1
$\begingroup$

$$Var[\bar{X}] = \sigma^2/n $$ $$Var [\sum{X}_i] = n\sigma^2$$ $$lim_{n \to \infty} Var[\bar{X}] = 0 $$ wich means at $\infty$ we will always get the same $\bar{X}$ after every simulation. I understand this such as if I have an observation $X_i >E[X]$ I'm sure that I also have $X_i <E[X]$ in my$ \infty $ observations that will compensate it. is this correct?

What I don't understand is if $\bar{X}$ is constant in $ \infty $ why $\sum{X}_i = n\bar{X}$ isn't? That is to say why $$lim_{n \to \infty} Var[\sum{X}_i] \neq 0$$ I'm not looking for mathematical proof but the idea.

$\endgroup$
  • 1
    $\begingroup$ What is the limit of a constant times a number ($n$) that goes to $\infty$? $\endgroup$ – jbowman Dec 26 '18 at 3:16
  • $\begingroup$ I don't understand your question? just if the sample mean is constant why the sample sum isn't $\endgroup$ – Youssef Dec 26 '18 at 3:21
  • $\begingroup$ The sample sum equals the sample mean times $n$. If the sample mean is constant, what happens to the sample sum as $n \to \infty$? $\endgroup$ – jbowman Dec 26 '18 at 4:33
  • $\begingroup$ The sample sum tends to infinity but its variance is not 0 . The sum is infinity and variable, at the same time the sum divided by n is a constant. I find this counter-intuitive $\endgroup$ – Youssef Dec 26 '18 at 7:11
0
$\begingroup$

You asked for some intuition. Well, even when the law of large numbers hold, so that $\bar{X}_n \to \mu$ when $n \to \infty$, in practice you will never have $n=\infty$, so the variance of $\bar{X}_n$ is $\sigma^2/n$ which will be positive, maybe tiny but not zero (as long as $\sigma^2 >0$.)

If you multiply $\bar{X}_n$ with $\sqrt{n}$ so to get a constant variance not depending on $n$, you can think of that as looking at the small deviations from $\mu$ through a magnification lens, so as to see the details. Multiplying that with $\sqrt{n}$ another time to get the sample sum, you are looking at the first magnification lens through another magnification lens, and so now the image will be increasing in size with $n$.

$\endgroup$
  • $\begingroup$ Thank you, But if you magnify something that is decreasing (the variance of $\bar X$) it's true that it will give something bigger ( The variance of $S_n$) but magnifying won't make the two variances go in different directions ( one decreases and the other increases as n increases). This is why I don't find this explanation satisfying. Please explain me more if you think I still don't get your idea. $\endgroup$ – Youssef Feb 4 at 16:20

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.