Natural experiments: reading cause and effect from real life
Awarded to David Card, Joshua D. Angrist and Guido W. Imbens “for his empirical contributions to labour economics · for their methodological contributions to the analysis of causal relationships”.
What was the 2021 Nobel Prize in Economics awarded for?
The 2021 economics prize honours a quiet revolution in how we find cause and effect in the messy real world. David Card showed that natural experiments, like one state raising its minimum wage while its neighbour did not, can overturn settled beliefs; his study found that a higher minimum wage did not cut jobs. Joshua Angrist and Guido Imbens built the framework that says exactly which causal effect such experiments measure.
You want to know whether raising the minimum wage destroys jobs. You can survey the country after a raise, but the economy is also doing a hundred other things at once. How could the real world hand you something close to a clean experiment?
A draft lottery decides, by birth date, who is eligible for military service. Years later, why would economists treat that lottery as a gift?
Suppose you want to know if raising the minimum wage costs people their jobs. You cannot rewind time and run the country twice, once with the raise and once without. So how can you ever be sure what the raise actually caused?
Sometimes the real world sets up the test for you. In 1992 New Jersey raised its minimum wage. Right next door, Pennsylvania did not. Fast-food restaurants on the two sides of that state line were otherwise much alike. So one group of workers got the raise and a very similar group did not, almost as if someone had flipped a coin.
Let real life run the experiment
This year's prize honours scientists who learned to read these accidents of real life like proper experiments. Card and Krueger counted the jobs on both sides before and after. To everyone's surprise, the higher wage did not wipe out jobs. Angrist and Imbens worked out exactly what such a real-life test can, and cannot, tell you.
Economists usually cannot run randomised trials on whole economies. You cannot randomly assign some workers a higher minimum wage and leave others alone, then compare. The laureates' answer is the natural experiment: a situation in real life, created by a policy change, a rule, or chance, that splits people into a treated group and a comparison group much as a coin toss would.
David Card, with Alan Krueger, used exactly such a case. In April 1992 New Jersey raised its minimum wage from $4.25 to $5.05 an hour while neighbouring Pennsylvania held at $4.25. They surveyed 410 fast-food restaurants on both sides before and after the rise. Simple theory predicts the higher wage should cut jobs in New Jersey. Instead they found no such drop; employment in New Jersey held up and even edged ahead of Pennsylvania. The trick is called difference-in-differences: rather than compare the two states directly, you compare the change in jobs in New Jersey with the change in jobs in Pennsylvania. Anything that hit both states equally cancels out, leaving the effect of the wage rise.
A higher wage need not cost jobs
Card and Krueger's finding challenged the textbook claim that a minimum-wage rise always reduces employment. Their survey data showed New Jersey actually gained about 2.4 workers per restaurant relative to Pennsylvania. The work did not prove that wages can rise without limit, but it showed the question is empirical, not settled by theory alone.
Textbook prediction versus what the data showed
Change in jobs per restaurant in New Jersey relative to Pennsylvania after the wage rise.
Joshua Angrist and Guido Imbens asked a sharper question: when a natural experiment only nudges some people, what exactly are we measuring? Angrist's earlier work used the Vietnam draft lottery, a literal lottery, as an instrumental variable to study how military service changed later earnings. But not everyone drafted served, and some volunteered anyway. Angrist and Imbens proved that such a study estimates the effect only for the people whose behaviour the nudge actually changed, the so-called compliers. They named it the local average treatment effect. You cannot name those individuals, but you can measure the size of the group and the effect on it.
The credibility revolution in empirical economics rests on a simple shift: instead of imposing a structural model and hoping the assumptions hold, find a source of variation in the treatment that is as good as random and let it carry the identification. A natural experiment supplies that variation through institutions, policy discontinuities, or chance, so that treated and untreated units are comparable in expectation.
Card and Krueger's 1994 study is the canonical example. New Jersey's April 1992 increase from $4.25 to $5.05 served as the treatment; eastern Pennsylvania, where the minimum held at $4.25, served as the control. The difference-in-differences estimator takes the change in mean full-time-equivalent employment in New Jersey and subtracts the change in Pennsylvania. The identifying assumption is parallel trends: absent the wage rise, employment in the two groups would have moved together, so common shocks such as a recession, seasonal demand, or national fast-food trends difference out. Contrary to the competitive labour-market prediction, the estimate was around zero and if anything slightly positive, a result that launched a still-running literature.
