Official Poker Site Data Analysis and Discussion Thread

[Mike Kelley]

Originally Posted by king_nothing_

I feel like I must be missing something since I have yet to see someone correct this. Everything below 2.5% is not outside of 3 standard deviations. It's a little less than 2 standard deviations.

Thank you, That was annoying me too.

[Dire]

What exactly is stopping this from going forward? It seems like this test would be fairly easy to implement. Filter all hands to just be within the general period of examination, perhaps since the beginning of this year. Define regulars as players with more than 'x' (150k?) hands with a BB/100 > 1.5. Run that sample's net all-in equity, run and compare to the sample consisting of the net equity for 'everybody else.' Assuming there is a relevant discrepency, we could then begin looking for possible confounding variables.

As one aside, I think it is important to choose a BB/100 significantly above break even when segregating the players. If there were any rigging, it seems logical that it would mostly be targetted towards benefitting small losers at cost to the small/large winners since that would have numerous benefits to the site (more games, 'softer' games, more players, numerically more players getting 'lucky' than 'unlucky' for word of mouth discussion, negligable risk of detection - the sites always have plausible deniability and variance in their favor while, etc). Point being that just choosing > 0 as the point may easily end up with players who should be in one group ending up in the other.

[Elbow Jobertski]

I think the examination loses some legitimacy when it is framed as "is this site rigged" as much as what closer analysis would actually show, that the RNG is flawed in some way. At this point we are limited to a discussion of the "MicroBob effect" as if this is even close to significant it makes calculating expected results to a degree of accuracy that holds up after millions of trials impossible.

(This effect is generally the same thing as keeping track of live cards when playing stud, just that the immediate effect is much, much less because one is not seeing actual cards and just assigning a range.)

I remember the good old early days when the claim was that the sites dealt out extra playable hands to encourage action... At least that was easily falsifiable.

What would be interesting is a pokerstove type program that would allow a player to also put folding players on a range of hands.

[Dire]

Originally Posted by Elbow Jobertski

I think the examination loses some legitimacy when it is framed as "is this site rigged" as much as what closer analysis would actually show, that the RNG is flawed in some way. At this point we are limited to a discussion of the "MicroBob effect" as if this is even close to significant it makes calculating expected results to a degree of accuracy that holds up after millions of trials impossible.

Two things.

1. I expect that the whole bunching conversation is mostly just a red herring. To start with nobody (from my reading in this thread) has put fourth any suggestion on why it would favor bad players over good players for example.

2. For confounding variables that are sample independent (such as bunching would be) we have control samples. You can run analysis on equity scenarios on Full Tilt and Stars, for example. If the results differ significantly then it is highly likely that exactly one site is 'rigged'. If the results are the same then it shows that they are 'rigged' in identical fashion which is extremely unlikely, or that they are both legitimate which would be the likely conclusion.

[Dire]

Also, I think framing the question in terms of the validity of the RNG is improper. If there was any illegitimacy in the deal, it would not be this blatant. It would almost certainly be based on $ or equity and not via universally biased distributions. What I mean is that the search should not be geared towards finding things like your two card flushes only coming in 32% of the time instead of 34.97% of the time - that's just not going to be the case. But, for example, if 'regulars' flushes tended to come in 36% of the time in small pots or in pots against other regulars, but 34% of the time in larger pots against fish - then you would have very significant evidence of a biased deal, but assuming that the biasis were balanced then it would be statistically invisible. The only thing that would show up would be an unexplained downward trend in the 'regulars' net equity.

[GT30]

i did the same analysis that 1pokerboy did and took all of my 100nl hands from this year, 850k total, and i filtered out all of the showdown hands and it shows this:



i'm under expectation but nothing substantial

[GrannyMae]

Originally Posted by TakenItEasy

I think there are a lot of silent 2+2ers who have suspicions but have remained silent. Noticing statistical anomalies is one thing. Coming up with evidence proving that they are not anomalies is quite another

I do think that rigged discussions poping up with higher frequency is quite telling. One or two can be shrugged off as players tilting/running bad. However when I see more and more people around the web and in chat boxes noticing the same things I am seeing, that doesn't seem to be coincidence. Some might even go as far as to say it's bordering on common knowledge. I'd say most of the bad beats I see taken by a decent player is followed by some Stars comment hinting at rigging.

is it possible that we can get a programmer to chime in here, or even one of our software forum superstars?

this is the simple question i have for those in the know.

is it even possible/practical to include such rigging in the software? i have been told that to write code into the software to apply doomswitches and bad beats etc is nearly an impossible task due to the complexity of such coding??

anyone?

