on employee numbers. Its broader announcement makes the careful point: raising the minimum wage does not necessarily mean fewer jobs. That is a challenge to an automatic prediction, not a guarantee that every increase, everywhere, has the same result.

An answer with a boundary

Angrist and Imbens addressed another difficulty: an opportunity or rule does not change everyone’s behaviour.

A rule allowing pupils to leave school earlier will matter to some students. Others would continue to university regardless. Mixing these groups together can obscure whose response the experiment actually reveals.

As the Academy explains, their framework identifies, under specified assumptions, an average effect for people whose behaviour changes because of the natural experiment. This is the local average treatment effect. “Local” is doing important work: the answer concerns that responsive group, not automatically everyone who might encounter a similar policy.

The achievement is therefore twofold. Researchers gain a route to causal knowledge, and a clearer account of where that knowledge stops. Natural variation is not a magic solvent for confounding. A plausible comparison and defensible assumptions remain part of the result, not paperwork attached afterwards.

That is the coherenceist insight here: receive what reality offers, but keep a method for testing what you receive. Neither forcing society into a laboratory nor surrendering to “everything is complicated” serves the people living with the consequences. A useful account must let evidence correct a confident theory—and resist turning a bounded finding into a universal prescription.

We still have no spare humanity in the cupboard. We can, however, get better at learning from this one without pretending its members are interchangeable.