The Monty Hall Problem: Why Switching Doors Wins

Imagine you're standing on the set of a game show. In front of you are three closed doors. Behind one is a brand-new car. Behind the other two: goats. You pick Door Number 1. The host — who knows what's behind every door — swings open Door Number 3, revealing a goat. Then he asks you: "Do you want to switch to Door Number 2?"

Your gut says it doesn't matter. There are two doors left. The car is behind one of them. That's 50/50 odds, right?

Wrong. And the gap between intuition and reality here is one of the most instructive puzzles in all of mathematics.

The Concept

The Monty Hall Problem — named after the long-running game show Let's Make a Deal and its genial host Monty Hall — is a probability puzzle that has humbled Nobel laureates, ignited national newspaper wars, and convinced Paul Erdős, one of the most prolific mathematicians in history, only after he was shown a computer simulation.

The setup is simple. Three doors, one car, two goats. You pick a door. The host (who always knows where the car is, and always opens a losing door) reveals a goat behind one of the doors you didn't pick. Now you can stick with your original choice, or switch to the other remaining closed door.

The correct answer: always switch. Switching wins the car two-thirds of the time. Staying wins only one-third of the time.

Why It Matters

Before we dive into why this is true, it's worth understanding why it matters beyond winning game show cars.

The Monty Hall Problem is a clean, concrete illustration of conditional probability — how probabilities shift when you receive new information. This isn't just a math curiosity. Conditional probability is the engine behind:

Medical diagnosis — When a test comes back positive for a rare disease, the actual probability you have it depends heavily on how rare the disease is. Ignoring this leads to medical errors.
Bayesian machine learning — AI systems use Bayes' theorem (the same math that solves Monty Hall) to update beliefs as they receive new evidence.
Legal reasoning — Misapplying conditional probability in courtrooms has led to wrongful convictions — prosecutors sometimes present statistics as if prior probabilities don't exist.
Spam filters — Your email's spam filter is essentially running Monty Hall logic thousands of times per second, updating probabilities based on new evidence (this word appeared, that phrase didn't).

The Monty Hall Problem is a training ground for the kind of probabilistic reasoning that computers do natively but humans find profoundly counterintuitive.

The History: A Puzzle That Broke the Internet (Before the Internet)

The origins go back further than most people realize. In 1959, mathematician Martin Gardner described what he called the Three Prisoners Problem in his Mathematical Games column in Scientific American: three prisoners, one pardoned at random, a guard who reveals one of the others' fates. It's mathematically identical to Monty Hall, just dressed differently.

The game-show version was formally posed in 1975 by Steve Selvin, a biostatistics professor at UC Berkeley, who wrote a letter to the American Statistician describing it as a probability puzzle based on Let's Make a Deal. His follow-up letter — responding to critics — was the first time the phrase "Monty Hall Problem" appeared in print.

Then came 1990, and one of the most remarkable intellectual controversies in modern history.

Marilyn vos Savant, who held the Guinness World Record for highest recorded IQ and wrote the Ask Marilyn column in Parade magazine, received a reader's letter asking about the problem. She published her answer: yes, you should switch. Switching wins two-thirds of the time.

The response was unprecedented. Parade received approximately 10,000 letters. Roughly 1,000 were signed by people with PhDs — many on university letterhead — insisting she was wrong. Letters came in from mathematics professors, statisticians, and scientists. One letter read: "You blew it, and you blew it big! ... There is enough mathematical illiteracy in this country, and we don't need the world's highest IQ propagating more." Another: "You are the goat!"

She was right. They were wrong.

Vos Savant issued a challenge: teachers across America should have their students run the experiment with paper cups and pennies, hundreds of times, and report the results. The data came back confirming 2/3 wins when switching. Letters of apology began to arrive. Erdős himself — who had reportedly rejected the solution even when shown the math — finally accepted it only after seeing a computer simulation.

The New York Times ran a front-page story about the controversy in 1991.

The Details: Why 2/3?

Here is why switching wins two-thirds of the time, explained three different ways.

Explanation 1: Track Where the Car Was

When you first pick a door, there are three equally likely scenarios:

Scenario A: Car is behind Door 1 (your door). Probability: 1/3.
Scenario B: Car is behind Door 2. Probability: 1/3.
Scenario C: Car is behind Door 3. Probability: 1/3.

In Scenario A, if you switch, you lose. (You had the car; you left it.) In Scenario B, the host opens Door 3 (the only losing door he can open). If you switch to Door 2, you win. In Scenario C, the host opens Door 2. If you switch to Door 3, you win.

Switching wins in Scenarios B and C. That's 2 out of 3 scenarios. Switching wins 2/3 of the time.

Staying wins only in Scenario A. Staying wins 1/3 of the time.

