The gambler's fallacy: why nothing is ever due
After five reds in a row, black feels overdue. It is not. Independent events have no memory, and the feeling that they should even out is one of the oldest errors in probability.
What the fallacy is
The gambler's fallacy is the belief that a random outcome becomes more likely because it has not happened for a while, or less likely because it has just happened several times. It shows up as "black is due", "the six hasn't come up all night" or "that number already won, it won't come again".
For independent events this is simply wrong. A coin has no memory. A die does not know what it rolled last time. A wheel does not keep score. Each new trial has exactly the same probabilities as the first one.
The numbers
Five reds in a row
On a single-zero wheel the chance of red on any one spin is 18/37 = 48.65%.
The chance of five reds in a row, judged before the first spin, is (18/37)5 = 2.72%. That is unusual.
But once five reds have already happened, the chance that the sixth spin is red is still 18/37 = 48.65%. The five past spins are history. Only the next spin is uncertain.
The confusion comes from mixing two different questions. "How likely is a run of six reds, starting now?" has a small answer. "Given five reds, how likely is a sixth?" has the ordinary answer. Most of the improbability was already used up by the spins that have happened.
Every sequence is equally unlikely
Ten heads in a row from a fair coin has a probability of 1/1024. So does the sequence H T T H T H H H T T, and so does every other specific sequence of ten flips. Runs look special because we notice patterns, not because they are rarer than any other exact sequence.
Long runs are normal
People who are asked to write down a "random" list of coin flips tend to switch between heads and tails too often and avoid long runs. Real randomness is streakier than that. The table gives the exact chance of at least one run of the same face:
| Run of the same face | In 20 flips | In 100 flips | In 200 flips |
|---|---|---|---|
| 5 or more | 45.8% | 97.2% | > 99.9% |
| 6 or more | 23.7% | 80.7% | 96.5% |
| 7 or more | 11.5% | 54.2% | 79.9% |
| 8 or more | 5.4% | 31.5% | 54.2% |
In 100 flips, a run of six or more heads or tails is more likely than not. Seeing one is not evidence that anything is broken or "hot".
The opposite mistake: the hot hand
The reverse belief is that a streak will continue: a number is "hot", so back it again. For a fair coin, die or wheel, that is the same error turned around. Past results do not change future probabilities either way.
There is one honest exception. If you do not know whether a device is fair, a very long run can be evidence that it is biased. That is a question about the device, though, not about the next result being due. In a well-run game with checked equipment, the sensible assumption is that each trial is independent.
"It will even out in the end"
There is a true statement buried in the fallacy. Over a very large number of trials, the proportion of reds does settle near 18/37. But it does this by swamping early deviations with a huge number of new trials, not by correcting them. After 10 extra reds in the first 100 spins, the count of reds is not pulled back; the surplus simply becomes a smaller and smaller share of the total. The law of large numbers guide works through this with exact figures.
Believing in "due" outcomes is risky because it encourages larger bets after losses. The betting systems guide shows where that leads.