It is Question 17 in Ross's book (Introduction to Probability Models-11th edition)

Let $Y$ be a gamma random variable with parameters $(s, \alpha)$. That is, its density is

$$f_Y(y) = Ce^{-\alpha y}y^{s-1} \quad \forall\,y > 0,$$

where $C$ is a constant that does not depend on $y$. Suppose also that the conditional distribution of $X$ is given that $Y = y$ is Poisson with mean $y$. That is,

$$P(X=i\mid Y=y) = \frac{e^{-y} y ^ i}{i!} \quad \forall\, i \ne 0$$

Show that the conditional distribution of $Y$ given that $X = i$ is the gamma distribution with parameters $(s+i,\alpha+1)$.

Here is my steps:

  1. Find the joint density function

$$f_{X,Y}(i,y) = f_{X|Y}(i,y) f_Y(y) = e^{-y} y^i/i! \cdot Ce^{-\alpha y} y^{s-1} = Ce^{-(\alpha + 1)y} \cdot y^{s+i-1}/i!.$$

  1. Find $f_X(i)$ by $$ \int_0^\infty f_{X,Y}(i,y) \,dy = \int_0^\infty Ce^{-(\alpha + 1)y} \cdot y^{s+i-1}/i! \,dy ?$$

  2. Find $f_{Y|X}(y,i)$ by $f_{X,Y}(i,y) / f_X(i)$.

Then I am stuck at step 2, do not know how to proceed with that integral.

Thanks in advance.


Just continue step (3).

\begin{align} f_{Y \mid X}(y,i) &= \frac{f_{X,Y}(i,y)}{f_X(i)} \\ &= \large\frac{\frac{Ce^{-(\alpha + 1)y} \cdot y^{s+i-1}}{i!}} {\int_0^\infty \frac{Ce^{-(\alpha + 1)y} \cdot y^{s+i-1}}{i!} \,dy} \\ &= \underbrace{\frac{1}{\int_0^\infty e^{-(\alpha + 1)y} \cdot y^{s+i-1} \,dy}}_{\text{constant independent of }y} \cdot e^{-(\alpha + 1)y} \cdot y^{s+i-1} \end{align}

This is a constant multiplied by $e^{-(\alpha + 1)y} \cdot y^{s+i-1}$.

Recall that gamma distribution with parameters $(s+i,\alpha+1)$ has density $C_{(s+i,\alpha+1)} e^{-(\alpha + 1)y} \cdot y^{s+i-1}$ for all $y>0$. By the very definition of density function,

$$\int_0^\infty C_{(s+i,\alpha+1)} e^{-(\alpha + 1)y} \cdot y^{s+i-1} \,dy = 1.$$

Observe that

$$\int_0^\infty \frac{1}{\int_0^\infty e^{-(\alpha + 1)y} \cdot y^{s+i-1} \,dy} \cdot e^{-(\alpha + 1)x} \cdot x^{s+i-1} \,dx = \frac{\int_0^\infty e^{-(\alpha + 1)x} \cdot x^{s+i-1} \,dx}{\int_0^\infty e^{-(\alpha + 1)y} \cdot y^{s+i-1} \,dy}=1.$$

So $C_{(s+i,\alpha+1)} = \dfrac{1}{\int_0^\infty e^{-(\alpha + 1)y} \cdot y^{s+i-1} \,dy}$ and $f_{Y \mid X}$ is the density function of $\Gamma(s+i,\alpha+1)$.

  • $\begingroup$ Thanks. My questions seem stupid. 1. when deriving fY|X(y,i)fY|X(y,i), treat fX(i)fX(i) as a constant, because only i is defined and given? Is this always true? Please notice there is an i in exponent in that integral. 2.when Is the constant 'C' in your answer the same as the constant 'C' in $f_Y(y)$? $\endgroup$ Jan 19 '18 at 3:35
  • $\begingroup$ @evergreenhomeland (1) $f_{Y \mid X}(y,i) = f_Y(y \mid X = i)$. This means you condition on $X = i$. (Recall: $P(A \mid B)$ is read as probability of $A$ given $B$ in words.) (2) I'm sorry for the ambiguity. The constant $C$ in my answer is for $X \mid Y \sim \Gamma(s+i,\alpha + 1)$. I'm going to revise my answer so that it reads differently from $Y \sim \Gamma(s,\alpha)$ in the question. $\endgroup$ Jan 19 '18 at 4:11

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.