0
$\begingroup$

Let $\bar{X}_n = X_1 + \dots X_n$ where $X_i \sim N(0,1)$. We can easily verify that $\bar{X}_n \sim N(0, 1/n)$.

Thus $\text{Var}(\bar{X}_n) = 1/n$.

Let the density of $X \sim N(0,1)$ be denoted $\phi(x) = \frac{1}{\sqrt{2 \pi}} e^{-\frac{x^2}{2}}$.

Apparently we can write express $\bar{X}_n = \frac{1}{\sqrt{n}} Z$ where $Z \sim N(0,1)$ is the standard normal distribution.

My attempt so far is to note that $\phi(\bar{x}_n) = \frac{1}{\sqrt{2 \pi} \frac{1}{\sqrt{n}}} e^{-\frac{x^2}{2 \frac{1}{n}}}$

We can then define $Z=\frac{\bar{X}_n - \mu}{\sigma} = \frac{\bar{X}_n}{\frac{1}{\sqrt{n}}}$ as usual. Then be substiution we obtain:

$\phi(\bar{x}_n) = \frac{1}{\frac{1}{\sqrt{n}}} \phi(z) = \sqrt{n} \phi(z)$

Why do I keep getting the factor of $\sqrt{n}$ on the numerator rather than the denominator?

Thanks

$\endgroup$
1
$\begingroup$

You denote the prob. density by $\phi(x)$, I prefer $f_X(x)$.

If you like to rewrite this in terms of a new random variable $z$ you have to use a transformation, i.e. $$f_X(x) = f_Y(y) \left| \frac{dx}{dy}\right|$$ This is valid for continuous random variables, not for discrete random variables.

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

You have several errors caused by abuse of notations. First, let's define sample mean correctly: $$\bar{X_n}=\frac{1}{n}\sum_{i=1}^n X_i$$ Variance and mean are correct. The density of $\bar{X_n}$ can be correctly written as $p_{\bar{X_n}}(x)$ (or $\phi_{\bar{X_n}}(x)$ if you want, but I'll reserve $\phi$ for standard normal distribution). Your notation uses $\phi(\bar{x_n})$, as if you replace $x$ by $\bar{x_n}$ in the definition of $\phi(x)$. But, if we replace $\phi(\bar{x_n})$ by $p_{\bar{X_n}}(x)$, your density function for $\bar{X_n} $is correct, i.e. $$p_{\bar{X_n}}(x)=\frac{\sqrt{n}}{\sqrt{2\pi}}e^{-\frac{nx^2}{2}}$$ This can be written in terms of $\phi(x)$: $$p_{\bar{X_n}}(x)=\sqrt{n}(\sqrt{2\pi})^{n-1}\left(\frac{1}{\sqrt{2\pi}}e^{-\frac{x^2}{2}}\right)^n=\sqrt{n}(2\pi)^{(n-1)/2}\phi(x)^n$$

Another comment on how you're trying to relate the two: $\sqrt{n}\phi(z)$ is not equal to $\phi(\bar{x_n})$; first of all the variables are different, and if they are PDFs, then their integral should be $1$, but due to $\sqrt{n}$ one of them is not.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks. I follow your result that the x pdf can be written in terms of $\phi(x)^n$ but how does this show $\bqr{X}_n = n^{-1/2} Z$? See second to last line of page 1 in these notes for the claim: www.stat.cmu.edu/~larry/=stat705/Lecture2.pdf $\endgroup$ – user11128 Feb 24 '19 at 10:53
  • $\begingroup$ This doesn't show $\bar{X_n}=\frac{1}{\sqrt{n}}Z$, but I tried to answer your title question. For this one, in the notes, $\bar{X_n}$ is defined that way via $\overset{d}=$ operator. So, it's basically saying that if $\bar{X_n}$ is Normal with $(0,1/n)$, then a standard normal RV can be defined such that $\bar{X_n}=\frac{1}{\sqrt{n}}Z$. $\endgroup$ – gunes Feb 24 '19 at 11:10
  • $\begingroup$ Basically just applying Var(cX) = c^2 Var(X)? $\endgroup$ – user11128 Feb 24 '19 at 11:13
  • $\begingroup$ Yes, and it also applies the fact that normal RVs are still normal under linear transformation. Other RVs may not satisfy this property. For example, if $X$ is Bernoulli, $2X$ is not Bernoulli. $\endgroup$ – gunes Feb 24 '19 at 11:16

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.