Statistics › Section D · Discrete random variables › Topic D.5
The hypergeometric distribution
The binomial model could not handle drawing without replacement from a small population, where every draw changes what remains. The hypergeometric distribution is the model for exactly that: sampling from a finite two-group population and counting the group of interest.
Binomial's twin, without replacement
The binomial topic flagged the case it could not handle: drawing without replacement from a small population, where every draw changes the remaining chances. The hypergeometric distribution is the model for exactly that - sampling a fixed number of items from a finite population that holds two groups, and counting how many of the sample come from the group of interest. Committees drawn from a mixed pool, defectives found in a small crate, votes from a known ballot box.
The hypergeometric setting
Five characteristics define the setting. 1. The population holds two groups - junior and senior, defective and sound, red and blue. 2. One group is the group of interest, the count being tracked. 3. A fixed sample of \(n\) is drawn. 4. Draws are without replacement, so each pick changes the probabilities for the next. 5. Consequently the picks are dependent - these are not the independent yes-or-no trials of a binomial setting. The random variable \(X\) counts how many of the sample come from the group of interest.
Notation: three parameters
Write \(X \sim H(r, b, n)\), read as "\(X\) has a hypergeometric distribution": \(r\) is the size of the group of interest, \(b\) the size of the other group, and \(n\) the sample size drawn from the combined population \(r + b\). \(X\) runs from 0 (or from \(n - b\), when the sample is too big to avoid the interest group) up to the smaller of \(n\) and \(r\). Those three numbers are everything: every probability and the mean follow from them alone.
The formula
The probability of exactly \(k\) from the group of interest is
Read it as counting equally likely samples. The denominator \(\binom{r+b}{n}\) counts every possible sample of \(n\) from the whole population - all equally likely, since the drawing is random. The numerator counts the samples holding exactly \(k\) from the interest group: choose those \(k\) from the \(r\) available, and the remaining \(n-k\) from the \(b\) others. The combination notation is the same one the binomial formula used.
The mean
The expected count from the group of interest is its population share times the sample size:
If juniors make up \(\frac{9}{16}\) of a pool, a 4-person team holds \(4 \times \frac{9}{16} = 2.25\) juniors on average. The formula formalises a natural intuition: random selection preserves proportions, so the expected sample mirrors the population's mix - even though any single sample can land anywhere in the range. The spread has a formula too:
For the 4-person team that is \(4 \times \frac{9}{16} \times \frac{7}{16} \times \frac{12}{15} = 0.7875\), so \(\sigma \approx 0.89\). The last factor, below 1, is why sampling without replacement varies less than a binomial count would.
Worked example: a project team
A 4-person team is chosen at random from a pool of 9 juniors and 7 seniors. What is the probability of exactly 2 juniors? 1. Parameters: \(r = 9\), \(b = 7\), \(n = 4\). 2. Denominator: every possible team, \(\binom{16}{4} = 1820\). 3. Numerator: \(\binom{9}{2} \times \binom{7}{2} = 36 \times 21 = 756\) teams with exactly 2 juniors. 4. \(P(X = 2) = \frac{756}{1820} \approx 0.415\). 5. Check: the mean is \(4 \times \frac{9}{16} = 2.25\), just above 2, and 2 being the nearest whole value makes a probability near 0.4 plausible.
Worked example: the crate of bottles
A crate holds 20 bottles, 3 of them cracked; five are drawn without replacement. a) \(P(\text{no cracked}) = \frac{\binom{3}{0}\binom{17}{5}}{\binom{20}{5}} = \frac{6188}{15504} \approx 0.399\). b) \(P(\text{at least one cracked}) = 1 - 0.399 = 0.601\). c) The binomial approximation with \(p = \frac{3}{20} = 0.15\) would give \(1 - 0.85^5 \approx 0.556\) - an underestimate, because each sound bottle drawn raises the cracked share among those remaining. Check: \(0.399 + 0.601 = 1\), and the true answer exceeds the binomial's, exactly as the dependence predicts.
When the population is large
The binomial and hypergeometric models converge as the population grows. With replacement, or from a population so large that removing \(n\) items barely shifts the proportions, the two distributions agree closely - which is why the binomial topic could treat a big production run as binomial. The hypergeometric earns its keep exactly where that approximation fails: small committees, single crates, one classroom - anywhere each removal visibly changes what remains.
Common mistakes to avoid
First, using the binomial formula on a without-replacement draw from a small population - the dependence is real and the numbers differ, as 0.601 against 0.556 shows. Second, mixing up the parameters: \(r\) is the group of interest, not the whole population, which is \(r + b\). Third, forgetting the second combination factor \(\binom{b}{n-k}\) - the sample's other members must come from somewhere. Fourth, worrying when \(k\) exceeds \(r\) or \(n\): the numerator becomes 0, which is correct, not an error. Fifth, computing the mean as \(n \times r\) instead of \(n \times \frac{r}{r+b}\). Sixth, claiming independence of draws when justifying the model.
Key takeaways
Hypergeometric counts the group-of-interest members in a fixed sample drawn without replacement from a finite two-group population: \(X \sim H(r, b, n)\). The draws are dependent - not Bernoulli trials - and \(P(X = k) = \frac{\binom{r}{k}\binom{b}{n-k}}{\binom{r+b}{n}}\) counts equally likely samples. The mean is the interest group's share times the sample size, \(\mu = \frac{nr}{r+b}\). Small populations are where the model matters: the crate's true 0.601 against the binomial approximation's 0.556. As the population grows, the binomial takes over.
Flashcards
Flashcard
Draws without replacement from a finite two-group population
Test your understanding
Question 1 of 5
Choosing 5 students from one class of 30 without replacement and counting the freshmen is best described as:
Question 2 of 5
A 4-person team is drawn from 9 juniors and 7 seniors. What is P(exactly 2 juniors)?
Question 3 of 5
For that same team draw, what is the expected number of juniors?
Question 4 of 5
From the crate (20 bottles, 3 cracked, 5 drawn), what is the probability of at least one cracked bottle?
Question 5 of 5
The binomial approximation gives 0.556 for the crate's at-least-one probability, but the true hypergeometric answer is 0.601. Why?
