0 shares
Probability is the branch of mathematics most likely to be wrong in public. Not because it’s difficult, since most of what follows needs nothing beyond arithmetic, but because human intuition about chance is systematically and predictably faulty in ways that don’t improve with education.
The fifteen problems below have all produced sustained public disagreement. Several have produced wrongful convictions. They’re ranked roughly by how strongly they conflict with intuition rather than by mathematical difficulty.
If you find yourself certain that one of these answers is wrong, you’re in distinguished company. Hundreds of people with doctorates in mathematics publicly rejected the most famous entry on this list.
| # | Problem | What surprises people |
| 1 | The Monty Hall Problem | Switching doors doubles your chances |
| 2 | The Birthday Problem | 23 people give a better-than-even match |
| 3 | The Prosecutor’s Fallacy | A one-in-a-million match is not one-in-a-million guilt |
| 4 | The Base Rate Fallacy | A 99% accurate test can still be usually wrong |
| 5 | Simpson’s Paradox | A trend can reverse when groups are combined |
| 6 | The Gambler’s Ruin | Enough play makes ruin near-certain |
| 7 | The St Petersburg Paradox | Infinite expected value, worth about £10 |
| 8 | The Boy or Girl Paradox | How you ask changes the answer |
| 9 | Bertrand’s Box Paradox | Two-thirds, not one-half |
| 10 | The Two Envelopes Problem | A swap that always looks profitable |
| 11 | The Problem of Points | The question that created probability theory |
| 12 | The Inspection Paradox | Your bus wait is longer than average by design |
| 13 | Benford’s Law | Real data starts with 1 about 30% of the time |
| 14 | The Secretary Problem | Reject the first 37%, then take the best so far |
| 15 | Buffon’s Needle | Dropping needles calculates pi |
*Each entry is explained in full below.*
Ranked by the gap between the correct answer and what most people confidently believe the answer to be, weighted by how much real-world damage the misunderstanding has caused. Problems that have influenced court verdicts rank above curiosities.
Three doors. A car behind one, goats behind the other two. You pick a door. The host, who knows where the car is, opens a different door to reveal a goat and offers you the chance to switch. Should you?
Yes. Switching wins two-thirds of the time; staying wins one-third.
The clearest way to see it is this. Your original pick had a one-in-three chance of being right, and nothing the host did changed that. So there’s a two-in-three chance the car sits behind one of the other two doors, and the host has just told you which of those two it isn’t. All of that two-thirds probability collapses onto the single remaining door.
The crucial condition, which often gets left out of the telling, is that the host knows where the car is and always opens a losing door. If the host opened a door at random and happened to reveal a goat, switching would give you no advantage at all. When Marilyn vos Savant published the correct answer in 1990 she received thousands of letters insisting she was wrong, many from people with doctorates.
How many people have to be in a room before there’s a better-than-even chance two of them share a birthday? Twenty-three. At fifty-seven people the probability passes 99%.
Intuition says several hundred, because it’s quietly answering a different question: how many people before someone shares *my* birthday. That answer is indeed around 253.
The actual question is about any pair at all. Twenty-three people produce 253 possible pairs, and each pair is a separate chance for a match. The number of comparisons grows with the square of the group, which is why the threshold arrives so much earlier than it feels it should.
A forensic match has a one-in-a-million chance of occurring by coincidence. The defendant matches. Is there a one-in-a-million chance they’re innocent?
No, and confusing those two statements has sent people to prison.
The one-in-a-million figure is the probability of a match given innocence. What a court needs is the probability of innocence given a match, which is an entirely different quantity and depends on how many other people could have produced the same match. In a population of sixty million, roughly sixty people would match by chance. The match alone therefore places the defendant in a group of about sixty, not a group of one.
This isn’t a theoretical concern. Statistical reasoning of this kind has featured in several contested convictions, and the Royal Statistical Society has issued public guidance about it.
A disease affects one person in a thousand. A test is 99% accurate. You test positive. What’s the probability you have the disease?
Around 9%, not 99%.
Take ten thousand people. Ten have the disease and nearly all will test positive. Of the 9,990 who don’t, 1%, or about a hundred people, will test positive anyway. So roughly a hundred and ten positive results exist, of which ten are genuine. Ten out of a hundred and ten is about 9%.
The error is ignoring how rare the condition was to begin with. Same mistake as the prosecutor’s fallacy in a different costume, and it recurs in medicine, security screening and fraud detection with expensive consequences.
