2.22 设总体 ξ \xi ξ是含未知参数 θ \theta θ的分布,其密度函数为
p ( x ; θ ) = 2 θ x 3 e − θ / x 2 I ( 0 , ∞ ) ( x ) , θ > 0. p(x;\theta)=\frac{2\theta}{x^3}e^{-\theta/x^2}I_{(0,\infty)}(x),\theta>0. p(x;θ)=x32θe−θ/x2I(0,∞)(x),θ>0.
求 θ \theta θ的 F i s h e r Fisher Fisher信息量和当样本量为 n n n时, θ \theta θ无偏估计方差的 C − R C-R C−R下界。
参考答案:
设 x 1 , ⋯ , x n x_1,\cdots,x_n x1,⋯,xn是对应 X 1 , ⋯ , X n X_1,\cdots,X_n X1,⋯,Xn的样本值。
总体 ξ \xi ξ的密度函数为
p ( ξ ; θ ) = { 2 θ ξ 3 e − θ ξ 2 , x > 0 0 , x ≤ 0 ( θ > 0 ) p(\xi;\theta)=\left\{\begin{array}{rl}\frac{2\theta}{\xi^3}e^{-\frac{\theta}{\xi^2}},& x >0 \\ 0,&x\le 0 \end{array}\right.(\theta>0) p(ξ;θ)={ξ32θe−ξ2θ,0,x>0x≤0(θ>0)
似然函数为
L ( θ ) = { 2 n θ n ( ∏ i = 1 n x i ) 3 e − ∑ i = 1 n θ x i 2 , x i > 0 0 , x i ≤ 0. L(\theta)=\left\{\begin{array}{rl}\frac{2^n\theta^n}{(\prod\limits_{i=1}^{n}x_i)^3}e^{-\sum\limits_{i=1}^{n}\frac{\theta}{x_i^2}},&x_i>0\\ 0\qquad,&x_i\le 0.\end{array}\right. L(θ)=⎩⎪⎨⎪⎧(i=1∏nxi)32nθne−i=1∑nxi2θ,0,xi>0xi≤0. ( i = 1 , ⋯ , n , θ > 0 ) (i=1,\cdots,n,\theta>0) (i=1,⋯,n,θ>0)
对数似然函数为
l ( θ ) = ln L ( θ ) = n ln 2 + n ln θ − 3 ∑ i = 1 n ln x i − ∑ i = 1 n θ x i 2 , x i > 0 , ( i = 1 , ⋯ , n , θ > 0 ) \begin{array}{rl}l(\theta)&=\ln L(\theta)\\&=n\ln 2+n\ln \theta-3\sum\limits_{i=1}^{n}\ln x_i\\& -\sum\limits_{i=1}^{n}\frac{\theta}{x_i^2},x_i>0,(i=1,\cdots,n,\theta>0)\end{array} l(θ)=lnL(θ)=nln2+nlnθ−3i=1∑nlnxi−i=1∑nxi2θ,xi>0,(i=1,⋯,n,θ>0)
由
d l ( θ ) d θ = n θ − ∑ i = 1 n 1 x i 2 , \frac{dl(\theta)}{d\theta}=\frac{n}{\theta}-\sum\limits_{i=1}^{n}\frac{1}{x_i^2}, dθdl(θ)=θn−i=1∑nxi21,
得
d 2 l ( θ ) d θ 2 = − n θ 2 . \frac{d^2l(\theta)}{d\theta^2}=-\frac{n}{\theta^2}. dθ2d2l(θ)=−θ2n.
I X ( θ ) = − E ( d 2 l ( θ ) d θ 2 ) = n θ 2 . I_X(\theta)=-E(\frac{d^2l(\theta)}{d\theta^2})=\frac{n}{\theta^2}. IX(θ)=−E(dθ2d2l(θ))=θ2n.
所以, θ \theta θ的 F i s h e r Fisher Fisher信息量为
I ( θ ) = I X ( θ ) n = 1 θ 2 . I(\theta)=\frac{I_X(\theta)}{n}=\frac{1}{\theta^2}. I(θ)=nIX(θ)=θ21.
当样本量为 n n n时, θ \theta θ无偏估计方差的 C − R C-R C−R下界
1 I X ( θ ) = θ 2 n . \frac{1}{I_X(\theta)}=\frac{\theta^2}{n}. IX(θ)1=nθ2.
2.23写出下列分布族中,达到 C − R C-R C−R下界的基于样本 X 1 , ⋯ , X n X_1,\cdots,X_n X1,⋯,Xn的无偏估计的参数函数形式和估计量:
(1)二项分布族 { b ( k , p ) , p ∈ ( 0 , 1 ) } . \{b(k,p),p\in(0,1)\}. {b(k,p),p∈(0,1)}.
