The Gambler’s Fallacy: Why “It’s Due” Is Not Evidence

Five heads in a row. Surely the next flip has to be tails.

It feels reasonable. Balanced, even. As if probability has noticed the uneven score and is about to tidy up.

That feeling has a name: the gambler’s fallacy. It is the mistaken belief that a sequence of previous outcomes makes an independent opposite outcome more likely because things are “due” to balance out. The problem is not noticing a streak. The problem is inventing a correcting force that the process does not contain.

The coin is not keeping score

Imagine a fair coin flipped independently. Each flip has a one-half probability of heads and a one-half probability of tails. Five previous heads do not change the coin, the model, or the next flip’s probability.

The next head is still a one-in-two event.

Now compare that with a different question: before any flips happen, what is the probability of six heads in a row? It is one in sixty-four. Those two answers are both correct because they answer different questions. One concerns an entire unknown sequence. The other concerns one unknown event after five results are already known.

A surprising sequence is not a forecast

An exact sequence such as heads-heads-heads-heads-heads-heads has the same probability as heads-tails-tails-heads-tails-heads under the fair independent model. We tend to find the first more remarkable because its pattern is immediately visible.

But describing a result after it happens is different from specifying it beforehand. After any six flips, you can point to the exact sequence and say it had a one-in-sixty-four probability. Something had to occur. Its specificity does not automatically make the process suspicious.

Our storytelling machinery is excellent. Sometimes it is simply answering a question the evidence did not ask.

What the law of large numbers does not say

Over many independent repetitions, the proportion of heads tends towards one half. That does not require tails to become more likely after a run of heads.

Suppose an early sample contains six extra heads. As the sample grows, that fixed difference becomes a smaller share of the total even without a special burst of compensating tails. A proportional imbalance can shrink because the denominator grows.

The long run is not an appointment. There is no particular future flip at which a process must settle its account with you. The same confusion appears when people read return to player as a promise that a disappointing session must recover.

Sometimes past information really does matter

Independence is an assumption, not a magic word to paste onto everything.

If objects are removed from a bag without replacement, the bag’s contents change. If a machine’s design changes probabilities based on earlier events, that process is different from independent coin flips. If a physical coin is biased, the fair-coin model is a poor fit.

This is why good explanations name their assumptions. The UK Gambling Commission distinguishes random machines from compensated machines, rather than treating all machines as identical. “The past never matters” is just as careless as “the opposite result is due.”

The mirror-image mistake

A person can look at the same streak and conclude that it will continue because the process is “hot.” In a genuinely independent model, that story has no more predictive power than the balancing story.

One person sees five heads and predicts tails. Another sees five heads and predicts more heads. Both may be assigning meaning to the sequence without evidence that the underlying probability has changed.

The useful question is not which story sounds more convincing. It is what mechanism would make the next outcome depend on the previous ones.

A better way to read a streak

Separate observation from interpretation. “There were five heads” is an observation. “Therefore tails is more likely next” is a claim about the process. The claim needs support beyond the observation.

This distinction is useful well outside casinos: queues, sporting narratives, investment stories, and everyday coincidences all invite us to turn a small sample into a grand explanation.

A pattern can be real without being predictive. A feeling can be strong without being evidence. And a fair independent coin, however dramatic its recent history, has no idea what you think it owes you.