1
$\begingroup$

The Inverse-Wishart wikipedia article states that $ X \sim \operatorname{Wish}^{-1}(\Psi,\nu)$ if $X^{-1} \sim \operatorname{Wish}(\Psi^{-1},\nu),$ where $\Psi$ and $X$ are $p\times p$.

The mean of the Inverse-Wishart is given as $\frac{\Psi}{\nu-p-1}$, but the mean of the Wishart is $\nu(\Psi^{-1})$ (See here). Why don't these match up? Wouldn't the mean of the Inverse-Wishart be $\frac{\Psi}{\nu}$?

Is this just because of a difference in parameterizations?

$\endgroup$
2
$\begingroup$

Let's look at the simpler case of positive univariate random variables. I assume it's already clear that in general for a positive random variable whose variance is not $0$, that $\text{Cov}(X,1/X)< 0$. If you're happy to accept that as the case, then

$\text{Cov}(X,1/X) = E(X.1/X) - E(X)E(1/X) = 1-E(X)E(1/X) <0\,,$ so

$E(X) E(1/X) > 1\,,$ or

$E(1/X)>1/E(X)$

or you could just use Jensen's inequality to the same end.

(In fact we can show $1/E(X)<E(1/X)\leq 2/E(X)$, a rather neat little fact)

Generally we should not expect means to "match up" in the way you anticipated.

$\endgroup$

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.