Colin Camerer on Behavioral Economics and How People Make Decisions

That shift matters beyond academic theory. Models of choice can shape economic predictions and policy recommendations, so assumptions about people’s limits—whether in reasoning, attention, or self-control—can change the conclusions. Camerer’s work asks what those models look like when they account for human behavior.

Listen to or watch Rolex Planet Visionaries wherever you get your podcasts. #ad

Get 30% off your first year of the Base44 Builder plan at Base44.com/mindscape. #ad

Get 50% off your first year of HomeServe at homeserve.com/mindscape. #ad

Look your best and support men’s health at amazon.com/movember. #ad

Support Mindscape on Patreon.

Camerer earned a Ph.D. in behavioral decision theory from the University of Chicago. He is the Robert Kirby Professor of Behavioral Economics, the T&C Chen Center for Social and Decision Neuroscience Leadership Chair, and director of that center at Caltech. He has received a MacArthur fellowship and other awards. His book is Behavioral Game Theory: Experiments in Strategic Interaction.

• Camerer’s website
• Caltech profile
• Google Scholar publications
• Bluesky
• Wikipedia

People reason in steps, then stop

Our judgments don’t always start from a fixed answer waiting to be retrieved. Attention, memory and other mental processes can shift the numbers we give. That matters because behavioral economics is interested in more than showing that people make mistakes. It asks what they’re doing instead. Framing effects offer a clear example. Amos Tversky and Daniel Kahneman used the term for cases where changing the wording of an equivalent choice changes what people do. “90% fat-free” salad dressing sells better than “10% fat,” even though the labels describe the same thing. Advertisers and political consultants have long understood the practical force of that gap. Strategic reasoning also has limits. In the “two-thirds of the average” game, everyone secretly picks a number from 0 to 100. The winner is whoever gets closest to two-thirds of the group’s average. If you expect everyone else to choose around 50, 33 looks sensible. If you expect them to choose 33, you might pick 22. A further step points toward 15, then lower numbers. But the choices don’t usually descend very far. Across many lab studies, newspaper contests and samples from Germany, Singapore and elsewhere, researchers tend to see more 33s than 22s, and relatively few choices around 15. A New York Times contest drew about 50,000 participants, whose choices resembled those of students in other countries. Even groups of game theorists rarely settle on zeros and ones; fives and tens are more common. The audience changes the calculation. A game theorist may pick a very low number when playing with other game theorists, because they expect everyone to reason through several steps. Against an unfamiliar online crowd, that same person might choose 22 or 15. What matters is not just what you know, but what you think the other players know about one another. Researchers call that shared awareness “common knowledge.” Expertise doesn’t automatically travel, either. Chess players, despite their practice in anticipating moves, perform much like other people in these games. The evidence points to strategic thinking as a domain specific skill, with limited transfer from one kind of problem to another.

How much thinking does a game require?

