Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I derived a posterior distribution, which looks like this:

$$p(\alpha, \beta | y,t) \propto \frac{exp( - \sum_i (\alpha + \beta t_i))\prod_i (\alpha + \beta t_i)^{y_i}}{\prod_i y_i!}$$

How can I show this is a proper distribution of $\alpha$ and $\beta$?

What distribution would it be to make it integrate to 1?

share|improve this question
    
Since it's non-negative, integrate it out and show that the result is 1? –  gt6989b Jan 31 '13 at 18:24
    
My question is how do you integrate it to 1 –  user1769197 Jan 31 '13 at 18:25
    
Defined $I = \int p$, then $f = \frac{p}{I}$ is a density function of some distribution. The question is however whether $I$ is finite –  Ilya Jan 31 '13 at 18:26
    
I m looking for an answer which can show me computation steps. If you have any idea, please elaborate a little bit more. I spent the whole afternoon to do this but failed –  user1769197 Jan 31 '13 at 18:37
    
Got something from the answer below? –  Did Feb 9 '13 at 10:36

1 Answer 1

Assume that every $t_i$ is positive and every $y_i$ nonnegative, and consider the function $$ q(a,b)=\exp\left(-\sum_{i=1}^n(a+bt_i)\right)\cdot\prod_{i=1}^n(a+bt_i)^{y_i}. $$ The question seems to be to show that the integral of $q$ on $(0,+\infty)\times(0,+\infty)$ is finite.

To do so, call $t$ the maximal $t_i$ and note that $a+bt_i\leqslant a+bt$ and $\sum\limits_{i=1}^n(a+bt_i)\geqslant a+bt$ for every $(a,b)$ hence $q(a,b)\leqslant r(a+bt)$ with $$ r(x)=\mathrm e^{-x}x^y,\qquad y=\sum_{i=1}^ny_i. $$ The change of variable $x=a+bt$, $z=a$, yields $$ \iint r(a+bt)\mathrm da\mathrm db=\int_0^{+\infty}r(x)\int_0^x\frac{\mathrm dz}{t}\mathrm dx=\frac1t\int_0^{+\infty}xr(x)\mathrm dx=\frac{(y+1)!}t. $$ Thus there exists some constant $c$ such that $cq$ is a PDF. Note that $c\geqslant t/(y+1)!$ and that, when $n=1$, $c=t/(y+1)!$.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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