Assessment

[imported_CaseClosed326]

I just thought I would chime in on this thread even though I probably have nothing of true substance to add.

I really like your posts, you give a lot of good information and seems like to play a good game. I hope you can get some control and play within your bankroll. Good luck.

[mmbt0ne]

[ QUOTE ]
[ QUOTE ]
and we shouldn't be assuming a normal distribution to analyze the math in poker.

[/ QUOTE ]

why not?

[/ QUOTE ]

Shill for some reason decided to be really smart and pretty much say what I was going to say, but I'll add some things I have absolutely no mathematical evidence for.

For starters, the standard deviation needs a LOT of hands to actually be accurate. I don't mean like 50k, 100k, I mean like A LOT. mike l. has talked about how he believes winrate doesn't converge until about 700k hands. I don't know if I believe that, but I wouldn't be surprised at all to see the SD number at or above that.

A few reasons. First, the extremes of 100 hand samples are so much to make the SD value very inconsistent. Sometimes, you're playing 35 hands out of 100, and winning 50BBs with them. Sometimes, you play 9 hands out of 100, and lose the blinds each orbit, so you end up down 10BB or so. Sometimes you play more hands, and run like [censored] and drop more BBs.

Secondly, you aren't playing against the same people every time. Each time you sit down you aren't playing with the same lineup, in the same seats, with the same strategy as last time. All of these variables put together make all calculations much more imprecise than a lot of people give them credit for.

Thirdly, players almost always start off in a very tight, A-B-C poker style that limits the variance. As they move up, they begin to add more complex plays, in depth reads, adaptions to other people's constantly changing play, etc. At this point, they already think that they have a certain standard deviation, as they've put in quite a few hands, and PT is giving them a nice number they can look at and feel good about at night. However, as they're adding these other, more variance prone, plays they are actually playing at a much higher variance than they expect. Obviously, they don't see a change in their SD number either, since this sample is so miniscule compared to the rest of their database.


Anyway, as for the normal distribution, I still haven't seen any data that shows winrates are distributed in such a way. And honestly, it's probably impossible to get that data considering we aren't going to see the 5bb/100 losers with 70k hands logged in PT. I mean, if you really want to approximate something, go ahead, use it. But, I hope these people can understand that it's not that good of a guess.

[stinkypete]

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
and we shouldn't be assuming a normal distribution to analyze the math in poker.

[/ QUOTE ]

why not?

[/ QUOTE ]

Shill for some reason decided to be really smart and pretty much say what I was going to say, but I'll add some things I have absolutely no mathematical evidence for.

For starters, the standard deviation needs a LOT of hands to actually be accurate. I don't mean like 50k, 100k, I mean like A LOT. mike l. has talked about how he believes winrate doesn't converge until about 700k hands. I don't know if I believe that, but I wouldn't be surprised at all to see the SD number at or above that.

[/ QUOTE ]

whether or not you can determine your standard deviation accurately doesn't change the fact that results approach exactly a normal distribution as hands approach infinity, assuming that you don't change the way you play (and when you're talking about winrate in the first place, you're making that assumption). even if your winrate and sd change, it'll still look a whole lot like a normal distribution.

i think you'd be surprised at how close you get to a normal distribution even if you're considering 100 hand blocks. if someone has a database with 200k+ hands it would be pretty easy to graph it and demonstrate it. (it won't be solid proof of course, but it'll illustrate the point)

if you're going to be analyzing the math in the first place, the normal distribution is the model to use if you're going to analyze this, and i assume you do want to analyze it since you didn't flat out say "and we shouldn't be analyzing the math in poker"

[ QUOTE ]
A few reasons. First, the extremes of 100 hand samples are so much to make the SD value very inconsistent. Sometimes, you're playing 35 hands out of 100, and winning 50BBs with them. Sometimes, you play 9 hands out of 100, and lose the blinds each orbit, so you end up down 10BB or so. Sometimes you play more hands, and run like [censored] and drop more BBs.

[/ QUOTE ]

i don't see how any of this suggests anything but that variance is high. it certainly doesn't do anything to suggest that a normal distribution is a poor model.

[ QUOTE ]
Secondly, you aren't playing against the same people every time. Each time you sit down you aren't playing with the same lineup, in the same seats, with the same strategy as last time. All of these variables put together make all calculations much more imprecise than a lot of people give them credit for.

[/ QUOTE ]

there's no reason why these things can't be considered a natural part of the variance.