A trend that appears in every subgroup of a dataset can reverse when the subgroups are combined.
The classic illustration involves university admissions. A university can accept a higher proportion of female applicants in every single department, and still accept a lower proportion of female applicants overall. That’s not a contradiction and it isn’t a statistical trick. It happens when applicants are unevenly distributed across departments with very different acceptance rates.
The unsettling implication is that there’s no general rule for which version of the data is correct. Aggregated and disaggregated figures answer different questions, and choosing between them requires understanding the causal structure rather than doing more mathematics.
A player with finite money betting repeatedly against an opponent with far more money will eventually go broke, even in a completely fair game with no house edge whatsoever.
The reason is asymmetry of resources rather than unfairness of the game. Both players experience swings. Only one has a floor they can hit. Reaching zero ends the game permanently, while the wealthier opponent absorbs an equivalent swing and carries on.
Add a house edge and the outcome accelerates sharply. This is the mathematical statement of something every gambler eventually notices: the problem isn’t any individual bet, it’s the length of the sequence.
A coin is tossed until it lands heads. If that happens on the first toss you win £2, on the second £4, on the third £8, doubling each time. How much should you pay to play?
The expected value is infinite. Each possible outcome contributes exactly £1 to the expectation, and there are infinitely many of them.
Yet almost nobody would pay more than about £20, and they’re being sensible. The paradox exposed the gap between expected monetary value and actual usefulness, and prompted Daniel Bernoulli to propose that people value additional money less as they have more of it. That’s the origin of utility theory, and eventually of most of modern economics.
A family has two children. At least one is a boy. What’s the probability both are boys?
One-third, not one-half. The possible families are BB, BG and GB, since GG is excluded, and only one of those three is BB.
Now change one word. “The *older* child is a boy.” Now it’s one-half, because only BB and BG remain.
The problem is famous because the two versions sound almost identical and aren’t. It demonstrates that in probability the way information reaches you is part of the information, which is also the hidden mechanism in the Monty Hall Problem.
Three boxes: one holds two gold coins, one two silver, one one of each. You pick a box at random and draw a coin at random. It’s gold. What’s the probability the other coin in that box is also gold?
Two-thirds. The instinct is one-half, on the reasoning that the silver-silver box is eliminated so it must be a coin flip between the remaining two.
The error is treating boxes as equally likely when the evidence doesn’t support that. There are three gold coins in play and two of them sit in the gold-gold box. Given that you drew a gold coin, you’re twice as likely to be holding one from that box.
Published by Joseph Bertrand in 1889, it’s the direct ancestor of the Monty Hall Problem and fools people in exactly the same way a century later.
Two envelopes, one containing twice as much money as the other. You pick one and are offered a swap. Reasoning that the other holds either double or half, giving an expected value of 1.25 times what you’re holding, you swap. Then the same reasoning applies again, and you swap back forever.
Something is clearly wrong, and pinning down exactly what has occupied mathematicians for decades. The short answer is that the calculation treats the amount in your envelope as a fixed quantity while simultaneously allowing the total to vary, which quietly assumes a distribution of possible amounts that cannot exist.
It’s here because it’s the clearest demonstration that an argument can be individually plausible at every step and collectively nonsense.
Two players are partway through a series of games for a prize when they have to stop. One leads. How should the stake be divided?
The question was put to Blaise Pascal in 1654, and his correspondence with Pierre de Fermat about it is generally regarded as the founding document of probability theory.
Their insight was that the division should reflect not the games already played but the ways the remaining games could unfold. Counting the possible futures, rather than describing the past, is the conceptual move on which every entry in this article ultimately rests.
Worth noting what prompted the birth of an entire mathematical discipline: a dispute about how to split a gambling pot.
Buses are scheduled every ten minutes. You arrive at a random moment. Your average wait is longer than five minutes, often considerably longer.
You’re more likely to arrive during a long gap than a short one, precisely because long gaps occupy more of the timeline. Your arrival samples the intervals in proportion to their length rather than uniformly.
The same effect explains why class sizes feel larger than the published average, why your friends have more friends than you do on average, and why waiting rooms seem permanently full. It isn’t pessimism. It’s sampling.
In many naturally occurring datasets the leading digit is 1 about 30% of the time and 9 only about 5% of the time, rather than each digit turning up in roughly 11% of cases.
The effect appears in populations, river lengths, physical constants, stock prices and financial statements. It arises from data that spans several orders of magnitude and grows multiplicatively.
