The birthday problem, explained

Plain-language answer #

In a room of just 23 randomly chosen people, the probability that some two of them share a birthday is already just over 50% — better than even odds, in a room most people would guess needs to be far larger before a shared birthday becomes likely. By 70 people, it is over 99.9%. This is the birthday problem, one of the most famous demonstrations that intuition badly underestimates how many “opportunities” a group creates for a match.(Diaconis & Mosteller, 1989)

Why it matters #

The birthday problem is the clearest small, checkable example of the “multiple opportunities” logic that runs through almost every coincidence story: the question people intuitively answer (“what’s the chance someone shares my birthday?”) is not the question actually being asked (“what’s the chance any two people in the room share a birthday?”). The gap between those two questions is the entire explanation for why the answer feels so counterintuitive.

Worked example: two different questions, two different answers #

Question 1: “What’s the probability someone else shares my specific birthday?” With 22 other people, each with roughly a 1-in-365 chance of matching your one fixed date, the probability that at least one does is low — around 6% in a room of 23. This matches most people’s intuition, because it’s the question they’re implicitly answering.

Question 2: “What’s the probability that any two people in the room share any birthday?” Now every pair counts, not just pairs involving you. A room of 23 people contains 23 × 22 / 2 = 253 distinct pairs, and any one of those 253 pairs matching is enough. The exact calculation works by finding the probability that no two people share a birthday — multiplying, for each person added, the probability their birthday avoids every date already taken — and subtracting that from 1:

These figures assume birthdays are spread uniformly across 365 days (ignoring February 29 and small real-world seasonal variation) and are independent across people — assumptions that are close to, but not exactly, true of real populations.(Diaconis & Mosteller, 1989) The probability is a model page works through this calculation’s structure in more depth, and the Birthday Room interactive lets you vary the room size and see the running simulation against the exact formula.

Common misconception #

“253 pairs still isn’t that many, so 50% still seems too high” underestimates how the pair count grows. Adding one more person to the room adds as many new pairs as there are existing people — the number of pairs grows roughly with the square of the group size, not linearly with it. That quadratic growth, not any special property of birthdays, is what makes the crossover point so much lower than intuition expects.

Limits and open questions #

Real birthdays are not perfectly uniform across the year (some months have modestly higher birth rates than others in most populations), which makes real shared-birthday probabilities very slightly higher than the idealized uniform-distribution calculation predicts, because any deviation from uniformity increases the chance of a match. The idealized figures above are useful for building intuition, not for a precise real-population estimate.

Related #

• Probability is a model, not a verdict introduces the same worked example as part of a broader point about reference classes and models.
• Interactive: Birthday Room lets you run the simulation yourself, with a seeded, reproducible random draw.

Key takeaways #

• The birthday problem crosses even odds at just 23 people because it asks about any matching pair, not a match with one specific person.
• The number of pairs in a group grows roughly with the square of the group size, which is why the probability of a match rises so much faster than intuition expects.
• The same “count every pair, not just the one you’d notice” logic explains why many real coincidences are less surprising than they first appear.