If $X_i \sim$ uniform$(0,\theta)$, how can I show that the estimator $(n+1) \max(X_1, X_2, \ldots , X_n)/n$ is a method of moments estimator for $\theta$? For example, we know the first sample moment is $\sum_{i=1}^{n} X_i / n$ and that the first population moment is $\theta / 2$. Thus, the estimator we get from the method of moments is $2 \bar{X}$. I don't know how to get $(n+1) X_{(n)} / n$ by the method of moments.

  • $\begingroup$ What is the expected value of $X_{(n)}$? $\endgroup$ – dsaxton Mar 8 '16 at 23:57
  • $\begingroup$ One does not estimate the values of a random variable. Are you asking for the distribution of the maximum? For prediction of the maximum? Are you considering actual data or are you asking a theoretical question about random variables? $\endgroup$ – whuber Mar 9 '16 at 0:26
  • $\begingroup$ @dsaxton The expected value of X(n) should be n*theta/(n+1) $\endgroup$ – purod Mar 9 '16 at 1:37
  • $\begingroup$ @whuber What I want to ask is that if theta is not known, we want to find an unbiased estimator for it. Using other method, I get estimate of (n+1)*x(n)/n. However, I am asked to get this estimator with method of moment. I know that mean = theta/2 then an estimator for theta by method of moment should be 2*mean. I don't know how to get estimator (n+1)*theta/n with method of moment $\endgroup$ – purod Mar 9 '16 at 1:40
  • $\begingroup$ It's a trivial method of moments estimator based on a sample of size one from a "stretched" beta distribution over $(0, \theta)$ (the distribution of the maximal order statistic). This isn't really the idea behind method of moments though, generally you work with the underlying distribution and not the distribution of a sample-based statistic. $\endgroup$ – dsaxton Mar 9 '16 at 1:46

Let $X_1, \dots X_n \sim U(0, \theta)$. Let $X_{(n)} = \max\{X_1, \dots, X_n\}.$ The traditional way to answer a question "What is the method of moments estimator?" is to do what you did in your statement of the question.

However, you are being asked to show a statistic is a method of moments estimator. Since, a method of moments estimator for a parameter is unique to a distribution, it is the distribution that you have to change. What I mean is,

Define $Y = X_{(n)}$. Now you will find the method of moments estimator for $\theta$ that characterizes the distribution of $Y$ (not the distribution of $X_1$).

You already have $E(Y) = \dfrac{n \theta}{n+1}$, and that gives you the desired result.

  • $\begingroup$ Thank you ! The understanding of the method is really important... $\endgroup$ – purod Mar 10 '16 at 13:37

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.