Call-Range Modeling

RE SNG P Tools... On the easy/donkey games (I play the $11s a Party)

Along the lines of garbage-in = garbage-out, the most important element of analyzing push/fold decisions in the late game is putting the villains on an accurate calling range. My question then is: What's a good model for determining call ranges? If nobody has had to show a hand in the last 10-15 deals what can be inferred about any given villain's calling range?

I know that when the villian or villians don't have you covered their call range will be tighter, unless they are "desperate". And that if they have you "comfortably covered", their range will be wider/looser. So how do we define "desperate" and "comfortably covered"? I think stack ratio (villain:hero) of <.5 is "desperate" and >1.5 is "comfortably covered". And that these cause of loosening of around 10 percentage points. So if I apply this as a modifier to their observed VP$IP (or a fraction thereof since calling != pushing) over the last 10-20 hands, would that give a reasonable estimate of their calling range? How would you tweak these values in the model?

I understant that "It all depends". I'm looking for a first-order approxiamtion based simply on recent VP$IP and stack ratios of hero to villain(s). Further modifiers based on number of players about to bust, does a particular villain close the action, etc. make it really complicated. I know.

So please be gentle. And try to stick to a simple model. Replies with "this is useless, it all depends" will not add much to the discussion. Remember this is for "stupid" games, low-stakes on Party.

Thank you all in advance. I hope EastBay replies...

[HesseJam]

I think that is a good start. But I have the following problems with it:

1) Do VPIPs reliably converge to their true value after 10-20 hands?

2) How do you monitor VPIPs? PT or just an estimate by observation (he pushed/called 5 times within the last 20 hands)

3) you say that their calling range changes relatively to the grade of their coverage of the stack of the pusher. How likely is it that this grade remained the same with regard to the different pushers during the last 20 hands. Not very likely I think because they will have pushed against different opponents and their grade of coverage had been different also. Alternatively, you would need a coverage-grade related observation over quite many of the last hands. But then, you also have to take into account a wide range of blind sizes that blurr your observations.

All in all, a lot of work juxtaposed to the expected result, I think.

Now, if you observe a very aggressive player and realize that he obviously only pushes and calls if he has a certain amount of coverage (scared to bust) then you should adjust your pushing and calling range accordingly. By how much I don't know but I doubt that there is lots of value doing this with a formula. I would just open up pushing a lot more if I had him covered and call slightly less.

[microbet]

Have you done any work on this yet? My first comment is "baby steps." The way I will approach the problem, when I get around to it, is to collect the data and find out how many BBs people call allins with with what hands. After that, I'll worry about categorizing the types of players.

1) I agree that VPIPs do not converge to their true value after 10-20 hands. But we can't increase the sample because blinds/stacks/etc will have changed too much if we go back any further.

2) I track the VPIP with PT. Hopefully I have no prior data on them that "blurrs" the stat for the current tourney. Maybe there is a setting I don't know about, but this is really annoying. When I have prior level 4-6 history on a player I can't tell what their VPIP is for the current tourney... For this reason, I'm working on a little history parser to tell me in reall time what their VPIP is for current toruney. Which does lead into 3...

3) I realize VPIP is not best measure. It has to be qualified as to were they pushing into stacks they had covered or not, or even better were they calling pushes from stacks they could cover or were dominated by. With soem parsing scripts I could make a tool that tallies those situations.

Given possible history parsing work from #3 I hope to develop a model where I can plug in things like: %push into smaller stack, %push into larger stacks, %call smaller stack, and %call bigger stack. All of these over the last 20 or so hands (blind levels again...). I really would like to have a robust way of putting villains on their call reanges. Of course when they are calling, we have added benefit of seeing what they called with, so we know that call range is AT LEAST X%. Maybe someone could make good argument that statiscally we now know their call range is at least X% + some percentage. (ie... what are chances we saw their absolute worst possible calling hand in range?)

So...
Thanks for the feedback.

not sure a simple solution to an extremely complex problem will actually be useful, but the excercise can be. I have worked extensively on this problem and have found it to be a fascinating study that has helped my game tremendously.

First of all, the villian's stack ratio in relation to that of hero's is actually less of a factor when calling off their stack than the their ratio in relation to the blinds. Villian can have <.5 of Hero's stack and still not be desperate in relation to the blinds or other stacks.

I also think you have to at least find a simple model that identifies your image. Even just whether or not the table perceives you as "desperate".

But in the end, without a vastly more complex model, I think you will find the results to be quite inaccurate.