Why would a statistician need to know how to do delta - epsilon proofs?