The Gambler's Fallacy
The coin does not owe you a tails. Why the intuition is so strong anyway.
The coin does not owe you a tails. After ten heads in a row, the next flip is still exactly fifty-fifty, and no amount of history changes it.
Almost everyone knows this and almost everyone feels otherwise, which is what makes it worth writing about. The fallacy is not an information deficit — it survives knowing the correct answer.
This page covers where the intuition comes from, one case where a similar-sounding intuition is actually correct, and the inverted version that is equally wrong.
The core error
The gambler's fallacy is the belief that independent events are somehow linked — that a run of one outcome makes the other more likely next. It is applied to roulette wheels, coin flips, lottery numbers and random pickers with equal confidence.
The coin has no memory. It has no mechanism for storing what happened, and no mechanism for adjusting. Every flip is a fresh event with the same probabilities, and a sequence of ten heads followed by a head is exactly as likely as ten heads followed by a tail.
The apparent support from the law of large numbers is where it gets its respectability, and that support is a misreading. Averages converge by dilution, not by correction: a deficit is not repaid, it becomes proportionally smaller as more trials accumulate around it.
Where the intuition comes from
One well-supported explanation is that people hold a representativeness heuristic: a sequence is judged random if it looks like a typical random sequence, and a run of ten heads does not. Since the sequence does not look random, something must be about to correct it.
That heuristic is a good general-purpose tool that fails badly here. It works because most real processes do have memory — a queue that has been long is likely to stay long, a machine that has failed is more likely to fail again. Independence is the unusual case, and our defaults are tuned for the common one.
There is also a simple counting confusion. A run of ten heads is genuinely unlikely in advance, about one in a thousand. Once nine heads have happened, the tenth is not unlikely at all — it is fifty-fifty. Conflating the probability of the whole sequence with the probability of the next event is the error in its purest form.
Before and after
- Probability of ten heads in a row, stated in advance: 1 in 1,024.
- Probability of the tenth head, given nine already: 1 in 2.
- These are different questions. The fallacy answers the first when asked the second.
The case where the intuition is right
There is one important situation where "a run makes the opposite more likely" is correct, and confusing it with the fallacy causes as much trouble as the fallacy itself: sampling without replacement.
Draw cards from a deck without putting them back, and the deck genuinely does have memory. After five red cards, the next card is more likely to be black, because there are now more black cards remaining. That is not a fallacy, it is arithmetic.
The distinction is whether the outcome is removed from the pool. Coin flips, dice rolls and roulette spins put nothing back because there was never a pool. Card draws, bingo calls and no-repeat pickers do remove, and their odds genuinely shift.
- With replacement (coins, dice, roulette): no memory, the fallacy is a fallacy.
- Without replacement (cards, bingo, no-repeat draws): genuine memory, odds shift.
- The test is whether the outcome is removed from the pool afterwards.
- A no-repeat picker on this site is the second kind, which its page states.
The inverted version
The mirror image is the hot-hand belief: that a run of one outcome makes the same outcome more likely next. In a genuinely independent process it is exactly as wrong as the gambler's fallacy, and it is applied just as confidently.
Interestingly, the two contradict each other and people hold both, applying the gambler's fallacy to mechanical processes like roulette and the hot hand to human ones like basketball shooting. That is not entirely unreasonable — human performance genuinely can have momentum, and the empirical question about basketball is a live one — but it is a different claim, and it does not transfer to a coin.
For anything on this site, both are wrong. A random draw has no momentum and no debt.
What to do about it
The practical response is not to argue people out of the intuition, which does not work, but to make the numbers visible. A run of three in a thirty-name picker is expected to happen at a calculable rate, and calculating it in front of somebody is more persuasive than explaining independence.
The second response is to change the mechanism where the complaint is legitimate. If a class does not accept that repeats are fair, a no-repeat picker removes the repeats — which is a real change in the odds, honestly disclosed, and it ends the argument.
That is a better outcome than being correct. The draw was already fair; what was missing was the room's acceptance of it, and a coverage guarantee buys that directly.
Frequently asked questions
After ten heads, is tails more likely?
No. The coin has no memory and no mechanism for correction. The next flip is exactly fifty-fifty.
Doesn't the law of large numbers say it evens out?
It says averages converge, and they do so by dilution rather than correction. A deficit is never repaid; it becomes proportionally smaller.
Why does the intuition feel so strong?
Because most real processes do have memory, and independence is the unusual case. Our defaults are tuned for the common one.
Is there a case where a run does change the odds?
Yes — sampling without replacement. After five red cards from a deck, black genuinely is more likely, because there are more black cards left.
How do I tell the two apart?
Ask whether the outcome is removed from the pool. Coins and dice put nothing back; cards, bingo calls and no-repeat pickers do.
What about the hot hand?
For an independent process it is equally wrong, and it contradicts the gambler's fallacy. People apply one to machines and the other to humans.
How do I convince someone a picker is fair?
Calculate the expected rate of the run they saw. That is more persuasive than an explanation of independence, which rarely lands.
Should I switch to a no-repeat picker if people complain?
It is a reasonable response. It genuinely changes the odds, which you should disclose, and it removes the complaint rather than arguing about it.
Tools that use this
Related guides
Coin Flip Streaks and What They Mean
How long a run of heads you should expect, as a function of how many times you flip.
The Law of Large Numbers, Practically
What 'it evens out' actually promises, on what timescale, and what it never promises.
Why Humans Are Bad at Being Random
Asked to be random, people avoid repeats, alternate too often, and dodge the edges.