Suppose that $W_1,W_2,...,W_n$ are uncorrelated unbiased estimators of a parameter $\theta$.

Consider $W=\sum_{i=1}^na_iW_i$ such that $E(W)=\theta$ and $Var(W_i)=\sigma^2_i$, where the $a_i$'s are constants. So it is trivial to see that $\sum_{i=1}^na_i=1$. Call all such unbiased estimators $W$ "linear unbiased estimators" of $\theta$.

Let $W^{\star}$ be the linear unbiased estimator having minimum variance. It is easy to see, by a simple application of Cauchy-Schwarz Inequality, that $W^{\star}=\dfrac{\sum_{i=1}^nW_i/\sigma_i^2}{\sum_{i=1}^n(1/\sigma_i^2)}$ and $Var(W^{\star})=\dfrac{1}{\sum_{i=1}^n\dfrac{1}{\sigma_i^2}}$. The corresponding optimal weights are $a_i^{\star}=\dfrac{\dfrac{1}{\sigma_i^2}}{\sum_{i=1}^n\dfrac{1}{\sigma_i^2}}$.

Let $W'$ be any other linear unbiased estimator of $\theta$ such that $a_i\geq0$ for all $i\in\{1,2,...,n\}$.

Then Tukey's Inequality is: $\dfrac{Var(W')}{Var(W^{\star})}\leq\dfrac{1}{1-\lambda^2}$ where $\dfrac{1+\lambda}{1-\lambda}=\dfrac{b_{\max}}{b_{\min}}$, with $b_{\max}=\max_i(\dfrac{a_i}{a_i^{\star}})$ and similarly $b_{\min}=\min_i(\dfrac{a_i}{a_i^{\star}})$, $i\in\{1,2,...,n\}$.

How can this inequality be proved?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.