I have a couple of questions when reading the chapter 7 The Multivariate Normal Model of Peter Hoff's "A First Course in Bayesian Statistical Methods".

First, could anyone give me any resource about "$z^i$ is a gamma random variable given z is a mean-zero univariate normal random variable"? (first paragraph on page 110).

My second question is, on the same page 110, it samples a covariance matrix $\Sigma$ from an inverse-Wishart distribution in the three steps below:

  1. sample $\textbf{z}_1,\dotsc,\textbf{z}_{v0}\sim i.i.d. \text{multi normal}(0,\textbf{S}_0^{-1})$ (According to text, $\textbf{z}_{v0}$ is a $p \times1$ matrix, correct me if I am wrong);
  2. calculate $Z^TZ=\sum_{i=1}^{v0}z_i z_i^T$ (according to text, this should be sum of $v_0$ $z_i z_i^T$, this result will be a $p\times p$ matrix);
  3. set $\Sigma=(Z^TZ)^{-1}$.

Then in the next paragraph, the book points out that the expectation is $\Sigma^{-1}$ and $\Sigma$ are:



Here is what I am confused, the books continues comments: choosing $v_0=p+2$ and $\textbf{S}_0=\Sigma_0$ makes $\Sigma$ only loosely centered around $\Sigma_0$. In this way, I think the result of $E[\Sigma]$ will be the same as making $\textbf{S}_0=(v_0-p-1)\Sigma_0$. However, the book comments that the distribution $\Sigma$ concentrated around $\Sigma_0$ by making $\textbf{S}_0=(v_0-p-1)\Sigma_0$. I feel the comments contradicts with each other.

Thank you so much for helping me solve the problems.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.