Borrowing randomness from a lottery
When treatment is not randomly assigned but some instrument is, you can still recover a causal effect. Angrist's draft-lottery study used draft eligibility, assigned by birth-date lottery, as an instrument for military service: it shifts the chance of serving but plausibly affects earnings only through service. Two-stage least squares then isolates the service-driven component of earnings. He estimated veterans earned roughly 15 percent less than comparable non-veterans.
The deeper methodological problem Angrist and Imbens solved is what an instrument identifies when its effect varies across people. In 1994 they showed, under a monotonicity assumption (the instrument pushes everyone the same direction or not at all), that instrumental variables recover the local average treatment effect: the average effect among compliers, those whose treatment status is actually moved by the instrument. Always-takers and never-takers contribute nothing to the estimate. So Angrist and Krueger's estimate that an extra year of schooling raises income about nine percent applies to those induced to stay in school by compulsory-schooling laws, not to everyone. LATE reframed a long fight over instrumental variables: an estimate is not biased simply because it differs across studies; different instruments move different compliers and so legitimately recover different local effects.
This is why the prize split two ways. Card showed that carefully chosen natural experiments can overturn settled beliefs about wages, immigration, and schooling. Angrist and Imbens supplied the framework that says precisely which causal effect such designs estimate, turning natural experiments from clever anecdotes into a disciplined tool used across the social sciences.
Where this toolkit reshaped economics
- Minimum wage: difference-in-differences across a state border showed employment effects are small, turning a theoretical certainty into an empirical question.
- Immigration: Card's natural-experiment studies found that an inflow of immigrants did not, on the whole, depress the wages of native-born workers, though earlier immigrants could lose out.
- Schooling: instruments such as compulsory-schooling rules estimate the return to an extra year of education for those whose schooling the rules actually changed, about nine percent in Angrist and Krueger's work.
- Open frontier: every natural experiment still needs a credible as-good-as-random source and a defensible parallel-trends or exclusion assumption, which is where most disputes now live.
The placebo got sick, not the patient
Reviewing Card and Krueger's data, economist John Kennan remarked that the result was like a drug trial in which the drug had no effect but the placebo made people ill: employment, if anything, slipped in Pennsylvania, the control, rather than in New Jersey, where the wage actually rose.
Check yourself
What makes a natural experiment useful for finding causes?
Card and Krueger compared fast-food jobs in New Jersey, which raised its minimum wage, and Pennsylvania, which did not. What did they find?
A natural experiment only changes the behaviour of some people. Whose causal effect does it actually measure?
Key terms
- Natural experiment
- A real-life situation, created by a policy, a rule, or chance, that splits people into a treated group and a comparison group much as random assignment would, allowing a causal effect to be read off.
- Difference-in-differences
- A method that compares the change in an outcome for a treated group with the change for a control group, so that anything affecting both groups equally cancels out.
- Instrumental variable
- An outside factor, such as a draft lottery, that nudges whether someone gets a treatment without affecting the outcome any other way, used to recover a causal effect.
- Local average treatment effect
- The causal effect measured among compliers only: the people whose treatment status is actually changed by the instrument or natural experiment, not the whole population.
- Compliers
- The subgroup whose behaviour responds to the natural experiment. A study using that experiment estimates the effect for them, not for those who would have acted the same way regardless.
- Parallel trends
- The key assumption behind difference-in-differences: that without the treatment, the treated and control groups would have followed the same path over time.
The laureates
Card, born in Guelph, Canada in 1956 and now at UC Berkeley, used natural experiments to test core labour-market questions. His studies from the early 1990s, including the New Jersey minimum-wage comparison with Alan Krueger, showed that a higher minimum wage need not cut jobs and that the effects of immigration and schooling differ from the textbook story.
Angrist, born in Columbus, Ohio in 1960 and based at MIT, pioneered the use of instrumental variables drawn from natural experiments, including the Vietnam draft lottery, to pin down causal effects. With Imbens he clarified what such designs actually estimate.
Imbens, born in the Netherlands in 1963 and now at Stanford, supplied much of the statistical foundation for causal inference from natural experiments. With Angrist he proved in 1994 that instrumental variables recover the local average treatment effect, the effect on the people whose behaviour the experiment actually changed.
Sources
Facts are pinned from the official Nobel Prize API. The explanations were written from these sources: