Quantifying the accuracy of tells
 

I believe most people who have read Caro's book of tells will agree that the "accuracy" figures for the tells are essentially meaningless except for purposes of comparison.

I can think of at least one reason why this is true: they fail to take into account background probability. If it's 99% likely that your opponent is strong based on previous action, and you get a tell that, with 80% raw accuracy, tells you he's strong, have you learned much of anything? Not really.

The purpose of this thread is to create a new system for quantifying tells & using them in decisions. It should have these properties:
0) Mathematically sound
1) Useful for computation about what play to make (& ideally simple enough to use at the table)
2) Addresses the issue of background probability
3) Boils down to a number (or a few numbers) that could be listed for each tell in a book like Caro's.
4) Addresses the other issues raised on the first 2 pages of chapter 2 in Caro's book.

I have some ideas I'll state later.

Re: Quantifying the accuracy of tells
 

This is impossible. Live play tells cannot ever be quantitized for a mathematical formula. You could use some statistics, but still then it would be very hard to use in a live game.
Experience will tell you how accurate tells are. Also tells differ a lot from person to person. For example, some people are nervous just about playing poker, while others never seem to have any of the tells associated with nervousness, so these tells tell you nothing except if you know your player.
Betting patterns on the other hand would be suitable for this, but they would differ too much from player to player as well. I think common sense and good judgement will do a lot better.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
This is impossible. Live play tells cannot ever be quantitized for a mathematical formula.

[/ QUOTE ]

Why not? Your objection seems to be "they vary from person to person" but that doesn't mean that you can't quantify them. Just that you have to do it once per person. Which is essentially what Caro's trying to do by providing 3 numbers for varying qualities of players.

The question here is NOT how difficult it is to actually quantify tell behavior for a givn person, but what the correct mathematical structure is ie. if someone were going to hand you an accurate "tells sheet" on your opponent, how would you like the information formatted so you could make use of it.

Re: Quantifying the accuracy of tells
 

Reads and tells are part of the game that fits in "art" and not science. It is impossible to put tells into some mathematical formula and you would have no edge, and in fact, it would probably a negative effect on your game.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Reads and tells are part of the game that fits in "art" and not science. It is impossible to put tells into some mathematical formula and you would have no edge, and in fact, it would probably a negative effect on your game.

[/ QUOTE ]

Hardly.

The most difficult part of using tells correctly is integrating the information gained from them with other more conventional information in a sound manner.

For example, if you see an "80% accurate" tell that says X, and you believe X is going to be true in the background 99% of the time, do you now believe P(X) = 80%? 99%? 99.8%?

You can hardly claim that the difference between those numbers is irrelivant. So it's probably worth thinking about this after all.

Re: Quantifying the accuracy of tells
 

Where do you come up with the 80%

Where do you come up with ANY of those percentages?

On that same topic, when people say "I probably call 20% and raise 80%" do you actually think they are anywhere near that? Doubtful.

Reads are not math unless its a betting pattern. You can make up random percentages to try and make it LOOK mathematical...and by all means, if you make one, stick to it. Because when you start making read specific plays based on a mathematical equation, you are going to get run all over.

The second someone picks up on what you're doing, they now can impact your decisions by a "tell."

If its math, its exploitable in poker. If you are making folds based on math, and you play a single player who picks up on it, you're toast.


I see a lot of people on this forum who are hardcore into poker math and thats fine. But math isnt everything in poker and it never will be. Why do you think people say things like "a computer could never beat better players"? Because the computer would be making every decision based on math and patterns which the real live player could then exploit.

Nothing wrong with doing this, im sure its a good exercise, but its not realistic.

/rant

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Where do you come up with the 80%

Where do you come up with ANY of those percentages?

On that same topic, when people say "I probably call 20% and raise 80%" do you actually think they are anywhere near that? Doubtful.

Reads are not math unless its a betting pattern. You can make up random percentages to try and make it LOOK mathematical...and by all means, if you make one, stick to it. Because when you start making read specific plays based on a mathematical equation, you are going to get run all over.

The second someone picks up on what you're doing, they now can impact your decisions by a "tell."

If its math, its exploitable in poker. If you are making folds based on math, and you play a single player who picks up on it, you're toast.


