I am working on an exercise asking which of the two following estimators; $X1, X2 $for the population mean of a normal population with parameters $ \mu, \sigma$ is best and why

$X1:=\frac {X_1+X_2+....X_{2n}}{2n}$,

$X2:=\frac {X_1+X_2+....X_n}{n}$

I computed the expectation and both are unbiased estimators. I tried to find the variance of the estimators, but the computations became unwieldy:

$Var(X1)= E[(X1- \mu)^2]=$

$\frac {1}{4n^2}E[( \Sigma _{i=1}^{2n}X_i-\mu)^2]=$

$E[\frac{1}{4n^2} (X_1+X_2+...+ X_{2n})^2 - 2\mu X1+ \mu^2]$

and in order to clear this I need to compute $E[X_i^2] $ ,

Which I don't see how to do. I tried standardizing but did not see it helping. Without this computation, I cant decide the best estimator for the pop mean based on neither variance nor bias-variance tradeoff.

Any ideas?



It's possible to derive the Expectation of $X_i^2$, however there's a much simpler way to calculate your variance:

$$ Var(X1) = Var(\frac{1}{2n} \sum_{i=1}^{2n} X_i) \\ = \frac{1}{4n^2} \sum_{i=1}^{2n} Var(X_i) \\ = \frac{1}{4n^2} 2 n \sigma^2 \\ = \frac{\sigma^2}{2n} $$ This comes from the fact that the variance is linear.

  • $\begingroup$ Thank you Erin. I guess we need indepence to use this formula, otherwise we need to use covariance/correlation, right? $\endgroup$ – MSIS Apr 8 '20 at 23:43
  • $\begingroup$ That's true. But the exercise you are working on is of typical nature where draws of the population are supposedly stochastically independent. If you are interested in getting the expectation of X_i^2 I suggest to look at the theoretical variance by integration and reformulate that. Then you'll see that E[X_i^2] is Var(X) + mu^2. $\endgroup$ – Erin Sprünken Apr 8 '20 at 23:47
  • 1
    $\begingroup$ Erin, I selected your answer. One last question, please : If we have two estimators E1, E2 of the same parameter and they are both unbiased and they have the same variance/s.e, what criteria would we use to select one over the other. Or maybe that should be a separate question? $\endgroup$ – MSIS Apr 11 '20 at 4:29
  • 1
    $\begingroup$ I'd recommend opening another question for this since other people may have a different view or more experience. However, If they're both unbiased and equal variances, then I'd choose the one which is easier to compute (or faster to compute on a PC). Although, what I teach to my students, is that unbiasedness is not the non-plus-ultra. It is really nice to have and sometimes important. However, there are cases where a biased estimator has a lower MSE. Just as an additional information. $\endgroup$ – Erin Sprünken Apr 11 '20 at 14:29

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.