Card, Angrist & Imbens (2021): When Life Runs the Experiment for You

The Problem of Cause and Effect

Economics is full of questions that sound simple but are fiendishly hard to answer:

• Does raising the minimum wage destroy jobs?
• Does immigration lower wages for native workers?
• Does an extra year of school increase your earnings?

The difficulty isn’t a lack of data — it’s that correlation isn’t causation. States with higher minimum wages may have stronger economies to begin with. Cities with more immigrants may be more dynamic. People who stay in school longer may be more motivated regardless. The real cause hides behind a tangle of confounding factors.

The gold standard for establishing causation is a randomized experiment — like the RCTs that won Banerjee, Duflo, and Kremer their 2019 Nobel. But you can’t randomly assign minimum wages to states, or randomly send immigrants to cities, or randomly force teenagers to stay in school. Many of the most important economic questions are ones where experiments are impossible.

The 2021 Nobel honored three economists who found another way. David Card showed that nature sometimes runs the experiment for us — policy changes, historical accidents, and geographic boundaries create situations that mimic random assignment. Joshua Angrist and Guido Imbens built the rigorous statistical framework for extracting causal conclusions from these “natural experiments” — and for knowing exactly how far those conclusions extend.

Together, they launched what became known as the credibility revolution in economics.

Card: Overturning Conventional Wisdom

David Card’s genius was spotting situations where real-world events created something close to a controlled experiment — and then using them to challenge beliefs that economists had held for decades.

The minimum wage study that shocked the profession

For generations, economists overwhelmingly agreed: raise the minimum wage, and employers will hire fewer workers. It was textbook economics — higher labor costs mean less labor demanded. The only debate was how big the job losses would be.

In 1992, Card and his colleague Alan Krueger (who died in 2019 and would likely have shared the Nobel) noticed a natural experiment unfolding. New Jersey raised its minimum wage from $4.25 to$5.05. Neighboring Pennsylvania didn’t. The two states had similar economies and similar fast-food industries. This was as close to a controlled experiment as you could get.

Card and Krueger surveyed fast-food restaurants on both sides of the border, before and after the increase. Their finding: employment in New Jersey’s fast-food sector didn’t fall — it actually rose slightly relative to Pennsylvania.

The result was explosive. It didn’t just question the magnitude of job losses from minimum wages — it questioned whether they existed at all. The study prompted enormous debate and extensive follow-up research. The emerging consensus: moderate minimum wage increases have surprisingly small effects on employment, for several reasons:

• Companies can pass costs to customers through small price increases
• In areas where one employer dominates (monopsony), wages were already below market-clearing levels, so the minimum wage actually pushes employment closer to efficient levels
• Higher wages reduce turnover, saving on hiring and training costs
• Higher wages can increase worker productivity through efficiency wages

The Mariel Boatlift: Immigration doesn’t do what you think

In 1980, Fidel Castro suddenly allowed Cubans to leave. Within months, 125,000 Cubans emigrated to the United States, with most settling in Miami. The city’s labor force grew by 7% virtually overnight — a massive and completely unexpected shock.

Card compared Miami’s labor market to four similar cities that didn’t receive the influx. If immigration depresses wages and takes jobs from natives, Miami should have been devastated. The result: wages and employment for low-skilled workers in Miami were essentially unaffected. The labor market absorbed the shock with little trace.

How? Immigrants create demand as well as supply — they buy food, rent apartments, use services. Industries expanded to serve the larger population. And immigrants often complement native workers rather than substitute for them — they take different jobs, work different hours, or bring different skills.

Returns to education

Do people earn more because they went to school longer — or do the same traits that keep kids in school (motivation, ability, family background) also lead to higher earnings? Card used natural experiments to separate cause from correlation:

• Differences in compulsory schooling laws across states and time periods created variation in how long people stayed in school — variation that had nothing to do with individual motivation
• Children who grew up near a college were more likely to attend — proximity as a source of random variation
• These natural experiments showed that the causal return to education was at least as large as — and possibly larger than — naive estimates suggested. Education wasn’t just a signal of pre-existing ability; it genuinely increased earnings

Angrist and Imbens: The Framework for Causal Inference

Card’s natural experiments were powerful, but they raised a deeper question: exactly what causal conclusions can we draw from them?

