2
$\begingroup$

Suppose we have random variable $Y$ has the Poisson distribution with parameter $\theta$, and $\theta$ has a Gamma prior distribution, i.e. $$\begin{aligned} & \text{Data: }\hspace{1cm} y|\theta \sim \text{Poisson}(\theta)\\ & \text{Prior: } \hspace{1cm} \theta \sim \text{Gamma}(\alpha,\beta) \end{aligned}$$ We want to show the marginal distribution of a random sample ${\bf y}=y_1,\cdots,y_n$ follows the Negative Binomial distribution.

My attempt: $$\begin{aligned} f({\bf y})& =\int f({\bf y},\theta)d \theta\\ & =\int f({\bf y}|\theta)f(\theta)d\theta\\ & =\int \left[\prod^n_{i=1}f(y_i|\theta)\right]f(\theta)d\theta\\ & \propto \int \left[\prod^n_{i=1}e^{-\theta}\theta^{y_i} \right] \theta^{\alpha-1}e^{-\beta\theta} d\theta\\ & =\int\theta^{\sum^n_{i=1}y_i+\alpha-1}e^{-(n+\beta)\theta} d\theta\\ & =\frac{\Gamma(\sum^n_{i=1}y_i+\alpha)}{(n+\beta)^{\sum^n_{i=1}y_i+\alpha}} \end{aligned}$$ But I did not see why it is a Negative Binomial distribution, since if $$y \sim NB(r,p)$$ then the pmf is $$f(Y=y)=\begin{pmatrix} y+r-1 \\ y \end{pmatrix}p^y(1-p)^r$$

$\endgroup$
2
$\begingroup$

Let $Y\sim \mathcal{P}(\lambda) \underset{\lambda}{\wedge} \Gamma(a,b)$ and $f$ denote the pdf of the gamma distribution with parameters $a$ and $b$. \begin{align*} P(Y=y) & = \int_0^{+\infty} e^{-\lambda} \frac{\lambda^y}{y!} f(\lambda) d\lambda \\ & = \frac{b^a}{\Gamma(a)y!} \underbrace{\int_0^{+\infty} \lambda^{y+a-1} e^{-(1+b)\lambda} d\lambda }_{\displaystyle =\frac{\Gamma(y+a)}{(1+b)^{y+a}}} \\ & = \frac{\Gamma(y+a)}{\Gamma(a)y!} \left(\frac{1}{1+b}\right)^y \left(\frac{b}{1+b}\right)^a \\ P(Y=y) & = \binom{y+a-1}{y} p^y (1-p)^a \end{align*} where $p:=\frac{1}{1+b}$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.