参考答案:
设 x 1 , ⋯ , x n x_1,\cdots,x_n x1,⋯,xn是对应 X 1 , ⋯ , X n X_1,\cdots,X_n X1,⋯,Xn的样本值。
总体
ξ = d b ( k , p ) , p ∈ ( 0 , 1 ) . \xi\xlongequal{d}b(k,p),p\in(0,1). ξdb(k,p),p∈(0,1).
总体 ξ \xi ξ的分布律为
p ( ξ = x ) = C k x p x ( 1 − p ) k − x , p(\xi=x)=C_k^xp^x(1-p)^{k-x}, p(ξ=x)=Ckxpx(1−p)k−x, x = 0 , 1 , ⋯ , k , p ∈ ( 0 , 1 ) . x=0,1,\cdots,k,p\in(0,1). x=0,1,⋯,k,p∈(0,1).
似然函数为
L ( p ) = ∏ i = 1 n p ( X i = x i ) = ∏ i = 1 n C k x i p x i ( 1 − p ) k − x i = ( ∏ i = 1 n C k x i ) p ∑ i = 1 n x i ( 1 − p ) n k − ∑ i = 1 n x i , \begin{array}{rl}L(p)&=\prod\limits_{i=1}^{n}p(X_i=x_i)\\&=\prod\limits_{i=1}^{n}C_k^{x_i}p^{x_i}(1-p)^{k-x_i}\\&=(\prod\limits_{i=1}^{n}C_k^{x_i})p^{\sum\limits_{i=1}^{n}x_i}(1-p)^{nk-\sum\limits_{i=1}^{n}x_i},\end{array} L(p)=i=1∏np(Xi=xi)=i=1∏nCkxipxi(1−p)k−xi=(i=1∏nCkxi)pi=1∑nxi(1−p)nk−i=1∑nxi,
x i = 0 , 1 , ⋯ , k , p ∈ ( 0 , 1 ) . x_i=0,1,\cdots,k,p\in(0,1). xi=0,1,⋯,k,p∈(0,1).
对数似然函数为
l ( p ) = ln L ( p ) = ∑ i = 1 n ln C k x i + ∑ i = 1 n x i ln p + ( n k − ∑ i = 1 n x i ) ln ( 1 − p ) , x i = 0 , 1 , ⋯ , k , p ∈ ( 0 , 1 ) . \begin{array}{rl}l(p)&=\ln L(p)=\sum\limits_{i=1}^{n}\ln C_k^{x_i}+\sum\limits_{i=1}^{n}x_i\ln p\\&+(nk-\sum\limits_{i=1}^{n}x_i)\ln (1-p),\\&x_i=0,1,\cdots,k,p\in(0,1).\end{array} l(p)=lnL(p)=i=1∑nlnCkxi+i=1∑nxilnp+(nk−i=1∑nxi)ln(1−p),xi=0,1,⋯,k,p∈(0,1).
令
d l ( p ) d p = 1 p ∑ i = 1 n x i − 1 1 − p ( n k − ∑ i = 1 n x i ) = 0 , \begin{array}{rl}\frac{dl(p)}{dp}&=\frac{1}{p}\sum\limits_{i=1}^{n}x_i\\&-\frac{1}{1-p}(nk-\sum\limits_{i=1}^{n}x_i)=0,\end{array} dpdl(p)=p1i=1∑nxi−1−p1(nk−i=1∑nxi)=0,
则
p = ∑ i = 1 n x i n k = x ˉ k . p=\frac{\sum\limits_{i=1}^{n}x_i}{nk}=\frac{\bar{x}}{k}. p=nki=1∑nxi=kxˉ.
所以, p p p的估计量
P ^ = X ˉ k . \widehat{P}=\frac{\bar{X}}{k}. P =kXˉ.
