0
$\begingroup$

Suppose $X_{i}|\theta_{i} \sim $ $f_{1}(\theta_{i},t_{i})$ , $\theta_{i}|\alpha,\beta \sim$ $f_{2}(\alpha,\beta)$. And again $\alpha \sim f_{3}$ and $\beta \sim f_{4}$. Now we are given a bunch of $X_{i}$'s and their corresponding $t_{i}$'s, let $n$ be the number of samples given. Using hierarchical bayesian model we can write down the joint posterior distribution of $(\theta, \alpha, \beta)$ given $X$. Now when I try to find the marginal posterior distribution of $(\alpha, \beta)$ given $X$, we will have to integrate the joint posterior distribution with respect to $\theta_{i}$'s here $i=1,2,....,n$. I am really confused what the limits of integration would be ?

$\endgroup$
0
$\begingroup$

You have that:

\begin{equation} \begin{split} P[\{\theta_i\}_{i=1}^n,\alpha,\beta|X] & = \frac{\prod_{i=1}^n(P[X_i|\theta_i]P[\theta_i|\alpha,\beta])P[\alpha]P[\beta]}{P[X]}\\ \end{split} \end{equation} Then integrating out $\{\theta_i\}_{i=1}^n$, we get: \begin{equation} \begin{split} P[\alpha,\beta|X] & = \int \dots \int P[\{\theta_i\}_{i=1}^n,\alpha,\beta|X] d\theta_1 \dots d\theta_n\\ & = \frac{ \int \dots \int\prod_{i=1}^n(P[X_i|\theta_i]P[\theta_i|\alpha,\beta])P[\alpha]P[\beta]d\theta_1 \dots d\theta_n}{P[X]}\\ & = \frac{ \prod_{i=1}^n(\int P[X_i|\theta_i]P[\theta_i|\alpha,\beta] d\theta_i)P[\alpha]P[\beta]}{P[X]}\\ \end{split} \end{equation}

The limits of integration are whatever the support of the posterior of each $\theta_i$ is.

$\endgroup$

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.