Project - Compiled stats for Short Handed Games

[CallMeIshmael]

[ QUOTE ]
with enough hands you can get a decent idea about some more of the obscure stats. i have no idea off the top of my head what your win at sd when calling a river bet should be. if i could look at stats over like a million hands of 20-30 different 1bb+ winners, i'd be interested in seeing what that average number is.

[/ QUOTE ]


This is exactly the point


Leader seems to believe that there is common knowledge on a lot of these stats. And, if he is right, then there is little merit in doing something like this.

But, part of me finds it hard to believe that this is true of some of the less discussed stats for many, and even some of the main ones for newer players


(also, just to note, average values for something like this probaly means less than simply a look at how far spread out the data is. Saying that "the average is X" doesnt say as much about where you should look for possible problems as does "most players fall between Y and Z"

[CallMeIshmael]

Leader,

we should probably draw this discussion to a close, but there is one thing in your post that really needs to be discussed

[ QUOTE ]
that the optimal VPiP is between 25-33

[/ QUOTE ]

[ QUOTE ]
Our purpose is to achieve the tightest range possible.

[/ QUOTE ]

NO IT ISNT

Our purpose is to look at how people are currently playing. No one is looking to keep the ranges super tight, or find optimal values of ANYTHING.


If the data shows that people are winning with a stat that varies from 30% to 50%, then it shows that people can vary that up quite a bit, and still come out ahead.



Also, "I'm working to develop a method to use past observation to find correct game theoretical strategies in poker games"

Ive done quite a bit of work on this stuff myself. PM me if you wanna compare/discuss methods of abstraction/equil. finding, etc


EDIT:

also to note, wrt to:

[ QUOTE ]
It's almost certain that the optimal VPiP is between 25-35 for 4-6 handed based on various factors. PFR between 17-23. AF between 1.8-2.5. FtRB between 30-40. Other stats are less certain

[/ QUOTE ]

you are aware, that you arent in the target audience, right?

[CallMeIshmael]

[ QUOTE ]
Why should we care about optimal play?

[/ QUOTE ]

Optimal play is a good foundation upon which to derive maximal play. And, obvious academic interest. But, I never said optimal play was *that* important to learn.


[ QUOTE ]
Further, many of the PT stats that are often asked about are not independent variables.

[/ QUOTE ]

Yeah, I put that in a disclaimer in the other thread.

[ QUOTE ]
What is averaging my stats with some one, like say Chezlaw, who is also a +EV player playing a VERY different style going to tell us?

[/ QUOTE ]

Again, averages probably arent the most important part. However, it WILl tell people that there are winning players playing vastly different WTSDs,etc.


[ QUOTE ]
If all that is being suggested is a survey of typical ranges for winning players of various stats, I guess that's ok as far as being a quick diagnostic for if someone is doing something horrendously wrong.

[/ QUOTE ]

Right. That was the whole point. Its not meant to help anyone who already has a pretty good grasp of the basics. Its meant as an easy reference for people who are new to poker/PT and dont quite have the common knowledge of the regulars yet.

[Leader]

[ QUOTE ]
Leader,

we should probably draw this discussion to a close, but there is one thing in your post that really needs to be discussed

[ QUOTE ]
that the optimal VPiP is between 25-33

[/ QUOTE ]

[ QUOTE ]
Our purpose is to achieve the tightest range possible.

[/ QUOTE ]

NO IT ISNT

Our purpose is to look at how people are currently playing. No one is looking to keep the ranges super tight, or find optimal values of ANYTHING.


If the data shows that people are winning with a stat that varies from 30% to 50%, then it shows that people can vary that up quite a bit, and still come out ahead.

[/ QUOTE ]

Could be that or it could be that the stat hasn't converged and, in fact, every one you surveyed has a true percent of 39-41 with a SD of 5. Anyway enough of that.

[ QUOTE ]
Also, "I'm working to develop a method to use past observation to find correct game theoretical strategies in poker games"

Ive done quite a bit of work on this stuff myself. PM me if you wanna compare/discuss methods of abstraction/equil. finding, etc

[/ QUOTE ]

I will certainly take you up on this though I first need to complete my background research on game theory.

[ QUOTE ]
EDIT:

also to note, wrt to:

[ QUOTE ]
It's almost certain that the optimal VPiP is between 25-35 for 4-6 handed based on various factors. PFR between 17-23. AF between 1.8-2.5. FtRB between 30-40. Other stats are less certain

[/ QUOTE ]

you are aware, that you arent in the target audience, right?

[/ QUOTE ]

I do. However there are quite a few posters around that have this knowledge. If people are interested in acquiring it they can just search for stat threads myself other vets have replied to.

[sethypooh21]

[ QUOTE ]

Again, averages probably arent the most important part. However, it WILl tell people that there are winning players playing vastly different WTSDs,etc.


[/ QUOTE ]

Again, and not to nitpick, but I think this is part of the problem with respect to the target audience - when you divorce a stat like WTSD from the context in which it exists in 'real world' terms, you're quite likely to reach unfortunate conclusions. The truly important thing is to have a properly integrated style, and examining each stat in isolation is going to detract, rather than assist in that goal.

[CallMeIshmael]

Leader,

I got to thinking about this discussion today, and had some free time this afternoon, and wrote a little program to simulate some stats


Essentially, my opinion, with regard to the debate about using a method like that in the OP vs looking at the stat posts of good players over larger samples of data, is that the method of using the more accurate data sets overshoots the need for accuracy and underestimates the need for a larger sample size of *players.*

I had said this above, and you said you disagreed. And, at that point, there really wasnt much to say, since its a quantitative debate and neither of us really had any good quantitative data.