由于
E ( P ^ ) = E ( X ˉ k ) = 1 k E ( X ˉ ) = p , E(\widehat{P})=E(\frac{\bar{X}}{k})=\frac{1}{k}E(\bar{X})=p, E(P )=E(kXˉ)=k1E(Xˉ)=p,
所以, P ^ = X ˉ k \widehat{P}=\frac{\bar{X}}{k} P =kXˉ是 p p p的无偏估计。
V a r ( p ^ ) = V a r ( X ˉ k ) = 1 n 2 k 2 ∑ i = 1 n V a r ( X i ) = p ( 1 − p ) n k \begin{array}{rl}Var(\widehat{p})&=Var(\frac{\bar{X}}{k})\\&=\frac{1}{n^2k^2}\sum\limits_{i=1}^{n}Var(X_i)\\&=\frac{p(1-p)}{nk}\end{array} Var(p )=Var(kXˉ)=n2k21i=1∑nVar(Xi)=nkp(1−p)
d 2 l ( p ) d p 2 = − 1 p 2 ∑ i = 1 n X i − 1 ( 1 − p ) 2 ( n k − ∑ i = 1 n X i ) \begin{array}{rl}\frac{d^2l(p)}{dp^2}&=-\frac{1}{p^2}\sum\limits_{i=1}^{n}X_i\\&-\frac{1}{(1-p)^2}(nk-\sum\limits_{i=1}^{n}X_i)\end{array} dp2d2l(p)=−p21i=1∑nXi−(1−p)21(nk−i=1∑nXi)
I X ( p ) = E ( − d 2 l ( p ) d p 2 ) = E ( 1 p 2 ∑ i = 1 n X i + 1 ( 1 − p ) 2 ( n k − ∑ i = 1 n X i ) ) = n k p ( 1 − p ) \begin{array}{rl}I_X(p)&=E(-\frac{d^2l(p)}{dp^2})\\&=E(\frac{1}{p^2}\sum\limits_{i=1}^{n}X_i\\&+\frac{1}{(1-p)^2}(nk-\sum\limits_{i=1}^{n}X_i))\\&=\frac{nk}{p(1-p)}\end{array} IX(p)=E(−dp2d2l(p))=E(p21i=1∑nXi+(1−p)21(nk−i=1∑nXi))=p(1−p)nk
因为
V a r ( p ^ ) = 1 I X ( p ) Var(\widehat{p})=\frac{1}{I_X(p)} Var(p )=IX(p)1
所以 P ^ \widehat{P} P 作为 p p p的无偏估计其方差达到了 C − R C-R C−R下界。
(2)正态分布族 { N ( μ , σ 0 2 ) , μ ∈ R } . \{N(\mu,\sigma_0^2),\mu\in R\}. {N(μ,σ02),μ∈R}.
参考答案:
设 x 1 , ⋯ , x n x_1,\cdots,x_n x1,⋯,xn是对应 X 1 , ⋯ , X n X_1,\cdots,X_n X1,⋯,Xn的样本值。
总体
ξ = d N ( μ , σ 0 2 ) , ( σ = σ 0 ) \xi\stackrel{d}{=}N(\mu,\sigma_0^2),(\sigma=\sigma_0) ξ=dN(μ,σ02),(σ=σ0)
总体 ξ \xi ξ的密度函数为
p ( ξ ; μ ) = 1 2 π σ 0 e − ( ξ − μ ) 2 2 σ 0 2 , μ ∈ R . p(\xi;\mu)=\frac{1}{\sqrt{2\pi}\sigma_0}e^{-\frac{(\xi-\mu)^2}{2\sigma_0^2}},\mu\in R. p(ξ;μ)=2πσ01e−2σ02(ξ−μ)2,μ∈R.
似然函数为
L ( μ ) = ∏ i = 1 n p ( x i ; μ ) = ∏ i = 1 n 1 2 π σ 0 e − ( x i − μ ) 2 2 σ 0 2 = 1 ( 2 π ) n 2 σ 0 n e − ∑ i = 1 n ( x i − μ ) 2 2 σ 0 2 , μ ∈ R \begin{array}{rl}L(\mu)&=\prod\limits_{i=1}^{n}p(x_i;\mu)\\&=\prod\limits_{i=1}^{n}\frac{1}{\sqrt{2\pi}\sigma_0}e^{-\frac{(x_i-\mu)^2}{2\sigma_0^2}}\\&=\frac{1}{(2\pi)^{\frac{n}{2}}\sigma_0^n}e^{-\frac{\sum\limits_{i=1}^{n}(x_i-\mu)^2}{2\sigma_0^2}},\mu\in R\end{array} L(μ)=i=1∏np(xi;μ)=i=1∏n2πσ01e−2σ02(xi−μ)2=(2π)2nσ0n1e−2σ02i=1∑n(xi−μ)2,μ∈R
对数似然函数为
l ( μ ) = ln L ( μ ) = − n 2 ln ( 2 π ) − n 2 ln ( σ 0 2 ) − ∑ i = 1 n ( x i − μ ) 2 2 σ 0 2 , μ ∈ R . \begin{array}{rl}l(\mu)&=\ln L(\mu)\\&=-\frac{n}{2}\ln(2\pi)-\frac{n}{2}\ln(\sigma_0^2)\\&-\frac{\sum\limits_{i=1}^{n}(x_i-\mu)^2}{2\sigma_0^2},\mu\in R.\end{array} l(μ)=lnL(μ)=−2nln(2π)−2nln(σ02)−2σ02i=1∑n(xi−μ)2,μ∈R.
