2
$\begingroup$

Suppose $X_{1}, X_{2}, ....$ is a random sample from a probability distribution with mean $\mu < \infty$ and $\sigma^{2} < \infty$. Let $\bar{X}_{n} = \frac{\sum X_{i}}{n}$. From the Central Limit Theorem we can state that: $\sqrt{n}(\bar{X}_{n} - \mu) \xrightarrow {d} N(0,\sigma^{2})$

All right, but now, i want to know how to explain (intuitively) the next statements:

$(\bar{X}_{n} - \mu) \xrightarrow {d} N(0,\frac{\sigma^{2}}{n})$

$\bar{X}_{n} \xrightarrow {d} N(\mu,\frac{\sigma^{2}}{n})$

$ \sum X_{i} \xrightarrow {d} N(n\mu,\sigma^{2})$

Is it because the Slutsky theorem? (about continuous functions)

Thank's a lot!

|cite|improve this question
$\endgroup$
  • 6
    $\begingroup$ None of those three statements, understood in its usual mathematical sense, makes any sense at all, because the limiting procedures implied by the arrows render "$n$" meaningless. $\endgroup$ – whuber♦ May 18 '17 at 19:08
  • $\begingroup$ @whuber thanks a lot! I just wrote an answer, i think is in the right direction right? $\endgroup$ – S. Cow May 18 '17 at 21:40
4
$\begingroup$

This is a wrong statement

$$(\bar{X}_{n} - \mu) \xrightarrow {d} N\left (0,\frac{\sigma^{2}}{n}\right)$$

because it is translated "the left-hand side converges to a normal random variable as $n$ goes to infinity. But as $n$ goes to infinity,the right hand side acquires zero variance and becomes a degenerate random variable, a constant, not a normal distribution. Let alone that we should write something like $$(\bar{X}_{n} - \mu) \xrightarrow {d} N\left (0,\frac{\sigma^{2}}{\lim_{n\to \infty}(n)}\right)$$

The following statement is correct (although notation is not universal)

$$(\bar{X}_{n} - \mu) \sim_{\text{approx}}N\left (0,\frac{\sigma^{2}} {n}\right),\;\;\; n<\infty$$

and how "close" to this normal random variable is it will depend on the sample size in combination with the properties of the distribution $X$ follows.

|cite|improve this answer
$\endgroup$
1
$\begingroup$

I was thinking about something:

Suppose we have a suficiently big $n$, so we can state that:

$\sqrt{n}(\bar{X}_{n} - \mu) \sim N(0,\sigma^{2})$

Fix n, and we can ask ourselves the next question:

How is $(\bar{X}_{n} - \mu)$ distributed?

Since is a linear transformation of a normal, it still a normal and the parameters are:

$E[\bar{X}_{n} - \mu] = 0$ and $Var(\bar{X}_{n} - \mu) = \frac{\sigma^{2}}{n}$

So finally: $(\bar{X}_{n} - \mu) \sim N(0,\frac{\sigma^{2}}{n})$

For the others random variables is straightforward.

|cite|improve this answer
$\endgroup$
  • 2
    $\begingroup$ The intuition is good, but the mathematics isn't right. A sequence of random variables typically approaches a limiting distribution like the Normal without ever being Normal for any $n$. The intuition is that for "sufficiently large" $n$, the variables will be "sufficiently close to" Normal. The answer to "How is $\bar X_n-\mu$ distributed?" is still worked out the same way, but the justification is totally different: since $\sqrt{n}(\bar X_n-\mu)$ has a mean of $0$ and variance of $\sigma^2$, $\bar X_n$ has a mean of $\mu$ and variance of $\sigma^2/n$ and is still approximately Normal. $\endgroup$ – whuber♦ May 18 '17 at 21:46

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.