Quote:
Originally Posted by Sol Reader
btw I would like to add, iirc, the higher your winrate, the lower your sample is required for your sample winrate to be close to your true winrate (% wise). I'm not a math buff, but I'm fairly certain.
As such, someone who has 8-12bb/100 might want like a 500k+ hand sample to know what his winrate is, but someone with 30+bb/100 can probably be sure his winrate is around that range with a much smaller sample.
This is sort of true.
There is nothing inherent about the math that makes an observed win rate of a higher magnitude inherently more dependable.
However, when we make observations for an observed winrate of X bb/hr with a sample size Y hours, we get a prediction of an actual winrate with P% confidence level of X bb/hr +/- Z bb/hr.
Z is a number, not a percentage, so it appears less significant as X increases. But the prediction still has the same spread.
FOr example, lets say my observed winrate is 2bb/hr. I take my sample size and determine that this is within 5 bb/hr of my expected winrate 90% of the time. This tells me little about my prospects as a player, since i might be a 3bb/hr loser or a 7bb/hour winner.
Now lets say that for the same sample my observed winrate is actually 500 bb/hr. Now the +/- 5bb/hr makes little difference, since I am crushing the game one way or the other. However, it still takes the same sample size to get me to the same confidence level of 90% +/- 5bb/hour.
That is to say, a higher observed winrate isnt inherently a more accurate predictor of true expectation.
Quote:
Originally Posted by parisron
300 hours is nothing live, I have had 100-200 hour break even stretches over the course of years playing live. You need a large sample (at least 1000 hours) with some run bad and/or long break even stretch mixed in to ball park long term winrate and even that isn't enough.
OK pet peeve. It is not nothing. We might not be able to pinpoint a players expected winrate to the level of accuracy you would be comfortable with, but with a 300 hour sample size, we have inherently more information than we did with a 0 hour sample size.
This might seem like an incredible piece of nittery, but i have sort of a point. Some random guy wants to play for a living. He has to decide what level to play at. He has to make some sort of estimate as to what his hourly rate might be. Saying "300 hours is nothing" means that he plays for months and in your mind should not even begin to come to a conclusion on whether he is a profitable player.
In a game inherently based upon statistics, we make predictions and decisions based upon the information available to us -- even when that information might not be as complete or as informative as we would like it to be.
Last edited by AEPpoker; 08-18-2012 at 09:48 AM.