I've got a simple question buy I'm not sure how to solve it. It's a bit long.

Suppose you've got $n$ iid random variables $X_1$, $\dots$, $X_n$ from the normal distribution with unknown mean $M$ and unknown precision (inverse variance) $H$. Then we've got the likelihood function for data $X_1=x_1$, $\dots$, $X_n=x_n$ $$L_n(\mu,h)\propto h^{n/2}\exp\left(-\frac{1}{2}h\left(n(\bar{x}-\mu)^2+S\right)\right), $$ where $\bar{x}$ is the mean of $x_1$, $\dots$, $x_n$ and $S=\sum_i(x_i-\bar{x})^2$.

Now, a bivariate prior distribution for $(M,H)$ is specified, in terms of hyperparameters $(\alpha_0,\beta_0,m_0,\lambda_0)$, as follows. The marginal distribution of $H$ is $\Gamma(\alpha_0,\beta_0)$ with density $$\pi(h)\propto h^{\alpha_0-1}e^{-\beta_0h}$$ for $h>0$, and the conditional distribution of $M$ given $H=h$ is normal with mean $m_0$ and precision $\lambda_0h$.

Now I should find the posterior joint distribution of $(M,h)$ given data $X_1=x_1$, $\dots$, $X_n=x_n$ and give the updated hyperparameters $(\alpha_n,\beta_n,m_n,\lambda_n)$ in terms of the prior hyperparameters and the data.

Could somebody please tell me how to do this? Thank you very much.


Your prior is just a Normal-gamma distribution. The likelihood is a standard normal likelihood, but obviously taken as the product of the $n$ conditionally independent observations (with the terms in the exponent rewritten by completing the square). The prior is conjugate, so the posterior distribution is also a Normal-gamma distribution. I would advise you to work out & slog through the proportional form & posterior parameters through $$ \text{Posterior} \propto \text{Prior} \times \text{Likelihood}, $$

but if you want to verify your answer you can here (under Normal likelihood with unkown but exchangeable mean and precision).


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.