The Monty Hall Problem: Why Switching Wins and Your Brain Hates It

You pick one of three doors. Behind one is a car; behind two are goats. The host — who knows what's behind every door — opens one of the other two doors to reveal a goat. Now he offers you a choice: stay with your original door, or switch to the remaining closed door. What should you do? The mathematically correct answer is to switch, which doubles your probability of winning from 1/3 to 2/3. In September 1990, when Marilyn vos Savant published this answer in Parade magazine, approximately 10,000 readers wrote in to tell her she was wrong. Nearly a thousand of those letters came from people with PhD credentials. She was right, they were wrong, and the dispute became one of the most instructive demonstrations in the history of human reasoning about probability.

The Setup: A Puzzle with a History

The Monty Hall problem did not originate with Marilyn vos Savant or with the game show Let's Make a Deal. Statistician Steve Selvin first formalized it in a letter to the American Statistician in February 1975, posing it as a probability puzzle and providing a solution. He named it after Monty Hall, the host of Let's Make a Deal, which had run the broad format (though not this exact scenario) since 1963. After receiving pushback from readers, Selvin published a follow-up letter in August 1975 — the first time the phrase "Monty Hall Problem" appeared in print.

Source: Monty Hall problem — Wikipedia

Source: Steve Selvin — Wikipedia

The puzzle then largely circulated in statistics and mathematics circles for 15 years, generating occasional debate, before Craig F. Whitaker sent a version of it to Marilyn vos Savant's "Ask Marilyn" column in Parade magazine. Vos Savant, listed in the Guinness Book of World Records as having the world's highest recorded IQ, answered it in the September 9, 1990 issue.

Her answer: yes, you should switch. The remaining closed door has a 2/3 probability of hiding the car.

The letters started arriving within days.

The Protocol: What the Math Actually Says

The problem seems to present a 50/50 choice. After the host opens a door, two doors remain. Surely each has a 1/2 probability of hiding the car?

This intuition is wrong, and understanding exactly why it is wrong is the educational content of the puzzle.

When you first picked a door, you had a 1/3 chance of being correct. That means there was a 2/3 chance the car was behind one of the other two doors. This 2/3 probability does not evaporate when the host opens one of those two doors. It concentrates in the remaining unopened door.

Think of it this way: imagine the game with 100 doors instead of three. You pick one door. The host, who knows where the car is, opens 98 of the other 99 doors to reveal goats, leaving one door closed. Would you switch? Almost everyone says yes — because clearly the car is probably in that one door the host carefully left unopened, not in your original random pick. The three-door version is identical in structure; it just compresses the intuitive force of the asymmetry.

Source: The Monty Hall Problem — Parade

The key to the correct answer is that the host's behavior is not random. The host always opens a door with a goat and always offers the switch. This behavior carries information. If you picked the car initially (1/3 probability), the host can open either remaining door — he has flexibility. If you picked a goat (2/3 probability), the host has no flexibility — he must open the one specific other door that has a goat, leaving the car behind the door he does not open. So the door the host chose to leave closed is enriched with probability from the cases where you were wrong.

Source: Monty Hall Problem — Britannica

Formally, in Bayesian terms:

• P(car behind your door) = 1/3
• P(car behind either remaining door) = 2/3
• After host opens one goat door: P(car behind the other remaining door) = 2/3
• Staying wins if you originally picked the car: P = 1/3
• Switching wins if you originally picked a goat: P = 2/3

Since you are twice as likely to have picked a goat as the car on your first choice, switching wins twice as often as staying.

The Results: 10,000 Letters and What They Said

The volume and character of the negative mail vos Savant received is worth dwelling on, because it tells us something important about expert overconfidence in fields adjacent to one's specialization.

The Parade magazine received an estimated 10,000 letters in response to her answer. Vos Savant reports that close to 1,000 were signed by people identifying themselves as PhD holders, many on the letterheads of mathematics and statistics departments.

Source: The Time Everyone "Corrected" the World's Smartest Woman — Priceonomics

A representative sample of the letters:

A Ph.D. at the University of Florida wrote: "You blew it, and you blew it big! Since you seem to have difficulty grasping the basic principle at work here, I'll explain. After the host reveals a goat, you now have a one-in-two chance of being correct. Whether you change your selection or not, the odds are the same."

A professor of mathematics at George Mason University wrote: "You made an error, but look at the positive side. If all those Ph.D.s were wrong, the country would be in very serious trouble."

A cognitive scientist at the National Institutes of Health wrote: "You are utterly incorrect. How many irate mathematicians are needed to get you to change your mind?"

Source: The Monty Hall Problem: The Two Goats, Three Doors Question — Parade

Vos Savant published three more columns on the problem in 1990 and 1991, each providing additional explanations and mathematical proofs. The mail did not abate — if anything, it intensified after each column. She then proposed a nationwide classroom experiment: have students run 400 trials each with paper cups and pennies, tracking switching versus staying outcomes.

