If an item follows normal distribution, average also follows normal distribution. What about minimum and maximum?

  • You might want to look into this book. – mpiktas Apr 18 '11 at 6:57
  • 1
    @user4211, do you ask about distribution of minimum and maximum of any sample distribution, or only normal? – mpiktas Apr 18 '11 at 6:58
up vote 13 down vote accepted

You should have a look at the order statistics. Here is a very brief overview.

Let $X_{1}, \ldots X_{n}$ be an i.i.d. sample of size $n$ drawn from a population with distribution function $F$ and probability density function $f$. Define $Y_{1}=X_{(1)}, \ldots, Y_{r} = X_{(r)}, \ldots, Y_{n}=X_{(n)}$, where $X_{(r)}$ denotes the $r$th order statistic of the sample $X_{1}, \ldots X_{n}$, i.e., its $r$th smallest value.

It can be shown that the joint probability density function of $Y_{1}, \ldots, Y_{n}$ is

$f_{X_{(1)}, \ldots, X_{(n)}}(y_{1}, \ldots, y_{n}) = n! \prod_{i=1}^{n} f(y_{i})$ if $y_{1} < y_{2} < \ldots < y_{n}$ and $0$ otherwise.

By integrating the previous equation we get

$f_{X_{(r)}}(x) = \frac{n!}{(r - 1)! (n - r)!} f(x) (F(x))^{r-1} (1 - F(x))^{n - r}$

In particular, for the minimum and maximum, we respectively have

$f_{X_{(1)}}(x) = n f(x) (1 - F(x))^{n-1}$

$f_{X_{(n)}}(x) = n f(x) (F(x))^{n - 1}$

  • +1, I've edited a small mistake in the second last formula. – mpiktas Apr 18 '11 at 6:59
  • @mpiktas: Thank you for the edit! – ocram Apr 18 '11 at 7:01
  • Thanks ocram, the answer is impressive so I checked as good answer but now can you make it in plain english thanks :) By the way how do you put equation in stackexchnage ? – user4211 Mar 5 '12 at 8:07
  • What do you mean exactly? You asked for the pdf's of the minimum and of the maximum, and these two are given by $f_{X_{(1)}}$ and $f_{X_{(n)}}$, respectively. So, if you draw many many samples and compute the min for each, then you end up with a random variable with pdf $f_{X_{(1)}}$. Is it ok? – ocram Mar 6 '12 at 9:21

You might also want to read up on the generalized extreme value (GEV) distribution. It turns out that as $n\rightarrow\infty$, the (shifted and scaled) distribution of the maximal value of the sample converges to one of the three special cases of the GEV distribution.

  • Great link will read it – user4211 Mar 5 '12 at 8:08

The sum of Gaussians is Gaussian. That is why the average is normal. The distribution of any non-linear function of (finitely many) Gaussians need not be Gaussian, and it usually isn't. Such is the case of the maximum function. To approximate the maximum of a multivariate Gaussian, Hothorn is a good place to start.

  • very interesting will read hothorn – user4211 Mar 5 '12 at 8:08

Your Answer

 
discard

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.