Project - Compiled stats for Short Handed Games

[Leader]

[ QUOTE ]
"the WSEx player might be excluded because he's a breaking even player at the table while making 2BB/100 in rakeback. Thus he's by any objective measure a more successful player then someone making 1.4, but he's not included."

These players, IMHO, should be excluded. They are playing in a manner that allows them to break even in their play, while having the house pay them 2bb/100. They are earning an hourly wage for playing poker. Just like a shill.

I don't see why we would be interested in having the stats of shills included with the stats of players earning their big bets on their own.

[/ QUOTE ]

Poker is about making money. If you choose to play in a place that charges less rake, that is a poker decision. The games at that place are likely to be tougher. These players aren't playing to break even. They are playing to win the most money possible in games that aren't beatable for 1.4 BB/100 after the rake.

In the end, poker greatness is about obtaining a large combination of skills some of them executed at the table some away from it. The only objective measure of that skill is how much money the player has put into his pocket. If the greatest tactical player sits in the worst games, he isn't the greatest poker player.

[timoK]

[ QUOTE ]
What we need to do this study is a group of willing, winning players ie. greater than 1.4BB/100 hands and over 20K hands.

[/ QUOTE ]

this samplesize is soooo random.

a friend once showed me some datamining results from stars a few weeks ago:
some of the highly respected posters here showed a winrate of 2BB/100 over 50k hands in one month and in the next month the were at -1 BB/25k hands

20k hands might be ok for swingless full ring, but for shorthanded you better look at 200k hands.
also people are right: you better look at your winrate after Rakeback or Boni-stuff

[CallMeIshmael]

(sorry, was just told about this now)

[ QUOTE ]
I applaud CMI's effort but is method is badly flawed.

[/ QUOTE ]

Do tell.

[ QUOTE ]
I have two semesters of college level statistics and am currently working in a paid research position which will produce a publishable academic paper on survey methodology and coverage bias. So let's say I got you by a nose.

[/ QUOTE ]

And Ive got that by a nose, though a lot of my experience is on the more biometric side of things.

[ QUOTE ]
[ QUOTE ]
Almost everyone who uses PT wants to know what others results are

[/ QUOTE ]

They shouldn't.

[/ QUOTE ]


thats just silly.

When people are first learning to play poker/use PT, knowledge of roughly standard values for some of the bigger stats is certainly something they should want to know about. If people are playing a VPIP 15% above or 10% below the average value for a group of winning players AND this person isnt making money, its a great spot to look to first.



Now, like someone did back in 2005 in a thread leading up to the final post, you've made the mistake of thinking that its meant to be a rigorous statistical study. Its not. Im hoping to publish it. Its not a post of "here is how to play poker."


At the time the post was made, there were a lot of "how are my stats?" posts in the forum. These posts are particularily boring, and tend to lead to little or no meaninful discussion. Having a page of averaged values for a group of winning players is going to put an end to those posts. That was the point of the post. Im not sure if those posts are a problem here, since I dont read this forum.

No one says that you must play within any range of stats. But, if you are looking to plug holes in your game, *glaringly* errant stats are a good place to start. They even allow the player to look for hands to post. For example, if someone is calling the river WAY more than most others, they'd probably like to post a few of their marginal river calls and get feedback.

Of course I couldnt be certain that all the players that submitted were winning players. Though, since nearly all who submitted were choosen by me based on ~1 year of quality posting in addition to the required win rates, I'd bet a large sum of money that a very large % of them were. The parlay of both being able to explain decisions well and beat the games for a longish time despite being a poor player is unlikely.

[Leader]

[ QUOTE ]
(sorry, was just told about this now)

[ QUOTE ]
I applaud CMI's effort but is method is badly flawed.

[/ QUOTE ]

Do tell.

[/ QUOTE ]

I did second post in this thread.

[ QUOTE ]
[ QUOTE ]
I have two semesters of college level statistics and am currently working in a paid research position which will produce a publishable academic paper on survey methodology and coverage bias. So let's say I got you by a nose.