Its practical importance is forensic. Fabricated figures tend to distribute leading digits far more evenly than real data does, because people inventing numbers try to look random. Benford analysis is now a standard tool in fraud and accounting investigations, and has been used in election forensics.
You interview candidates one at a time, must decide immediately, and can’t recall a rejected candidate. How do you maximise your chance of hiring the best one?
Reject the first 37% automatically, then hire the first candidate better than everyone you’ve seen so far. That gives roughly a 37% chance of selecting the very best, which is remarkably high given how little information the rule uses.
The figure is 1/e in both places, one of the more elegant results in applied probability. Variants of the problem turn up wherever irreversible decisions have to be made in sequence, from house-hunting to hiring.
Drop a needle onto a floor of evenly spaced parallel lines. The probability that it crosses a line depends on pi, which means dropping enough needles and counting the crossings will estimate pi experimentally.
Posed by Georges-Louis Leclerc, Comte de Buffon in the eighteenth century, it’s one of the first problems in geometric probability and an early ancestor of Monte Carlo methods, the technique of solving deterministic problems by running large numbers of random trials.
A strange and rather beautiful result: a physical experiment involving no circles at all, producing the most famous constant associated with them.
Three errors recur across almost every problem here.
The first is ignoring how information arrived. Monty Hall, Bertrand’s Box and the Boy or Girl Paradox all turn on the process that generated the evidence rather than on the evidence itself.
The second is ignoring base rates. The prosecutor’s fallacy and the medical testing problem are the same error, and both get made routinely by intelligent people under professional conditions.
The third is confusing expected value with sensible behaviour. St Petersburg is the pure case, but the same confusion underlies most enthusiasm for betting systems.
Probability theory was invented to settle a gambling dispute, and the connection has never really broken. Every casino game is an applied probability problem in which one party understands the mathematics and the other frequently doesn’t.
The specific relevance of this list is that the errors it catalogues are the same errors that make gambling losses larger than they need to be. The gambler’s ruin explains why session length matters more than any individual bet. The inspection paradox explains why other people’s wins seem more common than yours. And failing to account for how information arrived is the mechanism behind almost every system ever sold.
Fifteen problems, three underlying errors. That ratio is the interesting part. These aren’t fifteen separate quirks of the human mind requiring fifteen separate corrections. They’re a handful of faults appearing repeatedly in different costumes, which is why solving Monty Hall doesn’t inoculate anyone against Bertrand’s Box even though the two are structurally identical.
The gambling relevance is narrower than it first appears, and more useful for being narrow. You won’t beat a slot machine by understanding Benford’s Law. What these problems do give you is a reliable warning sign: whenever a piece of reasoning about chance feels obviously correct and arrives quickly, that’s precisely the condition under which intuition has historically been wrong. The problems on this list all felt obvious to somebody, including to a great many people who were paid to know better.
Which is the practical takeaway. Not fifteen answers to memorise, but the habit of slowing down at exactly the moment you feel most certain.
What is the answer to the Monty Hall Problem?
Always switch. Switching wins two-thirds of the time because your original choice had a one-in-three chance of being correct, and the host revealing a losing door concentrates the remaining two-thirds onto the single unopened door.
Why is the Birthday Problem only 23 people?
Because the question asks about any shared birthday among all possible pairs. Twenty-three people form 253 pairs, and each pair is a separate opportunity for a match.
What is the prosecutor’s fallacy?
Confusing the probability of evidence given innocence with the probability of innocence given evidence. A one-in-a-million forensic match in a population of sixty million implies roughly sixty coincidental matches, not near-certain guilt.
Does the gambler’s ruin apply to casino games?
Yes, and more severely. The principle shows that a player with limited funds facing an opponent with far greater resources will eventually reach zero even in a fair game. A house edge accelerates it.
What is Benford’s Law used for?
Detecting fabricated data. Genuine datasets spanning several orders of magnitude show a leading digit of 1 around 30% of the time, while invented figures tend to distribute leading digits far too evenly.
Who invented probability theory?
It’s generally traced to the 1654 correspondence between Blaise Pascal and Pierre de Fermat about how to divide the stake in an unfinished gambling contest.
Gambling should be treated as paid entertainment and never as a way to make or recover money. Set deposit and loss limits before you play. Free, confidential support is available at BeGambleAware.org and on the National Gambling Helpline, 0808 8020 133. GAMSTOP lets you self-exclude from every UK licensed online operator at once.
24 August 2026
24 August 2026
24 August 2026
Casino Bonuses 
