Hot-Hand Fallacy: Why Four in a Row Feels Like a Fifth

A simulated shooter who makes half of all shots will produce runs. In a hundred independent attempts at even odds, a streak of four or five is not unusual; it is what independence looks like from close up. The fallacy is to read such a run as a property of the shooter at that moment, so that the fifth attempt inherits something from the four before it. For a process that carries no information from one attempt to the next, it does not.

Tversky and Kahneman called the underlying intuition the law of small numbers: people expect a short stretch of a random process to look like its long-run proportions. A sequence that holds close to half and half looks random, and a sequence with a run of five looks caused. Since runs appear anyway, the intuition has to attribute them to something, and in sports the available something is the player's form.

The hot-hand belief and the gambler's fallacy are mirror images drawn from the same misreading. Faced with four hits, one intuition says the streak will continue because the shooter is hot, the other that a miss is due because the proportions must even out. Which one arrives depends on whether the process is believed to have a state worth tracking. Croson and Sundali found both in the same casino: players bet against streaks on the wheel, which they knew was random, while betting more after their own wins.

The status of the word fallacy has changed, and a page about this belief has to say so. The 1985 paper reported no positive dependence between shots in professional records, and for three decades the hot hand was the textbook case of a pattern that existed only in the viewer. The 2018 correction by Miller and Sanjurjo removed that footing: once the estimator is fixed, the controlled shooting data of that same paper support a modest hot hand rather than none. What remains a fallacy is the inference from a short streak in a process you have no other evidence about. What no longer stands is the blanket claim that basketball streaks are nothing but an illusion.

Thomas Gilovich, Robert Vallone and Amos Tversky published "The hot hand in basketball: On the misperception of random sequences" in Cognitive Psychology in 1985. What this page draws on is the part of that paper about the belief itself, because the belief is what the demo above asks you to act on.

The authors put the question to 100 knowledgeable basketball fans. Ninety-one percent said a player has a better chance of making a shot after having just made two or three in a row, which is exactly the call the demo puts in front of you after four hits. The belief was not held loosely or as a figure of speech: the fans treated the run as information about the next attempt, and Tversky and Gilovich reported in 1989 that coaches and players talked about it the same way.

The same respondents were then shown sequences of 21 hits and misses, presented as a player's record, and asked to label each one. A sequence that switches sides about half the time, which is what an independent 50-50 process produces, was read as streak shooting by most of them, and the sequences they picked as the clearest examples of chance switched sides far more often than chance does. So the belief does not begin with streaky records. It begins with an expectation of what a chance record ought to look like, and that expectation is wrong in a known direction. The page on the clustering illusion takes that sequence-reading error as its own subject.

The other half of the 1985 paper has not held up as well. Gilovich, Vallone and Tversky also looked for shot-to-shot dependence in professional and controlled shooting records and reported none, and for three decades that settled the question. Joshua Miller and Adam Sanjurjo showed in 2018 that the estimator is biased downward, because picking out the attempts that follow a run drops sequences in a way that pulls the figure below the shooter's own rate, and that correcting it moves the controlled Cornell result from about 3 percentage points to about 13 in favor of a hot hand. So the belief this page is about is not simply false. What stays a fallacy is treating a short run as a state when nothing else is known about the process.

The reading of a streak as a state appears wherever outcomes arrive in sequence. The rows below are illustrations, not findings from the studies cited on this page.

The useful question is not whether streaks are real but whether this process can carry a state from one attempt to the next.

• Ask the mechanism question first. A wheel, a coin or a lottery machine carries nothing. A shooter with warm muscles, a trader with better information or a machine that is wearing out might. Only in the second kind of case can a streak mean anything.
• Expect streaks from independence. In 100 even-odds attempts, a run of five is likely to appear somewhere. A run is not evidence by itself.
• Count the base rate before the streak. The question is whether this shooter hits more than usual after a hit, which needs the usual rate as a comparison.
• Beware of selecting on the streak. Looking only at the attempts that followed a run is the exact move that biased the 1985 measure, and it biases informal judgments the same way.
• Keep the size in mind. Where a hot hand has been measured carefully it is modest, not a different player, so decisions should move a little rather than a lot.
• Fix the rule in advance. A stake, a position size or an assignment decided before the run removes the question.

Three biases concern runs in data. They differ in what the observer concludes.

• It is not the gambler's fallacy, which is the opposite call: that a run makes the other outcome due. The two come from the same misreading of chance and appear in the same people, one about a person's form and the other about a machine's accounts.
• It is not the clustering illusion, which is the earlier step of seeing a pattern in random data at all. The hot-hand fallacy is the prediction that follows once a run has been read as meaningful.
• It is not the illusion of control, the belief that your own actions influence a chance outcome. A fan who thinks a shooter is hot is making a claim about the shooter; a fan who thinks their own seat changed the shot is not.