[/ QUOTE ]

And Ive got that by a nose, though a lot of my experience is on the more biometric side of things.

[/ QUOTE ]

Fair enough.

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
Almost everyone who uses PT wants to know what others results are

[/ QUOTE ]

They shouldn't.

[/ QUOTE ]


thats just silly.

When people are first learning to play poker/use PT, knowledge of roughly standard values for some of the bigger stats is certainly something they should want to know about. If people are playing a VPIP 15% above or 10% below the average value for a group of winning players AND this person isnt making money, its a great spot to look to first.

Now, like someone did back in 2005 in a thread leading up to the final post, you've made the mistake of thinking that its meant to be a rigorous statistical study. Its not. Im hoping to publish it. Its not a post of "here is how to play poker."


At the time the post was made, there were a lot of "how are my stats?" posts in the forum. These posts are particularily boring, and tend to lead to little or no meaninful discussion. Having a page of averaged values for a group of winning players is going to put an end to those posts. That was the point of the post. Im not sure if those posts are a problem here, since I dont read this forum.

[/ QUOTE ]

They aren't generally, but in any case, this method still seems very unreliable to me. General ranges for stats are pretty common knowledge IMO and there are several threads linked to in the sticky which provide this info. There's also a thread in which 3 winning mid stakes players post 100+K samples.

[ QUOTE ]
No one says that you must play within any range of stats. But, if you are looking to plug holes in your game, *glaringly* errant stats are a good place to start. They even allow the player to look for hands to post. For example, if someone is calling the river WAY more than most others, they'd probably like to post a few of their marginal river calls and get feedback.

Of course I couldnt be certain that all the players that submitted were winning players. Though, since nearly all who submitted were choosen by me based on ~1 year of quality posting in addition to the required win rates, I'd bet a large sum of money that a very large % of them were. The parlay of both being able to explain decisions well and beat the games for a longish time despite being a poor player is unlikely.

[/ QUOTE ]

Even if we assume that they're winning players, you still have a boatload of problems. In order for an average to mean anything, you have to averaging numbers from the same source. The players are playing in, often times, vastly different games. There's also the problem that many stats haven't converged over 20K or even 40. So whatever average/range you get is going to have a good amount of random error in it.

I'm not going to tell anyone how to use their time, but IMO there's way too much noise in this method for anything much to be concluded from the results.

[CallMeIshmael]

Leader,


I'll reiterate that I think you are too focused on methods that would be applicable in an academic setting. Stat classes are great at telling you what to do, but not when to do it. I cant remember when I read it, but I once heard that a 99% CI is considered very low for some areas of medicine, though its well above the commonly used 95%.
In a case like this, a bit of random noise doesnt have that great of a negative affect.

When you say "In order for an average to mean anything, you have to averaging numbers from the same source," thats wrong. Optimal play is the same at party, stars and AP (save small variations in rake structure). Maximal play might be a bit different because of typical opposition, but not so vastly different that stats like VPIP, VPIP/pos,PFR, etc, are going to so different as to be meaningless. You're still governed by the game structure. How vastly different WPEX is, Im unsure. I dont play there, so I really cant comment)

"There's also the problem that many stats haven't converged over 20K or even 40"

cut your DB into 20k chunks, filter it correctly (obv you cant have one full of 3 handed and one 6 handed) and see how much stuff like PFR, VPIP in positions, etc, vary. BB/100 data varies a ton. A lot of others ones dont.


Beyond that, "General ranges for stats are pretty common knowledge IMO and there are several threads linked to in the sticky which provide this info. There's also a thread in which 3 winning mid stakes players post 100+K samples. "


How can there be common knowledge or, for that matter, merit to these 100k posts, that dont face the same pitfalls you describe? Surely between the choice of 3 highly likely winners with 100k hands, and, say, 20, players, all with a very good shot at being winners, with DBs ranging from 30k to 200k, the latter is going to give people better ideas.


