0
$\begingroup$

Suppose that $Y_1,...,Y_n$ is an IID sample from a uniform $U(\theta, 1)$ distribution. The method of moments estimator for $\theta$ is $\tilde \theta=2\bar Y-1$. The standard error of $\tilde \theta$ is

$$\sigma_{\tilde \theta}=\frac{1-\theta}{\sqrt{3n}}$$

Find an unbiased estimator of $\sigma_{\tilde \theta}$ and show that it is unbiased.

So normally I know how to find the estimator of a parameter of a distribution, but I don't quite understand how I'm supposed to find an estimator of the standard deviation of another estimator of a parameter of a distribution... so I really don't know how to go about this. What do I do?

$\endgroup$
  • $\begingroup$ Sorry, but those are all basically given. I don't understand how those quantities are helping me find this second estimator. $\endgroup$ – ereHsaWyhsipS Feb 9 '18 at 4:40
  • $\begingroup$ If you let $\hat{\sigma}_{\tilde{\theta}} = \frac{1-\tilde{\theta}}{\sqrt{3n}}$ and $\tilde{\theta}$ is unbiased, then $\hat{\sigma}$ is an unbiased estimator of $\sigma_{\tilde{\theta}}$. I think $\tilde{\theta} = 1-2\bar{Y}$ might be the correct method of moments estimator, though? Otherwise it's expectation is $-\theta$ $\endgroup$ – Ryan Warnick Feb 9 '18 at 19:08
2
$\begingroup$

My Comment (now deleted) was an attempt, perhaps not well worded, to get you to look carefully at everything again.

First verify that $E(Y_i) = \frac{\theta + 1}{2}$ and $Var(Y_i) = \frac{(1-\theta)^2}{12},$ so that $E(\bar Y) = \frac{\theta+1}{2},\,$ $Var(\bar Y) = \frac{(1-\theta)^2}{12n}$ and $SD(\bar Y) = \frac{1-\theta}{\sqrt{12n}}.$

Then, setting $\bar Y = \frac{\tilde \theta+1}{2}$ we obtain the MME of $\theta$ as $\tilde \theta = 2\bar Y - 1.$ Morover, $\tilde \theta$ is an unbiased estimator of $\theta$ because $E(2\bar Y - 1) = 2E(\bar Y) -1 = \theta.$ [Earlier typos above fixed, thanks to Comment by @ereHsaWyhsipS.]

As a reality check on the above, I use a simulation of the special case $\theta = .25,\, n = 10.$

set.seed(209)  # retain 'set.seed' to repeat same simulation, omit for fresh simulation
th = 1/4; n = 10
a = replicate(10^6, mean(runif(n, th, 1)))  # one million sample averages
th.mme = 2*a - 1
mean(th.mme);  th
## 0.2499683  # aprx E(MME of theta)
## 0.25       # exact $(MME of theta) = theta, MME unbiased

Also $Var (\tilde \theta) = Var(2\bar Y - 1) = 4Var(\bar Y) = \frac{(1-\theta)^2}{3a}$ and $SD(\tilde \theta) = SD(\tilde \theta) = \sigma^* = \frac{1-\theta}{\sqrt{3n}}.$ (This is your $\sigma_{\tilde \theta}.)$ Another reality check:

sd(th.mme);  (1 - th)/sqrt(3*n)
## 0.136847    # aprx SD(MME of theta)
## 0.1369306   # exact

A reasonable first try toward an estimator $\widetilde{\sigma^*}$ of $\sigma^*$ is simply to use $\tilde \theta$ for $\theta$ in the numerator to get $$\widetilde{\sigma^*} = \frac{1 - \tilde\theta}{\sqrt{3n}} = \frac{1 - (2\bar Y - 1)}{\sqrt{3n}} = \frac{2(1 - \bar Y)}{\sqrt{3n}}.$$ Then $E(\widetilde{\sigma^*}) = \frac{2 - (1+\theta)}{\sqrt{3n}} = \frac{1-\theta}{\sqrt{3n}},$ so that $\widetilde{\sigma^*}$ is unbiased for $\sigma^*.$

mean(2*(1-a)/sqrt(3*n))
## 0.1369364     # agrees aprx with above, suggesting sg.mme unbiased
$\endgroup$
  • 1
    $\begingroup$ This is a uniform distribution. The expectation is $\frac{1+\theta}{2}$. $\endgroup$ – ereHsaWyhsipS Feb 9 '18 at 21:34
  • $\begingroup$ Thanks. Fixing typos. Simulations OK $\endgroup$ – BruceET Feb 9 '18 at 23:32

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.