29.07.2014 Views

Asymptotic Methods in Statistical Inference - Statistics Centre

Asymptotic Methods in Statistical Inference - Statistics Centre

Asymptotic Methods in Statistical Inference - Statistics Centre

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

135<br />

• Rao-Blackwell Theorem: We can always reduce<br />

the variance of an unbiased estimate by condition<strong>in</strong>g<br />

on X () .<br />

Proof: Let = ( 1 ) be any unbiased<br />

estimator of and def<strong>in</strong>e<br />

˜ = h |X ()<br />

i<br />

<br />

Then (Double Expectation Theorem)<br />

[ ˜] = X()<br />

½ |X()<br />

h<br />

|X()<br />

i ¾ = [] =<br />

Furthermore the decomposition<br />

[] = h ³ h ³<br />

|X ()´i<br />

+ |X()´i<br />

shows that<br />

[] ≥ h ³ |X ()´i<br />

= <br />

h<br />

˜ i <br />

This <strong>in</strong>equality is strict unless<br />

0 = h ³ |X ()´i<br />

½ ∙ ³ h i´2<br />

= X() |X() − |X() |X()¸¾<br />

= <br />

½<br />

<br />

∙ ³<br />

− ˜´2<br />

|X()¸¾<br />

which holds iff ³ = ˜´ =1.<br />

∙ ³<br />

= − ˜´2¸

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!