All Chapters

The Missing Chapter

The Gambler's Fallacy

The roulette wheel has no memory, but you do.

An extension of Jordan Ellenberg's "How Not to Be Wrong"

Chapter 19

The Night the Wheel Went Mad

On the evening of August 18, 1913, at the Monte Carlo Casino, something happened that shouldn't have happened — except that it absolutely should have, and eventually must.

The roulette ball landed on black. Then it landed on black again. And again. The gamblers at the table noticed around the tenth spin. By the fifteenth, word had spread across the casino floor. People rushed over, pushing stacks of chips onto red. It has to come up red now, they reasoned. Black can't keep going forever.

But it did keep going. Black came up twenty, twenty-one, twenty-two times. The betting on red grew frenzied. Gamblers doubled and tripled their stakes, certain that the cosmic balance of probability was about to snap back in their favor. Millions of francs piled onto red.1

Black came up twenty-six times in a row.

The gamblers were ruined. The casino had a very good night. And probability theory sat in the corner, quietly sipping its drink, unsurprised.

"The wheel has neither conscience nor memory."
— Joseph Bertrand, Calcul des probabilités (1889)

What happened that night in Monte Carlo is the most famous illustration of the gambler's fallacy: the belief that if a random event has occurred more frequently than expected in the past, it's less likely to happen in the future. Red is "due." The coin "owes" you a heads. The universe keeps a ledger.

It doesn't. The roulette wheel has no memory. Each spin is independent of every spin that came before it. The probability of black on spin twenty-seven, given that the previous twenty-six were all black, is exactly what it always is: 18/37, or about 48.6%.2

This is one of those facts that almost everyone knows intellectually and almost nobody feels in their bones. And that gap — between what we know and what we feel — is where the fallacy lives.

Monte Carlo, August 18, 1913 — 26 consecutive blacks ? Spin 27: Still 48.6% black, 48.6% red

Twenty-six black spins in a row. Spin 27 doesn't care about any of them.

• • •
Chapter 19

Independence: The Hardest Easy Idea

The mathematical concept behind the gambler's fallacy is independence, and it is deceptively simple. Two events are independent if the occurrence of one tells you nothing about the occurrence of the other. A coin flip doesn't know what happened last time. A roulette wheel doesn't keep score.

If events A and B are independent, then:

Independence
P(A ∩ B) = P(A) × P(B)

The probability of both happening is just their individual probabilities multiplied together.

This means the probability of 26 blacks in a row is (18/37)26, which is about 1 in 136 million. Extraordinary! But here's the critical distinction: that's the probability calculated before the first spin. Once you've already seen 25 blacks, the probability of the 26th being black is just… 18/37. The past doesn't change the future.

This is where human intuition revolts. We are pattern-completion machines. Our brains evolved to find streaks, to extract signal from noise, to notice that "every time it rains after those clouds appear, the river floods." In most of life, patterns do predict. But a roulette wheel is not a rain cloud. It is, by careful engineering, the purest expression of randomness that money can buy.

Amos Tversky and Daniel Kahneman called this the representativeness heuristic. We judge the probability of a sequence by how "random" it looks. A sequence like HTHTHT looks more random than HHHHHH, even though both are equally likely. We expect random processes to be self-correcting — to look like our mental image of randomness even in small samples.3

They called this the "law of small numbers" — our mistaken belief that even short sequences should reflect the overall probabilities. The law of large numbers guarantees that proportions converge over thousands of trials. It says absolutely nothing about the next ten.

Try it yourself. Flip some coins and watch what happens to the streaks.

🪙 Coin Flip Streak Tracker

Flip coins and watch streaks form. Long streaks are expected — not evidence of a broken coin.

Total Flips
0
Heads
0
Tails
0
Longest Streak
0
Current Streak
0
Expected Longest*

*For n fair coin flips, the expected longest streak ≈ log₂(n). After 100 flips, expect a streak of ~7.

• • •
Chapter 19

The Inverse Fallacy

There's a less famous cousin of the gambler's fallacy that runs in the opposite direction. Philosopher Ian Hacking called it the inverse gambler's fallacy.4

Imagine you walk into a room and see someone roll two dice. They come up double sixes. You think: Wow, they must have been rolling for a long time to get that.

