# Probability distribution of the median

Suppose we have $2k + 1$ points $a_1, ..., a_{2k+1}$. Each point is uniformly distributed between 0 and N. What is the distribution of the median (i.e. of the k+1-th point) ?

What happens if $a_1, ..., a_{2k+1}$ is a random set of $2k+1$ distinct points from $1, 2, ..., N$ ?

-
-1 because this looks like a homework problem (especially the second question). –  Daniel Moskovich Dec 24 '09 at 0:36
Who assigns homework over Christmas break? –  Greg Kuperberg Dec 24 '09 at 12:54

I assume the points are meant to take integer values between $0$ and $N$.

The second question is easier. The probability that the median is $j$ is $$\binom{j}{k} \binom{N-j}{k} / \binom{N+1}{2k+1}.$$ (I hope I'm not answering homework here. If I am, remember that you need to justify this statement to get full marks.)

The probability that $j/N$ is between $x$ and $x+dx$ is roughly $$\frac{(2k+1)!}{(k!)^2} x^k (1-x)^k dx ,$$ as $N$ goes to infinity with $k$ fixed.

When you allow repetition, you should have the same asymptopics, but the problem is just messy enough that I can't get a simple answer. My best formula is

$$\frac{(2k+1)!}{(k!)^2} \frac{1}{(N+1)^{2k+1}} \int_{x=0}^1 (x+j)^k (N-j+x)^k dx.$$

Bounding the integral by the minimum and maximum values of the integrand gives reasonable bounds. I'll provide a proof if needed.

-

have you checked the "order statistics" article on Wikipedia.

-