I see a lot of people on this forum who are hardcore into poker math and thats fine. But math isnt everything in poker and it never will be. Why do you think people say things like "a computer could never beat better players"? Because the computer would be making every decision based on math and patterns which the real live player could then exploit.

Nothing wrong with doing this, im sure its a good exercise, but its not realistic.

/rant

[/ QUOTE ]

Rant duely ignored.

Anyone with a firmer grasp of theory want to contribute anything useful?

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
Where do you come up with the 80%

Where do you come up with ANY of those percentages?

On that same topic, when people say "I probably call 20% and raise 80%" do you actually think they are anywhere near that? Doubtful.

Reads are not math unless its a betting pattern. You can make up random percentages to try and make it LOOK mathematical...and by all means, if you make one, stick to it. Because when you start making read specific plays based on a mathematical equation, you are going to get run all over.

The second someone picks up on what you're doing, they now can impact your decisions by a "tell."

If its math, its exploitable in poker. If you are making folds based on math, and you play a single player who picks up on it, you're toast.


I see a lot of people on this forum who are hardcore into poker math and thats fine. But math isnt everything in poker and it never will be. Why do you think people say things like "a computer could never beat better players"? Because the computer would be making every decision based on math and patterns which the real live player could then exploit.

Nothing wrong with doing this, im sure its a good exercise, but its not realistic.

/rant

[/ QUOTE ]

Rant duely ignored.

Anyone with a firmer grasp of theory want to contribute anything useful?

[/ QUOTE ]Why dont you just simply tell me how i'm wrong?

Making edge decisions based on a mathematical equation is exploitable. Correct?

Rather than just ignoring what I said, why not tell me how I'm wrong.

Re: Quantifying the accuracy of tells
 

You're "wrong" in that your previous 2 posts contained nothing that accuratly represents anything about poker. Or perhaps I should say you were neither right nor wrong per se as the posts made little sense. If you really want an example:

[ QUOTE ]

Making edge decisions based on a mathematical equation is exploitable. Correct?


[/ QUOTE ]

This statement is more or less meaningless.

What is an "edge decision"? Why do you believe it is explitable if the decision is made using an equation? What if that "equation" represented the Nash equilibrium play in a given situation? What if it was a last to act call/fold decision, and as such your opponent cannot act again and thus can't exploit it.

In other words, your comment was more or less nonsensical.

Re: Quantifying the accuracy of tells
 

"What if it was a last to act call/fold decision, and as such your opponent cannot act again and thus can't exploit it."

Speaking of nonsensical.

We are talking about tells. If a person KNOWS how you are going to act based on a piece of information, falsely presenting that information in a situation where they want you to act in a certain way is now exploitable because you're doing the same thing based on math.

Its the same as giving bad odds, or pot committing someone.

Re: Quantifying the accuracy of tells
 

Your posts are so non-sensical as to be a complete waste of time. Sorry.

Re: Quantifying the accuracy of tells
 

I think, unless you're building some kind of pokerbot or pokre playing computer that is able to collect loads of statistics that will also be based on measurements that might indicate a tell, like blinkrate, whatever, you are going to play far worse trying to crunch numbers on tells. As I said, online it is a lot easyer, because you'll have betting patterns and software to analyse. As far as classical tells, you'll need a camera and a computer to be able to collect all these statistics and proberly process them. I don't think this is allowed and even if you do manage you will probably be more accurate by experience and intuition.

But if the question is; suppose we were to make this machine, what would it look and work like? I think it would be a camera with lots of recognition and interpretation software connected to it that will statistically determine whether a pattern in behaviour that it has found will tell anything about the intentions and hand of a player. It will return a couple of possible hands for the opponent with the likelyness next to it, so it will combine hand reading with tells (which is quite logical). Since the results are derived from earlyer hands the background probabilities will be already in there.

Anyways, as stated avobe, I think it is practically impossible, except for SF movies, to approach tells in a very mathematical way you are proposing and be better off than good old human judgement.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Your posts are so non-sensical as to be a complete waste of time. Sorry.

[/ QUOTE ]

Posts like this are a complete waste of time and will not be tolerated.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Anyways, as stated avobe, I think it is practically impossible, except for SF movies, to approach tells in a very mathematical way you are proposing and be better off than good old human judgement.

