Multinomial Distribution
The Multinomial Distribution: Counting Outcomes Across Several Categories
The multinomial distribution extends the binomial idea from two outcomes to k outcomes. In n independent trials, each trial lands in exactly one of k categories with fixed probabilities p1,p2,…,pk (where p1+⋯+pk=1). If X1,…,Xk count how many trials fall into each category (so X1+⋯+Xk=n), the joint probability function is:
Classification probabilities for each of n = 12 bottles inspected on a production line.
Marginal Distribution of the Number of Passing Bottles (n = 12, p = 0.85)
Marginal and conditional structure
- Each individual Xi on its own is binomial: Xi∼Binomial(n,pi), because we can lump every other category into "not category i".
- Given the total in two categories, say XA+XB=m, the conditional distribution of XA is Binomial(m,pA+pBpA), the other categories drop out because we already know the trials landed in A or B.
Worked example
A symmetric four-sided die is rolled n=5 times, so each face has p=1/4. Find P(2,1,1,1), i.e. face 1 appears twice and faces 2, 3, 4 appear once each.
Which of the following experiments can be correctly modelled with a multinomial distribution?
Using the graph of the marginal distribution of the number of passing bottles (with n=12, p=0.85), what is the probability that at least 10 of the 12 bottles pass?
Classification probabilities for each of n = 12 bottles inspected on a production line.
Marginal Distribution of the Number of Passing Bottles (n = 12, p = 0.85)