The Gambler’s Fallacy: Is Tails Ever “Due”?

A fair coin has landed heads five times in a row. What’s the chance the next flip is tails?

It feels like it should be more than 50%. Five heads in a row is a lot, so surely tails is due. But it isn’t. It’s exactly 50%, just as it always was, and believing otherwise has a name: the gambler’s fallacy.

Why it’s wrong

Each flip of a fair coin is independent. The coin has no memory of what it did before, and nothing in the world is keeping score. The chance of heads on the next flip is 1 in 2, whatever came before.

Five heads in a row is unusual: it happens about 1 time in 32 starting from any given flip. But once it’s happened, it’s done. The question is only about the next flip, and the next flip is a fresh 50/50. You can check this with the toy at the top, which flips a coin thousands of times and looks only at what comes straight after five heads. It comes out about half and half. Every time.

The night at Monte Carlo

On 18 August 1913, at the casino in Monte Carlo, the ball on a roulette wheel landed on black 26 times in a row. As the streak grew, gamblers piled their money on red, sure it was overdue. Each time, black came up again. By the end, the story goes, they had lost millions of francs.

A run of 26 blacks on a wheel with a single zero is a tens-of-millions-to-one event. But look at what each gambler was betting on: not the chance of a 26-run, but the next spin, which at every point was a fresh 18-in-37 for red. The wheel doesn’t remember the previous 25.

Why it feels so right

If the fallacy is so plainly wrong, why does nearly everyone feel it? The pull is strong and comes from a few places.

We expect short runs to look like long ones. We know that over thousands of flips, heads and tails come out about even. So we expect even a short stretch to be balanced. Kahneman and Tversky called this the belief in the law of small numbers. But a short run has no obligation to look like the long-run average. Small samples swing a lot, and that’s just what they do.

Real randomness is streakier than we think. Ask people to write down a “random” sequence of 100 coin flips and they alternate too often and avoid long runs. In 100 real flips, a run of six or seven in a row is quite likely. We judge randomness by how it looks, and true randomness often doesn’t look random to us.

It sounds like the law of large numbers. There’s a true idea nearby: over many flips, the share of heads settles close to 50%. So it seems tails must catch up. It doesn’t. The share settles because the lucky streak gets drowned out by thousands of later flips, not because anything corrects it.

This is the single most useful thing to take away. Imagine you start with 60 heads in the first 100 flips, 10 above average. If the coin then behaves fairly, after 10,000 more flips you’d expect about 5,060 heads in 10,100, or 50.1%. The share has gone almost all the way back to half. But the 10 extra heads are still there. They were never repaid. They’re just a rounding error now.

An experiment: can you fake randomness?

Try this with a friend. Ask them to write down 50 coin flips from their head, as H and T, while you flip a real coin 50 times and write those down too. Then swap the sheets and try to tell which is which.

The giveaway is usually the runs. People who invent a sequence rarely let a streak go beyond three, because a long run “doesn’t look random”. Real flips are much less polite: in 50 flips you’ll typically see a run of five or six. The human sequence also alternates too often. Statisticians use exactly this to catch fabricated data, because made-up numbers tend to be too evenly mixed to be real.

It’s a neat way to feel the fallacy from the inside. The intuition that says “that’s too many heads in a row” is the same one that says “tails is due”. It’s a sense of what randomness should look like, and it’s wrong in a consistent direction.

It isn’t only gamblers

The gambler’s fallacy turns up wherever people judge a series of independent cases, and it’s been caught in some surprising places.

In 2016, Daniel Chen, Tobias Moskowitz and Kelly Shue studied what happened in three settings. US asylum judges were a few percentage points less likely to grant asylum after granting the previous two cases. Loan officers were less likely to approve an application after approving the last. Baseball umpires were more likely to call a pitch a ball after calling the one before a strike. In each case the sequence of cases should have been independent, so there was no reason for the previous decision to matter. But the decision makers seemed to expect a balance, so after a run of one kind, the other started to feel overdue.

Lottery players show it too. In a study of a Maryland lottery, Charles Clotfelter and Philip Cook found that after a number was drawn, people bet on it noticeably less over the following weeks, as if it had used up its luck.

The opposite mistake

Some people make the reverse error: believing a streak will continue. A basketball player on a “hot hand”. A roulette colour that’s “running hot”. A fund manager who’s beaten the market three years running.

For a coin or a fair wheel, this is just as wrong. The next flip is still 50/50 whatever came before. It’s the same underlying error, treating independent events as connected, in the other direction.

Humans aren’t coins, though. Whether basketball players really get hot streaks has been argued over for decades. A famous 1985 study concluded the hot hand was a myth, but in 2018 Joshua Miller and Adam Sanjurjo found a subtle statistical flaw in the way streaks had been counted and argued that a small hot-hand effect is real. Skills can run in streaks because of focus or confidence. A coin can’t.

Where it matters

• Gambling. Betting systems that raise stakes after losses assume a win is “due”. It isn’t, and the losses can pile up fast.
• Decisions. A manager who turned down three good proposals in a row shouldn’t feel the next one deserves a yes just to balance things.
• Life choices. “We’ve had two boys, so the next one will probably be a girl.” For each birth, it’s still close to 50/50.
• Health. “I’ve had bad luck this year, so I’m due good luck.” Luck isn’t a bank account.
• Sport and trading. “The team has lost five in a row, they must win soon.” Maybe they’re just not very good.

How to resist it

The cure is one question, asked whenever you feel “due”: does the thing that produces the next result know what happened before? A coin doesn’t. A roulette wheel doesn’t. Dice, lottery balls and the sex of a baby don’t either. If the answer is no, the past tells you nothing about the next one.

When it’s not independent, the logic flips. Draw cards from a deck without putting them back and the chances really do change, because the deck remembers: if the first two cards are aces, there are fewer aces left. Process matters. The fallacy is applying a “deck” intuition to a “coin” situation.

If you’d like to see how your own instincts hold up, the probability test has streaks, shared birthdays and a few traps that look a lot like this one, and the thinking traps test covers its neighbours.