A natural experiment is a study where some real-world event, not a researcher, divides people into groups that can be compared to answer a cause-and-effect question. A policy change, a lottery, a disaster, or a geographic boundary creates conditions that mimic a controlled experiment, even though nobody designed it that way. Researchers then step in after the fact to analyze what happened.
How Natural Experiments Work
In a traditional experiment, researchers randomly assign people to a treatment group and a control group, then measure the difference. A natural experiment follows the same logic, but the “assignment” happens on its own. Some external event splits a population into exposed and unexposed groups in a way that’s random or close enough to random that researchers can draw meaningful conclusions about cause and effect.
The key concept is called “as-if random” assignment. The people affected by the event didn’t choose to be affected, and the factors that determined who ended up in which group are unrelated to the outcome being studied. This mimics what a lab experiment achieves on purpose. If a state raises its minimum wage and the neighboring state doesn’t, workers on either side of the border didn’t sort themselves based on that policy. They just happened to live where they live. That makes the comparison scientifically useful.
Why Researchers Can’t Always Run Experiments
Randomized controlled trials are considered the gold standard for proving that one thing causes another, because the researcher controls everything. But many of the most important questions in medicine, economics, and public policy can’t be studied that way. You can’t randomly assign people to poverty to study its health effects. You can’t force half a population to breathe polluted air. You can’t deny healthcare to a control group for years just to see what happens.
Natural experiments solve this problem. They let researchers evaluate changes to a system that would be impossible or unethical to manipulate on purpose. And when the conditions are right, they can produce effect estimates close to those of a true randomized trial.
The Cholera Study That Started It All
One of the earliest and most famous natural experiments happened in London in 1854, when physician John Snow investigated a cholera outbreak in the Soho neighborhood. Snow traced roughly 600 deaths over 10 days to people who lived near or drank from the Broad Street water pump. Brewery workers and poorhouse residents in the same area, who relied on their own local wells instead of the public pump, escaped the epidemic entirely.
Snow pushed the analysis further. He compared London households served by two different water companies: one drew water from a sewage-contaminated stretch of the Thames, while the other pulled water from an uncontaminated section upstream. Infection rates among households receiving the contaminated water far exceeded those of the other group. Nobody assigned these households to one water company or the other. The assignment was effectively random, determined by where pipes happened to run. Snow used that accident of infrastructure to prove that cholera spread through contaminated water, not through the air as most doctors believed at the time.
A Minimum Wage Study That Changed Economics
In the early 1990s, economist David Card and his colleague Alan Krueger used a natural experiment to challenge a longstanding assumption: that raising the minimum wage would destroy jobs. When New Jersey raised its minimum wage and neighboring Pennsylvania did not, the two states created a near-perfect comparison. Card and Krueger surveyed 410 fast-food restaurants on both sides of the border before and after the increase.
The result surprised the economics profession. Employment in New Jersey’s fast-food restaurants didn’t fall. It actually rose relative to Pennsylvania’s, with New Jersey stores gaining about 13% more full-time-equivalent workers than their Pennsylvania counterparts. The study’s “difference in differences” approach, comparing the change over time in one group to the change in the other, became a foundational tool in modern economics.
This work was part of the reason Card received the 2021 Nobel Prize in Economics. He shared it with Joshua Angrist and Guido Imbens, who developed the statistical framework for drawing cause-and-effect conclusions from natural experiments. One of their key insights was that natural experiments typically reveal the effect only among people whose behavior actually changed because of the event, not for the entire population. That narrower but more honest conclusion is a defining feature of how natural experiments are analyzed today.
The Oregon Health Insurance Lottery
In 2008, Oregon opened a limited number of spots in its Medicaid program for low-income adults and selected applicants by lottery from a waiting list of 90,000 people. The randomness of the lottery created a textbook natural experiment: researchers could compare those who won the lottery (and gained access to insurance) against those who didn’t, with the two groups being essentially identical except for the luck of the draw.
The findings were striking. Within one year, lottery winners were 25 percentage points more likely to have insurance than the control group. That insurance coverage led to a 35% increase in the chance of having an outpatient doctor visit, a 30% increase in hospital admissions, and a 15% increase in prescription drug use. People with coverage were also 30% more likely to follow through on recommended preventive care like mammograms and cholesterol screenings.
The financial effects were just as clear. Insurance reduced the probability of having a medical bill sent to collections by 25% and cut the likelihood of any out-of-pocket medical spending by 35%. People reported better physical and mental health. Because the lottery was genuinely random, these differences could be attributed to insurance itself rather than to the kinds of people who seek out coverage.
Common Sources of Natural Experiments
Natural experiments arise from several types of real-world events:
- Policy changes across borders. When one jurisdiction changes a law and a neighboring one doesn’t, the border becomes a dividing line between treatment and control groups. Minimum wage studies, smoking bans, and gun regulations have all been studied this way.
- Lotteries and administrative cutoffs. The Oregon Medicaid lottery is one example. Draft lotteries, school admission lotteries, and visa lotteries all create random assignment without any researcher involvement.
- Natural disasters and environmental events. Earthquakes, floods, and industrial accidents expose some communities but not others, allowing researchers to study health and economic effects.
- Arbitrary thresholds. A student who scores 501 on a scholarship cutoff of 500 is nearly identical to one who scores 499, but they receive very different treatment. These sharp cutoffs let researchers compare people just above and below the line.
Limitations to Keep in Mind
Natural experiments are powerful, but they have real weaknesses. The biggest is that researchers can’t verify the “as-if random” assumption with certainty. In a true experiment, randomization is built into the design. In a natural experiment, you have to argue that the event divided people randomly enough, and that argument can always be challenged. Researchers typically check whether the two groups look similar on measurable characteristics before the event, but hidden differences can still exist.
There’s also a timing problem. Because the event has already happened, researchers work with whatever data is available rather than designing data collection in advance. This can mean incomplete records, missing variables, or samples that are too small to detect meaningful effects.
Another limitation is scope. The results of a natural experiment apply most directly to the specific population and context where the event occurred. A minimum wage study in New Jersey’s fast-food industry tells you something important, but it doesn’t automatically tell you what would happen in a different state, a different industry, or with a different size of wage increase. Generalizing from one natural experiment to broader conclusions requires caution.
Despite these constraints, natural experiments remain one of the most valuable tools in science for answering questions that can’t be studied any other way. They turn the messy, uncontrolled events of real life into opportunities for rigorous evidence about what causes what.