The EV of different playing styles - Part Two

[rharless]

Good point.

[MicroBob]

not really. the players who raise pre-flop are presumably doing so with good hands.
the players who don't raise pre-flop are presumably NOT raising their better hands properly.

that is what the data tells us.

this stuff is absolutely awesome btw. thanks so much for putting it together.

if you really wanted to look at post-flop play i might suggest tying in factors such as W$SD and % of hands taken to SD.

looking at these in combination with the previous tight/loose, aggressive/passive factors would be interesting as well.
i imagine that would only take you around half-a-year to put together.


come to think of it.....how long did it take you to put all of this data together in the first place?? looks like a mountain of work to me.

[Jeremy'sSpoken]

Perhaps the aggression factor PT calculates at every stage could be used along with a % of seeing SD. I know these can all be exported to excel rather simply. What you do with after that, I have no idea.

[rharless]

not really. the players who raise pre-flop are presumably doing so with good hands.

Well, I suppose I should qualify -- I think it is a good point for the players that have a lower volume of hands. I can play as many as 300 or 400 hands where my cards seem really really hot or really really dry. Although I only used players with 40 hands or more, those with under 300 hands could have their stats still skewed by the flow of their cards.

On the other hand, removing players with less than 300 hands leaves me little data to work with and also really gets me away from a "zero sum" database. When you get into volumes of hands that high, the majority of losers are dropped out of the analysis.

how long did it take you to put all of this data together in the first place??

The first one, in my first post, took about ten hours, as it took awhile to figure out how I wanted to approach it.

The second one in this thread, probably about fifteen hours. It took a few hours to coordinate getting data submissions. Then it took an hour or two to load the data and remove duplicate hands. Then I just ran the same queries as I did last time, so that part was fast, but I did spend some time putzing around with thresholds. Finally, formatting and assembling the post took kinda long, actually. That's why it went up at 1am.

The real "cost" in assembling this data is the time lost at the tables. At Barry or MG can attest to, I haven't been online very much lately. My absence only partly due to this, though. [img]/images/graemlins/smile.gif[/img]

I like all of your ideas for data analysis. I will pursue them sometime undetermined time in the future. It definitely won't be right away but I'll mull it over in my mind and eventually be curious enough to work out some more numbers.

[rharless]

Yes, I think the aggression factor is ideal. The problem, however, is that the post-flop aggression factor requires a significant number of hands to be ball-park accurate, especially for a tight VPIP player. Say we have just 100 hands for that player -- this means that they get to the flop about 20 times in those 100 hands (this includes their blinds). Calculating their post-flop AF on those 20 hands is really a shot in the dark. I don't even know if 500 hands would be enough to rely on AF as being in the right ballpark.

I think actually many post-flop measures (W$SD, Went to Showdown, etc) would only start to be accurate for players with hundreds of hands. I'd have to cut my analysis off to remove players with fewer than 500 or 1000 hands or so. I'd have a very small data set to work with after the fact.

[Bozeman]

Let it be said that I love this analysis. However, microboy's complaint is even more valid than you concede. Suppose you have a set of players that all play the same hands, and play them all in the same way. After N hands, half of these players will be more aggressive than median, half will be less. The more aggressive players will do better, and only because they have had better hands. So this analysis has a systematic error in favor of players who are labelled aggressive. This error will decrease in magnitude with number of hands, but it will always be present, and not decrease as quickly as you might think.

On the other hand, there is a systematic error that favors players identified as loose. It is entirely possible that these errors alone explain the better performance of medium aggressives than tight mediums, or LAG vs. TP.

Craig

[rharless]

Craig, Do you mean HajiShirazu's complaint? If not, I am very confused.

I agree thoroughly with your comments. I am wondering at what threshold of hands, would you feel that based on numbers alone, you could correctly characterize a player as LAG or TP etc?

My feeling was usually that after four orbits or so, I can "feel" if a player fits a certain profile, but that is insufficient for this analysis... because when I am playing, I can observe the quality of hands that are shown down, and get a feel for someone who appears to exploit stealing opportunities. But, that shortsighted view is what let me to use a cutoff of 40 hands. The threshold probably has to be much higher.

[Bozeman]

You're right, I was too lazy to open another window and reread the whole thread, so I guessed about who had said what.

The biggest problem, I think, is that AA is so much better than all other hands. After 200 hands, ~1/2 the people will have never gotten AA, and their PF raise will likely be down .5% and their win rate ~2 BB, while those who have had it twice will be .5% more aggressive and ~2 BB better. No other hand will have nearly this affect, unless there are TA's who raise with 72 every time.

Craig

[Izverg04]

Thank you for providing this data! This is certainly information
that I came looking for to this forum.

My one remark is that while many have noted how you can have
systematic bias in your data due to your selection (Nplayed>40,
different cut-offs for classification), nobody has said so far the
obvious: your numbers are very much statistically limited despite
the number of hands you and your friends have managed to put
together! Your findings can be misleading without giving the
error bars.

If I try to calculate the errors for the example most interesting to
me at this moment, $2/$4 and lower, I get using your data (see
below for the details of the calculation):
<font class="small">Code:</font><hr /><pre>

Micro and 2-4

|| Tight Normal Loose
----------------------------------------------------------------------------
Aggressive || 2.5+-1.3 2.0+-1.2 -1.8+1.5
Normal || 1.5+-0.8 2.1+-1.0 -1.1+-1.5
Passive || -0.2+-0.7 -1.4+-0.8 -4.0+-0.8
----------------------------------------------------------------------------

This should warn you against taking these numbers too literally.
For example, you can't determine from your data whether the
tight/passive player is a donor at the small stakes table (and I am
pretty sure this is not the case).

Thanks again for your effort.

Izverg, aka "the new geek on the board".

----------------------------------------------------------------------------------------------------------

To determine the errors, the only input you need is the standard
deviation of EV per hand which strongly depends on the style of
play. I haven't found the relevant numbers by searching this
forum, so I have to rely on my educated guessing. I think that
players can expect the following st.dev. (in BB per 1 hand
played):

|| Tight Normal Loose
---------------------------------------------------------------
Aggressive || 2.5 3.0 3.5
Normal || 2.0 2.5 3.0
Passive || 1.5 2.0 2.5

You can question my assumptions, but obviously variance will be
larger the more hands you play, and the more aggressively you
play.

Now calculating the error for the win/per 100 hands that you
calculate by averaging results from N hands is easy:

Sigma=sigma(1 hand)/sqrt(Nhands) x 100 hands.

[Toonces]

There's another flaw too. The results of this study show that at most levels, the most profitable players are tight-aggressive, and the least profitable players are loose-passive. In addition, it is fairly well-known that tight-aggressive is the best strategy, so I assume people that the "best players" would rarely buck that trend.

But this is different than the implication that if you switch your game to tight-aggressive, your game will improve. To this, the study really can't help. It also doesn't tell you if it was the "tight-aggresiveness" that leads to the profit as much as it is just the style that the most skillful players use.


For example, a similar study at a live pokerroom might show that players wearing UltimateBet caps have a higher than average EV. But that doesn't mean that you will improve just by putting on a UB cap. Or that the link between UB caps and profit isn't just a correlation based on UB giveing caps to top players.