[ QUOTE ]
Thirdly, players almost always start off in a very tight, A-B-C poker style that limits the variance. As they move up, they begin to add more complex plays, in depth reads, adaptions to other people's constantly changing play, etc. At this point, they already think that they have a certain standard deviation, as they've put in quite a few hands, and PT is giving them a nice number they can look at and feel good about at night. However, as they're adding these other, more variance prone, plays they are actually playing at a much higher variance than they expect. Obviously, they don't see a change in their SD number either, since this sample is so miniscule compared to the rest of their database.

[/ QUOTE ]

you're right. when you change your game, your results will no longer follow the same normal distribution. they will still follow some normal distribution. the normal distribution that you can use to model your play will change as you improve. this is obvious. but again - there's no better model. are you just saying that we shouldn't try to analyze these things at all?

[ QUOTE ]
Anyway, as for the normal distribution, I still haven't seen any data that shows winrates are distributed in such a way. And honestly, it's probably impossible to get that data considering we aren't going to see the 5bb/100 losers with 70k hands logged in PT. I mean, if you really want to approximate something, go ahead, use it. But, I hope these people can understand that it's not that good of a guess.

[/ QUOTE ]

no, it won't be that good of a guess. for all the same reasons that saying your expectation in a game is 2.5BB/100 because you've beat it for 2.5BB/100 over 100k hands isn't that good of a guess. but people do want to model their winrates. why not model the variance along with it?

[stoxtrader]

I always kind of assumed the kurtosis of this distribution was higher than in a normal distribution.

seriously.

[stinkypete]

[ QUOTE ]
I always kind of assumed the kurtosis of this distribution was higher than in a normal distribution.

seriously.

[/ QUOTE ]

and that's a correct assumption. just look at the distribution of a single hand and it's obvious.

despite that though, as the number of hands increases, the kurtosis approaches that of the normal distribution.

[CallMeIshmael]

You CAN use the normal to approximate.

By the CLT, anything that is sampled repeatedly many times converges to normal. Even if what is being sampled isnt normal (this is why the 'you can only lose 12BB but make much more in any hand' argument isnt valid)

[kurosh]

I played some 2/4 and 3/6, ran bad to $800. I decided to play $22 SNGs. I ran bad some more, lost 7 or 8 in a row and at one point, my balance was around $600. Then I won and won and won and won. I won 3 in a row, got 2nd and 3rd in the next two. I played the rest of the day and finished at around $1400. I went back to where I feel my game is strongest, SH limit and set a loss limit of $400 for myself at 3/6 6max. I ran good. A few days and a couple thousand hands later, I had 3k.

I felt like I was playing very well. I was betting and folding at all the right times. I took the 3k and decided to take a 50BB shot at 5/10. I came up a few hundred at first, but then got crushed badly and quit at around 2500. I've been grinding at 3/6 for a bit and, counting rakeback, I will have 4250. I think I will take another shot at 5/10 soon.

Thanks for all the support.

[pzhon]

[ QUOTE ]
I've played SNGs and they have the highest variance except for MTTs. You can read threads about it in the SnG forum.

[/ QUOTE ]
Most of those threads are filled with ignorance. People whine about variance in SNGs even when it is significantly lower than the variance they would experience in limit ring games.

Of course, if your ROI/win rate is low, then you will experience larger swings. This is true if you are a marginal winner in any form of poker. It is behind the constant assertions that the variance is huge in 6-max games despite contradictory evidence from PokerTracker.

Many 1TT forum regulars play on Party, where the rapid blind increases decrease the advantage of winning players but shorten the tournaments, and they have adopted marginally winning styles they can execute while playing 8+ tables. These choices increase the sizes of swings, but they are not universal. For me, SNGs have about the same variance as NLHE, and significantly less than LHE.

[kurosh]

I started playing some shorthanded and HU 5/10. I run good and play a lot. I've been playing, at a minimum, 1k hands a day. My roll is now a healthy 7k online so that's plenty for 5/10 and hopefully will be enough for 10/20 again soon.

Yay!

[jba]

[ QUOTE ]
I started playing some shorthanded and HU 5/10. I run good and play a lot. I've been playing, at a minimum, 1k hands a day. My roll is now a healthy 7k online so that's plenty for 5/10 and hopefully will be enough for 10/20 again soon.

Yay!

[/ QUOTE ]

this is awesome to hear, keep it up