One way to describe these patterns is the level-k model, also called the cognitive hierarchy model. It starts with “level zero” players who aren’t reasoning strategically: they may choose at random, pick a salient number, or follow a personal hunch. In Rosemarie Nagel’s early experiment, she asked participants why they had chosen their numbers. One person said they picked 18 because they were 18 years old. That’s a level-zero choice, not a hidden chain of calculation. A different game helps researchers see where strategic reasoning stops. In the 11–20 game, each player chooses an integer from 11 to 20 and earns that amount. But a player who chooses one less than their opponent gets a 20-point bonus. If you expect the other person to choose 20, you can choose 19; if you expect 19, choose 18. Yet the logic doesn’t run neatly down to 11: if both players choose 11, neither receives a bonus, so choosing 20 becomes attractive again. The result is a mixed equilibrium, with choices cycling rather than converging on one answer. In this game, 20 is the obvious focal choice for players who aren’t thinking about how another person’s choice affects their payoff. They recognize the basic setup—choose the largest number—but don’t work through the strategic consequences. Across such experiments, people generally appear to reason about two or three levels deep. Language can make reasoning look more elaborate. A sentence such as “I know that you know that your wife didn’t feed the cat, so you need to text your daughter” contains several layers of knowledge. But that may reflect familiarity with everyday scripts, or schemas, rather than a general ability to think many steps ahead. People know what normally happens and prepare a plan B or plan C when it doesn’t. There’s an ongoing debate about whether language shows deeper strategic reasoning; the evidence doesn’t settle it. Across several games, a person’s apparent reasoning level does correlate with working memory and education. In older comparisons, Caltech students averaged about half a level more strategic thinking than students at nearby Pasadena City College. Caltech’s student body was highly selected for analytical ability, with a median math SAT score of 800; the researchers’ comparison controlled for geography, since the campuses are about a mile apart. Still, the difference was smaller than expected. There’s no comparable study of Caltech students against peers at Princeton, Stanford or the University of Chicago. The researcher declined to guess at the result, but contrasted Caltech with Harvard and MIT students he encountered during a sabbatical year—Harvard in the fall, MIT in the spring. His impression was that Harvard selects for ambitions beyond test performance: future leaders, young politicians and would-be magazine honorees. Strategic reasoning might help those students, he suggested, but it isn’t necessarily what they’re selected to excel at. A separate measure, the cognitive reflection test, asks people to resist an obvious but wrong answer. Consider a pond where algae doubles in area every day and covers the whole pond on day 48. On what day is it half covered? The tempting answer is day 24, which would fit linear growth. The correct answer is day 47. Daniel Kahneman called the quick, intuitive response “System 1” thinking; Shane Frederick, now a professor at Yale, developed the test. Frederick collected results from many schools, including Harvard, and also surveyed people in Boston Common. The test has three questions, each built around an easy wrong answer and a less obvious correct one, for a score from zero to three. MIT students averaged about 2.28; students at Claremont averaged around 1.6. Harvard students scored a little below the Boston Common sample. Frederick cautioned that the park visitors might not represent Boston residents: on the day of the survey, an orchestra or other event may have drawn an unusual crowd. The comparison is memorable, but it can’t tell us much about either population without a representative sample.

Testing the theory outside the lab

Behavioral economics has faced a fair challenge from the start: showing that people aren’t perfectly rational is easy. A useful theory must explain what replaces perfect rationality. Cognitive hierarchy is one attempt, and researchers have applied level-k models in fields including linguistics and adversarial control, where, for example, two fighter jets must decide how to patrol and deter an enemy aircraft. The model also makes predictions about how people interpret other people’s actions. Consider movie reviews. Between 2000 and 2009, researchers examined 600 wide-release, big-screen films in the United States and other countries. Studios sometimes sent critics screeners on Monday or Tuesday, hoping to have reviews ready for Friday’s release. A favorable review might supply a useful promotional quote; a vague one-word response such as “Interesting” was less helpful. About 10% of the films were “cold opens”: studios didn’t show them to critics before release. Critics reviewed them later, and those films received much lower scores on Metacritic, the review-aggregation site. In other words, the absence of an advance review was a negative signal. Some moviegoers, however, treat missing information as neutral: perhaps a review got lost, or a newspaper ran out of space. They fail to infer that the studio may have kept critics away because it expected bad reviews. That gap produced a measurable effect. The researchers predicted that cold-open films would draw extra viewers who didn’t read the lack of reviews as bad news. Their data showed about a 15% increase in box-office gross from this “sucker effect.” It’s a neat finding, though it doesn’t mean the audience was the only reason those films earned what they did. There’s a related complication: Hollywood can be quite good at predicting a movie’s performance. A separate, unpublished project by a Caltech researcher and colleague Mark Wise compared earnings with budgets. They found that money earned per dollar spent was similar for $1 million and $100 million films, as was the dispersion of results. Smaller films showed more variation, but there were also more of them, leaving the average similar. Studios often estimate opening-weekend revenue days or a week before release, drawing on analytics, test screenings and audience feedback cards. That makes cold openings look less like a mystery than a decision. If a studio expects a film to do badly, withholding it from critics can keep the warning out of view until audiences have already bought tickets. The paper’s examples included *Pluto Nash*, starring Eddie Murphy, then a major star, and *Killers*, with Katherine Heigl and Ashton Kutcher. *Killers* performed poorly, which helps explain why it’s less familiar.

