Negative multinomial distribution

Notation NM(x0,𝐩)
Parameters x0>0 — the number of failures before the experiment is stopped,
𝐩 ∈ Rm — m-vector of "success" probabilities,

p0 = 1 − (p1+…+pm) — the probability of a "failure".
Support xi∈{0,1,2,…},1≤i≤m
PMF Γ(∑i=0mxi)p0x0Γ(x0)∏i=1mpixixi!,
where Γ(x) is the Gamma function.
Mean x0p0𝐩
Variance x0p02𝐩𝐩′+x0p0diag⁡(𝐩)
MGF (p01−∑j=1mpjetj)x0
CF (p01−∑j=1mpjeitj)x0

In probability theory and statistics, the negative multinomial distribution is a generalization of the negative binomial distribution (NB(x0, p)) to more than two outcomes.

As with the univariate negative binomial distribution, if the parameter x0 is a positive integer, the negative multinomial distribution has an urn model interpretation. Suppose we have an experiment that generates m+1≥2 possible outcomes, {X0,...,Xm}, each occurring with non-negative probabilities {p0,...,pm} respectively. If sampling proceeded until n observations were made, then {X0,...,Xm} would have been multinomially distributed. However, if the experiment is stopped once X0 reaches the predetermined value x0 (assuming x0 is a positive integer), then the distribution of the m-tuple {X1,...,Xm} is negative multinomial. These variables are not multinomially distributed because their sum X1+...+Xm is not fixed, being a draw from a negative binomial distribution.

Properties

Marginal distributions

If m-dimensional x is partitioned as follows 𝐗=[𝐗(1)𝐗(2)] with sizes [n×1(m−n)×1] and accordingly 𝒑 𝒑=[𝒑(1)𝒑(2)] with sizes [n×1(m−n)×1] and let q=1−∑ipi(2)=p0+∑ipi(1)

The marginal distribution of 𝑿(1) is NM(x0,p0/q,𝒑(1)/q). That is the marginal distribution is also negative multinomial with the 𝒑(2) removed and the remaining p's properly scaled so as to add to one.

The univariate marginal m=1 is said to have a negative binomial distribution.

Conditional distributions

The conditional distribution of 𝐗(1) given 𝐗(2)=𝐱(2) is NM(x0+∑xi(2),𝐩(1)). That is, Pr⁡(𝐱(1)∣𝐱(2),x0,𝐩)=Γ(∑i=0mxi)(1−∑i=1npi(1))x0+∑i=1m−nxi(2)Γ(x0+∑i=1m−nxi(2))∏i=1n(pi(1))xi(xi(1))!.

Independent sums

If 𝐗1∼NM(r1,𝐩) and If 𝐗2∼NM(r2,𝐩) are independent, then 𝐗1+𝐗2∼NM(r1+r2,𝐩). Similarly and conversely, it is easy to see from the characteristic function that the negative multinomial is infinitely divisible.

Aggregation

If 𝐗=(X1,…,Xm)∼NM⁡(x0,(p1,…,pm)) then, if the random variables with subscripts i and j are dropped from the vector and replaced by their sum, 𝐗′=(X1,…,Xi+Xj,…,Xm)∼NM⁡(x0,(p1,…,pi+pj,…,pm)).

This aggregation property may be used to derive the marginal distribution of Xi mentioned above.

Correlation matrix

The entries of the correlation matrix are ρ(Xi,Xi)=1. ρ(Xi,Xj)=cov⁡(Xi,Xj)var⁡(Xi)var⁡(Xj)=pipj(p0+pi)(p0+pj).

Parameter estimation

Method of Moments

If we let the mean vector of the negative multinomial be 𝝁=x0p0𝐩 and covariance matrix 𝜮=x0p02𝐩𝐩′+x0p0diag⁡(𝐩), then it is easy to show through properties of determinants that |𝜮|=1p0∏i=1mμi. From this, it can be shown that x0=∑μi∏μi|𝜮|−∏μi and 𝐩=|𝜮|−∏μi|𝜮|∑μi𝝁.

Substituting sample moments yields the method of moments estimates x̂0=(∑i=1mxi¯)∏i=1mxi¯|𝐒|−∏i=1mxi¯ and 𝐩̂=(|𝑺|−∏i=1mx¯i|𝑺|∑i=1mx¯i)𝒙¯