So I'm a little stuck with what I feel is a basic question. The calculation is easy, but I've obviously missed a key concept of MLE.

The Question
Consider the family of models for the data X1,...,Xn i.i.d. unif[0,θ], θ ∈ Θ = R+; the true, unknown parameter governing the data generating process is $θ_{0} > 0$.

We are given the estimator for $θ_{0}$: $\tilde{\theta}_n = 2\bar{x}_{n} = 2\frac{1}{n}\sum^{n}_{i=1} X_{i}$

And we are asked to find the variance for $\tilde{\theta}_n$

The Answer
So I have the following solution supplied to me: $$ Var(2\bar{x}_{n}) = \frac{4}{12n}\theta^{2}_{0} = \frac{1}{3n}\theta^{2}_{0} $$
The Attempt #1
Because we have so much information, my first inclination was to forget estimation, and just calculate the distribution's variance in terms of $\theta$ directly: i.e. $$ \begin{align*} \mu_{2} = E[X^{2}] &= \int_{0}^{\theta} X^{2} f(x) dx \\ &= \int_{0}^{\theta} X^{2} \frac{1}{\theta} dx \\ &= \frac{\theta^{2}}{3} = Var(X) \end{align*} $$

and then sub this result into $Var(2\bar{x}_n)$, i.e.: $$ \begin{align*} Var(\tilde{\theta}_{n}) &= Var(2\bar{x}_n) \\ &= 4Var(\bar{x}_n)\\ &= \frac{4}{3}\theta^{2} \end{align*} $$

I'm assuming my reasoning was flawed here, because in calculating the second moment,second moment=!Variance I've calculated $Var(X)$ (or $Var(x_{i})$), and $Var(X) \neq Var(\bar{x}_n)$

The Attempt #2
I could just use the knowledge that for this uniform distribution $Var(X) = \frac{\theta^2}{12}$. i.e.

$$ \begin{align*} Var(\tilde{\theta}_{n}) &= Var(2\bar{x}_n) \\ &= 4Var(\bar{x}_n)\\ &= \frac{4}{12}\theta^{2}\\ &= \frac{1}{3}\theta^{2}\\ \end{align*} $$ This is a little closer, but I still lack the 'n' term in my answer, and am uncomfortable with this approach, for the same reason that my approach #1 failed.

As mentioned, I've completely overlooked/misunderstood a fundamental here. If it's obvious to anyone, a push in the right direction is greatly appreciated.

| cite | improve this question | | | | |
  • 1
    $\begingroup$ Note that this a method of moments question and has nothing to do with maximum likelihood. $\endgroup$ – Dason Apr 13 '15 at 13:37
  • $\begingroup$ You wrote: $$\require{cancel} \xcancel{\vphantom{\int\limits^\int_\int}\mu_2 = E[X^2] = \int_0^\theta X^2 f(x) \, dx}$$ That is wrong. It should say this: $$ \mu_2 = E[X^2] = \int_0^\theta x^2 f(x) \, dx. $$ One should be careful about which is capital $X$ and which is lower-case $x.$ Without attention to this distinction, there are elementary things that one cannot understand. $\endgroup$ – Michael Hardy Oct 25 '19 at 19:31

$$\newcommand{\Var}{\operatorname{Var}}\newcommand{\E}{\operatorname{E}} \Var(2\bar{X}_n) = 4\Var(\bar{X}_n) = 4 \Var\left(\frac{\sum_{i=1}^n{X_i}}{n}\right) =\ldots\text{?}$$

$$=\frac{4}{n^2} \Var\left(\sum_{i=1}^n{X_i}\right)$$

Then, as the $X_i$ are independent

$$=\frac{4}{n^2} \sum_{i=1}^n{\Var(X_i)}$$

To find $\Var(X_i)$ you can use the identity

$$\Var(X_i) = \E(X_i^2) - [\E(X_i)]^2$$

& will therefore have two expectations to find through simple integrations.

| cite | improve this answer | | | | |
  • $\begingroup$ That's an excellent push :) $\endgroup$ – Vince Apr 13 '15 at 12:50
  • 3
    $\begingroup$ $$ \begin{align*} Var(\tilde{\theta}_{n}) &= Var(2\bar{x}_n) \\ &= 4Var(\bar{x}_n)\\ &= 4 Var\left(\frac{\sum_i^n{X_i}}{n}\right) \\ &= \frac{4}{n^{2}} Var(x_{1} + \ldots + x_{n}) \\ &= \left( \frac{4}{n^{2}} \right ) n \left( \frac{\theta^{2}_{0}}{12} \right ) \mbox{from knowledge of }unif[0,\theta]\\ &= \frac{\theta^{2}_{n}}{3n} \\ \end{align*}$$ $\endgroup$ – Vince Apr 13 '15 at 12:50
  • $\begingroup$ my only remaining disconnect is that when I calculated the second moment in attempt #1...this doesn't seem to be knowledge that I needed, or used? $\endgroup$ – Vince Apr 13 '15 at 12:53
  • $\begingroup$ @Vince Well, you need to calculate $\operatorname{Var}(X_i)$ if you don't already know it. (And note that $\operatorname{E}(X_i^2)\neq\operatorname{Var}(X_i)$.) $\endgroup$ – Scortchi - Reinstate Monica Apr 13 '15 at 13:09
  • $\begingroup$ Understood, I need to use the centralized moment. I'm really happy with this question now. Many thanks for all of your help. $\endgroup$ – Vince Apr 13 '15 at 14:04

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.