But that's wrong for the same reason the gambler's fallacy is wrong. Each roll is independent. The probability of double sixes on this roll is 1/36 regardless of whether it's the first roll or the thousandth. Seeing an unlikely result doesn't tell you anything about how many trials preceded it.

Hacking pointed out that this fallacy has a surprising application in cosmology. Some people argue: "The universe's physical constants are extraordinarily fine-tuned for life. This is so improbable that there must be many universes, and we just happen to be in one where the constants worked out." But if each universe's constants are determined independently, then observing our fine-tuned constants tells us nothing about how many other universes exist — just as seeing a double-six tells you nothing about how long the dice have been rolling.

✗ "They must have been rolling a long time!" ✓ P(double six) = 1/36 on ANY roll. The inverse gambler's fallacy: rare outcomes don't imply many prior attempts.

Double sixes on the first roll is just as (un)likely as on the hundredth.

• • •
Chapter 19

The Hot Hand: When the Fallacy Isn't

Now here's where the story gets delicious.

In 1985, psychologists Thomas Gilovich, Robert Vallone, and Amos Tversky published a landmark paper studying the "hot hand" in basketball — the widespread belief that players go on shooting streaks, that a player who has just made several shots in a row is more likely to make the next one.5

They analyzed shooting records from the Philadelphia 76ers and conducted controlled experiments with the Cornell basketball team. Their conclusion was devastating: the hot hand was a myth. Players who had just made several shots were no more likely to make the next one. The apparent streaks that fans and players saw were exactly what you'd expect from random variation — the same kind of streaks you see in coin flips.

This became one of the most celebrated results in behavioral economics. It seemed to confirm everything we knew about the gambler's fallacy: humans see patterns in randomness, even in domains where they have massive expertise. Coaches, players, and fans were all fooled. The paper was cited thousands of times. It became a staple of every "your brain is lying to you" popular science book.

And then, thirty-three years later, it turned out to be wrong.

In 2018, economists Joshua Miller and Adam Sanjurjo published a paper revealing a subtle but devastating statistical error in the original analysis.6

The error is counterintuitive. Suppose you flip a fair coin 100 times and look at all the flips that came immediately after a streak of three heads. You might expect about 50% of those to be heads. But in fact, the expected proportion is less than 50%. This is because of a selection bias: by conditioning on streaks, you systematically oversample from the ends of runs, where reversals are more likely.

This means the original Gilovich-Tversky analysis was biased against finding a hot hand. When Miller and Sanjurjo corrected for this bias, the hot hand emerged clearly from the same data. Players do shoot better when they're on a streak — not by a huge amount, but by a statistically significant one.

The hot hand is real. The fans knew it all along. And the scientists, for thirty-three years, were the ones committing the error.

This is one of the most remarkable reversals in the history of social science. The very paper that was supposed to prove humans are bad at spotting randomness turned out to be an example of scientists being bad at statistics. There's a lesson in humility here that goes beyond basketball.

The deeper lesson: not every streak is an illusion. In roulette, where the mechanism guarantees independence, streaks genuinely are meaningless. But in basketball, where a human being's confidence, muscle memory, and rhythm affect their performance, some dependence between shots is perfectly plausible. The trick is knowing which domain you're in.

• • •
Chapter 19

The Market's Memory

Nowhere does the gambler's fallacy wreak more financial havoc than in the stock market. The belief in mean reversion — that what goes up must come down, that a stock that has fallen for five days is "due" for a bounce — is the gambler's fallacy wearing a pinstripe suit.

Now, to be fair, some financial instruments do exhibit mean reversion. Interest rates tend to drift back toward long-term averages. The price-to-earnings ratio of the overall market has historically reverted. But individual stock prices on short time scales? The evidence for mean reversion is thin at best. A stock that dropped 20% this week is, in many cases, more likely to drop further (momentum) than to bounce back (reversion).7

The problem is that mean reversion and the gambler's fallacy feel identical but operate on completely different mechanisms. True mean reversion happens when there's a structural force pulling values back toward equilibrium — like how a rubber band snaps back when stretched. The gambler's fallacy just assumes the universe is keeping score, which it is not.

