What makes a question hypergeometric
You have a pool of things. Some of them are what you want. You take a handful, and you do not put any of them back. That is the whole shape of it.
The not-putting-back is the important part. Once you have drawn a heart, there are fewer hearts left and fewer cards left, so the odds for the next draw are different from the first. The hypergeometric distribution is what you get when you account for that shift exactly rather than pretending it does not happen.
The formula
Read it as counting rather than as probability. The bottom is every way to choose your draw from the population. The top is every way to choose k of your successes and fill the rest of your hand from the failures. Divide the favourable arrangements by all arrangements and you have the answer.
Worked example
Exactly two hearts in a five card hand.
N = 52 cards, K = 13 hearts, n = 5 drawn, k = 2 wanted.
P = C(13, 2) × C(39, 3) ÷ C(52, 5)
P = 78 × 9,139 ÷ 2,598,960
P = 0.274280, or about 27.4%
For comparison, a binomial calculation with p = 13/52 gives 0.263672. Close, but not the same number, and the gap has a direction worth understanding.
Hypergeometric against binomial
The binomial distribution answers the same shape of question with replacement: put the card back and shuffle before each draw. That keeps every trial identical, which makes the maths simpler and the answer wrong for a deck.
Two facts describe the relationship completely:
- The means are identical. Both come to n × K ÷ N. On average, the two distributions agree exactly.
- The variances are not. The hypergeometric variance is the binomial variance multiplied by (N − n) ÷ (N − 1), which is always less than one. Drawing without replacement is less variable.
That second multiplier should look familiar if you have used the sample size calculator: it is the finite population correction, the same factor that lets a survey of a small town need fewer responses than a survey of a continent. It is the same idea seen from the other end.
Which way a binomial answer goes wrong
Because hypergeometric is more tightly concentrated, a binomial substitute does not simply run high or low. It understates the middle and overstates the tails.
| Five card hand | True (hypergeometric) | Binomial says | Off by |
|---|---|---|---|
| At least 1 heart | 0.7785 | 0.7627 | 2% low |
| At least 3 hearts | 0.0928 | 0.1035 | 12% high |
| At least 4 hearts | 0.0112 | 0.0156 | 39% high |
The tails are where people actually ask questions. Nobody wonders about an average hand; they wonder about a great one. So the binomial shortcut fails hardest at exactly the question you most wanted answered, and it fails in the flattering direction.
The practical rule: if your draw is under about 5 percent of the population, removing a few items barely changes what is left and the two distributions converge. Above that, use the exact one. It is not harder, and this page does it for you.
Where this comes up outside card games
- Acceptance sampling. Pull 20 units from a batch of 500 and ask how likely you are to catch a defect if 15 are bad. The batch is finite and you do not put units back.
- Audit and compliance. Sampling 40 invoices from 600 to estimate an error rate is the same question wearing a suit.
- Lotteries. Balls are drawn without replacement, so matching a given number of them is exactly this formula.
- Ecology. Capture-recapture estimates of animal populations rest on it.
- Deckbuilding. The opening hand calculator and the manabase calculator are both this distribution with the questions pre-asked.
If your trials really are independent, with a fixed probability that does not move as you go, the binomial calculator is the right page instead.