28.01.2015 Views

PDF of Lecture Notes - School of Mathematical Sciences

PDF of Lecture Notes - School of Mathematical Sciences

PDF of Lecture Notes - School of Mathematical Sciences

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.

2. STATISTICAL INFERENCE<br />

Examples<br />

(1) x 1 , x 2 , . . . , x n i.i.d. N(µ, σ 2 ), σ 2 known.<br />

n∏ 1<br />

Then f(x; µ) = √ e (−1/(2σ2 ))(x i −µ) 2<br />

2πσ<br />

=⇒ S =<br />

i=1<br />

{<br />

}<br />

= (2πσ 2 ) −n/2 exp − 1 n∑<br />

(x<br />

2σ 2 i − µ) 2<br />

i=1<br />

{ ( n∑<br />

)}<br />

= (2πσ 2 ) −n/2 exp − 1<br />

n∑<br />

x 2<br />

2σ 2 i − 2µ x i + nµ 2<br />

i=1<br />

i=1<br />

{ (<br />

)} (<br />

= (2πσ 2 ) −n/2 1<br />

n∑<br />

exp 2µ x<br />

2σ 2 i − nµ 2 exp − 1<br />

2σ 2<br />

= exp<br />

{<br />

(<br />

1<br />

2µ<br />

2σ 2<br />

i=1<br />

)} {<br />

n∑<br />

x i − nµ 2 exp<br />

i=1<br />

n∑<br />

x i is sufficient for µ.<br />

i=1<br />

(2) If x 1 , x 2 , . . . , x n are i.i.d. with<br />

f(x) = exp{A(θ)t(x) + B(θ) + h(x)},<br />

then f(x; θ) = exp<br />

=⇒ S =<br />

{<br />

A(θ)<br />

n∑<br />

t(x i )<br />

i=1<br />

− 1<br />

2σ 2<br />

n∑<br />

i=1<br />

x 2 i<br />

)<br />

}<br />

n∑<br />

x 2 i − n 2 log(2πσ2 )<br />

i=1<br />

} {<br />

n∑<br />

n∑<br />

}<br />

t(x i ) + nB(θ) exp h(x i )<br />

i=1<br />

i=1<br />

is sufficient for θ in the exponential family by the Factorization Theorem.<br />

Theorem. 2.2.5 Rao-Blackwell Theorem<br />

If T is an unbiased estimator for θ and S is a sufficient statistic for θ, then the quantity<br />

T ∗ = E(T |S) is also an unbiased estimator for θ with Var(T ∗ ) ≤ Var(T ). Moreover,<br />

Var(T ∗ ) = Var(T ) iff T ∗ = T with probability 1.<br />

Pro<strong>of</strong>.<br />

(I) Unbiasedness:<br />

θ = E(T ) = E S {E(T |S)} = E(T ∗ )<br />

=⇒ E(T ∗ ) = θ.<br />

92

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

Saved successfully!

Ooh no, something went wrong!