[/ QUOTE ]

We talk about this a little in our book. (p67-68) Suppose that you observe some particular tell (model it as a binary flag - either you observe the tell or you don't). What you are looking for is the probability that some event (like "has a bluff" or "has the nuts") is true given that this tell is observed.

If you call the event A and the event of observing the tell T, then you want p(A | T). Now p(T) = p(A | T) + p(!A | T) -- the second term here is the critical one to the reliability of a tell which is the chance of a false positive.

Now if you want to quantify the accuracy of a tell, you can build a model of the tell, which includes a series of outcomes T_i and A_i for hands which you see, and some inferential model of hands which you don't see. Then you can use the data to estimate the a priori p(A | T) and p(!A | T) and the likely errors on each of these. Then you go play a ton of hands to reduce the errors to an acceptable level.

I responded to this post specifically because I would suggest that the last sentence is kinda both right and wrong. The human mind is pretty good at this kind of thing as a rough cut, but most human minds seem to a) undervalue the error of their predictions, b) weight more recent evidence a lot too heavily -- rather than a little more heavily which is probably right, and c) ignore evidence that doesn't catch its attention.

So as usual, I'd suggest that some combination of analysis and intuition would result in stronger tell-reading overall than either alone.

jerrod

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Your posts are so non-sensical as to be a complete waste of time. Sorry.

[/ QUOTE ]No, I think you just didnt think this post through before you made it, and now that not a single person has said "oh wow why didnt we ever think of turning tells into MATH!" you're getting all angry.

When you say you believe a tell to be "80% correct" or "correct 80% of the time" you're guessing without actual tracked data. Even then, a tell is not something that just happens whenever you want it to. To say there would be ANY situation where, if your opponent knew what you were doing, they would not be able to exploit it is ludicrous.

Being last to act does not change the fact that if you are using math, you will be folding based on math, and if your opponent knows this, he can exploit it from any position at any time as long as he knows what to do when. Then you have to change your formula all over again, and after maybe 3 sessions it will be completely useless.

But hey, good luck, have fun falling off that horse.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Being last to act does not change the fact that if you are using math, you will be folding based on math, and if your opponent knows this, he can exploit it from any position at any time as long as he knows what to do when.

[/ QUOTE ]look, i don't necessarily agree that tells will be mathetmatically quantifiable with enough accuracy to be useful but this statement is clearly even less useful. its like you're trying to tell someone teaching you pot odds that its useless because "math is explotable".

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
b) weight more recent evidence a lot too heavily -- rather than a little more heavily which is probably right

[/ QUOTE ]

Your general approach is very interesting, but I think that this particular observation needs further conditioning.

If your opponent is human, then recent events are more likely to weigh heavily if the opponent is losing, tilting, or otherwise demonstrates a tendency toward emotional decision making.

This seems to be the default experience of most people which is why they tend to overweight this way in general.

If your opponent is a computer or very good player, then I think that this weighting needs to be reduced considerably.

To OP,

I think that in order to create a usable system for live play, a strong memory is an absolute requirement. I happen to be lucky in that I have a near photographic memory. (It used to be somewhat selective as well, but that has been overcome through conditioning and practice.)

I do believe that a system could be designed that would work for a computer player who had a large enough database of hands that included shown down hands. Stats like %won at showdown when raising the turn could be revised to %Bluff Raising the turn after enough showdowns were seen.

With datamining, I think that players with similar vpip, pfr, AF, etc. could be grouped into categories, assigned hand ranges, and back test the whole thing against a large hand history database. I believe that useful general conclusions could be drawn about the contents of opponent's hands just by analyzing this data alone (correlated with opponent's subsequent actions).

Of course, this would require a large sample size of hands, but that is not too difficult to come by. (Obviously, computer players assume online play)

I use a system myself where I will sometimes 'pay' for information by deliberately raising a new player who obviously has a strong hand and then elliciting responses from him which include hand movements, smiling, and hopefully speaking.

I then file this information away in memory where it can be retrieved later in the session for comparison to play in future hands. Among the things that I try to discover is whether the previously observed tell(s) are deliberate or not.

