6
$\begingroup$

I understand that in CLT, we have

$\sqrt{N} \left( \frac{\bar{X}_N - \mu}{\sigma} \right) \overset{d}\to N\left(0,1\right) \qquad (*)$

where $\bar{X}_N := \frac{1}{N} \sum_{i=1}^N X_i$ is the sample mean, and $\sqrt{N}$ inflates the $\frac{\bar{X}_N - \mu}{\sigma}$ to a $\mathcal{O}_p(1)$, while without the $\sqrt{N}$, $\frac{\bar{X}_N - \mu}{\sigma} \overset{p}\to 0$.

$\textbf{My Question is:}$ since $\sqrt{N} \left( \frac{\bar{X}_N - \mu}{\sigma} \right) = \frac{\bar{X}_N - \mu}{\frac{\sigma}{\sqrt{N}}} = \frac{\bar{X}_N - \mathbb{E}\left[\bar{X}_N\right]}{SD\left(\bar{X}_N\right)}$, is it correct to view CLT as $\frac{\bar{X}_N - \mathbb{E}\left[\bar{X}_N\right]}{SD\left(\bar{X}_N\right)} \overset{d}\to N\left(0,1\right)$? If yes, are we considering the $\sqrt{N}$ from expression $(*)$ as a part of the standard deviation of the sample mean, as opposed to an inflater, and does that mean as long as we can show that $SD\left(\bar{X}_N\right) \in \mathcal{O}_p(1)$, we do not need to consider about the rate of inflation (e.g. $\sqrt{N}$) separately?

$\endgroup$
  • $\begingroup$ What do you mean by $ SD\left(\bar{X}_N\right)$? $\endgroup$ – Viktor Dec 5 '17 at 15:27
3
$\begingroup$

Formally speaking, the CLT is in principle, indeed concerned with the limiting behavior of standardized sums of random variables, say $S_n=(1/n)\sum X_i$,

$$\frac{S_n - \mathbb{E}\left[S_n\right]}{\sqrt{\text{Var}(S_n)}}$$

So indeed the $\sqrt n$ term comes from the standard deviation of the sum.

Still, thinking of it as an "inflater", as the OP says, is not without conceptual merits (although it is a "derived" inflater, not one we determine separately).

$\endgroup$

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.