‽ Paradox Lab

Home › Probability › Bertrand's box paradox: the other coin is probably gold

Bertrand's box paradox: the other coin is probably gold

You pull a gold coin from a random box. Is the other coin gold? Not 50/50.

The puzzle

There are three boxes. One holds two gold coins, one holds two silver coins, and one holds one of each. You choose a box at random and, without looking, take out one coin. It's gold.

What's the chance the other coin in that box is also gold? Most people say 1/2: it must be either the gold–gold box or the gold–silver box. The real answer is 2/3.

📦 Draw a coin

Gold–goldGoldGold
Silver–silverSilverSilver
Gold–silverGoldSilver

The boxes are shown open here so you know what's in them. When you draw, they close.

📈 Simulate thousands of draws

Of the draws that came up gold, how often was the other coin gold too?

Count the coins, not the boxes

The mistake is to think of the two possible boxes as equally likely. They're not, because the gold–gold box gives you two ways to draw a gold coin and the gold–silver box gives you only one.

Label the three gold coins: G1 and G2 in the gold–gold box, and G3 in the mixed box. Drawing any one of them is equally likely. If you drew G1, the other coin is G2 (gold). If you drew G2, the other is G1 (gold). If you drew G3, the other is silver. That's 2 gold out of 3.

The simulator shows the same thing: about two-thirds of gold draws come from the gold–gold box.

Bayes' rule version

P(gold–gold box | drew gold) = P(drew gold | gold–gold) × P(gold–gold) ÷ P(drew gold) = (1 × 1/3) ÷ (1/2) = 2/3

Seeing a gold coin is evidence. It's certain under the gold–gold box but only a coin-toss under the mixed box, so it shifts the odds towards gold–gold.

Where it comes from

French mathematician Joseph Bertrand posed it in his 1889 book on probability, as a warning about counting possibilities carelessly. Bertrand himself used the "two boxes left, so 1/2" argument as an example of reasoning that looks right but isn't. The same trap sits underneath the Monty Hall problem and the "boy or girl" puzzle.

Questions

What is the answer to Bertrand's box paradox?

2/3. Given that you drew a gold coin, there is a 2/3 chance the other coin in the box is gold.

Why isn't it 1/2?

Because the gold–gold box is twice as likely to produce a gold coin as the mixed box. Drawing gold is stronger evidence for the gold–gold box.

How is it related to the Monty Hall problem?

Both come from treating outcomes as equally likely when they aren't. In both, the information you receive is more likely under one possibility than another, and that shifts the odds away from 50/50.

Try another paradox

Monty Hall problem

Switching wins 2 times in 3. Play it, then run 10,000 games.

Probability

False positive paradox

A 99% accurate test can be wrong most times it says "positive". See 1,000 people.

Statistics