If I find a tricky enough player that he actively manipulates his tells to affect my perception of his level of thinking, then I become duly cautious when involved in a hand with this opponent and take appropriate measures to determine if he is, in fact, a better player than me.

The point is that the general variables for each tell change with the opponent's state of mind. Determining whether or not his apparent state of mind is genuine is, IMHO, most of the battle.

I hope this makes sense.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
Being last to act does not change the fact that if you are using math, you will be folding based on math, and if your opponent knows this, he can exploit it from any position at any time as long as he knows what to do when.

[/ QUOTE ]look, i don't necessarily agree that tells will be mathetmatically quantifiable with enough accuracy to be useful but this statement is clearly even less useful. its like you're trying to tell someone teaching you pot odds that its useless because "math is explotable".

[/ QUOTE ]Far from it. Pot odds are using math TO exploit. By giving a person incorrect odds, you force them to play incorrectly.

By using math to make edge decisions, where, say a "80% accurate" read will force you call based on odds, but 90% is a fold, you set yourself up to be toyed with by an observant villain.

Math is insanely useful in poker, when it gives you the best of it. By turning reads into math you will most likely be taking AWAY the edge you have against decent players.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Reads and tells are part of the game that fits in "art" and not science. It is impossible to put tells into some mathematical formula and you would have no edge, and in fact, it would probably a negative effect on your game.

[/ QUOTE ]

It is an art, but it can be applied scientifically.

Actually, you can put it into a mathematical formula. The thing you're not doing is leaving room to adjust. You're saying that every time it will be the same %. People change so you change with them depending on your read.

If someone asked you how sure you thought he was bluffing, and you said 'pretty sure'. With some thought, you could put an actual % number to your read. It might be wrong, but you can still put a number to 'your' feeling of whether he's pulling a move or not. I mean, you know it's likely he's bluffing, so it's not 0%. There is a chance he does 'have it' so it's not 100%. 'Pretty sure' would probably be well over 50%, probably more like 70 or something.

With some thought, you can put a number to the likelihood. Then you could take that number and compare it to the pot size.

Now, no one really goes literally that in depth with it. Most will just go with their feel, and that's likely close enough. It's kind of like people who have a feel for the pot size(by sight) without knowing the literal size of the pot.

But if you really thought about it, you can put a % number for that situation as it's happening right then.

b

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
By turning reads into math you will most likely be taking AWAY the edge you have against decent players.

[/ QUOTE ]

I guess we just throw Bayes Theorem out the window then...

Since that's based on the read of the possible hands an opponent could be holding.

b

Re: Quantifying the accuracy of tells
 

A system like this would be very easily expoited, because it will not be very flexible and therefore it would be easy to fake tells as soon as you get an idea of what the system will output.
I also doubt that it would be very reliable (as is human judgement), so it will only be effective in close situations.
I agree that a system like this (if performed by a human) might add to the accuracy of his tells, but only to a certain extend and so will experience and intuition. Especially if experience and intuition VS a system like this will contradict each other, I think experience will be correct more often. While if experience has no real strong tells the system might be correct more often.
It will be a nice complementation when you are A. in doubt of what to do and B. have no intuitive tells that help you.

So besides coming up with a system that a human can use in practise at a live game (which is hard) it will only add a little value in few situations and will be easy to exploit as soon as the system is known to your opponents.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
By turning reads into math you will most likely be taking AWAY the edge you have against decent players.

[/ QUOTE ]

I guess we just throw Bayes Theorem out the window then...

Since that's based on the read of the possible hands an opponent could be holding.

b

[/ QUOTE ]But if you say, 70%, and at 69% you're folding instead of calling, you're now just guessing.

Id like for someone to give an example on how this would actually be turned into an equation and would then be reliable rather than some loose number combined with a loose guess of a players hole cards.

Re: Quantifying the accuracy of tells
 

The responses you are getting are karma for your cuffs posts. Want to reconsider?

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
The responses you are getting are karma for your cuffs posts. Want to reconsider?

[/ QUOTE ]Do you think that the OPs idea is possible?

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
The responses you are getting are karma for your cuffs posts. Want to reconsider?

[/ QUOTE ]Do you think that the OPs idea is possible?

