I have this problem

Given a loss function

\begin{align}l(a,\theta) = |a-\theta|\end{align} for a ∈ A = {4,5,6} and $\theta$$\theta$ = {4,5,6}

Assume the prior distribution is

\begin{align}Pr(\theta=i)=\pi_i=\begin{cases} 0.5, & \text{for i=4}.\\ 0.3, & \text{for i=5}.\\ 0.2, & \text{for i=6}.\\ \end{cases}\end{align}

How would I find the posterior distribution of $\theta$ given a sample of 3 observations on a RV X: $X_1=x_1=4,X_2=x_2=0,X_3=x_3=3$. Note that $X_1,X_2,X_3$ are IID Poisson so

\begin{align} f(x_i|\theta=\frac{e^{-\theta}\theta^{x_i}}{x_i!}) for x_i=0,1,2,.. \end{align}

How would I find the posterior distribution of θ?

Thoughts so far: I have done other examples where the prior is a continuous distribution and applied it in the formula $f(\theta|x)=\frac{\pi(\theta)f(x|\theta)}{f(x)}$, but I am not sure regarding a discrete prior and what to do with the 3 sample observations On a side note, would the prior distribution be the natural conjugate for $\theta$?


If the prior distribution $\pi$ of $\theta$ and the conditional distributions $p(\ \mid\theta)$ of the observations are discrete, the posterior distribution $q(\ \mid x)$ of $\theta$ given some observations $x=(x_i)$ is $$ q(\theta\mid x)\propto p(x\mid\theta)\pi(\theta)=\pi(\theta)\prod_ip(x_i\mid\theta), $$ that is $$ q(\theta\mid x)=\frac{\pi(\theta)}{z(x)}\prod_ip(x_i\mid\theta),\quad z(x)=\sum_\alpha\pi(\alpha)\prod_ip(x_i\mid\alpha). $$ In your case, the sum over $\alpha$ has three terms for $\alpha$ in $\{4,5,6\}$ and the product over $i$ has three terms for $x_i$ in $\{0,4,3\}$.

Can you carry on from here? (Note that the loss function is irrelevant.)

  • $\begingroup$ @ Did: What is the "loss function"? Why are the left side of the bars in $(|\theta)$ and $q(|x)$ left out? $\endgroup$ – Parseval Dec 3 '18 at 13:01
  • $\begingroup$ @Parseval "Loss function" is referring to a previous version of the question, which I just reinstated. // The notation $q(\ \mid x)$ stands for the function $\theta\mapsto q(\theta\mid x)$, as opposed to a value $q(\theta\mid x)$ of this function at a specific argument $\theta$. $\endgroup$ – Did Dec 3 '18 at 14:03

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.