A natural experiment isn’t a true randomized trial. The variation it exploits (a policy change, a geographic boundary) affects some people more than others, and people respond differently. What precisely does the estimated effect represent?

The problem of heterogeneity

Imagine using Vietnam War draft lottery numbers as a natural experiment to study the effect of military service on earnings. The draft lottery randomly assigned numbers to young men — low numbers got drafted, high numbers didn’t. This creates random variation in who serves.

But not everyone with a low number actually served (some got exemptions), and some with high numbers volunteered. The treatment group and control group aren’t perfectly clean. Moreover, the people who would serve regardless of their number (volunteers) might be very different from people who only serve because they were drafted. The effect of service might be different for each group.

The LATE theorem

In their landmark 1994 paper, Angrist and Imbens solved this problem with elegant precision. They showed that a natural experiment with an instrumental variable (like the draft lottery number) identifies a specific causal effect: the Local Average Treatment Effect (LATE) — the causal effect for compliers, the people whose behavior was actually changed by the instrument.

In the draft example:

• Always-takers: Would serve regardless of their draft number (volunteers). The instrument doesn’t affect them
• Never-takers: Would never serve regardless of their draft number (got exemptions). The instrument doesn’t affect them either
• Compliers: Served because they got a low number, and wouldn’t have served otherwise. The LATE measures the effect of military service specifically for this group

This was a profound clarification. It meant:

• Honesty about what you’ve measured: The LATE doesn’t tell you the effect for everyone — it tells you the effect for the people whose behavior the instrument actually changed. This is a narrower but much more credible claim
• Transparent assumptions: Angrist and Imbens made explicit the assumptions needed for the instrument to work — assumptions that could be scrutinized, debated, and tested
• A bridge between economics and statistics: Their potential outcomes framework merged the instrumental variables tradition from economics with the causal inference framework from statistics, creating a common language for both fields

Beyond LATE

The Angrist-Imbens framework gave empirical researchers a rigorous protocol:

1. Find a natural experiment that creates plausibly random variation
2. Identify the instrumental variable — the source of this variation
3. State your assumptions explicitly
4. Estimate the LATE — the causal effect for compliers
5. Be transparent about the limitations — who the result applies to and who it doesn’t

This protocol became the standard for empirical work across all of economics — and increasingly in political science, sociology, epidemiology, and other fields.

The Credibility Revolution

Before Card, Angrist, and Imbens, much of empirical economics relied on statistical models with strong assumptions that were difficult to verify. Researchers would run regressions, “control” for confounding variables, and claim causal relationships — but the results were only as good as the assumptions, which were often questionable.

The credibility revolution changed the culture:

• Research design first: Instead of building elaborate statistical models, identify a source of plausibly random variation — a natural experiment — and let the design do the work
• Transparency over sophistication: A simple comparison across a policy boundary is often more convincing than a complex model with dozens of control variables
• Humility about scope: The LATE framework taught researchers to be honest about what their evidence shows — not overclaiming universal effects from local evidence
• Replicability: Natural experiments can be re-examined by other researchers using the same data and design, making results more robust to scrutiny

The revolution spread far beyond the topics Card originally studied. Today, researchers use natural experiments to study the effects of policing on crime, air pollution on health, social media on political behavior, and hundreds of other questions where randomized experiments are impossible.

The 2021 Nobel Prize was awarded to Card “for his empirical contributions to labour economics” and to Angrist and Imbens “for their methodological contributions to the analysis of causal relationships.”

Explain It to a Child

Imagine you want to know if eating breakfast makes kids do better on tests. You can’t just compare breakfast-eaters to breakfast-skippers, because maybe the kids who eat breakfast also have parents who help them study. You need a trick. What if one day, a snowstorm closes half the roads, so some kids can’t get to the cafeteria for breakfast and others can — and it’s basically random who lives on which road? Now you can compare the two groups fairly, because the snowstorm decided who ate breakfast, not the kids or their parents. That’s a “natural experiment.” Card found situations like this in real life — like when one state raised its minimum wage and the neighboring state didn’t — and used them to discover surprising things, like that raising the minimum wage didn’t actually cost jobs. Angrist and Imbens figured out the math for exactly what you can learn from these natural experiments, and what you can’t. Together, they taught economists to stop guessing about what causes what and start looking for situations where life accidentally runs the experiment for them.