100k hands.... open for inspection/selection/detection/interjection

[Jay Riall]

[ QUOTE ]
I doubt this is in NL200 games, where variance is exponentially lower than nosebleed stakes

[/ QUOTE ]

Obv, but there is still a very decent amount of variance, an amount that means no-one is just a 'lifetime 5ptbb/100 winner' or w/e.

[Nick Royale]

[ QUOTE ]
[ QUOTE ]
I don't think people saying 100-200k hands means nothing in terms of winrate have a clue about statistics. If you have you're welcome to prove me wrong.

[/ QUOTE ]

They obviously mean something, but people who think they have a 'true winrate' which they will maintain within 1-2 points in the future are seriously misguided. You're welcome to prove me wrong though. You realize cts has been breakeven over 100k hands before and has also run at 10ptbb/100 over the same sample in the same games?

[/ QUOTE ]
I tried to hand over the task because I suck at statistics, but I've read some and I think we can create a pretty good confidence interval at least over 200k hands.

The thing is I think stuff like cts breaking even over 100k hands makes people think wr over 100k hands means nothing, when it actually means quite a lot.


I'm going to make an attempt at this anyways. Someone please correct me if I'm wrong.

Assume your winrate is 8ptbb over 100k hands. The standard deviation/100 hands is about 50ptbb:

wr +- sd/sqrt(100000/100) = 8 +- 50/32 = 8 +- 1.5


Same wr over 200k hands:

wr +- sd/sqrt(200000/100) = 8 +- 50/45 = 8 +- 1.1


Cliff notes: This means that if your wr is 8ptbb/100 over 100k hands you can say with 90% confidence (or 95%, don't really remember) that your wr is in the intervall 6.5-9.5. The same intervall over 200k hands is 6.9-9.1.

[Jay Riall]

Okay, but I don't get how you can say this when people are constantly posting those EVgraphs where they run ridiculously over or under expectation over a very big number of hands.

[Nick Royale]

[ QUOTE ]
Okay, but I don't get how you can say this when people are constantly posting those EVgraphs where they run ridiculously over or under expectation over a very big number of hands.

[/ QUOTE ]
I've never looked close at those graphs because I don't get them [img]/images/graemlins/tongue.gif[/img]

Over how big samples?

Btw, if a 200nl player runs at a wr 1.5ptbb/100 or more lower than expected over 100k hands (which he should 5% of the time if my calcs are correct), his results will be $6000 lower than they should. There's tons of players on the forum and players running extremely bad probably like to show their ev-graph. That's the best explanation I can see (still not sure if my calc is correct though).

[_dave_]

[ QUOTE ]

The standard deviation/100 hands is about 50ptbb


[/ QUOTE ]

Is a fairly big assumption.

May well be correct, I have no idea - But I do remember the SD stat given in PT is a random mishmash of numbers bearing little similarity to standard deviation.

See Pokey's fantastic post on this matter for details.

[Mjafish]

[ QUOTE ]
Okay, but I don't get how you can say this when people are constantly posting those EVgraphs where they run ridiculously over or under expectation over a very big number of hands.

[/ QUOTE ]

rigged?

[Acemanhattan]

[ QUOTE ]
OP, i guess u have tilting problems?

[/ QUOTE ]

Have had tilting probs, still have massive spewing problems.

[Nick Royale]

[ QUOTE ]
May well be correct, I have no idea - But I do remember the SD stat given in PT is a random mishmash of numbers bearing little similarity to standard deviation.


[/ QUOTE ]
My PT number for SD was 44. Isn't this number reliable?

[Acemanhattan]

[ QUOTE ]
86o should not be a red number. [img]/images/graemlins/confused.gif[/img]

[/ QUOTE ]

Seat 1: Hubbo ($177 in chips)
Seat 2: AngeL_fH ($228.35 in chips)
Seat 3: Aloon74 ($315.50 in chips)
Seat 4: Acemanhattan ($374.20 in chips)
Seat 5: kabutze ($198 in chips)
Seat 6: burle ($138.60 in chips)
kabutze: posts small blind $1
burle: posts big blind $2
*** HOLE CARDS ***
Dealt to Acemanhattan [8s 6h]
Hubbo: folds
AngeL_fH: folds
Aloon74: folds
Acemanhattan: raises $6 to $8
kabutze: folds
burle: calls $6
*** FLOP *** [Ad 6s Jd]
burle: checks
Acemanhattan: bets $12
burle: calls $12
*** TURN *** [Ad 6s Jd] [5s]
burle: checks
Acemanhattan: checks
*** RIVER *** [Ad 6s Jd 5s] [Kc]
burle: checks
Acemanhattan: bets $354.20 and is all-in
burle: calls $118.60 and is all-in
*** SHOW DOWN ***
Acemanhattan: shows [8s 6h] (a pair of Sixes)
burle: shows [Jh Js] (three of a kind, Jacks)

obviously rigged [img]/images/graemlins/mad.gif[/img]

[Acemanhattan]

[ QUOTE ]
Btw, if a 200nl player runs at a wr 1.5ptbb/100 or more lower than expected over 100k hands (which he should 5% of the time if my calcs are correct), his results will be $6000 lower than they should. There's tons of players on the forum and players running extremely bad probably like to show their ev-graph. That's the best explanation I can see (still not sure if my calc is correct though).

[/ QUOTE ]


So the opposite may be true, and I may be running 1ptbb/100 higer than should be expected.....