How do we distinguish a reasonable decision from a mistake? With little time and incomplete information, a simple rule may help with one decision and fail at another task. Evaluating the choice requires understanding the problem and the person's available resources.
A mistake compared with which rule?
A friend changes her mind after hearing a rumor repeated by three people. You might call that sensible: three voices seem more persuasive than one. But what if all three heard the story from the same source? The repetitions do not provide three independent pieces of evidence. Ulrike Hahn uses problems like this to show why research on human rationality cannot begin only by recording what people do. It must also ask what would count as a good judgment in that situation.
That extra question makes the subject unusual in psychology. A study may describe how often someone changes an answer, but the word rational compares that change with a standard about how beliefs or choices ought to be made. Philosophy, statistics, economics and computer science therefore enter the discussion alongside experiments. None of those disciplines provides one universal test for every act; each offers tools for a particular kind of problem.
The review resists a global verdict that humans are either rational or irrational. Even if a response differs from an ideal answer, the significance depends on the task, the information available and the consequences. A tiny numerical departure in a laboratory may matter less than a repeated real-world error; an apparent bias may disappear once we understand how a participant interpreted the question. The interesting work starts when we specify the comparison.
Logic, probability and choice answer different questions
Logic tells us when a conclusion follows necessarily from premises. It is powerful for identifying contradictions, yet many everyday claims are uncertain rather than necessarily true or false. Probability deals with degrees of belief and how evidence should change them. If you say rain is 70 percent likely and no rain is 40 percent likely for the same place and time, the two statements cannot both express coherent probabilities.
Decision theory adds preferences to beliefs: it considers which option is worth choosing when outcomes are uncertain. A monetary gamble is easy to write down, but the money's personal value may differ from its face value. Game theory then studies choices whose best outcome depends on what other people choose. The standards differ because drawing a conclusion, revising a belief, selecting a gamble and cooperating with someone are not the same task.
The abstract rules do not tell researchers what a participant knew or valued. To judge a choice, one must establish the probabilities and outcomes as the decision-maker understood them. To judge an argument, one must establish what the speaker meant. A laboratory problem whose intended meaning is clear to its designer may not be so clear to the person answering. Calling the answer an error before checking this gap risks mistaking a communication problem for a reasoning problem.
An ideal answer can be useful without being feasible
A person cannot calculate every possible consequence of a decision, remember every report they have heard or map every link between sources. Does that make an ideal rule irrelevant? Hahn argues that an ideal can reveal how hard a problem is and where a workable shortcut might fail. But a standard that ignores human limits cannot by itself say what a person reasonably should have done under pressure and with scarce time.
This is the territory of bounded rationality: asking how people can choose well with limited time, attention and computation. Sometimes a simple strategy approximates a complicated ideal. Sometimes it works only in a particular environment. A calculator can extend a person's ability; a reliable institution can make an otherwise impossible information problem manageable. Human capacity is not fixed independently of tools and social support.
The review also separates normative and descriptive claims. A speed limit does not prove how fast anyone drives, and observing a driver's speed does not create the limit. Yet the limit can influence behavior, while observed behavior can help a researcher infer what rule someone is trying to follow. This relationship between what ought to happen and what does happen must be argued, not assumed. Rational models can explain behavior without declaring every observed act optimal.
A supposed bias can come from the test itself
Studies of judgment have reported familiar errors: neglect of prior probabilities, overconfidence and the impression that a detailed combination of events is more likely than one of its parts. These findings sparked valuable research. They also led to disputes about what participants thought the questions meant and which standard should assess their answers. The existence of disagreement is not evidence that every error vanishes. It is a reason to test the task as carefully as the person.
Hahn gives a particularly striking example from claims about optimism. A method used to show that people underestimate their own future risks can generate the same apparent pattern in simulated agents whose probability judgments are accurate. A related claim about updating more after good news than bad news can also emerge from the structure of the measurement. If an ideal agent looks biased on a test, the test cannot simply be treated as a window onto a human flaw.