Obviously, if there is already common knowledge of a lot of stats, then compiling a group players #s together is irrelevant. (again, Im unsure of common knowledge on this forum. sparky PMed me asking if I would do this for shorthanded, and normally dont read it). But, Im curious as to the origin of this common knowledge, and if you think there is as much noise in the methods used to obtain it as would be present in a compilation as described in the OP.

Someone linked wookie's stat post, which is great. But Id say the methods of something like the OP would produce data that would be much better than something like that post. And, if that post is part of this common knowledge, then I think your point is incorrect.

(fwiw, Im not really even all that interested in doing this; it was mainly sparky's interest. but, Im still going to defend the merit of the results it would obtain)

[sethypooh21]

CMI,

Why should we care about optimal play? The only reason the games are beatable for most of us is that our opponents are playing in a decidedly sub-optimal style.

Further, many of the PT stats that are often asked about are not independent variables. I have a markedly high WTSD. This is partially a function of me being skeptical, but it's also a function of the fact that I play good values in big pots, as well as probably an empirical effect of the turn-river lines I take in many common spots. 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? The mean is likely to produce the stats of a mediocre, generic, TAG grinder.

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. But Max EV play is so condition dependent that I don't think any sort of compilation is going to tell us a whole lot about anything.

[milesdyson]

when you average a lot of hands the differences in playing styles will be somewhat neutralized. this wasn't supposed to be anything but a quick diagnostic. i don't understand the big deal.

[ QUOTE ]
I have a markedly high WTSD. This is partially a function of me being skeptical, but it's also a function of the fact that I play good values in big pots, as well as probably an empirical effect of the turn-river lines I take in many common spots. 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? The mean is likely to produce the stats of a mediocre, generic, TAG grinder.

[/ QUOTE ]
as opposed to what? hip unbluffable lagtag or solid thinking tag? [img]/images/graemlins/smile.gif[/img] no one's gonna see either of these through your stats, but again that's not the point.

[sethypooh21]

[ QUOTE ]
when you average a lot of hands the differences in playing styles will be somewhat neutralized. this wasn't supposed to be anything but a quick diagnostic. i don't understand the big deal.

[/ QUOTE ]

If that's all it is, then we have no worries, but I wonder how much an actual compilation is going to tell us. I mean, if you asked a random sampling of the 25 better players on this forum to rattle off appropriate ranges for VPIP/PFR/AF WTSD, etc...I think there would be pretty uniform answers...

Also, does this mean you think that I'm a hip, unbluffable LAGTAG? [img]/images/graemlins/grin.gif[/img] (I guess one out of three ain't bad)

[milesdyson]

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.

[Leader]

[ QUOTE ]
Leader,


I'll reiterate that I think you are too focused on methods that would be applicable in an academic setting. Stat classes are great at telling you what to do, but not when to do it. I cant remember when I read it, but I once heard that a 99% CI is considered very low for some areas of medicine, though its well above the commonly used 95%.

[/ QUOTE ]

"Stats class" actually represents a small portion of my knowledge base on statistics. As I said, I do have current real world experience in this area.

Here's a brief summery of my correct research from an email I sent today:

[ QUOTE ]
Since I have some time, I figured I'd send you an update on how my summer research is going. I'm working with a great group of students and professors to try develop techniques to reduce coverage bias in internet surveys.

We're doing this by examining the Oct. 2003 Current Population Survey (CPS) internet and computer supplement put out by the Department of Labor. Our idea is to split the data into three groups, those with internet access, those without internet access, and those with internet access only away from their homes i.e. at work or at a public library.

This last group is key because we want to use it to represent those without internet access. We assume, in essence, that we took an internet survey and received the CPS data not including those without internet access. Our plan is to use a binary logistic regression to obtain the predicted propensity to have internet away from an individual’s home vs. at their home based on various demographic variables. We would then turn this value into a weight, rake the data back to external population controls including raking those that have the internet away from the home to represent those with no internet, and compare our results to the full dataset that includes those who do not have internet.

