STATISTICAL INFERENCE PART III EXPONENTIAL FAMILY, FISHER INFORMATION, CRAMER-RAO LOWER BOUND (CRLB) 1 EXPONENTIAL CLASS OF PDFS • X is a continuous (discrete) rv with pdf f(x;), . If the pdf can be written in the following form k f ( x; ) h ( x )c( ) exp( w j ( ) t j ( x )) j1 then, the pdf is a member of exponential class of pdfs of the continuous (discrete) type. (Here, k is the number of parameters) 2 REGULAR CASE OF THE EXPONENTIAL CLASS OF PDFS • We have a regular case of the exponential class of pdfs of the continuous type if a) Range of X does not depend on . b) c() ≥ 0, w1(),…, wk() are real valued functions of for . c) h(x) ≥ 0, t1(x),…, tk(x) are real valued functions of x. If the range of X depends on , then it is called irregular exponential class or range-dependent exponential class. 3 EXAMPLES X~Bin(n,p), where n is known. Is this pdf a member of exponential class of pdfs? Why? n x f ( x; p) p (1 p) n x ; x 0,1,...,n; 0 p 1 x n p (1 p) n exp(x ln( )) 1 p x n h ( x ) x t1 ( x ) x for x 0,...,n; c(p) (1 p) n for x 0,...,n; p w1 (p) ln( ) 1 p for 0 p 1 for 0 p 1 Binomial family is a member of exponential family of distributions. 4 EXAMPLES X~Cauchy(1,). Is this pdf a member of exponential class of pdfs? Why? f ( x; ) ( (1 [ x ]2 ))1 h(x) 1 1 exp{ ln(1 x 2 2x 2 )} ; c( ) 1; ln(1 x 2 2x 2 ) t1 ( x ) w1 ( ) Cauchy is not a member of exponential family. 5 EXPONENTIAL CLASS and CSS • Random Sample from Regular Exponential Class Y n t j (X i ) is a css for . i 1 If Y is an UE of , Y is the UMVUE of . 6 EXAMPLES Let X1,X2,…~Bin(1,p), i.e., Ber(p). This family is a member of exponential family of distributions. t1 ( x ) x n for x 0,...,n n t1( x i ) x i i 1 is a CSS for p. i 1 X X is UE of p and a function of CSS. is UMVUE of p. 7 EXAMPLES X~N(,2) where both and 2 is unknown. Find a css for and 2 . 8 FISHER INFORMATION AND INFORMATION CRITERIA • X, f(x;), , xA (not depend on ). Definitions and notations: x; ln f x; ln f x; x; 2 ln f x; x; 2 f x; f x; 2 f x; f x; 2 9 FISHER INFORMATION AND INFORMATION CRITERIA The Fisher Information in a random variable X: I E x; 2 V x; E x; 0 The Fisher Information in the random sample: I n nI Let’s prove the equalities above. 10 FISHER INFORMATION AND INFORMATION CRITERIA d f x; dx 1 d f x; dx 0 A A f x; dx 0 A f x; dx 0 A ln f x; f x; x; f x; f x; 2 x; x; f x; 11 FISHER INFORMATION AND INFORMATION CRITERIA E X ; x; f x; dx 0 A E X ; x; f x; dx A f x; 2 x; f x; dx A f x; 2 f x; dx x; f x; dx A A 2 0 E x; 12 FISHER INFORMATION AND INFORMATION CRITERIA 2 E x; E x; V x; The Fisher Information in a random variable X: 2 I E x; V x; E x; 0 The Fisher Information in the random sample: I n nI Proof of the last equality is available on Casella & Berger (1990), pg. 310-311. 13 CRAMER-RAO LOWER BOUND (CRLB) • • • • Let X1,X2,…,Xn be sample random variables. Range of X does not depend on . Y=U(X1,X2,…,Xn): a statistic; does’nt contain . Let E(Y)=m(). 2 m V Y The Cramer - Rao Lower Bound I n • Let prove this! 14 CRAMER-RAO LOWER BOUND (CRLB) CovY , Z 1 • -1Corr(Y,Z)1 1 V Y V Z 2 Cov Y , Z • 0 Corr(Y,Z)21 0 1 V Y V Z CovY , Z 0 V Z 2 V (Y ) • Take Z=′(x1,x2,…,xn;) • Then, E(Z)=0 and V(Z)=In() (from previous slides). 15 CRAMER-RAO LOWER BOUND (CRLB) • Cov(Y,Z)=E(YZ)-E(Y)E(Z)=E(YZ) EY.Z u x1, x 2 ,, x n x1, x 2 ,, x n ; f ( x1, x n ; )dx1dx 2 dx n f x1,, xn ; u x1,, xn f x1,, xn ; dx1 dxn f x1,, xn ; u x1,, xn f x1,, xn ; dx1dxn m 16 CRAMER-RAO LOWER BOUND (CRLB) • E(Y.Z)=mʹ(), Cov(Y,Z)=mʹ(), V(Z)=In() CovY , Z 0 V Z 2 V (Y ) The Cramer-Rao Inequality (Information Inequality) m 2 V Y The Cramer - Rao Lower Bound I n 17 CRAMER-RAO LOWER BOUND (CRLB) • CRLB is the lower bound for the variance of an unbiased estimator of m(). • When V(Y)=CRLB, Y is the MVUE of m(). • For a r.s., remember that In()=n I(), so, m 2 V Y The Cramer - Rao Lower Bound nI 18 LIMITING DISTRIBUTION OF MLEs • ˆ : MLE of • X1,X2,…,Xn is a random sample. 2 asymptotically m m ˆ ~ N m , RCLB m nI 1 ˆ ~ N , large n nI asympt . ˆ 1 nI d ˆ nI N 0,1 19 EFFICIENT ESTIMATOR • T is an efficient estimator (EE) of if – T is UE of , and, – Var(T)=CRLB • Y is an efficient estimator (EE) of its expectation, m(), if its variance reaches the CRLB. • An EE of m() may not exist. • The EE of m(), if exists, is unique. • The EE of m() is the unique MVUE of m(). 20 ASYMPTOTIC EFFICIENT ESTIMATOR • Y is an asymptotic EE of m() if lim E Y m n and lim V Y CRLB n 21 EXAMPLES A r.s. of size n from X~Poi(θ). a) Find CRLB for any UE of θ. b) Find UMVUE of θ. c) Find an EE for θ. d) Find CRLB for any UE of exp{-2θ}. Assume n=1, and show that (1) x is UMVUE of exp{2θ}. Is this a reasonable estimator? 22 EXAMPLE A r.s. of size n from X~Exp(). Find UMVUE of , if exists. 23 Summary • We covered 3 methods for finding UMVUE: – Rao-Blackwell Theorem (Use a ss T, an UE U, and create a new statistic by E(U|T)) – Lehmann-Scheffe Theorem (Use a css T which is also UE) – Cramer-Rao Lower Bound (Find an UE with variance=CRLB) 24