An unexpected result may also point to a problem with the measure. A normative model can act as a control for an experimental method: what would a well-calibrated reasoner look like under these instructions and response scales? If the measure itself creates an apparent error, changing the wording or analysis may be more informative than inventing a psychological defect. Conversely, a carefully validated departure from the standard remains something to explain. The standard and the measurement must both earn their place.
Information reaches us through other people
The rumor at the beginning reveals a problem that laboratory probabilities often simplify away. To weigh testimony, one needs to know whether different speakers witnessed the event or copied one another. Three independent reports of a bear near a trail are stronger evidence than three retellings of one person's sighting. But in an actual social network, the routes a story traveled are usually hidden and continually changing.
Hahn argues that an ideal instruction to ‘weigh every source correctly’ may be impossible even in principle in realistic networks. A person lacks a map of everyone who passed on the claim; a formal model may also struggle with ordinary patterns of dependence. That does not make accuracy unimportant. It shows why reliable beliefs can depend on the surrounding information environment, not only on the diligence of one isolated thinker.
The implication for misinformation is more specific than telling people simply to think harder. It invites questions about provenance, independence of sources, incentives for repetition and the institutions that help verify claims. A person may use a sensible rule with the information visible to them and still reach a poor conclusion because the environment conceals a shared origin. The review makes rationality social without erasing individual responsibility.
A group can correct errors or create new ones
Several careful individuals do not automatically make a careful group. When people vote, share information or respond to one another strategically, the collective result may follow rules that differ from any member's reasoning. A group can pool independent knowledge and improve a judgment. It can also repeat one rumor until it looks like consensus or produce a choice nobody considered best in isolation. The bridge between individual and collective rationality therefore needs its own study.
The review discusses formal results showing that individually consistent preferences do not always combine into a consistent group preference under common voting conditions. It also points toward research on deliberation, collective intelligence and the circumstances under which groups amplify errors or correct them. These are not interchangeable outcomes. The independence of information, the way a decision is aggregated and the power of speakers can determine which one appears.
Calling a group irrational can be as premature as calling a person irrational. One must ask what the group's task was, which rule selected the outcome and what information moved among members. A unanimous-looking answer may hide copied testimony; a disagreement may expose useful independent evidence. Rationality research becomes more informative when it examines those mechanisms instead of assuming that crowd wisdom or crowd folly is the default.
The harder test lies outside the puzzle
A puzzle can establish a sharp standard because the relevant probabilities and answers are given. Daily life is less accommodating. People must search for information, interpret language, decide when to stop searching and act while others are doing the same. A good laboratory result may not transfer automatically; a poor result may reflect a task that withholds cues available in the world. Hahn asks researchers to connect their findings to the decisions people actually face.
That connection requires choosing the right level of claim. Showing that an answer violates a probability rule under a specified interpretation says something precise. It does not yet show that the person reasons badly across situations, would make a harmful decision, or cannot improve with feedback and tools. Similarly, a model that describes a population's average choice may explain a pattern without serving as advice to a particular individual.
Return to the three people repeating a rumor. The useful question is not whether your friend belongs to a rational or irrational kind of person. It is whether the reports were independent, what she knew about their sources and what consequences followed from acting on them. A standard of rationality helps her ask better questions only when it is joined to an account of the real task. That is the review's lasting challenge to easy verdicts.
What to keep
- Description and norms are intertwined.
- Rationality has multiple frameworks.
- Real-world relevance is a demanding test.
The nuance
What this story does not let us claim
Individual and collective rationality can diverge; a group can correct errors or amplify them.
Go deeper: concepts and optional reading practice
Reversible cards
Test the concept
Topic assessment
Check whether you can explain the idea, not merely recognize it.
Every answer is grounded in the explanation above. Some questions introduce new cases to test whether you can apply and analyze the ideas, not merely recall them.
Source & scope
Human Rationality. Ulrike Hahn. Annual Review of Psychology, vol. 77 (2026).
This editorial guide is an independent explanation. Consult the original publication for the studies, references and full context.
Open the original review ↗