Redefining Aggression - PokerAce Hud

[PokerAce]

I'm looking to create a more accurate aggression stat for the next release of PokerAce Hud. There was a thread here a couple weeks ago on how non-useful or actually harmful the aggression factor in PokerTracker can be.

Here is my idea (with help from a friend) for the new aggression stat. It would be represented as:

(times_bet + times_raised) / (times_bet + times_raised + (times_checked - times_checkraised))

This totally removes calls out of the stat. Basically, it's nothing but the pure aggression stats.

I'm going to do some testing to see what kind of numbers I get, but I would like to get your opinions on this formula. Do you think it will be an accurate representation of aggression?

Also, what should I name this stat? I've gotten one suggestion, "aggression frequency." I like it, but I'm always open to suggestions.

[pokergrader]

Well checking and calling are similar actions in terms of aggression, so it is difficult to have one and not the other...

[PokerAce]

Here's what I'm trying to accomplish. Basically, I'm turning aggression into a percentage. 0% would be the most passive play possible. 100% would be the most aggressive play possible.

Would this not be pretty accurate to define aggression?

I think we need to define what is aggressive and what is passive.

Betting = Aggressive
Raising = Aggressive
Check Raising = Aggressive

Calling = Passive
Check Calling = Passive
Check Folding = Passive

Check without Call/Fold = Neutral


After discussing things with my friend some more, we came up with this new calculation:

(times_bet + times_raised) / (times_bet + times_raised + times_called + times_folded)


Let's do an example. We'll ignore preflop play for now. This player is aggressive, so he check raises the flop, bets the turn, but check folds the river.

That's 1 bet, 1 raise, and 1 fold. Doing the formula:

(1 + 1) / (1 + 1 + 0 + 1) = 67%

If the player bets the river, then it's:

(2 + 1) / (2 + 1 + 0 + 0) = 100%

If the player calls the river, then it's:

(1 + 1) / (1 + 1 + 1 + 0) = 67%


Another example. A player check calls the flop, check calls the turn, and makes his flush so he check raises the river:

That's 2 calls and 1 raise. That would be:

(0 + 1) / (0 + 1 + 2 + 0) = 33%

If he bets the river, then it's:

(1 + 0) / (1 + 0 + 2 + 0) = 33%

If he misses and check folds the river, then it's:

(0 + 0) / (0 + 0 + 2 + 1) = 0% (no aggression at all)


Let's say we have a maniac who bets his flush the whole way. Bet the flop, bet the turn, bet the river.

That's 3 bets, nothing else:

(3 + 0) / (3 + 0 + 0 + 0) = 100% aggression.


So does this make sense or am I completely out of my mind?

[APerfect10]

Aggression factor can sometimes be misleading. For example, if a player bets one flop and folds 99 their aggression factor would be infinite.

With this new formula that PokerAce and myself have come up with, if a player bets 1 flop and folds 99, their aggression frequency would be 1% which is extremely passive.

[garion888]

Hey PA,

I liked the first formula where there was a correction term in the denominator for the amount of times one check-raised. I happen to think that a check raise is qualitatively extremely agressive. Perhaps the numbers didn't flesh that out. I was wondering why you removed the correction term.

Thanks
-J

PS... I [censored] love PAHUD.

[PokerAce]

The problem with making a check raise count for more aggression than a normal raise is that if someone does nothing but check raise all the time, their aggression would be way higher than 100%.

Basically, if you want to know how often someone check raises, you'll need to look at the check raise stats. In the new version, you'll get check raises by street.

[theRealMacoy]

yo dude,

I am very glad to hear someone is taking on this problem!

I like what you have done so far, although the issue with calling frequency may still be a problem in interpreting true meaning (see below...sorry this is a bit long and rambling but it is definately something i have been pondering for awhile).

First of all, I completely agree that the CR stats should be kept separate and measured by themselves (I would most definitely like to see them street by streeet.... those are going straight to the pop-up stats on my HUD).