[/ QUOTE ]

Yes. Perhaps my answer is not as defined as the OP would like it to be, but in simplistic terms that even I can understand we already apply Bayes Therm in this manner.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
[ QUOTE ]
The responses you are getting are karma for your cuffs posts. Want to reconsider?

[/ QUOTE ]Do you think that the OPs idea is possible?

[/ QUOTE ]

Yes. Perhaps my answer is not as defined as the OP would like it to be, but in simplistic terms that even I can understand we already apply Bayes Therm in this manner.

[/ QUOTE ]But that doesnt seem to be what he's asking for.

Re: Quantifying the accuracy of tells
 

It's impossible to quantify tells with accuracy, because not only do the differ between people, but within the same person and the tells constantly change.

We might pickup a tell in someone that proves to be 100% true that session, and the next session is 100% false.

Re: Quantifying the accuracy of tells
 

youre trying to mix two different schools of poker. Youre mixing the new age, mathematics, statistics and game theory school with the old fashioned texas gambler "i jus don' think u got it" school. When I pick up a tell on an opponent, I dont act on it based on math. If I have a fairly strong and consistant tell that my opponent is weak, I will bet or raise regardless of my cards. And if i pick up a tell that my opponent is strong I throw everything away except a nearly unbeatable hand. Tells and reads are more about your instincts than math; try to rationalize what your gut tells you, and you'll probably just confuse yourself and make an incorrect decision more often than not.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
youre trying to mix two different schools of poker. Youre mixing the new age, mathematics, statistics and game theory school with the old fashioned texas gambler "i jus don' think u got it" school. When I pick up a tell on an opponent, I dont act on it based on math. If I have a fairly strong and consistant tell that my opponent is weak, I will bet or raise regardless of my cards. And if i pick up a tell that my opponent is strong I throw everything away except a nearly unbeatable hand. Tells and reads are more about your instincts than math; try to rationalize what your gut tells you, and you'll probably just confuse yourself and make an incorrect decision more often than not.

[/ QUOTE ]Exactly.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
[ QUOTE ]
youre trying to mix two different schools of poker. Youre mixing the new age, mathematics, statistics and game theory school with the old fashioned texas gambler "i jus don' think u got it" school. When I pick up a tell on an opponent, I dont act on it based on math. If I have a fairly strong and consistant tell that my opponent is weak, I will bet or raise regardless of my cards. And if i pick up a tell that my opponent is strong I throw everything away except a nearly unbeatable hand. Tells and reads are more about your instincts than math; try to rationalize what your gut tells you, and you'll probably just confuse yourself and make an incorrect decision more often than not.

[/ QUOTE ]Exactly.

[/ QUOTE ]

While it is true, I don't think it's all that relevant.

What is a tell? It's a noisy signal of the hand strength of the opponent. This is clearly possible to model.

Do I think that you can mentally come up with a model in your head and use it at the table, doing all the Bayesian updating? No.

Despite the tone of OP's responses in this thread being unacceptable (because it only hinders discussion and derails threads), his original post is actually pretty good. Responses like those above demonstrate why - this is an area of poker theory that hasn't really been touched. Caro clearly overrates the usefulness of tells, while the 2+2 guys tend to say that they aren't worth much at all (before Mason jumps in here, I will say that they do describe at various times when a tell could be useful, an example is in SSHE if I'm not mistaken). Nobody as far as I know has done any analysis on the subject. This is probably the reason behind responses claiming it can't be done. There are interesting questions such as how reliable does the tell have to be, once one defines reliability of a tell.

In approaching the question, one must consider different ways of modeling noise.

You could model it as a spread. Say you are playing [0,1] poker. With some probability one of the players gives off a tell. A tell in this case would be a random signal from some interval around your card. So say your card is a .8. In giving off the tell, it gives the other person a signal that is a random number between .78 and .82. This would do well at representing the situation where somehow you can tell broadly what type of hand your opponent has - say you can see if he likes his hand or not.

You could extend this by just giving a region. Some area is revealed, and the player has a hand somewhere in there and any hand in the region is equally likely. This would allow for situations where the vibe you're getting is that she either has a great or terrible hand.

The other obvious option is to set it up so that with probability p the player's actual hand is revealed and with probability 1-p a totally random card is revealed. This is probably the most realistic in terms of what people think of as tells. Her hands are shaking, does she have the nuts she's representing in her betting or did she just have a cup of coffee? You could also combine this with the spread - with probability p the signal is totally random, with probability 1-p it's within some range of the true value.

