Mathematics
The Birthday Problem and Its Probability Surprises
Quick fact
With just 23 people, the probability that at least two share a birthday is about 50.7%. With 50 people, it jumps to about 97%.
Why this is interesting
You walk into a room with 22 strangers. Would you bet that at least two of you share the same birthday? Most people would guess no, but the odds say it's more likely than not.
Read the full explanation
Understanding The Birthday Problem and Its Probability Surprises
The birthday problem asks: in a group of n people, what is the probability that at least two people have the same birthday? Our intuition says birthdays are spread out, so collisions are rare, but the reality is the opposite. The key is that we're not checking one person against the group; we're checking every pair. The number of pairs grows quickly: with 23 people, there are 253 pairs. To solve, we flip the question: instead of calculating the chance of a match directly, we calculate the chance that no one shares a birthday, then subtract that from 1. For two people, the probability they have different birthdays is 364/365. For a third person to also avoid both, it's (364/365) × (363/365), and so on. This product gets smaller very fast. By the time you reach 23 people, this product has fallen to about 0.493, so the chance of a match is 0.507—better than 50%.
A deeper explanation
The mechanism behind this surprise is combinatorial: the probability of a match rises rapidly because the number of possible pairs grows quadratically. With n people, there are n(n−1)/2 pairs, and each pair has a 1/365 chance of being a match. For small n, these events are nearly independent, so the expected number of matches is about n(n−1)/(2×365). For 23 people, that expected value is 253/365 ≈ 0.69, already substantial. The exact calculation uses the product of probabilities that each new person's birthday is different from all previous ones: P(no match) = ∏{i=1}^{n-1} (1 - i/365). This product can be approximated using the exponential function: 1 - x ≈ e^{-x} for small x, giving P(no match) ≈ exp(-n(n-1)/(2×365)). This approximation shows why the threshold is so low: the exponent grows quadratically, so the probability of a match approaches 1 very quickly. The same reasoning applies to any system where collisions can happen, from hashing algorithms to cryptographic attacks, which is why understanding this problem matters beyond party trivia.