The host's action doesn't make it 50/50 — it concentrates probability onto the door you didn't pick.

Explanation 2: The 100-Door Version

This is the clearest intuition pump for most people. Imagine 100 doors. Behind one is a car. You pick Door 1. The host then opens 98 doors, all revealing goats, leaving only your door and Door 72 closed.

Would you switch?

Of course you would. The host just told you an enormous amount of information. There was only a 1% chance the car was behind your door to begin with. There's now a 99% chance it's behind Door 72.

The 3-door version is the same logic, just compressed. Your initial pick carries 1/3 probability. The other door carries 2/3. The host opening a losing door doesn't change your initial pick's probability — it just consolidates all the remaining probability onto the one remaining door.

Explanation 3: Conditional Probability (Bayes' Theorem)

For those who want the rigorous version: the Monty Hall Problem is a textbook application of Bayes' theorem, which describes how to update a probability estimate given new evidence.

Before the host acts, P(car behind Door 2) = 1/3.

After the host opens Door 3 (revealing a goat), we update using Bayes:

P(car behind Door 2 | host opened Door 3) = P(host opens Door 3 | car behind Door 2) × P(car behind Door 2) / P(host opens Door 3)

If the car is behind Door 2, the host must open Door 3 (it's the only losing door available). So P(host opens Door 3 | car behind Door 2) = 1.

Working through the arithmetic: the posterior probability that the car is behind Door 2, given that the host opened Door 3, is 2/3.

This is why the Monty Hall Problem matters for machine learning. Every time a recommendation algorithm updates its model based on new user behavior, or a spam filter adjusts after seeing a new word pattern, it's doing essentially this — computing a posterior probability from a prior and new evidence.

The Crucial Assumptions

Here's where it gets interesting: the 2/3 answer depends on specific rules for the host's behavior.

The host must: 1. Always reveal a goat (never the car) 2. Always offer the switch 3. Know where the car is

If the host were choosing randomly which door to open — sometimes accidentally revealing the car — the math changes completely. If the host opens a door at random and happens to show a goat, it actually does become 50/50.

This is the source of much confusion. People imagine a host acting randomly, or wonder whether the host might be trying to trick them. In the classic formulation, the host's intent is known and fixed. His action carries information precisely because it's constrained.

The psychologically destabilizing thing about Monty Hall is that our intuitions were built for a world where revealing a door means "we've eliminated an option, now there are fewer." But probability doesn't work like that. Information from a constrained, knowledgeable source changes the landscape differently than random revelation.

Surprising Connections

The Monty Hall Problem echoes across several fields in ways you might not expect:

Quantum mechanics: Physicists have described measurement collapse in quantum systems using analogies to Monty Hall — the act of measurement (observation) constrains what's possible, concentrating probability in ways that seem counterintuitive from a classical viewpoint.

DNA ancestry testing: When you receive information that certain genetic markers are absent, the conditional probabilities of your ancestry shift — not always in intuitive directions. Genetic counselors are trained specifically to reason about this.

Drug trial analysis: In adaptive clinical trials, when one arm of a trial is showing poorer results (and thus patients are being shifted away from it), the statistics must account for the fact that this reassignment carries information — a real-world Monty Hall effect.

The two-envelope paradox: A close cousin to Monty Hall, where you're given two envelopes with money and told one has twice as much as the other. Should you switch after opening one? (The answer: it depends on what you know about how amounts were chosen — again, prior information matters.)

The Lesson

The Monty Hall Problem's deepest lesson isn't about game shows. It's about the relationship between knowledge and probability.

Probability isn't just "how many options are left." It's a measure of uncertainty given what you know. When an expert who knows the answer provides constrained information, that information reshapes the landscape in ways that pure counting can't capture.

This is why conditional probability is one of the most important — and most consistently misunderstood — ideas in all of mathematics. Our brains are pattern-recognizers that evolved in a world of direct causes and visible effects. We are not naturally wired to track how probability flows when constrained sources selectively reveal information.

The door didn't change. The car didn't move. But your information did. And that changes everything.

Takeaways

Always switch. In the classic formulation, switching wins 2/3 of the time; staying wins 1/3. This is mathematically certain, confirmed by simulation and experiment.
The host's knowledge is the key. The puzzle only works because the host knows where the car is and always reveals a goat. A randomly-acting host produces different math.
Probability is about information, not counting. The counterintuitive result comes from not accounting for the information the host's constrained action provides.
This confounded 1,000 PhDs. When vos Savant published the correct answer in 1990, thousands of academics wrote in to say she was wrong. She wasn't. The cognitive bias here is powerful and universal.
The underlying logic runs the world. Bayesian reasoning — updating probabilities with new evidence — is the foundation of modern spam filters, medical diagnosis, AI recommendation systems, and much more.