1
$\begingroup$

I am having a hard time trying to solve this problem. I don't know how to start it. Any help would be greatly appreciated.

Let T be any unbiased estimator of $\tau(\theta),$ and let $W$ be a sufficient statistic for theta. Define $\phi(W)=E[T|W]$. Show that $\phi(W)$ is an unbiased estimator of $\tau(\theta),$ and $\mathrm{Var}(T)=\mathrm{Var}\left(\phi(W)\right) + E\left[\mathrm{Var}(T|W)\right]$.

The only thing I know is that this may be a part of the Rao-Blackwell Thm.

$\endgroup$
3
$\begingroup$

We want to show that $E(\phi(W))=\tau(\theta)$.
Start with the definition:
$$ \begin{align*} E(\phi(W))&=E[E(T|W)]\\ &=E[T]\\ &=\tau(\theta) \end{align*} $$

where the last equality holds, because $T$ is an unbiased estimator for $\tau(\theta).$ Hence $\phi(W)$ is unbiased for $\tau(\theta).$

For the second part, use the fact that for any two random variables, $X$ and $Y$,
$$\mathrm{Var}(X)=\mathrm{Var}\left[E(X|Y)\right]+E\left[\mathrm{Var}(X|Y)\right].$$


Added:
As noted by @guy, for the first part, we also need to show that $\phi(W)$ is an estimator of $\theta$. But this is obvious. See comments below.

$\endgroup$
  • $\begingroup$ One thing to add: you have shown $\phi(W)$ is unbiased, but not that $\phi(W)$ is an estimator, i.e. we need $\phi(W)$ not to depend on $\theta$. It is trivial, of course, but this is where sufficiency comes in, and it is easy miss. $\endgroup$ – guy Jul 9 '11 at 22:08
  • $\begingroup$ @guy: I thought it was obvious since $W$ is a sufficient statistic for $\theta$, $\phi(\theta)$ is a function only of the sample and does not depend on $\theta$. Thanks for pointing it out though. I'll add to my answer. $\endgroup$ – Nana Jul 9 '11 at 22:18

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.