Finally, for the above you could allow for the possibility of the sender giving off fake tells. Maybe that could actually lead to the randomness. You could also make it not certain that she successfully gives off the tells as well.

I think solving the above games, in the three situations where the sender is unaware of giving off the tell, aware but can't help it, and not only aware but can give off fake tells somehow would be interesting.

Jared

Re: Quantifying the accuracy of tells
 

Okay, here's why you can't turn a tell into math

You raise to 200 in a 25/50 game with a pair of queens. The BB reraises you all in for 700 more. You pick up a tell he has that indicates strength.

What does that tell you? It depends on how your opponent thinks. Does he consider AK to be strong? What about pocket jacks?

Theres so much intuition involved in the deciphering of tells, based on how your opponents are playing, stack size, pot size, etc., that I dont think you can boil it down to an exact science.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
Okay, here's why you can't turn a tell into math

[/ QUOTE ]

You might want to read Jerod's new book, Mathematics of Poker. I'm not going to lie, I had a tough time with it and had to put it down before getting a mathematics refresher course but it might help you understand why there is some validity to the concept.

Re: Quantifying the accuracy of tells
 

I think many responders have missed the point of OP's post/idea. The idea of a mathematical framework for analysing and using tells is not ridiculous nor (necessarily) impractical as some of the responses have suggested. When responders say things like 'tells are art not maths', or 'it's all about instinct', or 'how can you know a tell is 80% accurate?' - they have not quite grasped what (I think) OP is trying to get at.

Instead of responding aggressively to doubters, why doesn't Splawndarts post some of the ideas he promised in his OP? Then the discussion might move on somewhat. Splawndarts - Yes, they have probably misunderstood what you have suggested. Yes, you might understand poker theory better than some of them. But no, you shouldn't be using that sort of tone. If nothing else it comes across as very arrogant.

It's a pity, because I have read quite a few of your posts in the last month or two and have been impressed by quite a bit of what you have to say. But your tone has been a bit unfortunate in quite a few of these other posts too. You are clearly an intelligent, thinking player who has plenty to contribute to these forums. But I think you need to sort your attitude out a little bit.

On a related note, don't you feel a bit 'uncomfortable' posting things like:
[ QUOTE ]
The purpose of this thread is to create a new system for quantifying tells & using them in decisions...

[/ QUOTE ]
It all sounds like a rather grand scheme, worthy of investigation by a top poker theorist. I wasn't aware that you were one yourself - maybe I am wrong!

Also, hasn't this sort of thing already been at least outlined in the Mathematics of Poker?

All this is meant in nicest possible way of course...

Re: Quantifying the accuracy of tells
 

Hi,
> The second someone picks up on what you're doing, they
> now can impact your decisions by a "tell."

That's anyhow an issue, no matter if you are using math or not?

> If its math, its exploitable in poker.
Rather the opposite?
You may have a look to some introduction (just google it) to game theory (optimal and exploitive play) -- I would recommend Mathematics of Poker - if one plays optimal strategy it cannot be exploited. Tells will basically contribute to know where your opponent does not play optimal, so you can actually exploit him. Well, these things are already pretty complex without the additional complexity of quantifying tells ;-) but that doesn't make them exploitable or wrong or not applicable.

> But math isnt everything in poker and it never will be
Here I agree :-) I also play sometimes poker for fun

> Why do you think people say things like "a computer
> could never beat better players"?

Who told you that? The only statement I know of is that while this is alrady true for chess it's not YET the case for poker. But NEVER is not the case.

> Because the computer would be making every decision based
> on math and patterns which the real live player could
> then exploit

It's actually the opposite ;-)

Have fun

Re: Quantifying the accuracy of tells
 

For the B&M games that's a fair conlusion.

For online games

1) Tells are not many and pretty basic, quantifying them may make more sense, in a way of having some additional correlations (e.g. time to act, or use of advance action buttons) that you can use for potential exploitive strategy.

2) You also have it in your book, you can hardly collect enough stats on your opponens to get statistically useful results, and for online tells you need more tools (than the standard hand history). While your general statement is correct, I think in many cases stats on opponents helps not so much for a specific opponent, but rather to find properties of the whole players field, and this can be useful for a default exploitive strategy when you don't know your opponents.

