Suppose I have data $y$ (N observations) which follows a normal distribution:

$y \sim N(\alpha+\beta*\mu,\sigma^2)$

while $\alpha$ and $\beta$ are known parameters. I want to update $\mu$ and the prior is $\mu \sim N(\mu_0,\sigma^2_0)$.

My question is what is the posterior distribution for $\mu$?

I know the posterior when $X \sim N(\mu,\sigma^2)$ (here). But in my case, $y$ is a shifted and scaled version of $X$.

For my own derivation, I treat the above as a regression problem, and I get

$\mu \sim N(\frac{\beta^T(y-\alpha)+\mu_0/\sigma^2_0}{\beta^T\beta+1/\sigma^2_0},\frac{\sigma^2}{\beta^t\beta+1/\sigma^2})$

However, the above formula does seem correct. For instance, suppose I have two data, $y_1$ and $y_2$. Updating with $y_1$ then $y_2$ vs updating with $y_2$ then $y_1$ will result in different posteriors.

Any help is appreciated!


1 Answer 1


You can do a standard update for distribution parametrized by $\mu’ = \alpha + \beta\mu$. Next, given that it's a location-scale transformation you can just transform it. But your likelihood is characterized by $\mu$ not $\mu’$, so you cannot use the distribution for $\mu$ in the next update, but the one for $\mu’$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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