Lecture12_570-1统计笔记.docVIP

 PAGE \* MERGEFORMAT 13 Lecture 12 (Regular) Exponential Family The density function of a regular exponential family is: f x; θ=cθhxexp j=1kwjθtjx, θ = θ1,…,θk Example. Poisson(θ) f x; θ=exp?(-θ)1x!explnθ*x Example. Normal. Nμ,σ2. θ = μ,σ2 (both unknown). fx; μ,σ2)=12πσe-12σ2x-μ2 =12πσexp-12σ2x-μ2 =12πσexp-12σ2(x2-2xμ+μ2) =12πσexp-μ22σ2exp-12σ2(x2-2xμ) Theorem (Exponential family sufficient Statistic). Let X1,…,Xn be a random sample from the regular exponential family. Then TX=i=1nt1Xi,…,i=1ntkXi is sufficient for θ = θ1,…,θk. Example. Poisson(θ) Let X1,…,Xn be a random sample from Poisson(θ) Then TX=i=1nXi is sufficient for θ . Example. Normal. Nμ,σ2. θ = μ,σ2 (both unknown). Let X1,…,Xn be a random sample from Nμ,σ2 Then TX=(i=1nXi, i=1nXi2) is sufficient for θ = μ,σ2. Exercise. Apply the general exponential family result to all the standard families discussed above such as binomial, Poisson, normal, exponential, gamma. A Non-Exponential Family Example. Discrete uniform. P (X = x) = 1/θ, x = 1,...,θ, θ is a positive integer. Another Non-exponential Example. X1, . . . , Xniid U 0, θ, T = Xn. Universal Cases. X1,…,Xnare iid with density f . ? The original data X1,…,Xn are always sufficient for θ. (They are trivial statistics, since they do not lead any data reduction) ? Order statistics T = X1,…,Xn are always sufficient for θ. ( The dimension of order statistics is n, the same as the dimension of the data. Still this is a nontrivial reduction as n! different values of data corresponds to one value of T . ) 3. Theorem (Rao-Blackwell) Let X1,…,Xn be a random sample from the population with pdf f (x; θ). Let TX be a sufficient statistic for θ, and UX be any unbiased estimator of θ. Let U*X=EUX|T, then U*X is an unbiased estimator of θ, U*X is a function of T, Var(U*)≤Var(U) for every θ, and VarU*Var(U) for some θ unless U*=U with probability 1 . Rao-Blackwell theorem tells us that in searching for an unbiased estimator with