With only 23 people in a room, the probability that at least two share a birthday is already just over half, at about 50.7 per cent. This sounds implausible because we often interpret the question as, “Does anyone share my birthday?” That would involve only 22 comparisons. The actual question asks whether any pair in the group has the same birthday.
There are 23×22÷2 = 253 possible pairs among 23 people. The number of comparison opportunities grows quickly, but 253÷365 is not the correct calculation because those pairings are not independent. The exact method instead calculates the probability that every birthday is different. Ignoring leap day, and assuming birthdays are uniformly distributed across 365 days and independent, the first person may have any birthday, the second must avoid one occupied date, the third must avoid two, and so on:
Probability of all different birthdays = 365/365×364/365×363/365×…×343/365 ≈ 49.27%.
The probability of at least one shared birthday is therefore 1 − 49.27% = 50.73%. This is not a genuine logical paradox. It reveals a weakness in intuition: we notice the number of people but underestimate how the number of possible pairs grows roughly with the square of the group size. Real birthdays are not perfectly uniform, and twins and similar cases weaken the independence assumption, but the model’s central explanation remains sound.
References
Purdue University, Birthday and Matching Problems
https://www.stat.purdue.edu/~mlevins/docs/stat516/lecture_notes/Lecture2a_STAT516.pdf
University of Cambridge, Case Studies for CST IA Probability
https://www.cl.cam.ac.uk/teaching/1112/Probabilty/Cases.pdf
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.