[Dire]

Originally Posted by GrannyMae

is it possible that we can get a programmer to chime in here, or even one of our software forum superstars?

this is the simple question i have for those in the know.

is it even possible/practical to include such rigging in the software? i have been told that to write code into the software to apply doomswitches and bad beats etc is nearly an impossible task due to the complexity of such coding??

anyone?

It would be trivial code wise. The hard part is developing a method that is not easily detactable through statistical analysis (this is a math/statistics problem, not a coding problem) and the restriction of information - in that no more than a handful of people involved with the company should have any idea that this even exists.

[TakenItEasy]

Originally Posted by Dire

The hard part is developing a method that is not easily detactable through statistical analysis (this is a math/statistics problem, not a coding problem)

I think banning datamining makes it difficult, though not imposible to detect.

To get truely reliable results, it would take a coordinated effort among a large number of random players.

One seems to preclude the other.

[TakenItEasy]

Originally Posted by Dire

It would be trivial code wise. The hard part is developing a method that is not easily detactable through statistical analysis (this is a math/statistics problem, not a coding problem) and the restriction of information - in that no more than a handful of people involved with the company should have any idea that this even exists.

Seems simple enough to just use a 2nd RNG to boost a bad beat by whatever percentage you want.

If A beats B > 3:1 and RNG2=5% than bad_beat, else random_card

Of course, it's not really that simple. Functions understanding hands and odds are required but still easy enough.

[TakenItEasy]

I think a poll like this would give us some meaningful information:

I am a long term winning player on Poker Stars, for the last two month's:

[ ] I am a TAG and my results are well above average
[ ] I am a TAG and my results are above average
[ ] I am a TAG and my results are average
[ ] I am a TAG and my results are below average
[ ] I am a TAG and my results are well below average
[ ] I am a LAG and my results are well above average
[ ] I am a LAG and my results are above average
[ ] I am a LAG and my results are average
[ ] I am a LAG and my results are below average
[ ] I am a LAG and my results are well below average

I tried but somehow I messed it up. It would get better results from a more respected poster anyway.

[tiltymcfish0]

Originally Posted by king_nothing_

I feel like I must be missing something since I have yet to see someone correct this. Everything below 2.5% is not outside of 3 standard deviations. It's a little less than 2 standard deviations.

actually I'm not sure how it morphed into 2.5%, I think it was actually much worse as in 0.25% which iirc is > 3 SD away from mean in pEV

[king_nothing_]

Originally Posted by tiltymcfish0

actually I'm not sure how it morphed into 2.5%, I think it was actually much worse as in 0.25% which iirc is > 3 SD away from mean in pEV

.25% would be 2.8 standard deviations below the mean. He said "2.5%" in serveral posts, though.

http://davidmlane.com/hyperstat/z_table.html

Use the bottom calculator.

Mean = 0
Sd = 1
Shaded area = .0025
Then click the "below" option.

[Devilboy666]

Originally Posted by TakenItEasy

I think a poll like this would give us some meaningful information:

I am a long term winning player on Poker Stars, for the last two month's:

[ ] I am a TAG and my results are well above average
[ ] I am a TAG and my results are above average
[ ] I am a TAG and my results are average
[ ] I am a TAG and my results are below average
[ ] I am a TAG and my results are well below average
[ ] I am a LAG and my results are well above average
[ ] I am a LAG and my results are above average
[ ] I am a LAG and my results are average
[ ] I am a LAG and my results are below average
[ ] I am a LAG and my results are well below average

I tried but somehow I messed it up. It would get better results from a more respected poster anyway.

No it won't. People with 'unusual' results are more likely to respond which skews the data enough to make it useless for this case.

[tiltymcfish0]

Originally Posted by king_nothing_

.25% would be 2.8 standard deviations below the mean. He said "2.5%" in serveral posts, though.

http://davidmlane.com/hyperstat/z_table.html

Use the bottom calculator.

Mean = 0
Sd = 1
Shaded area = .0025
Then click the "below" option.

That calculator doesn't work on the computer I'm using now... but only 2.8 std deviations wtf? Wow guess I forgot everything I thought I once knew about stats.

2.5% anyhow I guess it don't matter but thanks.