I mucked around with this and it says my true win rate lies within 21BB/100 and 5.5 BB/100 within a 95% confidence interval.

My inputted data:

hands: 20824

BB/100 13.42

standard deviation: 57.79

I'm no expert on this, but this calculation method seems flawed. Only the very, very best would be able to win at 21BB/100, certainly not some amateur like me, and I'd say that no body could win @ 30BB/100+ @ small stakes with the rake as it is. So since we know it can't get that high, can't we use this to shape the data? Being a decentish reg I'd my win-rate could never be less than -7BB/100 either.

Yes. Having a prior distribution of the populations win-rates (i.e. the average win-rate and the standard deviation) would make it possible. Let's say that the true population win-rate is -1.5BBs / 100 and the standard deviation is 40BB / 100. Let us further assume that win-rates are normally distributed.

Est. True Win-rate =
-1.5 / 40² + 13.42 / 57.79² Divided by
1/40² + 1/57.79²

The standard deviation around this estimate would be:
sqrt[ 1 / (1/40² + 1/57.79²) ]

So for your numbers we would get an estimated true win-rate of 3.33 with a standard deviation of 32.89. Of course, if you have a different population distribution or if certain assumptions cannot be met, these results won't be accurate.

At least this is the sort of calculation I argue for in another thread that no one seems to have destroyed yet: http://forumserver.twoplustwo.com/25...cture-1161363/