So, for me although there is no huge potential in this topic, for me I found some food for thoughts.

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
For the B&M games that's a fair conlusion.

For online games

1) Tells are not many and pretty basic, quantifying them may make more sense, in a way of having some additional correlations (e.g. time to act, or use of advance action buttons) that you can use for potential exploitive strategy.

2) You also have it in your book, you can hardly collect enough stats on your opponens to get statistically useful results, and for online tells you need more tools (than the standard hand history). While your general statement is correct, I think in many cases stats on opponents helps not so much for a specific opponent, but rather to find properties of the whole players field, and this can be useful for a default exploitive strategy when you don't know your opponents.

So, for me although there is no huge potential in this topic, for me I found some food for thoughts.

[/ QUOTE ]All an online tell is is a betting pattern. Nothing more. Tells tend to be physical or attitude/voice changes, not a betting pattern....usually they are seperated.

Re: Quantifying the accuracy of tells
 

First of all I've made and saved a lot of money (well, for me, lets say a lot of blinds [img]/images/graemlins/wink.gif[/img] ) on physical tells like blink rate, chip placement, attention, patience (or the lack of it), talkativeness, and so on. They are worth a lot and I think most internet players underestimate this.

As I stated before they can probably be recognised by some smart software and from statistics the accuracy can be determined if it's combined with betting patterns, etc. I think this is very possible, only hard to implement on a live table and easy to exploit by people that know the software, because it will be near impossible to create something that will adapt quickly enough to fake tells. I'd buy one of these machines and learn to make it say the opposite [img]/images/graemlins/smile.gif[/img]. A human operated system (with charts or memorized formulas and data, whatever) might help a player that is already good at detecting and using tells be a very little bit better. Not worth it imo.

Oh, about online tells... I think there are two (inacurate) real tells online that are not betting patterns:
1 the time someone takes for an action
2 the chatbox

Re: Quantifying the accuracy of tells
 

[ QUOTE ]
It's a pity, because I have read quite a few of your posts in the last month or two and have been impressed by quite a bit of what you have to say. But your tone has been a bit unfortunate in quite a few of these other posts too. You are clearly an intelligent, thinking player who has plenty to contribute to these forums. But I think you need to sort your attitude out a little bit.

[/ QUOTE ]

I disagree. This topic was valid, but some of us believe that Splawn Darts is someone who is too intelligent for his own good, who intentionally produces incorrect data hidden ever so slightly in his posts so he can revel in his genius as he argues with others - think Comic Book Man on the Simpsons. In short the wording of his posts seem to be intentional, he knows what he is doing, and if it continues I suspect a IP level perm-ban will be in the cards for him in the future. Fortunately the guy is obviously smart, I hope this experience helps him wake up and smell the coffee.

Re: Quantifying the accuracy of tells
 

Most of the posters here who have strongly disagreed with the original poster are probably doing a less formal analysis that is very similiar to OP's so called "mathematical" method. If I understand his point, he's saying that the best way to use a tell is to attach a % accuracy to it, and then incorporate that in you hand analysis, and range anaylsis.

Say we pick up a great tell, we notice the chatterbox in seat 5 shuts up and gets serious each and every time he is dealt a whopper hand.

He goes dead silent. He then raises, which should help confirm the accuracy of your tell.. It get folded around to non-math guy, who mucks his a little drawy but poor blind, thinking, "He's got a whopper hand."

OP analysis is just more deliberate, like so: Its 90% likely that the tell is accurate; if it is, his hand range is AA, KK, QQ, or JJ, in equal shares, all against which I have insufficient odds/equity without other callers, so I muck.

I think its just a logical way to keep your thoughts focused, and integrate a physical tell to standard hand analysis.

Re: Quantifying the accuracy of tells
 

Your idea is sound, but like everything, the devil is in the details. The only way to approach this is by writing out a large number of situations and going through the math of likely scenarios in each one. There are simply too many variables and hidden assumption for a more general system to be viable. There also needs to be debate about the reliability of various tells.

So, who's going to start the hard work? I don't play live much but it'd be interesting to see this written out.