Hands needed to rely on statisitc

what is in your aopinion a good sample size is necessary to be able to rely on the puck open % stat being an accurate microcosm of that players tendency?

I dunno much about all that other stuff but a dagger is pretty good on him.

http://www.husng.com/content/interpreting-small-sample-sizes-bayesian-estimators

I filter my hud to only show the most recent 300 hands I dont care about hands played 2+ days ago I consider that almost worthless esp when it comes to statistic, very often I see certain players fluctuate between 20vpip for 2-3 hours, they lose 4 or so all in pots, now that vpip goes up to 30 as does the big smile on my face

I think someone emphasized that not until 1000-2000 hands does a stat really specifically define something about someone, the kind if information you can take to the bank.

On April 19 2013 20:48 Twisted wrote: I dunno much about all that other stuff but a dagger is pretty good on him.

I'd like to see sample size in the microcosm of low level games, I'm guessing the player tendency will be not to micro accurately enough to be able to rely on the puck open in teamclashes

That's awesome Tom.

On April 19 2013 21:48 TalentedTom wrote: I filter my hud to only show the most recent 300 hands I dont care about hands played 2+ days ago I consider that almost worthless esp when it comes to statistic, very often I see certain players fluctuate between 20vpip for 2-3 hours, they lose 4 or so all in pots, now that vpip goes up to 30 as does the big smile on my face

Interesting

+1
I use the double click to show only stats at the table a lot how do you filter for last X hands from all tables?

On April 20 2013 06:32 LemOn[5thF] wrote: +1 I use the double click to show only stats at the table a lot how do you filter for last X hands from all tables?

For HEM2 I think it's: Home tab/HUD options/Stat Appearance/Min samples

I used to lock it at 50 just so I wasnt missing any crazy juicy aggro fish playing every hand. I mean it's pretty impossible for someone to run 75/45 after 20-30 hands at a full table without there being something going on.

You can do a lot with just two hands.

On April 20 2013 13:46 cariadon wrote: You can do a lot with just two hands.

even more with three

obv Bayesian Inference is the way to go.

while you wait for your 100+ hands sample sizes, someone who has a clue already stacked the fish and is on to the next one

--- Nuked ---

On April 19 2013 20:48 Twisted wrote: I dunno much about all that other stuff but a dagger is pretty good on him.

I lol'd

A players VPIP and PFR can be though of a sequences of bernoulli trials, that is they can be assumed to be independed trials in which probability of success (ie. putting money in the pot or raising) is always p. Assuming independence probably is not strictly justified, but in probability is all about making up unrealistic models which regardless give useful results.

So now that we have defined these as binomial trials we can use statistical methods to analyze them.

Skipping to the interesting stuff, you can use Wilsons Score Confidence Interval to estimate 95% CI to VPIP and PFR. http://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval

The formula (for 95% CI, for other CIs you would replace 1.96 with the relevant critical value of standard normal distribution) would be: upper bound = (p + (1.96/2n) + 1.96*squareroot(((1-p)p)/n)+(1.96/(4n*n))))/(1+(1.96/n)) where p is the estimated vpip or pfr and n is the amount of hands played. Lower bound would be (p + (1.96/2n) - 1.96*squareroot(((1-p)p)/n)+(1.96/(4n*n))))/(1+(1.96/n)) or see this from some site, its messy here and i probably mistyped somewhere.

This could be done with way easier calculation using asymptotic confidence interval, but the problem with that is that asymptotic method has poor properties with small sample sizes and with p = 0 it gives a single point (namely 0) as interval.

So with this we can see for example that if vpip = 15 over 50 hand, we can be confident (given our unrealistic and silly assumptions) that the true VPIP of this player is between 6 and 26 %. Over 100 hands that would be 9-23, over 200 11-21 and over 300 hands 11-19.

for vpip 0 over 20 hands we get 0 to 16 %. Note that for hand amounts smaller than 20 these start to get meaningless, because in our reasoning we rely on the central limit theorem which applies when n is "large" ("large" not being well defined, but usually taken to be like 30).

You can easily program the formula to excel (or if you use R you don't have to program anything) and obtain these yourself.

On April 19 2013 23:29 NewbSaibot wrote: I think someone emphasized that not until 1000-2000 hands does a stat really specifically define something about someone, the kind if information you can take to the bank.

Well depends how specific information you want. With 2000 hands you can get a reasonable amount of hands for situations that do not occur all the time, but for average VPIP or PFR 2k hands won' help that much. VPIP for example is essentially as accurate after 300 hands as it is after 2k hands. (for a player with VPIP 20%, 95% CI is 16-25 and 18-22 respectively).

^^ I concur, do you guys also concur?

i think it depends on how precise the info you need is.

If over 200 hands some guy is 19% VPIP, its fairly safe to assume he isn't a 70% VPIP Spewing monkey, you don't need 2K hands for that.

But he could possibly be a running good 15% VPIP rock or a running bad 25% VPIP tag.

VPIP is one of the slowest to converge. However, after 122 hands you can expect it to be within 2% of the actual value. This is on a big database btw