Calling Math Heads: "Villian is 27/11/1 over sample size..."

[Res-Judicata]

Pokey?

Anyway, I'm curious how I much confidence I should assign PT stats over a given sample size.

Is there some number beyond which I should I feel confident that these stats are an accurate reflection of a particular player? Right now, if I have > 1K hands, I assume stats are accurate.

Basically how should I approach someone's stats for whom I have 30 vs. 100 vs. 250 hands in my PT DB?

Thanks!

[Tickner]

PF stats converge very quickly. 100 hands gives you a good idea. 1000 hands will give you almost exact vpip/pfr stats.

Anything else, I dunno.

[CappyAA]

I usually feel comfortable looking at their stats over 200 hands. When it gets over 1000, I am almost positive their stats are telling the whole story. Basically I agree with Tickner, lol.

[Pokey]

I'm WAAAAAY out of the ordinary on preflop stuff -- my HUD starts showing me stats after 10 hands. My feeling is that:

1. Some information is better than no information, and
2. Even after 10 hands I have some idea of what I'm seeing.

For instance, if a player has a true VPIP of 50% or more, then the odds that they would go 10 hands without seeing more than 2 hands is under 21%. That means that if I see someone who has a VPIP of 20%, 10%, or 0% after 10 hands there's a 4-in-5 chance they're not a complete lagtard with a true VPIP of 50%. Conversely, the odds that someone with a true VPIP of 25% or less would see 4 or more hands out of 10 is under 23%. Therefore, if after 10 hands someone has seen 4 or more, I can be 3/4ths sure he's not a tighty.

The more hands you see, the better your read becomes -- that's why we datamine, that's why we observe, that's why we watch. But I'd rather have potentially faulty and questionable data than no data at all when I'm trying to make a decision at the tables.

Postflop data converges more slowly, just because you don't see as many postflop observations per hand. The higher a person's VPIP, the faster their postflop numbers will converge -- a person with a VPIP of 75% is giving you five times as many postflop observations per 100 hands as a person with a VPIP of 15%.

There's no "magic number" when a statistic starts to be relevant; it just gets gradually less and less noisy over time, and therefore it gets gradually more and more accurate. I'd say that once you've got 100 observations you'll have a reasonably accurate read on a player's VPIP and PFR (assuming they don't have multiple gears!) and once you get to 100/(VPIP) hands you'll have a reasonably accurate read on the player's postflop aggression, WtSD, W$SD, W$WSF, etc.

[cbloom]

One thing to be aware of is that the tighter someone is preflop, the slower their postflop stats converge. Some of the super-tighties that are like 5/4 preflop can have very messed up postflop stats since they hardly ever play postflop.

[kslghost]

Never met a freaking 5/4, but this is definitely true. In my opinion. Most stats begin to converge around 50 of whatever it is. So let's say you want PFR/VPIP, you only need a handful of orbits to get a general idea. You can't necessarily rely on them rock solid, but a person who's seen 50% of flops is generally going to be quite a loose player, etc.

Other stats begin to converge more slowly. W$SD is one of them, as this one has high variance over a short course of hands. W$WSF as well. It is hard to rely on these stats (these aren't THAT useful anyways as they often just tell you how hot a person might be running). I find WTSD to be one of the most important, as it tells you who is playing too deep, and once I have about 200-400 hands (depending on their VPIP), I feel that it becomes a more reliable statistic.

Positional stats converge after like 1000 hands though, and others even less, so start being aware of opponents who you've played with for a very long time and start peeking at how they play in position.

[MatthewRyan]

25 hands is the measure I use fwiw.