Site bias test

FlFishOn · #1 07-22-2006, 03:01 PM

Hypothesis: New online poker players are treated to more winning cards that random chance would allow. It may be better starters or favorable board cards.

Experiment #1: Open dozens of new accounts and... Not gonna do that one.

Experiment #2: Play single table $11 SNGs and make a note on every player every time you face them allowing you to identify fresh faces. After 300 hands, on every heads up hand VS a fresh face where either I or the FF is all in record the hands (and board cards, if any), the winner and then determine the exact expectation for the hand using twodimes.com. Compare the total wins VS the total expectation and look for a diverging trend. If things line up then all is well (maybe, but I'll accept this as confirmation of randomness). If things do not line up, well, here's where I need some ideas. Seems like if there is a linear divergance over time then there's likely a problem.

I have only just begun to gather data. Please comment on the experimental design and possibilities for data analysis.

Siegmund · #2 07-22-2006, 09:54 PM

if you use only hands when there are 2 players left in the SnG that may work. A hand that is heads up because someone went all in, someone called, and several other people folded is not the same thing as a hand where only two hands were dealt and they chose to play with each other.

It does sound like an awful lot of work (many thousands of hands before you see a significant difference, unless you believe the odds are REALLY heavily skewed) with no particular reason to believe such a trend exists.

FlFishOn · #3 07-23-2006, 07:55 AM

[ QUOTE ]
A hand that is heads up because someone went all in, someone called, and several other people folded is not the same thing as a hand where only two hands were dealt and they chose to play with each other.

[/ QUOTE ]

I don't see it. If any difference exists it is so tiny as to be of no import. Are you suggesting that the quality of the folded cards somehow would favor one hand or the other WRT the coming board cards? And if so, is the bias not unidirectional?

Siegmund · #4 07-23-2006, 04:20 PM

Well.... how tiny is tiny enough to be of no import to you?

You propose using only 300 hands (which means you will detect only a flamingly obvious bias that shifts the chances by more than about 5% on average). If you use that few, well, no worries, you'll only see random noise anyway.

Typically people who ask this type of question on the forums wind up collecting more and more data and eventually saying "hey, it's statistically significant now!"

Bunching effects can achieve statistical significance after anywhere from a few thousand to a few hundred thousand hands, depending on the playing styles ("play any ace" players common at LHE are the ones who cause it to emerge fastest.) Whether that bias will favour new players or not is hard to say -- it depends what hands both you and the average new player choose to play.

All I'm saying is that if you do a longish-term experiment and achieve statistical significance, we aren't going to believe that it did so because you've found the effect you say you are looking for.

FlFishOn · #5 07-24-2006, 09:18 AM

WHy would you assume the bunching effect to favor the new player or myself? Seems like is would wash out, being very nearly equal.

Also the 300 refers to the first hands I played seeding the notes, not collecting data.

numeri · #6 07-24-2006, 01:09 PM

Are you seriously this "experiment"? It's not realistic to expect to be able to collect all that data, and your data will be biased. (i.e. How is a "fresh face" defined?)

Have you thought about the sample size necessary to perform this type of statistical test? wikipedia Look down the list for the "One-sample z-test" that has a p in it. You'd have to do this test for every possible combination of hands against a so-called "fresh face". So if you have AA against KK, you would need ~40 examples of that to determine whether the KK wins more than it's 10%. Actually, you'd need more, since A[img]/images/graemlins/heart.gif[/img]A[img]/images/graemlins/diamond.gif[/img] vs K[img]/images/graemlins/heart.gif[/img]K[img]/images/graemlins/diamond.gif[/img] is different than A[img]/images/graemlins/heart.gif[/img]A[img]/images/graemlins/diamond.gif[/img] vs K[img]/images/graemlins/club.gif[/img]K[img]/images/graemlins/spade.gif[/img].

So anyway, I can't believe I just spent all this time. Your idea is not practical to study, and a waste of time anyway. Just play better.

FlFishOn · #7 07-24-2006, 02:25 PM

[ QUOTE ]
Just play better.

[/ QUOTE ]

My ROI is close to maximum in these. Thank you for your concern.

The rest of your ideas seem to show in incomplete grasp of the problem.

numeri · #8 07-24-2006, 03:12 PM

[ QUOTE ]
[ QUOTE ]
Just play better.

[/ QUOTE ] My ROI is close to maximum in these. Thank you for your concern.

[/ QUOTE ]
I guess I'm wondering what your concern is then. It sounded like you were complaining that you lost more to these "fresh faces" than you thought you should.

[ QUOTE ]
The rest of your ideas seem to show in incomplete grasp of the problem.

[/ QUOTE ]
That may be the case. What have I misunderstood?

numeri · #9 07-24-2006, 03:15 PM

[ QUOTE ]
The rest of your ideas seem to show in incomplete grasp of the problem.

[/ QUOTE ]
Also, your description of the problem seems to show in incomplete grasp of probability and statistics.

jackaaron · #10 07-25-2006, 10:31 AM

The only reason people think new players do well early is because new player generally suck. And, because they suck, they get in more dog situation than we could possibly count. Of course, since no hand is going to win 100 percent of the time pre-flop, these dogs will win, and we as good players will lose to these dogs from time to time.

If you really do this experiment you are completely wasting your time. Seriously. Bad players get in as dogs often and win, and I hope they keep doing it.