Pokertracker stats and confidence intervals

[sam h]

[ QUOTE ]
Even 37 hands can be useful. If it's 35/5/1.1 it's not very but if it's 55/1/3 than if can be. You just have to have a good understanding of confidence intervals and player types and stat ranges.

[/ QUOTE ]

I think this is a good point. The usefulness of small samples is much higher when you are dealing with fish with extreme scores on the VPIP/PFR. In a practical sense, it doesn't make that much difference whether someone has a true VPIP of 45 or 60 or a true PFR of 2 or 8. You want them in your game either way and will play against them roughly the same way anyway. But the difference between VPIPs of 24 and 36 (or PFRs of 12 and 18) is huge both in terms of game selection and in terms of adjusting play accordingly.

I think what these confidence intervals demonstrate is that for players that fall in the VPIP 20-35 range, a sample like 37 hands is pretty useless.

Of course, the best answer is just to collect more hands. So keep up with the datamining! [img]/images/graemlins/cool.gif[/img]

[sam h]

I agree with you that qualitative assessments are also problematic. The best strategy is going to be to combine the two in a useful way. I think that part of this is getting a good intuitive feel for what plays are worth making notes on and what are not. If the game has been pretty standard and a guy opens with A7o UTG, that is something I will note because - while it could be a random event as you note - it is still likely to say something about his opening range. Then one can combine those kinds of observations with your quantitative evidence and try to come up with the best composite picture you can.

[naphand]

There is much that can be learned about players quite quickly IMO just from a fairly basic tactical perspective.

The stats produced by PT/PV/GT+ do need to be qualified by an understandiong of their reliability, this is very important IMO. But it is also true that a player with V$IP 60% over 27 hands is far more likely to be loose than tight but the statement "far more" is accurately qualified by confidence intervals. Such stats, however, are supported when we also observe them limping 1st-in with 84s in the CO. Just seeing a few hands like this tells us this player has loose standards. Ditto for post-flop (though more variables to consider, of course).

We do not need to calculate these as we play, but an idea of reliability is very useful IMO. Too many "villain was 56.27 after 50 hands" type of thing. New players and experienced alike would do well to understand these confidence intervals as so much of what we do is based on partial information, or unreliable information or plain guesswork.

I agree that a "standard line" with "no reads" is correct, it is just a questions of where that "standard" starts to blur into more specific read-based play.

As for table selection, we do have individual stats AND the table overall (5-6 players). We are looking for loose players/tables and small stat samples may be unreliable but they do give us an idea of which tables are probably loose. A table with 26% seeing the Flop is much less likely to have a bunch of Fish than one with 52% seeing the Flop, even over just 20 hands.

Data-mining is most useful for identifying regular players, by definition this means the better players/winners. The fish turnover is high, and the ones that stick around very long tend to be relentlessly persued by the TAGs who have buddy-listed them... [img]/images/graemlins/grin.gif[/img]