thread seems rigged iyam

Have you noticed that the more work you do on your playing skills the more the rig the deal in your favour?

Hmmm, one of our rigtards has a shiny new gimmick account.

Is that you Smearsy?

Oh. Well, you're as ignorant as I am, then. That's no good. When you jumped into a question addressed to Pyro, I figured you were a statistics guy, too.

A) If any of the things you mention here have this large of an effect, then every book on poker is very wrong about the odds. If the probability of flopping a set differs wildly from 11.8% depending on whether you're playing full ring or heads up, low stakes or high stakes, how your opponents behave, and your own table image...then I think you've uncovered some major new poker theory.

B) Yeah, I'm not, either, but I can speculate with the best of them, which is always fun until a mathematician comes in and poops the party. But that's actually what I was wanting, here.

To find the standard deviation for flopping a set, I used the formula:

SQRT(npq)

where

n = number of trials
p = probability of flopping a set
q = 1-p

Not being a mathematician, I had to find that formula on the internet. I already knew you could calculate it if you have all the data points (by finding the mean, individually finding the differences, etc.), but I was a little surprised that simply knowing the number of trials and the probability of success allowed one to calculate the standard deviation for a data set. There has to be an assumption about the shape of the data, here, and I'm guessing the assumption in this case is that the underlying distribution is normal.

My question is...okay, I can't really articulate a specific question. I'd just like to read some educated speculation about why narrowing the data from all pocket pairs to just 9s and under had such a big effect. Sample size issue? Something to do with an erroneous assumption about the distribution? Card removal effects? Just a random outlier in the universe of poker results?

And is it even that much of an anomaly? 3.7 and 2.6 standard deviations are pretty far out there, and so is the difference between them, but with the number of people who play poker, we should expect quite a few outliers like that and several that are much worse.

Also, it would be interesting to know if the set-flopping results in Indy's huge data base form a normal distribution or something else.

how can somebody named moneymaker win a satellite and then the ME? see how online poker is rigged? all pokerstars' master plan. they probably rigged the live events too

If it is rigged, its rigged in my favor so I dont mind

As a newbie, dipping my toes into the online poker world for the first time, I would like to thank Ongame 'on the record' for rigging $5 turbo DON tables in my favour. This has helped build both my bankroll and my confidence.

I have no idea why Ongame have chosen to do such a thing, I certainly didn't ask, but I am very grateful all the same.

This is an estimate of the standard deviation, usually called the standard error of the mean. It assumes the population distribution is normal, not the sample distribution.

When doing estimates like this you should use a confidence interval to see what your estimate actually means. I do it with a T test for small samples. Using that, a sample of over 120 will have the same confidence level as a Z (normal) distribution, which means that the 5% level is at 1.96 standard deviations. For smaller samples, this number is higher, and can be found in a t-test table.

The formula is
p = expected
q = 1 - expected
n = sample size
act = actual

(act - p) / sqrt[(p*q)/n]

I am.

I'm saying those things do significantly impact card removal effects of broadway cards, yes. Just consider heads up vs. full ring. Think for a moment about how many hole cards ARE there heads up vs. a family pot into the flop full ring...

For full discloser my degrees are in computer science and economics. I'm close to a math degree as well (as well as physics.) So I can dabble in the maths. My lack of qualification in answering your question comes from the fact that your question was how STRONG the card removal effect should be. That's a poker question, not a math question.

I think maybe what suprised you about the standard deviation fall between the use of all pocket pairs and just 9's or below had more to do with the fact that the smaller sample size is going to reduce the number of standard deviations of its own accord EVEN IF THERE WAS NO CARD REMOVAL EFFECT. If some other pheonomon explains it the smaller sample size has it explaining it over a smaller "time frame" so it's less "bizzare."

My only intention in originally replying to you was to help point out to you some of you possible source of error.

By definition the standard deviation is smaller for smaller sample sizes when the ratio of actual/expected is constant. It's proportional to the square root of the sample size. But you also have the known removal effect here where broadway cards will show up on the board less often than low cards. So when you checked just 9s or less, you not only took out the removal effect, you also reduced the sample size, both of which should drop the standard deviation.

If your first ratio of 105/1260 or .0833 remained exactly constant on a smaller sample of n=673 (we know it changed very slightly to .086), then the SD drops from 3.8 to 2.8 from the sample size alone, without even removing the broadway cards. So that's the biggest part of the difference. The removal effect is small.

Over 1% of samples are expected to exceed 3 SD in a perfectly normal population. About 5% of samples should exceed 2 SD. And in small samples the estimating error makes these even larger.

Edit - in full ring games the removal effect is somewhere around 2% fewer broadway cards hit the board than would be expected randomly, due to players seeing more flops when holding broadway cards. Typical is ((.077-.0755) / .077) or 1.9% less.

One more comment on the set probability. Across all ranks, the probability is still 11.76%. The slope of the removal effect for the board crosses at 8s in very large samples of full ring that I've looked at, which hit at just about exactly that amount. Ace sets will still hit at around 11.5%, and deuces should hit at close to 12%. When looked at in the context of how often you see the flop with pair of deuces, say once in 440 hands, then this edge becomes .0026/440 or 1 / 169,000. So it is absolutely unexploitable and useless information in a real game, and really just interesting to know.

Saying this differs "wildly" from the calculated rate is kind of an overstatement. It doesn't differ enough to affect playing strategy at all.

http://www.pokerhandreplays.com/view.php/id/932405

I will take 0.118 as correct, cba to check that.

Each hand where you have a pocket pair is then a binomial trial with p = 0.118 of flopping 'set or better', 0.882 of flopping 'worse than a set'.

The variance of a single trial of a binomial variable is p(1-p), in this case, 0.104076. The variance of 1260 trials is therefore 131.13, and the s.d. of 1260 trials is sqrt(131.13) = 11.45.

You expected to see 1260 * 0.118 = 148.68 successes, you saw 105, a difference of 43.68. 43.68/11.45 = 3.8, so it looks like you calculated things correctly!

I think the s.d. of n trials of a binomial variable with probability p is *exactly* np(1-p), and no assumption is made about the distribution of n trials looking like a normal distribution. Variance is additive for any independent variables, regardless how they are distributed.

The only time normality is used is when we say something like "being 3.8 s.ds away from the mean implies that only x% of similarly sized samples would be further away"

wow... 3.8 sds... that's well beyond the 99.9% threshold... very unlucky run with those pocket pairs, weevil99...

Could you properly say the chances of this happening are less than 0.1%, or is there other information necessary before making such a statement?

I wasn't referring to the distribution of sample values in relation to the mean. I was referring to the distribution of the population being sampled (which is a random variable). I agree it is a binomial distribution rather than a normal distribution, since we're dealing with a discrete set and not a continous variable.

Yes, the chances of a different sample of 1260 trials producing less "sets or better" than Weevil's is less than 0.1%.

You don't need any other information - trying to estimate the probability using the CLT gives a number so far below 0.1% that it is obvious that calculating the exact answer explicitly via the binomial theorem will give an answer below 0.1%

fwiw if the sample distribution were exactly normal the chance of seeing a result more than 3.8 s.ds below the mean is 0.000072, or 0.0072%

This will not end well.

In before merge.

lol

I'm convinced. Thanks for bringing this to our attention.

That's all the proof I need

You almost had him there.

GUESS WHO IS BACK




Online poker could be rigged, but, then again, it might not be.

Who really cares anyways?

I am at the point of realization, if you want to play, play, if you think its rigged dont log on.

simple stuff ladies.