So, I set out to try to get an idea of how the different methods of getting ideas for theses stats compares.


I came up with the following model:


1. Assume comparision sample size (I used 25 vs 5)

2. Assume hand sample size, randomly sampled from (30k,100k) or (90k, 130k)
(ie. the 25 players have played 30k-100k hands, the 5 players have all played a ton of hands)

3. Assume that there is some range that winning players fall into. For example, winning players have a mean of, say, 15 for some stat, with a std of 2.5.

4. Assume that losers have some *other* mean/std for the same stat. (keep in mind, of course, that losing 2p2ers is different than general losers)

5. There is some chance of picking a winner/loser for each method, I used 0.85/0.99 for the two data sets

6. Sample players. So, for each player, flip a coin to see if he is a winner/loser (obv not a fair coin), get his number of random hands from the intervals above, and get his statistic value (from either the winner/loser distribution)

7. Assume this value is his is 'true' statistic, and use this, and a value for the standard deviation of the stat itself (which is obv different than the standard deviation of the stats of winning/losing players) to determine his 'observed' statistic

8. (repeat for all 30 players)

9. Now, using the OBSERVED statistics, try to estimate the mean and range of winning players.

10. Do the entire thing many times, (I did 50)



Results:

25 people:



5 people:



The bar on the left is the mean/2std error bar for the true winning player for the stat.

The 50 bars on the right indicate the mean/2stde error bars for the 50 samples of data obtained for the two methods.

If you notice, the method of including more players with fewer hands and a higher chance of being a loser, gives you a range closer to the true data. It also doenst get it drastically wrong as frequently.

Ironically, the method of including a few players with large data bases, gets the mean of both the mean and std closer than does allowing more players. However, that method does so much worse in the standard deviation of the standard deviation, that the small loss of accuracy is more than made up by the decrease in variance.


I hope this illustrates the concept. When you said above that the common knowledge range for some stat is X-Y, there is a good chance you have either overestimated or underestimated the true deviation of the upper and lower values for what *should* be the common knowledge range. You see the need for converged data, but have overlooked the need for lots of data to get ranges of players. If you know the true value of stat X for 5 players is 37, 38, 40, 39, 38, you might conclude that most players fall into a pretty narrow range, but it ignores the possibility that you have just, by chance, picked 5 players that happen to bunched together in a much wider distrubution.

Beyond that, it should be noted that altering the parameters does, of course, change the results, but not so greatly to change the conclusion.

Now, just to be clear: I am well aware that the model above isnt perfect, and can be nit picked in many areas. But, I think it does a good job in explaining the concept.

EDIT: to note, the above pictures are taken from parameter values that, imo, are probably shifted from reality in favour of the method of using a few large DBs, because the mean/std for losing players is shifted more than it is in reality from that of winning players, and the std for the stat is more than most stats would be.

[tagWAG]

Leader, are these stats really common knowledge? If so where is this knowledge shared? I'm new this game, but some of the recommended numbers from stox's book seem to be less than a perfect match for the recommendations in this link:

[ QUOTE ]
...There is always this too.

http://forumserver.twoplustwo.com/showfl...&PHPSESSID=

[/ QUOTE ]

Leader, do you have any links pls - compiled stats like this may not be perfect, but as someone who wants to get an insta-handle on typical numbers and ranges for winning players, I'd be interested in seeing the results.

[Leader]

This is certainly interesting. I have a few techincal questions.

[ QUOTE ]
Assume that losers have some *other* mean/std for the same stat.

[/ QUOTE ]

What did you use here?

[ QUOTE ]
a value for the standard deviation of the stat itself

[/ QUOTE ]

Did you take this from the binomial distribution?

Also, how did you write the program? Some kind of stat program or was it writen in a programing langauge.

Finally I have to point out that when I brought up those 3 100+k stat posts my point wasn't that they should be averaged. What I meant to impart was that if you want to look at statistics to become a better player, you're better off looking at the stats of a clearly winning player as a whole and trying to derive why that players stats are the way they are rather then the average of many players.

[sethypooh21]

[ QUOTE ]
you're better off looking at the stats of a clearly winning player as a whole and trying to derive why that players stats are the way they are rather then the average of many players.

[/ QUOTE ]

This can't be quoted enough.

[Leader]

[ QUOTE ]
Leader, are these stats really common knowledge? If so where is this knowledge shared? I'm new this game, but some of the recommended numbers from stox's book seem to be less than a perfect match for the recommendations in this link:

[ QUOTE ]
...There is always this too.

http://forumserver.twoplustwo.com/showfl...&PHPSESSID=

[/ QUOTE ]

Leader, do you have any links pls - compiled stats like this may not be perfect, but as someone who wants to get an insta-handle on typical numbers and ranges for winning players, I'd be interested in seeing the results.

[/ QUOTE ]

These will have to do you for now (read the threads all the way through).

http://forumserver.twoplustwo.com/showfl...rue#Post7070792

http://forumserver.twoplustwo.com/showfl...rue#Post6301267

http://forumserver.twoplustwo.com/showfl...rue#Post5473357

http://forumserver.twoplustwo.com/showfl...rue#Post4987345

http://forumserver.twoplustwo.com/showfl...rue#Post4762309