令
d l ( μ ) d μ = 1 σ 0 2 ( ∑ i = 1 n x i − n μ ) = 0 , \frac{dl(\mu)}{d\mu}=\frac{1}{\sigma_0^2}(\sum\limits_{i=1}^{n}x_i-n\mu)=0, dμdl(μ)=σ021(i=1∑nxi−nμ)=0,
则
μ = 1 n ∑ i = 1 n x i = x ˉ . \mu=\frac{1}{n}\sum\limits_{i=1}^{n}x_i=\bar{x}. μ=n1i=1∑nxi=xˉ.
所以, μ \mu μ的估计量
μ ^ = X ˉ . \widehat{\mu}=\bar{X}. μ =Xˉ.
由于
E ( μ ^ ) = E ( X ˉ ) = μ , E(\widehat{\mu})=E(\bar{X})=\mu, E(μ )=E(Xˉ)=μ,
所以, μ ^ = X ˉ \widehat{\mu}=\bar{X} μ =Xˉ是 μ \mu μ的无偏估计。
V a r ( μ ^ ) = V a r ( X ˉ ) = 1 n 2 ∑ i = 1 n V a r ( X i ) = σ 0 2 n \begin{array}{rl}Var(\widehat{\mu})&=Var(\bar{X})\\&=\frac{1}{n^2}\sum\limits_{i=1}^{n}Var(X_i)\\&=\frac{\sigma_0^2}{n}\end{array} Var(μ )=Var(Xˉ)=n21i=1∑nVar(Xi)=nσ02
d 2 l ( μ ) d μ 2 = − n σ 0 2 \frac{d^2l(\mu)}{d\mu^2}=-\frac{n}{\sigma_0^2} dμ2d2l(μ)=−σ02n
I X ( μ ) = E ( − d 2 l ( μ ) d μ 2 ) = n σ 0 2 I_X(\mu)=E(-\frac{d^2l(\mu)}{d\mu^2})=\frac{n}{\sigma_0^2} IX(μ)=E(−dμ2d2l(μ))=σ02n
因为
V a r ( μ ^ ) = 1 I X ( μ ) Var(\widehat{\mu})=\frac{1}{I_X(\mu)} Var(μ )=IX(μ)1
所以 μ ^ \widehat{\mu} μ 作为 μ \mu μ的无偏估计其方差达到了 C − R C-R C−R下界。
(3) Γ \Gamma Γ分布族 { Γ ( α , 1 2 α 2 ) , σ 2 > 0 } , α \{\Gamma(\alpha,\frac{1}{2\alpha^2}),\sigma^2>0\},\alpha {Γ(α,2α21),σ2>0},α为已知的.
参考答案:
设 x 1 , ⋯ , x n x_1,\cdots,x_n x1,⋯,xn是对应 X 1 , ⋯ , X n X_1,\cdots,X_n X1,⋯,Xn的样本值。
总体
ξ = d Γ ( α , 1 2 σ 2 ) , σ 2 > 0 \xi\stackrel{d}{=}\Gamma(\alpha,\frac{1}{2\sigma^2}),\sigma^2>0 ξ=dΓ(α,2σ21),σ2>0
总体 ξ \xi ξ的密度函数为
p ( ξ ; σ 2 ) = ( 1 2 σ 2 ) α Γ ( α ) ξ α − 1 e − 1 2 σ 0 2 ξ I ( 0 , ∞ ) ( ξ ) . p(\xi;\sigma^2)=\frac{(\frac{1}{2\sigma^2})^{\alpha}}{\Gamma(\alpha)}\xi^{\alpha-1}e^{-\frac{1}{2\sigma_0^2}\xi}I_{(0,\infty)}(\xi). p(ξ;σ2)=Γ(α)(2σ21)αξα−1e−2σ021ξI(0,∞)(ξ).
