Quote:
Originally Posted by grinder4all
So is a sample of 236 games enough to tell you you're a loser (i.e. -17 chips per game) but not enough to tell you you're a winner (i.e. my 400 tournaments with +56 chips per game?).
If so...how does that work?
Is there some maths around that explains precisely how many games you need to have 'x' amount of confidence about your winrate?
As for your last question, see Max Cut's above post.
Imagine that the losing sample of 236 games and the winning sample of 400 games both belonged to you.
Then you'd have two contrasting samples (if they were at the same buy-in level, which I don't remember, tbh) would mean nothing else than that your true winrate would be likely somewhere in between, i.e. you most likely ran cold in the losing sample but hot in the winning one.
Adding those two samples up, we'd get that you won ~18K chips over 636 tourneys (if I'm not mistaken) and overall your observed chip EV would be about 30 per tourney, and we could say with high confidence that 1) it's bigger than 10 chips (thus winning at least a bit post-rakeback), 2) it's lower than 50 chips. As I've heard that crushers win at least 100-ish chips at $2s and 70-ish at $10s, it means that you'd have room for improvement, that's all that I meant.
It's not very clear if KingFiveOff is indeed a loser, but it's quite clear that he's not a crusher and thus can improve his ROI easily by a bit of study.
But regarding your first question, you need to understand that the randomness in Spins is so huge that, over small samples, almost any player is usually in an 'in-between' state where they can't confirm the hypothesis of a being loser but also can't confirm the hypothesis of being a winner.
I.e. in statistics, a question about whether the EV of a random variable is positive with x% confidence often has three possible answers: 'Yes with x% confidence', 'No with x% confidence', and 'We don't know yet'.
Quote:
Originally Posted by ibavly
You need to 4x your sample to halve your confidence interval. So with the diminishing gains of increased volume and the likelihood of your true winrate to change over time, it seems that chipev is a highly flawed measure of skill that we shouldn't put very much stock into, it's simply the best measure we have available right now.
That's very true. I wouldn't rely solely on the chip EV if I were to evaluate potential horses (but I don't do this business). I'd dig deeper into their strategy knowledge, mental game quality, HUD-reading skills, etc.
Last edited by coon74; 01-17-2016 at 02:10 PM.