找到


10

X1,X2,...,Xn是具有pdf的iid随机变量

fX(xθ)=θ(1+x)(1+θ)I(0,)(x)

其中θ>0。给出1的UMVUE1θ并计算其方差

我了解了两种用于获得UMVUE的方法:

  • 克莱默罗下界(CRLB)
  • 莱曼-舍夫·特莱姆

我将尝试使用两者中的前者。我必须承认,我不完全了解这里发生的事情,而我的尝试解决方案是基于一个示例问题。我有一个fX(xθ)是一个完整的单参数指数族与

h(x)=I(0,)c(θ)=θw(θ)=(1+θ)t(x)=log(1+x)

由于w(θ)=1Θ上不为零,因此适用CRLB结果。我们有

log fX(xθ)=log(θ)(1+θ)log(1+x)

θlog fX(xθ)=1θlog(1+x)

2θ2log fX(xθ)=1θ2

所以

I1(θ)=E(1θ2)=1θ2

和CRLB为的无偏估计τ(θ)

[τ(θ)]2nI1(θ)=θ2n[τ(θ)]2

由于

i=1nt(Xi)=i=1nlog(1+Xi)

那么i=1nlog(1+Xi)任何线性函数,或者等效地,1的任何线性函数1ni=1nlog(1+Xi),将达到其期望的CRLB,因此将成为其期望的UMVUE。由于E(log(1+X))=1θ我们的UMVUE为1θ1ni=1nlog(1+Xi)

对于天然的参数,我们可以让η=(1+θ)θ=(η+1)

然后

Var(log(1+X))=ddη(1η+1)=1(η+1)2=1θ2

这是有效的解决方案吗?有没有更简单的方法?仅当E(t(x))等于您要估算的值时,此方法才有效吗?


4
在您表明pdf是单参数指数族的成员这一点上,很明显,该族的完整充分统计量为 因为正如您所说,E T / n = 1
T(X1,,Xn)=i=1nln(1+Xi)
根据Lehmann-Scheffe定理, θT/n1/θ的UMVUE。E(T/n)=1θT/n1/θ
StubbornAtom

那么,我有“自部分是非零..... θ 2w(θ)=1“是否无关紧要?θ2n[τ(θ)]2
Remy

2
并不是的; 使用CRLB更容易找到的方差。因此,要一次解决两个问题,您的论点就足够了。T
StubbornAtom

要找到方差这样,我会采取?因此,我以前做错了吗?θ2n[τ(θ)]2=θ2n(1θ2)2=1nθ2
雷米

是的,这就是的方差。精确地 T
StubbornAtom

Answers:


8

您的推理基本上是正确的。

样品(X1,X2,,Xn)的联合密度为

fθ(x1,x2,,xn)=θn(i=1n(1+xi))1+θ1x1,x2,,xn>0,θ>0lnfθ(x1,x2,,xn)=nln(θ)(1+θ)i=1nln(1+xi)+ln(1min1inxi>0)θlnfθ(x1,x2,,xn)=nθi=1nln(1+xi)=n(i=1nln(1+xi)n1θ)

Thus we have expressed the score function in the form

(1)θlnfθ(x1,x2,,xn)=k(θ)(T(x1,x2,,xn)1θ)

, which is the equality condition in the Cramér-Rao inequality.

It is not difficult to verify that

(2)E(T)=1ni=1nE(ln(1+Xi))=1/θ=1θ

From (1) and (2) we can conclude that

  • The statistic T(X1,X2,,Xn) is an unbiased estimator of 1/θ.
  • T satisfies the equality condition of the Cramér-Rao inequality.

These two facts together imply that T is the UMVUE of 1/θ.

The second bullet actually tells us that variance of T attains the Cramér-Rao lower bound for 1/θ.

Indeed, as you have shown,

Eθ[2θ2lnfθ(X1)]=1θ2

This implies that the information function for the whole sample is

I(θ)=nEθ[2θ2lnfθ(X1)]=nθ2

So the Cramér-Rao lower bound for 1/θ and hence the variance of the UMVUE is

Var(T)=[ddθ(1θ)]2I(θ)=1nθ2


Here we have exploited a corollary of the Cramér-Rao inequality, which says that for a family of distributions f parametrised by θ (assuming regularity conditions of CR inequality to hold), if a statistic T is unbiased for g(θ) for some function g and if it satisfies the condition of equality in the CR inequality, namely

θlnfθ(x)=k(θ)(T(x)g(θ))
, then T must be the UMVUE of g(θ). So this argument does not work in every problem.

Alternatively, using the Lehmann-Scheffe theorem you could say that T=1ni=1nln(1+Xi) is the UMVUE of 1/θ as it is unbiased for 1/θ and is a complete sufficient statistic for the family of distributions. That T is compete sufficient is clear from the structure of the joint density of the sample in terms of a one-parameter exponential family. But variance of T might be a little tricky to find directly.


One could also use the distribution of T to find its mean,variance.
StubbornAtom
By using our site, you acknowledge that you have read and understand our Cookie Policy and Privacy Policy.
Licensed under cc by-sa 3.0 with attribution required.