Edged out

I think I have a reasonable progression of logic that gets r/(r-1) in front of my sample variance. It's not formal proof by any means, but I don't think that's necessary since any competent statistician knows you have to do that to remove the bias of the sample moments.

Meanwhile, I shored up the arguments quite a bit and streamlined the discussion. I now have a section that is somewhat readable.

There are still some holes:

• I've pointed out that the edge case where the sampled blocks return no rows is pathological when using the most straightforward estimators. I haven't shown that my enhanced estimators do any better.
• I don't have any empirical results (or even a data set to get such results) for the most general case. That's my priority for tomorrow.
• There are still a few gaps in the logic that may or may not need closing - I'll let my adviser make that call.
• My list of references is still way shorter than I'd like.

So, there's work to do but, dang, this is starting to look like a real paper.