The experimental results poured in from hundreds of schools. They consistently showed the 2/3 switching advantage. The flood of mail reversed. She began receiving letters, including from some original critics, acknowledging she had been right.

Source: Monty Hall problem — Wikipedia

The Timeline

Paul Erdős: When Even the Best Get It Wrong

The Monty Hall problem's most celebrated victim of intuitive error was Paul Erdős, widely considered one of the most prolific and gifted mathematicians of the 20th century. Erdős collaborated on more papers than virtually any other mathematician in history and was famous for his ability to solve problems that stumped entire research teams.

When the problem was posed to him, Erdős gave the standard wrong answer: 1/2 probability for each remaining door. When presented with the correct answer and its proof, he refused to accept it. When presented with alternative proofs, he remained unconvinced.

Erdős was only persuaded when his friend and fellow mathematician Andrew Vázsonyi wrote a computer program and ran a Monte Carlo simulation 100,000 times in front of him. As the simulation results accumulated — switching winning approximately 66,700 times out of 100,000 — Erdős watched and gradually accepted that the answer was 2/3.

Source: 1 Paper 87-2010 Monty Hall, Paul Erdős, and Monte Carlo — MWSUG

Source: The Monty Hall Problem: Dividing Statisticians since 1975 — Medium

This is not an anecdote about Erdős being foolish. It is an anecdote about a deeply counterintuitive result that even the most powerful mathematical intuition cannot easily override. The problem exploits a specific cognitive architecture that humans appear to share regardless of mathematical sophistication: once we see two closed doors, we categorize the situation as a 50/50 problem, and we find it extremely difficult to revise that categorization based on reasoning about earlier probability states.

The Controversy: When Variant Assumptions Change the Answer

The vos Savant affair obscures an important nuance: the answer 2/3 is correct only given specific assumptions about how the host behaves. Changing those assumptions changes the answer.

The standard problem assumes:

1. The host always opens a door after your initial pick.
2. The host always reveals a goat (never the car).
3. The host always offers you the chance to switch.
4. If the host has two goat doors to choose from, he picks one at random.

Under these assumptions, switching gives 2/3 probability of winning.

What if the host is not constrained to choose randomly between two goat doors? If the host always opens the lower-numbered goat door when he has a choice, then the probability of winning by switching is no longer uniformly 2/3. It becomes conditional on which door the host opened — and in some cases equals 1/2. This is the "Monty Hall with a non-uniform host" version studied in the formal probability literature.

Source: The Monty Hall Problem is not a Probability Puzzle — arXiv

The lesson is not that vos Savant was wrong. It is that probability problems are highly sensitive to exact problem specification, and that many of the "1/2" advocates were implicitly solving a slightly different problem — one where the host's behavior is unconstrained — which genuinely does produce different answers. This is part of why the dispute proved so durable: some critics were making a legitimate point about problem specification even while being wrong about the standard formulation.

What the Problem Reveals About Human Reasoning

The Monty Hall problem has become a standard test case in cognitive psychology because it isolates a specific and robust reasoning failure: the tendency to treat updated probability as though it reflects only the current state, rather than incorporating the process by which that state was reached.

Psychologists call this related pattern "conservatism" in belief updating — the tendency to underweight information that arrived through a conditional process. The host's door opening is deeply informative, but our brains tend to represent it as a neutral event (now there are two doors, so 50/50) rather than a conditional one (the host had to choose this specific door, and that choice carries information about where the car is).

This failure is not random: it follows from the way humans quickly categorize situations and then resist revising the categorization. Two doors = binary choice = equal probability. The categorical representation overrides the conditional reasoning.

Source: Revisiting the Monty Hall Probability Puzzle — American Enterprise Institute

In everyday terms: imagine two job candidates remain after the initial screening. You might intuitively think each has a 50% chance of being selected. But if you know the screening process systematically eliminated weaker candidates while preserving information about the job's requirements, the remaining candidates are not equally likely to be selected — the one whose qualifications better match the screener's known criteria has a higher probability. The situation looks like a 50/50 but encodes a rich conditional structure.

This is the Bayesian perspective: probability is not a property of isolated choices but of the entire history of information that led to those choices.

Where It Stands Today

The Monty Hall problem has settled into the curriculum of probability, statistics, and cognitive bias courses worldwide. It appears in economics and decision theory because it cleanly demonstrates that humans systematically fail to update beliefs correctly even when the information structure is simple and the stakes are clear.

The practical application most often cited today is in medical testing. When a test result is positive, patients and sometimes physicians instinctively think "the test is positive, so I probably have the disease." But the probability of actually having the disease given a positive test (the posterior probability) depends critically on how common the disease is (the prior probability). A test with 99% sensitivity for a disease that affects 1 in 10,000 people will produce far more false positives than true positives. Like the Monty Hall problem, the intuitive "it's positive, so I have it" reasoning ignores the prior probability that established the base rate.

