Organizing a project to determine which sites are legit or rigged

[InTheDark]

[ QUOTE ]
For instance, some people say that they get sucked out too many times by people who go all-in. One possibility how a rigged site could try to accomplish that, is by filling up a possible straight - or flushdraw on the river in an all-in situation whereever that is possible and regardless of the hand of the all-in player.
This is easy to check and we would definitely not need millons of hands for that. A couple of hundred of relevant hands could already be enough to produce a significant result.

This was just an example, I am sure there are much more interesting checks to do.

[/ QUOTE ]

This is the route. Easy to test. Small quantity of hands needed to reach a hypothesis test. But....

What are you looking to do with the result?

No one will believe you. The site will blow you off. Or off you.

I've wrestled with this issue several times and here's the bottom line: You can test sites yourself and use that info yourself and that's all.

There are several sites I know bugger the turn and river cards.

Enjoy!

[InTheDark]

[ QUOTE ]
The way I see it is this -
If you analysed allin five hands to see if there was balancing you would calculate the EV for each hand and add it up to give your EV result - say your EV calculation says you should be up $3000 from these five hands.
Then you deal the cards and calculate your actual result - say you got $2000.
So you are up 2/3 of what you should be.

Now obviously that means less than nothing because the sample size is miniscule. But if you did it for a couple of million hands from several winning players if a result showed up that the actual winnings were only 2/3 of what they should be well then that would be extremely daming evidence. I don't get why some people are proposing complex statistical anaylsis, all that is need is an EV-Actual result comparison from a big enough sample of hands from winning players. If there is a big discrepancy then there is a problem.

[/ QUOTE ]

If you understood statistical methodology, you'd realize that a test of a hypothesis might require sampling only a few dozen or a few hundred hands. Every thinks 'millions' and they're usually ignorant.

[nineinchal]

[ QUOTE ]
[ QUOTE ]
The way I see it is this -
If you analysed allin five hands to see if there was balancing you would calculate the EV for each hand and add it up to give your EV result - say your EV calculation says you should be up $3000 from these five hands.
Then you deal the cards and calculate your actual result - say you got $2000.
So you are up 2/3 of what you should be.

Now obviously that means less than nothing because the sample size is miniscule. But if you did it for a couple of million hands from several winning players if a result showed up that the actual winnings were only 2/3 of what they should be well then that would be extremely daming evidence. I don't get why some people are proposing complex statistical anaylsis, all that is need is an EV-Actual result comparison from a big enough sample of hands from winning players. If there is a big discrepancy then there is a problem.

[/ QUOTE ]

If you understood statistical methodology, you'd realize that a test of a hypothesis might require sampling only a few dozen or a few hundred hands. Every thinks 'millions' and they're usually ignorant.

[/ QUOTE ]

This is not my expertise.I would require a professional statistician for this evaluation.

[Sciolist]

[ QUOTE ]
For example off the top of my head his ability to call down bluffs and fold strong but losing hands would increase.

So you take the hand histories and break them down into similar categories with the difference being the relative strength of the suspects hand. Then you compare them. If a player calls with A high when the opponent is bluffing but folds with A high when he is beat that is evidence.

[/ QUOTE ]
You'd never know this because you wouldn't know his holecards. All you will see are his holecards that go to showdown that he shows, as our sample size is going to be datamined.

[Sciolist]

[ QUOTE ]
You do realize that is normal? The percentage of bad players online is much greater then what you'd get at most casinos. If a larger number of players chase they will hit more often.

[/ QUOTE ]
The average online player is a LOT LOT LOT better than the average live player these days. You can't even compare. The reason people think the river always hits is because they are subconsciously used to the variance per hour in live games, and that number is way different online due to the number of hands per hour. It's also down to:

http://en.wikipedia.org/wiki/Confirmation_bias
http://en.wikipedia.org/wiki/Selection_bias

[Sciolist]

[ QUOTE ]
Party's net profit last year was $128M as it says on their site. And last year Party was bigger than Pokerstars so Pokerstars would've made nowhere near $400M

[/ QUOTE ]
PS at its peak is the same size as Party at its peak, infact I think slightly larger now because Party had a well defined time when they had the most players on, whilst PS has a couple as they have so many more Euro players now than Party ever did.

[Sciolist]

I suggest that you come up with a few definitions of what you want to check for. Then you check for them. Do something simple to start with, like "in all the hands that I was dealt, am I getting the proper distribution of holecards". You can't check that with a datamined sample by the way, as you only see showdown hands, not starting hands and there's a huge difference.

You will find that it's hard to define things closely enough to make them a useful test. You will find that when you prove one thing has the expected distribution, people will say "yes, but the site is actually rigged for abc".

You therefore will have lots of definitions of how a site might be rigged. You will be able to look in to some of them easily, and I suggest you start with those right now, rather than wait for a "definitive" way of measuring whether a site is random or not.

[Sciolist]

[ QUOTE ]
If you understood statistical methodology, you'd realize that a test of a hypothesis might require sampling only a few dozen or a few hundred hands. Every thinks 'millions' and they're usually ignorant.

[/ QUOTE ]
It depends what you measure. If you want to detect 0.1 ptbb variations in win rate, you need millions of hands. Infact, it's almost certainly impossible because the games change or the players you are measuring change over that period. On the other hand, if you want to know how often you are dealt aces, millions is massive overkill

[Freyalise]

[ QUOTE ]

If you understood statistical methodology, you'd realize that a test of a hypothesis might require sampling only a few dozen or a few hundred hands. Every thinks 'millions' and they're usually ignorant.

[/ QUOTE ]

Depends on what significance value you are using.

If you determine that the probability of observing whatever result you did is less than 1%, but you only looked at a few dozen hands, then your conclusion is not too exciting since there are many tens of thousands of sets of 'a few dozen hands' that you could have looked at and evidently 1% of them would fail the test assuming that the null hypothesis (whatever site has a perfectly random shuffle) is true.

By looking at a decent sample size, you could choose an extremely small significance value (like 1*10^-10) say, which would obviously be much more impressive if your data rejected the null hypothesis.

[Henry17]

[ QUOTE ]
[ QUOTE ]
For example off the top of my head his ability to call down bluffs and fold strong but losing hands would increase.

So you take the hand histories and break them down into similar categories with the difference being the relative strength of the suspects hand. Then you compare them. If a player calls with A high when the opponent is bluffing but folds with A high when he is beat that is evidence.

[/ QUOTE ]
You'd never know this because you wouldn't know his holecards. All you will see are his holecards that go to showdown that he shows, as our sample size is going to be datamined.

[/ QUOTE ]

You are right. When he folds we don't know what he had. The test about how many times he calls a bluff correctly could still be tested. I think that would be enough but you could probably design some tests about his behaviour vs flush draws and look for a difference when the opponents hits vs when he doesn't.

I haven't really given it much thought but there must be a few behaviours that are different for a player who can see hole cards vs a player who can't. I'd need coffee first before I can come up with some.