Hi crazygringo,

I was intrigued by your estimation procedure and was thinking of a way of playing around with it:

(1) Create a random bit vector of size 45,000. (2) Pick 40 positions in the vector at random and use those positions to estimate the proportion of ones in the original bit vector (so far this is easy). (3) Select an additional 120 positions and use the more sophisticated procedure to refine the estimate from part (2).

I was wondering if you have code or pseudo-code for how you implemented part(3) specifically how you choose the 120 words for extra testing. It seems there are a lot of different natural ways you can do it. Did you write an academic paper about this?

Thanks!