4
$\begingroup$

Suppose I have a discrete random variable $X$ which follows a geometric distribution on $x=0,1,2,...$ and I take a random sample from this distribution of size $n$. What is the asymptotic distribution of $\bar X$?

I already know that $E(X)=\frac{1-p}{p}$ and $V(X)=\frac{1-p}{p^2}$.

This seems like an application of central limit theorem, so I'm sure that $\bar X$ converges to a normal distribution. However, the part that's tripping me up is calculating the mean and variance of the normal distribution that it's converging to. How do you do this?

$\endgroup$
2
$\begingroup$

Note that $$ E\bar{X}=E\left(n^{-1}\sum_{i=1}^nX_{i}\right) =n^{-1}\sum_1^nEX_i =n^{-1}(nEX)=EX $$ since the $X_i$ are identically distributed. Similarly, $$ V(\bar{X})=V\left(n^{-1}\sum_{i=1}^nX_{i}\right) =n^{-2}\sum_1^nV(X_i) =n^{-2}(nV(X)) =n^{-1}V(X) $$ since the $X_i$ are independent and identically distributed.

$\endgroup$
  • 1
    $\begingroup$ If we're considering asymptotic convergence, then we're considering the limit as $n$ goes to infinity. So if that is the calculated mean and variance of the normal distribution that it's converging to, then how can we say: $X$~$N(E(X),\frac{V(X)}{n})$ when n is suddenly part of one of the parameters of the distribution? I'm not understanding. Or does this mean that the variance of the normal distribution that it converges to is zero? $\endgroup$ – jippyjoe4 Feb 14 '18 at 23:03
  • $\begingroup$ @jippyjoe4 : In MathJax one writes $X\sim N(E(X), \frac{V(X)} n),$ not $X$~$N(E(X), \frac{V(X)} n),$ i.e the binary relation symbol should be inside of MathJax, not outside. $\endgroup$ – Michael Hardy Feb 14 '18 at 23:35
  • $\begingroup$ Thanks for the tip. But I still don't understand what distribution, exactly, $\bar X$ is converging to from this answer. $\endgroup$ – jippyjoe4 Feb 15 '18 at 4:47
1
$\begingroup$

You have $$ \operatorname E( \, \overline X \, ) = \operatorname E\left( \frac {X_1+\cdots+X_n} n \right) = \frac 1 n \left( \operatorname E(X_1) + \cdots + \operatorname E(X_n) \right) = \frac 1 n \cdot n \operatorname E(X) = \operatorname E(X) $$ and \begin{align} \operatorname{var}\left( \sqrt n \cdot \overline X \right) = \operatorname{var} \left( \frac{X_1 + \cdots + X_n}{\sqrt n} \right) = \frac 1 n \left(\operatorname{var}(X_1) + \cdots + \operatorname{var}(X_n) \right) = \frac 1n \cdot n \operatorname{var}(X) \end{align}

So $\displaystyle \dfrac{\overline X - \operatorname E(X)}{\sqrt n}$ has expected value $0$ and variance $1,$ and approaches $N(0,1)$ as $n\to\infty.$

$\endgroup$
  • $\begingroup$ So then I understand what the mean of $\bar X$ converges to as $n$ goes to infinity; but what happens to the variance? Remember, I'm concerned specifically with the asymptotic distribution of $\bar X$, not any other function of it. $\endgroup$ – jippyjoe4 Feb 15 '18 at 4:49

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.