When multiple stats are combined into one there will often be interactions involved, which as a result, can make the meaning difficult to interpret.

Which leads to the main problem I have with the current PT aggression statistics is that they do not account for the looseness-tightness of the player (which, with certain players, drastically affects how one should correctly interpret the meaning and can actually be misleading if not taken into account).

Outlier type players are the most problematic, such as very loose or very tight players who either call relatively often or little respectively, which seriously affects the outcome of the current aggression metric and the resulting inference that should be made.

Basically, the inferences that we make are that a very aggressive player's bet/raise action is not necessarily indicative of a strong holding wheras a very passive player's bet is much more indicative of a strong hand (keeping the looseness/tightness of the player out of the picture for now and lets say equal for all).

Perhaps if you were able to remove or at least account for frequency of call actions this would likely distill this metric into a much better measure of aggression itself (without the interaction of looseness-tightness to worry about).


Let me back up though, I am trying to work out exactly what we are after here so bear with me....

...It seems to me that with the original aggression stats we were trying to interpret the average strength of the hand that the player is likely to be playing/betting.


So the question becomes, is there any way to measure the relative strength of a hand directly and in-turn use this in relation to bet/raise frequency?

..…Again, the goal is trying to ascertain the relative strength of the hand a player is betting/raising with...

In keeping with this method might to CALCULATE THE MEAN reletive strength of the hands (compared to the current board of course) a player is BETTING out with, is RAISING with and finally CR with? Overall and for each street….

You would likely need a fair number of hands on a player for these calculate means to have any meaning.

We should be able to work out the meaning of the spectrum of reletive hand strength numbers and put a chart togehter tying the numbers together with hands (e.g., high card, bottom pair, pair, top pair, two pair, top two pair.....etc).

So what would it all mean?...We should be able to take the means for PlayerA and infer that he BETS with only Top Pair or better, RAISES with Trips or better....

Pokergrader's luck calculation is the only thing that comes to mind so far for measuring relative hand strength, as I seem to recall takes this into consideration...?


...Anyway, just a couple of new ideas. I hope at least some of it was clear. I may way off here as i have been camping all weekend and poker has been the farthest thing from my mind.



With the current formula you have developed, it would be nice to get a picture of what the different types of players score (i.e., LAA, LPP, TPP, TPA, LPA, very very loose and very very tight, etc..). if you have run these I would love to see them. Perhaps, this issue is moot with your new formula.


cheers and keep up the good work,
the Real Macoy

ps. still loving your HUD

[Dan Mezick]

To attempt to isolate aggressiveness as a property, valued between 0 and 100, may be a suboptimal approach because of the high degree of intertwining of aggression along the range of very loose to very tight.

Aggressiveness from a loose player is very different from aggressiveness from a tight player. You welcome the LAG and avoid the TA type in most cases.

Consequently, since adjusting to Aggression requires context info to get the adjustment "just right", it may be suboptimal to isolate Agg as a property. Perhaps a value from 0 to 100 where 0 is loose passive (calling station) and 100 is super-tight super aggo may be closer to an optimal approach.

[PokerAce]

You are absolutely right that aggression from a loose player is different than aggression from a tight player.

However, I'm not trying to come up with a stat on how good a player plays, just how aggressive.

A loose aggressive player and a tight aggressive player will have pretty similar aggression frequency stats. Look at two examples I posted above:

54/16/1.3 = 42%
22/12/1.2 = 42%

The first player is very loose, and is no doubt a very bad player. However, you know that this player is semi aggressive. The second player is tighter, but he bets/raises/calls/folds just as frequently as the first player. When this player is in the hand, expect his actions to be very similar to the loose player.

This isn't meant to be a universal stat that displays how good a player is, just how aggressive they play their hands. You will need to use it in conjunction with other stats to determine the skill of the player.

[bhudson]

OK... isn't this just bet% + raise%?

(Which I think is fine as an indicator of aggression)