Key Insight

Before betting on reversion, ask: what is the mechanism? A roulette wheel has no mechanism for self-correction. An economy might. A basketball player's confidence definitely does. The question is never "is this a streak?" but "is this the kind of system where streaks carry information?"

• • •
Chapter 19

Spin the Wheel

The Monte Carlo incident feels impossible because we think in small samples. Let's fix that. Here's a roulette wheel. Spin it a few hundred times and watch the streaks form. Try betting against the streaks and see how you do.

🎰 Monte Carlo Roulette Simulator

Spin the wheel and track streaks. Try the "Bet Against Streaks" strategy — bet on the opposite color after every streak of 3+. Does it beat random betting?

Strategy:
Total Spins
0
Red / Black / Green
0 / 0 / 0
Longest Streak
0
Current Streak
0
Bets Won / Total
— / —
Win Rate
• • •
Chapter 19

The Birthday of a Fallacy

The gambler's fallacy is arguably as old as gambling itself, but it got its formal name and analysis in the twentieth century, when probability theory matured enough to rigorously explain why our intuitions fail.

Pierre-Simon Laplace described the fallacy (without naming it) in his 1814 Essai philosophique sur les probabilités, noting that gamblers at the lottery believed numbers that hadn't been drawn recently were "due." He was gently exasperated:8

"They do not reflect that in the conduct of chance, the past can have no influence on the future."

Two centuries later, we still don't reflect. The fallacy persists because it is wired into the architecture of human cognition — into the same pattern-matching machinery that makes us brilliant at language, at social reasoning, at science itself. We can't help looking for the pattern. The best we can do is notice when we're looking for a pattern in a place where there isn't one.

And sometimes — as the hot hand story reminds us — the best we can do after that is remain humble about our certainty that there's no pattern at all.

1814 Laplace 1913 Monte Carlo 1974 Tversky & Kahneman 1985 Hot hand "debunked" 2018 Hot hand vindicated! Two centuries of getting randomness wrong (and occasionally right).

A brief history of humans arguing with probability — and probability winning.

The roulette wheel has no memory. But you do. And your memory, for all its flaws, is the only tool you've got. Use it wisely — and maybe don't bet against a streak of twenty-six blacks.

Notes & References

  1. The Monte Carlo incident of 1913 is widely reported in probability literature. A detailed account appears in Darrell Huff, How to Take a Chance (W.W. Norton, 1959). The exact number of consecutive blacks (26) and the losses ("millions of francs") may be somewhat embellished in popular retellings, but the core event is well-documented in casino records.
  2. European roulette has 18 red, 18 black, and 1 green (zero) pocket, giving P(black) = 18/37 ≈ 0.4865. American roulette adds a second green (00), making P(black) = 18/38 ≈ 0.4737.
  3. Tversky, A. & Kahneman, D. (1974). "Judgment under Uncertainty: Heuristics and Biases." Science, 185(4157), 1124–1131. See also Tversky & Kahneman (1971), "Belief in the Law of Small Numbers," Psychological Bulletin, 76(2), 105–110.
  4. Hacking, I. (1987). "The Inverse Gambler's Fallacy: The Argument from Design. The Anthropic Principle Applied to Wheeler Universes." Mind, 96(383), 331–340.
  5. Gilovich, T., Vallone, R., & Tversky, A. (1985). "The Hot Hand in Basketball: On the Misperception of Random Sequences." Cognitive Psychology, 17(3), 295–314.
  6. Miller, J.B. & Sanjurjo, A. (2018). "Surprised by the Hot Hand Fallacy? A Truth in the Law of Small Numbers." Econometrica, 86(6), 2019–2047. The key insight is that conditioning on a streak of successes within a finite sequence creates a selection bias that depresses the observed proportion of subsequent successes, even for a fair coin.
  7. Jegadeesh, N. & Titman, S. (1993). "Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency." The Journal of Finance, 48(1), 65–91. This seminal paper documents the momentum effect: stocks that have performed well continue to outperform over 3-12 month horizons.
  8. Laplace, P.-S. (1814). Essai philosophique sur les probabilités. Translation from the French by F.W. Truscott and F.L. Emory (1902).