3
$\begingroup$

Given i.i.d random variable with mean $\mu$ and variance 1, $\bar{X}_n = \frac{1}{n}(X_1+\cdots+X_n)$, use the CLT to approximate the following probability:

$$P(|\bar{X}_n - \mu| \ge \frac{2}{\sqrt {n}})$$

My attempt:

$$P(|\bar{X}_n - \mu| \ge \frac{2}{\sqrt {n}}$$ $$=P(\sqrt n|\bar{X}_n - \mu| \ge 2) \;\;\;\text{by multiplying both sides by $\sqrt n$}$$ $$=P(|\sqrt n \bar{X}_n - \sqrt n\mu| \ge 2)$$

Let $Z = \sqrt n \bar{X}_n - \sqrt n \mu$, by CLT, we know that $\sqrt n \bar{X}_n - \sqrt n \mu \rightarrow N(,)$

This is the part where I'm not sure how to figure out what $\sqrt n \bar{X}_n - \sqrt n \mu \rightarrow$ to

Is it $N(\mu, 1)$ since we are given i.i.d random variable with mean $\mu$ and variance 1 or $N(0, 1)$ by the definition of CLT?

If anyone could help me in determining how to find what $\sqrt n \bar{X}_n - \sqrt n \mu \rightarrow$ to or in general how to determine by CLT the approximation normal distribution for questions similar to this that would be really appreciated.

$\endgroup$
3
$\begingroup$

Suppose $X_1,\ldots,X_n$ are uncorrelated (a weaker assumption than independence) and all have the same expected value $\mu$ and the same variance $\sigma^2<\infty$ (a weaker assumption that identical distribution). Then $$ \operatorname{E}\left( \frac{X_1+\cdots+X_n} {\sqrt n} - \mu \right) = 0 \text{ and } \operatorname{var}\left( \frac{X_1+\cdots+X_n} {\sqrt n} - \mu \right) = 1. $$ You don't need to know anything about the normal distribution, let alone the central limit theorem, in order to show that much.

If in addition $X_1,\ldots,X_n$ are normally distributed and independent, then so is $\dfrac{X_1+\cdots+X_n} {\sqrt n} - \mu$. And if they are not normally distributed but are independent, then the distribution of $\dfrac{X_1+\cdots+X_n} {\sqrt n} - \mu$ approaches a normal distribution as $n\to\infty$, and that normal distribution has the same expected value and the same variance.

$\endgroup$
  • $\begingroup$ You meant the variance is equal to $\sigma^2$ $\endgroup$ – lcv Dec 18 '15 at 8:47
0
$\begingroup$

EDIT: Since last I read this, there has been a change in the value of $\mu$. So, I will let $\bar X = \bar X_n$. I think you can determine that $\mu = E[\bar X] $, and $\sigma^2 = \text{Var}[\bar X] = \frac{1}{n}$. Then \begin{align*} P\left[|\bar X - \mu| > \frac{2}{\sqrt n}\right] &= P\left[\bar X-\mu > \frac{2}{\sqrt n} \cup\bar X-\mu <\frac{-2}{\sqrt n}\right]\\ &=P\left[\frac{\bar X - \mu}{\sqrt{1/n}}>\frac{(2+\mu)-\mu}{\sqrt{1/n}\sqrt{n}}\right]+P\left[\frac{\bar X - \mu}{\sqrt{1/n}}<\frac{(-2+\mu)-\mu}{\sqrt{1/n}\sqrt{n}}\right]\tag{1}\\ &=P\left[\frac{\bar X - \mu}{\sigma}>2\right]+P\left[\frac{\bar X - \mu}{\sigma}<-2\right]\\ &=P(Z>2)+P(Z<-2)\\ &=0.04550026 \end{align*} where $(1)$ is true since the events are disjoint, and $Z$ is a standard normal distribution. The CLT comes into play since for large $n$ $$\frac{\bar X -\mu}{\sigma}\to Z.$$

$\endgroup$

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.