In probability theory, a beta negative binomial distribution is the probability distribution of a discrete random variable
X
r
p
This distribution has also been called both the inverse Markov-Pólya distribution and the generalized Waring distribution[1] or simply abbreviated as the BNB distribution. A shifted form of the distribution has been called the beta-Pascal distribution.[1]
If parameters of the beta distribution are
\alpha
\beta
X\midp\simNB(r,p),
p\simrm{B}(\alpha,\beta),
X
X\simBNB(r,\alpha,\beta).
In the above,
NB(r,p)
rm{B}(\alpha,\beta)
Denoting
fX|p(k|q),fp(q|\alpha,\beta)
f(k|\alpha,\beta,r)
f(k|\alpha,\beta,r)=
1 | |
\int | |
0 |
fX|p(k|r,q) ⋅ fp(q|\alpha,\beta)dq=
1 | |
\int | |
0 |
\binom{k+r-1}{k}(1-q)kqr ⋅
q\alpha-1(1-q)\beta-1 | |
\Beta(\alpha,\beta) |
dq=
1 | |
\Beta(\alpha,\beta) |
\binom{k+r-1}{k}
1 | |
\int | |
0 |
q\alpha+r-1(1-q)\beta+k-1dq
Noting that the integral evaluates to:
1 | |
\int | |
0 |
q\alpha+r-1(1-q)\beta+k-1dq=
\Gamma(\alpha+r)\Gamma(\beta+k) | |
\Gamma(\alpha+\beta+k+r) |
If
r
f(k|\alpha,\beta,r)=\binom{r+k-1}k | \Beta(\alpha+r,\beta+k) |
\Beta(\alpha,\beta) |
f(k|\alpha,\beta,r)= | \Gamma(r+k) |
k! \Gamma(r) |
\Beta(\alpha+r,\beta+k) | |
\Beta(\alpha,\beta) |
f(k|\alpha,\beta,r)= | \Beta(r+k,\alpha+\beta) |
\Beta(r,\alpha) |
\Gamma(k+\beta) | |
k! \Gamma(\beta) |
Using the properties of the Beta function, the PMF with integer
r
f(k|\alpha,\beta,r)=\binom{r+k-1}k | \Gamma(\alpha+r)\Gamma(\beta+k)\Gamma(\alpha+\beta) |
\Gamma(\alpha+r+\beta+k)\Gamma(\alpha)\Gamma(\beta) |
More generally, the PMF can be written as
f(k|\alpha,\beta,r)= | \Gamma(r+k) |
k! \Gamma(r) |
\Gamma(\alpha+r)\Gamma(\beta+k)\Gamma(\alpha+\beta) | |
\Gamma(\alpha+r+\beta+k)\Gamma(\alpha)\Gamma(\beta) |
The PMF is often also presented in terms of the Pochammer symbol for integer
r
f(k|\alpha,\beta,r)= | r(k)\alpha(r)\beta(k) |
k!(\alpha+\beta)(r+k) |
The -th factorial moment of a beta negative binomial random variable is defined for
k<\alpha
\operatorname{E}l[(X)kr]=
\Gamma(r+k) | |
\Gamma(r) |
\Gamma(\beta+k) | |
\Gamma(\beta) |
\Gamma(\alpha-k) | |
\Gamma(\alpha) |
.
The beta negative binomial is non-identifiable which can be seen easily by simply swapping
r
\beta
r
\beta
The beta negative binomial distribution contains the beta geometric distribution as a special case when either
r=1
\beta=1
\alpha
\alpha
\beta
r
By Stirling's approximation to the beta function, it can be easily shown that for large
k
f(k|\alpha,\beta,r)\sim
\Gamma(\alpha+r) | |
\Gamma(r)\Beta(\alpha,\beta) |
kr-1 | |
(\beta+k)r+\alpha |
\alpha
The beta geometric distribution is an important special case of the beta negative binomial distribution occurring for
r=1
f(k|\alpha,\beta)= | B(\alpha+1,\beta+k) |
B(\alpha,\beta) |
This distribution is used in some Buy Till you Die (BTYD) models.
Further, when
\beta=1
X\simBG(\alpha,1)
X+1\simYS(\alpha)
In the case when the 3 parameters
r,\alpha
\beta
\alpha
\beta
r
X
BNB(r,\alpha,\beta)
r+\alpha
X+\beta
By the non-identifiability property,
X
\alpha
r
\beta