samplesize needed for statistic-reads (VPIP/PFR/AF)?

[trontron]

how many hands do you need, until your (statistical) reads on a villain become reliable? i'm talking about villains VPIP / PFR and AF.

i think you need a lot less hands for PFR and VPIP, and probably about 10times more hands for AF. i'd say 50hands is already very reliable for PFR and VPIP, but not for AF.

has anyone calculated this more precisely? (i.e. "how many hands do you need to be +-2percent sure with a confidence interval of 95%" or something like this)

thanks and sorry for my bad english!

[ottsville]

[ QUOTE ]
i'd say 50hands is already very reliable for PFR and VPIP,

[/ QUOTE ]

Fifty hands MAY give you an INDICATION if the player is to the extreme one way or the other. But, a 50 hand sample is too small to really determine with any accuracy the stats for a player.

Think about this..How many playable hands can you expect in a 50 hand sample? Raisable hands? How many times have you gone 30+ hands and never had a playable hand outside the blinds? Ever gotten AK 2x in a row followed by JJ?

[trontron]

of course you're right. my OP was unclear. what i wanted to say, is: if someone has stats like 15/10 or 70/3 after 50 hands, it is quite unlikely that these stats are totally wrong. of course, they could be 18/8 or 50/6 players "in reality", but it is unlikely, that the 70/3 guy is "in reality" a TAG player with 18/10 stats.

so:
[ QUOTE ]

has anyone calculated this more precisely? (i.e. "how many hands do you need to be +-2percent sure with a confidence interval of 95%" or something like this)

[/ QUOTE ]

[ottsville]

I would think to have a good confidence interval you would need several hundred hands for VPIP/PFR and probably twice as many for AF. However I wouldn't know how to statistically determine the number of hands needed...hopefully someone with a stronger mathematics background will chime in. You may try posting this in the "poker theory" forum.

[AKQJ10]

I really don't think it would be any different from any other confidence interval problem, although I've forgotten how to do those.