Why System 1 and System 2 never became brain maps

The appeal of Kahneman’s System 1 and System 2 is clear: one label for fast, automatic thought, another for slower, deliberate reasoning. But those labels describe behaviors, not distinct neural systems. Cameron is skeptical that neuroscience can get very far by sorting thought into two boxes based on traits such as speed, effort, or whether people can explain their reasoning. He recalls asking Daniel Kahneman whether neuroscientists had found evidence for a System 1/System 2 split. Kahneman said he had presented the idea, expecting it to inspire research, but it didn’t catch on. Neuroscientists already study networks such as the default mode and salience networks, along with specialized language areas. Against that detail, two broad systems may seem too crude. The wider debate over brain modularity—from Fodor to Cosmides and Tooby’s proposal of 78 evolved modules—underscores how hard it is to settle on a useful number of mental parts. Still, Cameron thinks the question deserves attention. If two systems really do explain a great deal, that would be striking. He speculates that System 1 might be present in infancy, while System 2 could emerge around puberty or even age 25. The conversation’s joke that some people never develop it lands because the underlying point remains unresolved: the labels are suggestive, but the brain evidence has not settled what they correspond to.

A career built around crossing boundaries

Cameron’s move from economics into neuroscience wasn’t an obvious academic path. He says he has gravitated toward unfamiliar subjects and unusual combinations, often working with senior collaborators who could teach him what he needed. Experimental economics was an early example. He learned about it from Caltech colleague Charlie Plott, a central figure in the field, when few economists were doing experiments. That work challenged a discipline that had long treated experiments as impossible. Cameron recalls that Samuelson and Nordhaus’s widely taught economics textbook still made that claim around 1985, despite hundreds of published experiments. Behavioral economics brought another round of resistance from journal referees. Cameron’s answer was to gather more evidence: test decisions with larger stakes, or run the studies in places where the same payment meant something different. In Vietnam around 2010, for example, the equivalent of $10 could amount to roughly two days’ wages. As Richard Thaler put it, Cameron says, the simplest way to win an argument is to run another experiment or regression. Neuroscience grew from the same willingness to cross disciplinary lines. In 1997, Cameron and George Loewenstein were at Stanford’s Center for Advanced Study in the Behavioral Sciences, or CASBS, alongside psychologist Drazen Prelec. Through Prelec’s wife, neuroscientist Danica Prelec, and Loewenstein’s friendship with cognitive neuroscientist John Cohen, they had links into neuroscience. They organized a meeting in Pittsburgh to bring economists and neuroscientists together and ask how brain research might address economic ideas such as utility maximization. Many of the economists mostly listened. Cameron suggests that economics has sometimes treated admitting confusion as a status risk. The attendees included Peter Diamond, Al Roth, and Paul Romer, all of whom later won Nobel Prizes; Cameron jokes that the meeting’s influence on their careers is a claim with a very small sample size. Romer wrote a short paper on emotion, but otherwise the economists largely remained interested observers. A later conference at Princeton, which Cameron missed, brought Angus Deaton into the discussion. Deaton’s question was more practical: economists had detailed data on what people buy, but less understanding of why preferences arise, change, or depend on who else is present. That gap still matters. Neuroeconomics has not yet supplied much material for a standard course on demand theory, Cameron says, but it can help shape a research agenda. Self-control offers one example. Researchers have data suggesting that the dorsolateral prefrontal cortex interacts with another brain region when people choose between tempting and healthy goods. The result may depend on how the brain represents health value. That is a narrower claim than a grand theory of economic choice, but it points toward the kind of question neuroscience can help sharpen. For anyone working across these fields, the lesson is less that neuroscience has already explained economics than that it can make economists ask better questions about where preferences come from. The evidence is still partial. The research agenda is not.