Regression t o estimate 'folding equity'

[A_PLUS]

For my econometrics class, I am doing a paper on small stakes NLHE. Pretty basic, I am trying to estimate the folding equity of a particular situation that comes up often in small stakes NL.

You are last to act after the flop, and the action is checked around to you.

I want to find "when a bet is made, what % of the time does the bettor win the hand right there".

Some of the Independent variables I have are:
-# of players that saw flop
-preflop action
-Check raise / raise ratio of (last 100 hands) of players in hand
-post flop tightness of table (% who see flop less SD%)(100 hands)
-Bet size relative to pot
-tilt flag (player lost X amt in last 20 hands)
-Flop Texture Flags (pair, 3 flush, 3 straight, A,etc)
-smallest stack in hand ($ of small stack / pot size)
-large stack in hand
-Rock flag (player in hand has seen <15% of flops in last 50, 100)


I have a few thousand observations where my scenario occurs. I am asking for anyone's thoughts on my variables, and possible regression techniques.

My initial thought is for a Logistic regression, but I am more of an economist than a statistician, so any advice would be helpful.

NOTE: I realize that this study is extremely limited, and there are much better ways to calculate many of the I.V.'s etc, etc. I am just using this as a way to get my feet wet in serious poker analysis.

[gaming_mouse]

A PLUS,

Logistic regression is fine for this problem -- but I honestly think it is no more than a nifty exercise. I highly doubt you will be able to get a model that would have any sort of accuracy in predicting on new data.

I mean, it is just so highly player and table dependent. I know you are taking some of that into account with the variables you've chosen, but it still won't be enough. And there are just so many things not accounted for -- bettor's table image is one of many that comes to mind. Also, some people will defend against a possible steal even if they are tight elsewhere -- how could you predict this without having data on it? Finally, your sample size is WAY TOO SMALL, I think.

Another approach you might want to look into is using a neural network. I don't much about the method myself, but I know the folks at U of A have been using it with some success to do opponent modelling.

HTH,
gm

[A_PLUS]

Thanks for the feedback.

Trust me, I know that this isn't going to be much more than a stat paper where I get to quote sklansky, malmuth, and Brunson as my prime theorists.

I thought about the table image variable, but for the sample I have chosen (.10/.25 NL) I am fairly certain that no one picks up on table image unless it is out of control loose.

I agree about sample size, unfortunately, I am working mostly on an older computer system, querying the data from Access was cumbersome due to memory issues.

Realistically, I think that with the players in the sample, I may get some (albiet very minor pokerwise) significant results. Results that would be interesting, but not much else.

For instance, 3+ oponents with a rag flop, folding equity is apx. zero.

I have read the UA neural networking stuff. They do great work, but thats not really what I am looking to accomplish.
In theory with enough manpower and computer power, you could theoretically create an optimal poker bot (poki/loki).

My goal is to use statistically analysis to better employ a fundamental game.

Every good player makes probability estimates based on game variables, most of which are nearly impossible to model. But, you may be able to make more realistic estimates if you know the stats of the general case of your situation.

You may take one thing from an analysis like mine, for instance. "I know there is only a 20% chance based on my model that this bluff will work, am I confident that this situation is unique? It forces you to think deeper into the game"

This analysis isn't supposed to be special, just a start, in what I hope to be meaningful analysis down the road.

[A_PLUS]

Poker aside, anyone have any suggestions on what to look out for with this type of logistic regression. It has been years since I have used this type of regression.