Suppose that $X_1,\cdots, X_n$ are independent random variables from uniform distribution on interval $(\theta_1,\theta_2)$, $\theta_2>\theta_1>0$. It is know that $T(X)=(X_{(1)},X_{(n)})$ is the complete sufficient statistic for $(\theta_1,\theta_2)$.

(a) Find the UMVUE of the range parameter $\tau(\theta)=\theta_2-\theta_1$.

(b) Find the UMVUE of the mean parameter $\tau(\theta)=(\theta_2+\theta_1)/2$.

  • $\begingroup$ This is phrased quite a lot like a homework problem. Can you tell us specifically what difficulties you've had with it. For part (a) common sense suggests that $X_{(n)}-X_{(1)}$ should be looked at, but obviously its expected value is less than $\theta_2-\theta_1$. But if the amount you need to multiply it by in order to get expectation $\theta_2-\theta_1$ does not depend on $\theta_1$ or $\theta_2$, then you can do something. For part (b), what function of $(X_{(1)},X_{(n)})$ would common sense suggest? ${}\qquad{}$ $\endgroup$ – Michael Hardy Mar 25 '15 at 22:58
  • $\begingroup$ Why $T(X)$ is complete statistic? I can just show it is a sufficient statistic. $\endgroup$ – username Nov 19 '18 at 6:34

You have $T(X)=(X_{(1)}=\min \mathbf{X},X_{(n)}=\max \mathbf{X})$ which is the sufficient statistic for the uniform distribution.

Then, the first UMVUE for $\theta_2-\theta_1$ is $((n+1)/n)(X_{(n)}-X_{(1)})$ and for the second $((n+1)/n)((X_{(n)}+X_{(1)})/2)$

The idea is as follows: Given two random variable with $X$ and $Y$ if they have minimum variance for the same true mean, the linear combinations of them is also minimum variance.

Why they are unbiased? have a look at this one

  • 3
    $\begingroup$ From order statistics I get $\mathrm{E}[X_{(i)}] = \theta_1 + (\theta_2 - \theta_1)\cdot i/(n+1)$ and thus $\mathrm{E}[X_{(n)} - X_{(1)}] = (\theta_2 - \theta_1)(n-1)/(n+1)$ and $\mathrm{E}[X_{(n)} + X_{(1)}] = \theta_2 + \theta_1$. Thus, I think the MVUE are $(n+1)/(n-1)(X_{(n)} - X_{(1)})$ and just $(X_{(n)} + X_{(1)})/2$. Wikipedia also states the MVUE of the mean is just the average of max and min sample. Care to comment? $\endgroup$ – GDumphart Mar 19 '18 at 19:06
  • $\begingroup$ Your final answers are incorrect, as rightly pointed out by the above comment. $\endgroup$ – StubbornAtom May 27 '18 at 22:14
  • $\begingroup$ @GDumphart please feel free to correct it. It seems I wrote it fast. I just took care of the bias correction with respect to the end point estimators as stated here: en.wikipedia.org/wiki/Minimum-variance_unbiased_estimator please see "other examples". $\endgroup$ – Seyhmus Güngören May 27 '18 at 23:58

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.