In my work I've come across the interesting binomial identity $$ \sum_{n\geq k} \frac{\binom{n}{k}}{\binom{m-1}{k}} \frac{\binom{m-1}{n} \binom{i-m-1}{j-n-1}}{\binom{i-2}{j-1}} = \frac{\binom{j-1}{k}}{\binom{i-2}{k}}. $$
That is, for some non-negative integer $k$ and hypergeometrically-distributed $n$, the expected value of $\binom{n}{k}/\binom{m-1}{k}$ has this form.
I'm sure it will fall to a tedious inductive proof or some generating function technique, but I was wondering if anybody knew of (a) a source that describes this identity or (b) a slick combinatorial or counting proof of such a statement?
Note that $$ \frac{\binom{n}{k}}{\binom{m-1}{k}}\binom{m-1}{n} =\frac{ (m-k-1)!}{(n-k)!(m-1-n)!}=\binom{m-k-1}{n-k} $$ Thus $$\eqalign{ \sum_{n\geq k}\frac{\binom{n}{k}}{\binom{m-1}{k}}\binom{m-1}{n}\binom{i-m-1}{j-n-1} &=\sum_{p\geq0}\binom{m-k-1}{p}\binom{i-m-1}{j-k-1-p}\cr &=\binom{i-k-2}{j-k-1} } $$ Where we used the fact that $$ \sum_{p\geq0}\binom{a}{p}\binom{b}{c-p}=\binom{a+b}{c} $$ Since this is the coefficient of $X^c$ in the product $(1+X)^a(1+X)^b$. The final step is to note that $$ \frac{\binom{i-k-2}{j-k-1}}{\binom{i-2}{j-1}}= \frac{\binom{j-1}{k}}{\binom{i-2}{k}} $$ which is straightforward.$\qquad\square$