For the most part, I’ve been developing a complete understanding of logistic regression for use in writing the final technical report as well creating a raking algorithm in spss. My experience with spss last semester was invaluable as a solid base for the more advanced techniques we're currently using.

[/ QUOTE ]

On a more independent basis, I'm working to develop a method to use past observation to find correct game theoretical strategies in poker games.

[ QUOTE ]
In a case like this, a bit of random noise doesnt have that great of a negative affect.

[/ QUOTE ]

I'm not arguing that it does. I don't see the useful positive effect though either. We also have different definitions of "a bit." A spread of +/- 5% here is huge in this context IMO.

[ QUOTE ]
When you say "In order for an average to mean anything, you have to averaging numbers from the same source," thats wrong. Optimal play is the same at party, stars and AP (save small variations in rake structure). Maximal play might be a bit different because of typical opposition, but not so vastly different that stats like VPIP, VPIP/pos,PFR, etc, are going to so different as to be meaningless. You're still governed by the game structure. How vastly different WPEX is, Im unsure. I dont play there, so I really cant comment)

[/ QUOTE ]

The problem is that you're assuming the base of knowledge is zero. My supposition is not that no information can be derived from this method. It is that no useful information can be drawn from it. If your method determines, in a statistically questionable way, that the optimal VPiP is between 25-33, it hasn't produced anything that isn't known.

[ QUOTE ]
"There's also the problem that many stats haven't converged over 20K or even 40"

cut your DB into 20k chunks, filter it correctly (obv you cant have one full of 3 handed and one 6 handed) and see how much stuff like PFR, VPIP in positions, etc, vary. BB/100 data varies a ton. A lot of others ones dont.

[/ QUOTE ]

Why would you use this method? All poker statistics are binomial random variables which over the sample sizes we're talking are normally distributed with mean p and variance np(1-p). The problem lies not primarily with VPiP and PFR in which n is large but with other variables in which n remains small even over large samples and in statistics which combine binomial random variables such as AF.

[ QUOTE ]
Beyond that, "General ranges for stats are pretty common knowledge IMO and there are several threads linked to in the sticky which provide this info. There's also a thread in which 3 winning mid stakes players post 100+K samples. "


How can there be common knowledge or, for that matter, merit to these 100k posts, that dont face the same pitfalls you describe?

[/ QUOTE ]

That particularly post is not the source of the common knowledge on this topic. 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 for the similar reason to why your method would be ineffective, n is small over large samples leading to significant variation among similar styles.

[ QUOTE ]
Surely between the choice of 3 highly likely winners with 100k hands, and, say, 20, players, all with a very good shot at being winners, with DBs ranging from 30k to 200k, the latter is going to give people better ideas.

[/ QUOTE ]

Depends on what the purpose is. Our purpose is to achieve the tightest range possible. 5 20K samples is significantly worse then one 100k sample because it adds a large amount of variance to the data. It is certainly easier to draw concrete conclusions with known variances from one large sample then the average of many small samples.

[ QUOTE ]
Obviously, if there is already common knowledge of a lot of stats, then compiling a group players #s together is irrelevant. (again, Im unsure of common knowledge on this forum. sparky PMed me asking if I would do this for shorthanded, and normally dont read it). But, Im curious as to the origin of this common knowledge,

[/ QUOTE ]

Many posters here have experience with more then one style as well as a theoretical understanding of where to apply a 25/17 style vs. a 32/23 style and the relative advantages/disadvantages.

[ QUOTE ]
and if you think there is as much noise in the methods used to obtain it as would be present in a compilation as described in the OP.

[/ QUOTE ]

Well a numeric comparison is probably impossible. Do I think the opinion and experience of a large number of very good players is more accurate then the method above? Yes, I do.

[ QUOTE ]
Someone linked wookie's stat post, which is great. But Id say the methods of something like the OP would produce data that would be much better than something like that post. And, if that post is part of this common knowledge, then I think your point is incorrect.

[/ QUOTE ]

There has been a significant increase in the knowledge base since that post. I've considered updating it, but haven't had the time.