Calculating conditional mean of 2 Normal

37 Views Asked by At

If $\theta$ is $N(\bar{\theta},\sigma^2_\theta)$, and $s=\theta+\epsilon$, where $\epsilon$ is $N(0, \sigma^2_\epsilon)$, how can I derive that $E(\theta|s)=\frac{\frac{1}{\sigma^2_\theta}\bar{\theta}+\frac{1}{\sigma^2_\epsilon}s}{\frac{1}{\sigma^2_\theta}+\frac{1}{\sigma^2_\epsilon}}$?

It wasn't explicitly stated but I think $\theta$ and $\epsilon$ are independent.

If you could suggest how I can go about deriving the conditional distribution of $\theta|s$ (e.g. via their individual pdfs?) rather than just where the conditional mean above came from, that would be greatly appreciated!

1

There are 1 best solutions below

0
On

The derivation follows from the standard definition of Bayes' rule with regards to obtaining a posterior from a prior on $\theta$, $\mu(\theta)$ and a distribution $f(s|\theta)$ - e.g. here $s=\theta+\epsilon$ so the distribution $s|\theta$ is $N(\theta, \sigma_{\epsilon}^2)$. The prior and the posterior are called conjugate distributions here because, given properties of the Normal distribution, the posterior distribution is in the same family as the prior distribution.