Trap/Cognitive Bias/No. 0413
Gambler’s Fallacy
The gambler’s fallacy, also called the Monte Carlo fallacy, is the belief that a random streak makes the opposite outcome more likely. In probability, independent events keep the same odds regardless of past results, so a run of heads does not make tails due.
Also called Monte Carlo Fallacy
- Evidence
- Well established
- Read
- 6 min
- Links
- 10 connections
01You've seen this when…
- in life
During a board game, you roll a six three times. On your next turn, you stop planning for a six because you figure you’ve used them up.
- at work
You and a coworker flip a coin each week to pick who gives the update. You have done three weeks in a row and expect next week to be theirs.
- out in the world
A roulette screen lists eight reds in a row. A player raises the bet on black to catch the reversal.
02The idea
A streak makes the opposite outcome feel overdue. After several heads, tails seems more likely. After a lottery number wins, another appearance seems less likely. The gambler’s fallacy is treating those feelings as information about the next result.
The key condition is independence: one outcome does not change the probability of another. If a fair coin’s tosses are independent, each toss has a one-half chance of tails, whether it follows a streak of five or twenty heads or starts the sequence. There is no account that chance needs to settle.
The law of large numbers says that over many tosses, the proportion of heads tends toward one-half. An early streak can become a smaller fraction of the total simply as more tosses accumulate. The difference between the total counts can stay the same or grow.
This is also different from predicting that a streak will continue, often associated with the hot hand fallacy. One expects reversal; the other expects continuation. For a known independent process, neither gets predictive help from the streak.
03Why it happens
- We expect a small sample to resemble the whole. A fair coin suggests an even mix of heads and tails. We expect even the next few tosses to show the balance of a long sequence. This is the law of small numbers bias: demanding too much balance from too little data.
- We judge randomness by its appearance. An alternating sequence looks more random than a solid block of heads, even though any particular sequence of the same length has the same probability under a fair coin. The representativeness heuristic substitutes resemblance for calculation.
- We turn a long-run tendency into a short-run promise. Knowing that heads and tails approach equal proportions makes it tempting to imagine a force pulling the next toss toward the missing outcome. Convergence describes proportions across a growing sequence. Each independent toss still follows the same probabilities.
- A visible history invites a forecast. A roulette display or list of lottery results gives you something concrete to analyze. Without checking how outcomes are generated, it’s easy to mistake a record of the past for a guide to the future.
04A worked example
Charles Clotfelter and Philip Cook examined betting in Maryland’s numbers lottery in a 1993 study. Players bet less on a recently winning number, and betting recovered over time.
What it looks like Sensible avoidance of a repeat. A number has just had its turn, so players shift their money elsewhere while it becomes eligible again in their minds.
What’s actually going on In independent drawings, a number has the same chance after winning as it had before. In a fair drawing with 1,000 possible three-digit numbers, each particular number has a 1-in-1,000 chance each time. The betting pattern reflected a change in players’ expectations while the drawing’s probabilities stayed fixed.
What would have helped Checking the lottery’s drawing rules before using its result history. To test this, calculate today’s odds with the previous result hidden. If revealing yesterday’s winner changes your estimate, identify what physical or procedural connection would justify that change. For independent drawings, there isn’t one.
05How to spot it
06What to do instead
- Calculate before studying the streak. Write down the next outcome’s probability from the rules of the process. Then ask whether the history actually changes it. Making this a routine is a cognitive forcing strategy.
- Look for a connection between outcomes. Are items removed after selection? Does someone enforce quotas? Does success change someone’s skill or fatigue level? Does it change the equipment? A connection between outcomes can change the odds. A feeling of imbalance alone leaves them unchanged.
- Keep probability separate from your losses. The next bet’s odds stay the same regardless of your need to recover money. Judge it by its expected value and by how much you can afford to lose. In roulette, the zero pockets give the house an edge that persists through any streak.
- Set stake limits before play. Decide your maximum loss and bet size before an outcome starts to feel overdue. Keep those limits fixed through a run of losses.
- Update on evidence about the process. If you learn that a coin is biased or a machine has changed, revise your estimate. That’s Bayesian updating: revising your estimate based on evidence about how outcomes are generated.
07When it isn’t a fallacy
Past outcomes can matter. Even when a process is unpredictable, independence is a property to check.
Drawing without replacement changes the odds. Suppose a bag holds four red and four blue marbles. Remove three reds, and the next draw has a four-in-five chance of blue. Blue has become more likely because removing the reds changed the contents of the bag.
An unknown process can be learned about. When a coin’s fairness is unknown, repeated heads may be evidence that it favors heads. Conditional on a particular fixed bias, tosses may be independent. Your estimate of that bias can change. A streak can therefore inform the next prediction through an updated estimate of the coin’s bias.
Regression to the mean predicts less extreme performance. An unusually high test score may be followed by a more ordinary score because the first included unusually good luck. Regression to the mean predicts less extreme performance in suitable settings, not a bad score owed as compensation.
Scheduled rotation, shuffled decks, changing equipment, and human decisions can all create dependence. The mistake is assuming reversal without establishing such a mechanism.
08Roots
Pierre-Simon Laplace found an everyday illustration for probability theory in lottery players. A number that had stayed absent seemed more likely to appear soon. In his 1814 Essai philosophique sur les probabilités, he discussed this temptation: people carried the history of earlier drawings into a calculation where that history did not belong. The error was familiar long before modern casinos put streaks on electronic displays.
In 1971, Amos Tversky and Daniel Kahneman connected the mistake to a broader expectation about small samples. Their work extended the concern to researchers’ judgments about sampling and replication, showing how readily even trained people expected small samples to represent a population reliably. A short run of coin tosses was supposed to display the coin’s fairness; a small study was supposed to display the population’s truth.
That connection moved the idea from gambling advice into the psychology of judgment. Clotfelter and Cook later showed that it also left a trace in actual lottery wagers. The lesson concerns a specific expectation about randomness: we often expect local balance even when earlier outcomes leave the next draw’s probabilities unchanged.
09How solid is this?
Well documented in laboratory judgments and lottery betting. Its prevalence depends on the task and context. Calling a prediction fallacious requires establishing that the relevant probabilities do not change with earlier outcomes.
10Connections
- Often confused with Regression to the Mean, Law of Large Numbers, Hot Hand Fallacy
- Countered by Bayesian Updating, Cognitive Forcing Strategy, Independence
- Can follow from Representativeness Heuristic, Law of Small Numbers Bias
- Part of Illusion of Control
- See also Expected Value
11Origin and sources
A longstanding probability error discussed by Pierre-Simon Laplace in Essai philosophique sur les probabilités (1814). Tversky and Kahneman connected it to belief in the law of small numbers in 1971.
- [1]Laplace, P.-S. (1814). Essai philosophique sur les probabilités.
- [2]Tversky, A., & Kahneman, D. (1971). Belief in the law of small numbers. Psychological Bulletin, 76(2), 105–110.
- [3]Clotfelter, C. T., & Cook, P. J. (1993). The "Gambler's Fallacy" in Lottery Play. Management Science, 39(12), 1521–1525.
Suggest an edit· Updated 2026-10-02