The EV of different playing styles

[rharless]

Hi MaxPower,

Thanks for your stats. The reason I did it in Access is because I was starting out trying to do a 9-box grid and it was just easier to write one query and run it in Access. Also in PT, you can't do a range (which I would need for my 9-box grid), so if for example my "average aggression" is defined as 5-8% PFR, you can't do that in PT.

Here are my LAG stats using the hand cutoffs you suggest and my VPIP (22%) and PFR (7%) cutoffs

at least 100 hands: EV +0.40 (33870 hands)

at least 200 hands: EV -0.90 (14729 hands)

I have mixed emotions in general about using higher cutoffs of hands, because I find in general, the more hands people play, the better player they tend to be regardless of the playing profile they fall under. The only need to use any cutoff at all is to have a reasonable confidence in profiling the playing style, which I feel can happen at about 40 hands or so with a player. So limiting it to say 200 hands might be limiting it to only "good" LAGs etc.

Also, not to be picky, but in each category you don't have 30k hands, and I have always heard that around 30k hands you can establish some reasonable range of your winrate. I think your results might change quite a bit after your db doubles.

[MaxPower]

Hi,

Thanks for your stats. The reason I did it in Access is because I was starting out trying to do a 9-box grid and it was just easier to write one query and run it in Access. Also in PT, you can't do a range (which I would need for my 9-box grid), so if for example my "average aggression" is defined as 5-8% PFR, you can't do that in PT.

Yes, the pokertracker filtering is limiting. Also, I am not entirely sure how some of the statistics on the Summary page are totaled.

I have mixed emotions in general about using higher cutoffs of hands, because I find in general, the more hands people play, the better player they tend to be regardless of the playing profile they fall under. The only need to use any cutoff at all is to have a reasonable confidence in profiling the playing style, which I feel can happen at about 40 hands or so with a player. So limiting it to say 200 hands might be limiting it to only "good" LAGs etc.

I agree than you need to set a cut-off to correctly classify the players. Over a 40 hand stretch I might look tight-aggressive, loose-passive or loose-aggressive depending on the 40 hands that I am dealt. I wanted to make sure that I was classifying players based on their playing style rather than on the hands they were dealt (although 100 still is low).

Also, not to be picky, but in each category you don't have 30k hands, and I have always heard that around 30k hands you can establish some reasonable range of your winrate. I think your results might change quite a bit after your db doubles.

Analyzing data is very tricky business. It is quite possible that there are outliers in both of our data sets than are have a large effect on the win rate for their groups. Perhaps I happened to play with a few maniacs who had a good run.

Anyway, I just thought I would share my data. I don't think my data is perfect by any means.

I'd be glad to help out if you want to pursue some more analyses. I'm a geek.

[rharless]

I think the differences in our results is somewhat the proof that you really just need thousands and thousands of hands to get a meaningful, reliable result.

I compared the summary totals in PT with the totals I got when I calculated the overall EV in Access and they were equal, so I am confident in PT's averages. Either that or Pat and I make the exact same mistakes [img]/images/graemlins/laugh.gif[/img]

I have gotten a couple of PMs from people who are willing to contribute their data sets, so once a few days go by, I will get in touch with you (and anyone else who PMs me) to let you know what's happening with the effort to combine of data into one larger dataset.

[Jim Easton]

[ QUOTE ]
But loading say 5x as many hands into it would be quite slow indeed.


[/ QUOTE ]

Yes, it is.

[jonahmavesin]

hey rharless,

awesome post. as a consultant, my mouth starts drooling at the idea of lots of data to start analyzing.

a histogram is just a graph of the results, and it would be useful here, since you are using relatively arbitrary cutoff points for segmenting (50th percentile) and then taking averages. Those are great first steps to get top line results as you did, but they can hide more interesting detail.

A simple histogram would be to just graph each quadrant against your criteria. So, a graph of TA players charting VPIP across X and EV on Y. This would theoretically give a brief look at the "optimal" VPIP level. Same thing could be done with PFR%, and across all four groups.

The problem is, the averaging IS necessary, to iron out the variation. With a sufficiently large Dbase, this isn't a problem. With yours, it probably is. The way a histogram deals with this is charting not a line graph with each discreet point, but charting a bar graph and grouping data into deciles or even quartiles. (e.g., for the TA graph, you'd have 10 bars - the first for the first 10% of players by VPIP or PFR, and so on, the height still average EV in the sample. the bars would form a graph with a peak at the "optimal" level.). The fewer the segments, the less variation will throw off the shape of the graph, but the less detail you'll see. Your four data points are essentially four historgrams, with segments size = 1 (i.e., it's one bar including 100% of players with that style, and it's height is the entire average of the group).

hope that helps. not sure if Access can draw graphs. wouldn't be hard to do using Excel though. especially if you start building bigger dbases, you should definitely play around with different views like this. I'd be happy to help.

[Bozeman]

OK, so postflop aggression is not easy. However, how about doing it with WENT TO SD %, which is a measure of postflop looseness?

Instead of dividing the data into quadrants, 9 or sixteen would be interesting, and 25+ would allow you to see the structure of various styles, and you could do a sort of visual average to get decent signal to noise, and see if there is any overall structure. Also might help you eliminate outliers.

Craig

[rharless]

how about doing it with WENT TO SD %, which is a measure of postflop looseness

This statistic is not available on the PT summary tab. It has "Won $ at Showdown" but not Went to SD. It is far easier for me to use the statistics that are already calculated on the summary tab, then to go into the database and derive the numbers from the raw data myself.

The Won at Showdown is more of a measure of river-looseness, I think, and also I find that statistic is not often meaningful for players that have <1000 hands in my DB.

[Norm]

Instead of a simple cutoff point, could we measure different ranges of looseness?

i.e. <20%,20-25%,25-30%,30%+ (or even tighter categories)

Then we can see at what point being loose really starts to cost you.

[rharless]

The problem is that I do not have enough hands to support a breakdown with this level of granularity. I thought one needed 30k hands to determine a true winrate. There is a recent thread in SS that says you need 80k hands to determine a win-rate.

If I split the hands up into that many categories, there is less than 20k hands in each category and so the result is not statistically valid. Interesting, but not valid and therefore asking for misinterpretation.

I have received PMs from several volunteers with hand databases that combined with mine, will reach about 250k hands. But, I will have to remove some data points due to duplicate hands, I am sure. I was really hoping to have a combined dataset on the order of 500k hands to work with.

I am still accepting and looking for PMs from people willing to contribute datapoints from their "Summary tab" of their PT databases.