Source: Monty Hall Problem: Solution Explained Simply — Statistics How To

The problem is also now used in AI and machine learning pedagogy. Bayesian reasoning — updating probability estimates as new information arrives, properly accounting for the conditional structure of that information — is foundational to many machine learning algorithms. The Monty Hall problem provides a pedagogically compact test of whether a student (or a model) is updating conditionally or merely counting.

The deeper point is this: the Monty Hall problem is not really about game shows. It is about the difference between the world's actual probability structure and the intuitive probability structure our brains impose on it. Those two things diverge more often than we expect, and they diverge in patterns that can be learned. That is exactly what probability theory is for.

When the Monty Hall Answer Changes: the Host's Information Matters

The 2/3 switching advantage is not a universal truth about three-door problems. It is the correct answer to one specific version of the problem — the version in which the host knows where the car is, always opens a goat door, and always offers the switch. Change any of those assumptions and the math changes too.

The clearest illustration is what probability theorists call the "Monty Fall" variant, introduced as a pedagogical contrast to the standard problem. In Monty Fall, the host does not know where the car is. After you pick a door, he slips, accidentally pushes open one of the other two doors, and it just happens to reveal a goat. Now you are offered the switch. Should you take it?

No — or rather, it does not matter. When the host's door-opening is random rather than informed, the situation is genuinely symmetric. The host had a 1/3 chance of accidentally revealing the car (which did not happen), and given that he revealed a goat at random, the two remaining closed doors are equally likely to hide the car. Switching wins 1/2 the time. Staying wins 1/2 the time. The 2/3 advantage has disappeared entirely.

Source: Monty Hall problem — Wikipedia

The Bayesian framing makes the mechanism visible. In the standard problem, the host's action is not a random event — it is a deliberate choice conditioned on the location of the car. That conditioning is what concentrates probability onto the remaining door. When the host is not conditioned on the car's location, his action carries no information about it, and the probability does not shift. As Selvin's original 1975 American Statistician letter established, the key variable is not the number of doors but the nature of the information the host's action encodes.

Source: Steve Selvin — Wikipedia

A variant with a biased host produces a third answer. If the host always opens the lower-numbered available door when he has a choice between two goat doors, then the probability of winning by switching is no longer uniformly 2/3 — it becomes conditional on which door the host opened. If the host opened a high-numbered door, he probably did not have a choice, which means you probably picked the car. The switching advantage is no longer flat across cases; it varies by door number.

The medical testing parallel is exact. When a diagnostic test comes back positive, the relevant question is not just "how accurate is this test?" but "what did the test actually know about me before it ran?" A test that is specifically looking for your risk factors carries more conditional information than a random screening. The posterior probability of disease depends on the prior probability (how common is it in people like you?) and on the test's detection process (did the test have access to information that changes what a positive result means?). This is identical to the relationship between the host's knowledge and what his door-opening reveals. A positive result from a poorly targeted test, like a door opened by a host who guesses randomly, is less informative than it appears.

The broader lesson is one that vos Savant's critics missed: the Monty Hall problem is not really a question about doors. It is a question about the difference between an event that happens and an event that was selected to happen. Those look identical from the outside — a door opened, a test came back positive, a fact was mentioned in conversation — but they encode radically different amounts of information. Tracking that difference is what Bayesian reasoning is for, and what common sense systematically fails to do.

FAQs

Q: What is the correct answer to the Monty Hall problem? A: You should switch doors. Switching wins the car with probability 2/3; staying wins with probability 1/3. This assumes the host always opens a goat door and always offers the switch.

Q: Why doesn't the probability become 50/50 after the host opens a door? A: Because the host's choice is not random — he always reveals a goat, which means his behavior is conditional on where the car is. The 2/3 probability that was distributed across the two unchosen doors at the start collapses onto the one remaining closed door, not equally across both remaining doors.

Q: Who first formulated the Monty Hall problem? A: Statistician Steve Selvin published it in a letter to the American Statistician in February 1975 and coined the name in a follow-up letter that August. Marilyn vos Savant's 1990 Parade column made it famous, but she did not originate the puzzle.

Q: Was Paul Erdős really wrong about the Monty Hall problem? A: Yes. Erdős, one of the most accomplished mathematicians of the 20th century, gave the intuitive but incorrect 50/50 answer and refused to accept correction until his friend Andrew Vázsonyi ran a 100,000-trial computer simulation in front of him.

Q: Does the answer change if the host doesn't always offer a switch? A: Yes. If the host is not constrained to always offer a switch, the problem changes substantially. The 2/3 answer applies only when the host's behavior follows the standard rules: always open a goat door, always offer the switch.

Q: What does the Monty Hall problem reveal about human reasoning? A: It demonstrates a systematic failure to perform Bayesian updating — incorporating the process by which information arrived, not just the current state. Humans tend to re-categorize a conditional situation as a fresh 50/50 choice, losing the information encoded in the host's constrained behavior.

Read next:

• Schrödinger's Cat: The Thought Experiment Physicists Misuse Most
• The Libet Experiment: Did Your Brain Decide Before You Did?
• Pierre-Simon Laplace: The Mathematician Who Mapped the Solar System