How can you tell if dice developed a bias?

from this thread about e-Craps:
https://wizardofvegas.com/forum/gambling/craps/12730-failed-the-wizard/2/#post212567

e-Craps has a past history board (like Roulette) that lists the past 100 #s rolled.
(for the sake of simplicity, lets say the #s presented are accurate.)

how can i tell if the dice have been worn out and are now biased? (either for the 7 or against the 7)

or is 100 rolls too small of a sample size?

Statistically, a sample of 100 outcomes is too small to have a high confidence. I recall the Wizard and others stating that samples in the 1000's would be required.

Quote: 100xOdds

from this thread about e-Craps:
https://wizardofvegas.com/forum/gambling/craps/12730-failed-the-wizard/2/#post212567

e-Craps has a past history board (like Roulette) that lists the past 100 #s rolled.
(for the sake of simplicity, lets say the #s presented are accurate.)

how can i tell if the dice have been worn out and are now biased? (either for the 7 or against the 7)

or is 100 rolls too small of a sample size?

You are doing the same thing that I have done for Interblock's roulette for the past six months or so (if you include the MegaStar brand, then I've spent 2+ years overall). I would love to share with you what I found but since you are the type of person to place words in my mouth[1], there's no effen way I would share with you some exploitable features I've discovered since it involves the "software". You see people write software and some software programs may not be robust.

In summary, the short answer is yes, there are sound mathematical techniques to discern bias.



[1]
"so you're saying the machine will stop the dice bouncing at a 7 or hit the point, depending on how many right and dark bettors there are at the moment?

for roulette, you're also saying the software takes into account the weight of the ball, gravity, deflection speed off a bumper,etc to estimate a zone where the ball should land?" (source: https://wizardofvegas.com/forum/gambling/craps/12730-failed-the-wizard/2/#post212566)

It's a little too open-ended a question, but the short answer is 100 rolls is way too small a sample size.

The actual answer depends on how biased you mean by biased, and how confident you want to be in your conclusion. Supposing that you want to test whether the dice are biased enough to eliminate the house edge on the pass line, and want to be say, 95% confident that you have found a bias, you will almost certainly need a lot more than 100 rolls.

A slightly backwards way of looking at the problem: there is a ~95% chance that fair dice will show between 9 and 23 rolls of '7' in a sample of 100. If you observe between 9 and 23 rolls of 7 in the history, you have no evidence of bias. If you look 20 times and see it fall out of the range once, you STILL have no evidence of bias.

Quote: Ayecarumba

Statistically, a sample of 100 outcomes is too small to have a high confidence. I recall the Wizard and others stating that samples in the 1000's would be required.

That is why many APs use the Law of Large Numbers.

Quote: dwheatley

It's a little too open-ended a question, but the short answer is 100 rolls is way too small a sample size.

The actual answer depends on how biased you mean by biased, and how confident you want to be in your conclusion. Supposing that you want to test whether the dice are biased enough to eliminate the house edge on the pass line, and want to be say, 95% confident that you have found a bias, you will almost certainly need a lot more than 100 rolls.

A slightly backwards way of looking at the problem: there is a ~95% chance that fair dice will show between 9 and 23 rolls of '7' in a sample of 100. If you observe between 9 and 23 rolls of 7 in the history, you have no evidence of bias. If you look 20 times and see it fall out of the range once, you STILL have no evidence of bias.

In econometrics, they have tests for this type of situations where you don't have large sample sizes. I believe the tests are logit or probit or something to that effect. The scientists used a logit-type model to verify the source of the the Challenger 0-ring failure.

There are these specialized statistic test if you are aware of them. Read here for an explanation http://www.stat.ufl.edu/CourseINFO/STA6167/logistregSFLM.pdf

If you look carefully at Output 10.1 in the pdf document, they only had 16 observations, which is much less than 100 and still is less than 20.

Quote: 100xOdds

from this thread about e-Craps:
how can i tell if the dice have been worn out and are now biased? (either for the 7 or against the 7)?

Just a minute here.... I thought those big floppy dice that are in the plastic dome are for DISPLAY purposes and that the real "dice" are some RNG chips somewhere in the machine.

Quote: FleaStiff

Just a minute here.... I thought those big floppy dice that are in the plastic dome are for DISPLAY purposes and that the real "dice" are some RNG chips somewhere in the machine.

Hell, if the machine could make the dice fall where it wants, then why would you believe the RNG was really random?

Quote: Ardent1

There are these specialized statistic test if you are aware of them. Read here for an explanation http://www.stat.ufl.edu/CourseINFO/STA6167/logistregSFLM.pdf

If you look carefully at Output 10.1 in the pdf document, they only had 16 observations, which is much less than 100 and still is less than 20.

I think you will have to bend over backwards to use a linear regression model to analyze possible dice bias.

For a significant 7 or no-7 actually *both* dices needs to be biased.

Quote: Ardent1

You are doing the same thing that I have done for Interblock's roulette for the past six months or so (if you include the MegaStar brand, then I've spent 2+ years overall). I would love to share with you what I found but since you are the type of person to place words in my mouth[1], there's no effen way I would share with you some exploitable features I've discovered since it involves the "software". You see people write software and some software programs may not be robust.

In summary, the short answer is yes, there are sound mathematical techniques to discern bias.



[1]
"so you're saying the machine will stop the dice bouncing at a 7 or hit the point, depending on how many right and dark bettors there are at the moment?

for roulette, you're also saying the software takes into account the weight of the ball, gravity, deflection speed off a bumper,etc to estimate a zone where the ball should land?" (source: https://wizardofvegas.com/forum/gambling/craps/12730-failed-the-wizard/2/#post212566)

is there a way to ban Ardent1 from my threads?
or at least a way to block his posts so that i dont see them?

Quote: 100xOdds

is there a way to ban Ardent1 from my threads?
or at least a way to block his posts so that i dont see them?

Yeah.

Click on Preferences. Then click the link to block members.

Of course, doing that you will run the risk of making yourself look foolish as another member did recently when he started a new thread to complain rather than reading the contents of a blocked post.

Sorry, but my understanding of e-Craps is minimal, does that board show only the total or the combination. The latter is important for the following suggestion.

I would look at the number of times a particular face does not appear on the dice. (for example in 89 "rolls" no 4 showed up on any die) In another thread dating back a while, I posted (and witnessed) such an event using real dice, and met someone who was keeping track over a period of time.

As someone else mentioned, 7/no-7 is a tougher challenge.