Quote:
Originally Posted by plexiq
I think 2) is missing the point a bit, as the blocker error being smaller than the std reduction is not sufficient at all.
If you reduce std by 10% while introducing a 0.5% bias, that's still a pretty terrible result. Players expect results and a/i EV lines to converge in the long run, if bias is introduced this will no longer be the case and that can be extremely misleading.
A player may be running below all-in EV and conclude he's playing fine and just running bad, when in reality the difference might be completely due to a biased all-in EV calculation. If that's the case he'll just continue "running bad" for eternity unless he changes his game.
Let's look at an example:
Regular SD: +/- 100$ per 100 hands
AIEV SD: +/- 90$ per 100 hands, with a +0.5$ per 100 hands bias
A 0.5$ bias to get a 10$ reduction in SD might look like a reasonable tradeoff on first sight, but let's see how this works out once you play more.
On a 10k sample:
Regular SD: +/- 1000$
AIEV SD: +/- 900$, +50$ bias (50$ bias for 100$ SD reduction)
On a 1mio sample:
Regular SD: +/- 10,000$
AIEV SD: +/- 9,000$, +5,000$ bias (5k bias for 1k SD reduction)
etc.
You are right, my reasoning was wrong.
So if we are winning 0.5bb/100 (which I find a high estimate tough) due to blockers our aiev becomes less accurate than net winnings after 40k hands.
For 6 max, it takes about 160k hands, not counting the fact that as the effect gets smaller we also have a harder time taking advantage of it