7

Suppose I pick $k$ integers without replacement from $\{1, \ldots, n\}$. Let $I$ be the value of the highest integer. A calculation with binomials reveals $$E[I] = \frac{k}{k+1}(n+1)$$ This is a very simple formula - does it have a simple calculation-free proof?

1 Answers1

6

Picking $k$ integers without replacement from $\{1, \ldots ,n\}$ involves breaking the interval $[0,n+1]$ into $k+1$ pieces where the length of each piece has the same distribution.

So the expected length of each piece (including the piece from the highest sampled integer to $n+1$) is $\dfrac{n+1}{k+1}$ and so the expected value of the highest sampled integer is $n+1 - \dfrac{n+1}{k+1}$.

Simple calculation will give your result.

So in general, the expected value of the $j$th sampled integer (counting from the bottom) is $j \dfrac{n+1}{k+1} $.

Henry
  • 157,058
  • Sorry by my ignorance, but I don't get it: what is the "same distribution" of each piece? (I don't understand the expected length stuff either, but hopefully this will be cleared when I get illuminated about the "same distribution" issue) Thanks. – Matemáticos Chibchas Feb 07 '13 at 20:08
  • Very nice! But how do you show that are no "edge effects," i.e., that the length of the first piece has the same as the length of the second? – Michael S. Feb 07 '13 at 22:04
  • @Matemáticos Chibchas: Each length is a random variable with a distribution. Each of these distributions is the same and has the same mean. – Henry Feb 08 '13 at 07:35
  • @pierre: Imagine seating $k+1$ people round a table with $n+1$ seats: there are no edge effects in the gaps between people. Then cut the table where the first person sits and straighten it into a bench of length $n+1$. – Henry Feb 08 '13 at 07:39
  • Lovely. Thank you for this fantastic answer. – Michael S. Feb 08 '13 at 23:41
  • @Henry Can you explain me please how one can conclude in an intuitive fashion that the distributions of the lengths are the same? If $X^{(1)}<X^{(2)}<\cdots<X^{(k)}$ are the elements of the $k$-sample, then your claim is that the random variables $X^{(i)}-X^{(i-1)}$, for $i=1,\dots,k$, are equally distributed (here $X^{(k+1)}=n+1$). Of course, this can be proved via calculations, but I would be glad to see an intuitive explanation. – Matemáticos Chibchas Feb 09 '13 at 10:06
  • @Matemáticos Chibchas: see my response to pierre. Initially the distributions are the same because of rotational symmetry. – Henry Feb 09 '13 at 21:02