似然函数为
L ( σ 2 ) = { ∏ i = 1 n ( 1 2 σ 2 ) α Γ ( α ) x i α − 1 e − 1 2 σ 0 2 x i , x i > 0 0 , x i ≤ 0 L(\sigma^2)=\left\{ \begin{aligned} \prod\limits_{i=1}^{n}\frac{(\frac{1}{2\sigma^2})^{\alpha}}{\Gamma(\alpha)}x_i^{\alpha-1}e^{-\frac{1}{2\sigma_0^2}x_i}&,x_i>0\\ 0&,x_i \le 0 \end{aligned} \right. L(σ2)=⎩⎪⎪⎨⎪⎪⎧i=1∏nΓ(α)(2σ21)αxiα−1e−2σ021xi0,xi>0,xi≤0 ( i = 1 , ⋯ , n ) (i=1,\cdots,n) (i=1,⋯,n)
对数似然函数为
l ( σ 2 ) = ln L ( σ 2 ) = − n ln Γ ( α ) + n α ln ( 1 2 σ 2 ) + ( α − 1 ) ∑ i = 1 n ln x i − ∑ i = 1 n x i 2 σ 2 . \begin{array}{rl}l(\sigma^2)&=\ln L(\sigma^2)\\&=-n\ln\Gamma(\alpha)+n\alpha\ln(\frac{1}{2\sigma^2})\\&+(\alpha-1)\sum\limits_{i=1}^{n}\ln x_i-\frac{\sum\limits_{i=1}^{n}x_i}{2\sigma^2}.\end{array} l(σ2)=lnL(σ2)=−nlnΓ(α)+nαln(2σ21)+(α−1)i=1∑nlnxi−2σ2i=1∑nxi.
令
d l ( σ 2 ) d σ 2 = − n α σ 2 + ∑ i = 1 n x i 2 σ 4 = 0 , \frac{dl(\sigma^2)}{d\sigma^2}=-\frac{n\alpha}{\sigma^2}+\frac{\sum\limits_{i=1}^{n}x_i}{2\sigma^4}=0, dσ2dl(σ2)=−σ2nα+2σ4i=1∑nxi=0,
则
σ 2 = x ˉ 2 α . \sigma^2=\frac{\bar{x}}{2\alpha}. σ2=2αxˉ.
所以, σ 2 \sigma^2 σ2的估计量
σ 2 ^ = X ˉ 2 α . \widehat{\sigma^2}=\frac{\bar{X}}{2\alpha}. σ2 =2αXˉ.
由于
E ( σ 2 ^ ) = E ( X ˉ 2 α ) = σ 2 , E(\widehat{\sigma^2})=E(\frac{\bar{X}}{2\alpha})=\sigma^2, E(σ2 )=E(2αXˉ)=σ2,
所以, σ 2 ^ = X ˉ 2 α \widehat{\sigma^2}=\frac{\bar{X}}{2\alpha} σ2 =2αXˉ是 σ 2 \sigma^2 σ2的无偏估计。
V a r ( σ 2 ^ ) = V a r ( X ˉ 2 α ) = σ 0 4 n α Var(\widehat{\sigma^2})=Var(\frac{\bar{X}}{2\alpha})=\frac{\sigma_0^4}{n\alpha} Var(σ2 )=Var(2αXˉ)=nασ04
d 2 l ( σ 2 ) d ( σ 2 ) 2 = n α σ 0 4 − ∑ i = 1 n x i σ 6 \frac{d^2l(\sigma^2)}{d(\sigma^2)^2}=\frac{n\alpha}{\sigma_0^4}-\frac{\sum\limits_{i=1}^{n}x_i}{\sigma^6} d(σ2)2d2l(σ2)=σ04nα−σ6i=1∑nxi
I X ( σ 2 ) = E ( − d 2 l ( σ 2 ) d ( σ 2 ) 2 ) = n α σ 0 4 I_X(\sigma^2)=E(-\frac{d^2l(\sigma^2)}{d(\sigma^2)^2})=\frac{n\alpha}{\sigma_0^4} IX(σ2)=E(−d(σ2)2d2l(σ2))=σ04nα
因为
V a r ( σ 2 ) = 1 I X ( σ 2 ) Var(\sigma^2)=\frac{1}{I_X(\sigma^2)} Var(σ2)=IX(σ2)1
所以 σ 2 ^ \widehat{\sigma^2} σ2 作为 σ 2 \sigma^2 σ2的无偏估计其方差达到了 C − R C-R C−R下界。
如果您在看完以后,无论是题目还是参考解答过程有什么问题,希望不吝指出!如果有更好的解题思路或者好的问题分享,欢迎通过后台发送给我们!