Random dice?

[Feanor79]

I have a general question... After browsing through this forum I could not find the answer. What program is used to general the numbers for dice? After playing for a few months I am getting more and more frustrated by the dice. My strategy seems to be mattering less and less, while the dice matter more and more, and this is really getting frustrating.

Before you tell me I don't know what I am talking about, I have had PhD level statistics courses and I am pretty adept at statistics. I know that these dice do not appear random nor do they appear independent. This is becoming so frustrating that I will not renew my membership as well as I will discourage my brother, who is currently a newbie, from buying a membership.

Maybe I just need to blow off some steam, and if so, I am sorry for any inconvenience. Still, it would be helpful to know what random number or random dice generating program this site uses.

[oVo]

The recently random dice have bent me over for a gracious buggering lately too. So blow off a bit of steam, then suck it up cupcake and dive back in... as it will balance out somewhere. Of course the problem there is that you won't actually notice the favors of good dice rolls because they just don't stand out like the crap ones.

There are pages upon pages of dice threads containing a wide variety of statistical analysis...
so refine your search. The truth is out there.

[Timminz]

The numbers come from http://www.random.org. The have had numerous studies done on their numbers. [player]e_i_pi[/player] has done an in-depth study of a sample file from random.org (mostly regarding "streakiness", I believe). It is posted in these forums somewhere, but I can't seem to find it right now.

[Velvecarrots]

You could also try downloading the dice analyzer, which keeps track of your attacking battle outcomes. As much as I want to complain after losing 10 straight armies in consecutive turns, I have to bring this page up which reminds me that the numbers don't lie.
















Yes, my 2v2 is incredible. On a game I already lost where I was trying to up my stats on some categories I won a 3v9.

Now go and make your confidence interval...

[jaseleo]

Random indeed
looking at those dice analyser stats wow they are them same as everyone else's 16.%

amazing 100 divided 6 =16.6 its so random that you have the chance to roll every dice the same percentage for every single turn you take making it erm not random at all !!!!

Im sure they call that set odds

[Feanor79]

Just because the numbers work out in the end, does not mean that the system is fair. To truly test the if the system is fair, you would have to run a repeated measures analysis. Building confidence intervals around final totals is the weakest test. For example: 1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,6,6,6,6 will give you even final averages, but noone would argue this is random. So, while a dice tracker can give you total analysis, only a history and a repeated measures analysis can determine if the distribution is random.

While intuition is not scientific, I feel that I will have a great days and get great rolls for a several days, then for several days I cannot beat a 16 v 1 to save my life. We all have these feelings, and I think, without the numbers I cannot test it, that the probabilities of everyone having these events randomly is quite low.

However, if the programers and the more experienced users say the system is right and stop whining, I will charge ahead and continue to be frustrated.

[Timminz]

I've found the links I was looking for. You might find these to be of interest.

analysis of CC Dice. Streaks, distribution, win/loss thread.

[player]e_i_pi[/player]'s Discussion of Statistical records of users dice results thread.

[Feanor79]

I appreciate the feedback and thank you for the links.

However, I am just not buying it. Case in point, just finishing up a a one on one game of Prison Riot (number 4677029). I had a 50 armies vs 11. I came out of that attack with 11 guys. The odds of that occurring are practically zero, as calculated at this website.

http://bartell.org/cgi-bin/risk.cgi

Then, since it was near the end of the game, I went on attacking with my 11 armies and did not lose a single one, which is nearly impossible too. This number generator seems too streaky. I know humans see patterns where there are none. I also know that I personally remember when I attack and lose or when I lose a game more then when I win a game.

[Timminz]

http://bartell.org/cgi-bin/risk.cgi

That's an interesting site. You might also like http://www.gamesbyemail.com/Games/Gambit/BattleOdds

[PLAYER57832]

Just because the numbers work out in the end, does not mean that the system is fair. To truly test the if the system is fair, you would have to run a repeated measures analysis. Building confidence intervals around final totals is the weakest test. For example: 1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,6,6,6,6 will give you even final averages, but noone would argue this is random. So, while a dice tracker can give you total analysis, only a history and a repeated measures analysis can determine if the distribution is random.

While intuition is not scientific, I feel that I will have a great days and get great rolls for a several days, then for several days I cannot beat a 16 v 1 to save my life. We all have these feelings, and I think, without the numbers I cannot test it, that the probabilities of everyone having these events randomly is quite low.

However, if the programers and the more experienced users say the system is right and stop whining, I will charge ahead and continue to be frustrated.

Programmers have looked into it.

As a statistician, you know that perceptive bias is as important as actual statistics in how we see things. That is, after all, a major reason for statistics .. to get beyond human biases.

Anyway, I strongly urge you to review the Strategy forum threads. Also, consider enrolling in the Training grounds/cooks forum classes.

[Feanor79]

A appreciate the feedback. I certainly don't mean to be rocking the boat, I am just claiming the analysis is wrong. I don't know how random.org generates their numbers. All I know is from what I have seen the numbers are too streaky. I did see some testimony on the site where researchers are using the generator to run monte-carlo simulations, and I know that non-random streaky patterns would trash a modeling simulations. I have done some modeling work with estimating leaf conductance and know that you can get some strange things with random number generators. I suggest that the analysis should be a repeated measures analysis. That way one can track patterns over time. If the events are completely random, and not based on something such as time or date or what ever else the programs use to generate their number patterns, repeated meas would show this. As far as I know SAS is the only program I know that can analyze repeated meas. You can't do it with excel or sigma stat or Systat or Coral stat.

I am just frustrated. I am tired of getting rolls that make my head spin in frustration where games slide by because of "(un)lucky" dice. I will check out those training pages (thanks Player57832). However, I should point out I use to be a Corp 1st class. I have some grasp of the game and strategy. It is frustrating to fall so far in rank. I know that rank does not matter. No game types or maps are open to higher ranking members then lower ranking members. Part of my fall in rank (or maybe most, or if you all are right, then all!) was foolish strategy or playing opponents who were much better then me. Part of it was just luck of the draw (an element to the game, I recognize). However, part of it was the streakiness of the dice. I would play a whole day an win every battle, even 6 vs 3 easily. The next day I could not win a 9 vs 2 for the life of me. It makes me think I should only play on odd days or something.

[Timminz]

I don't know how random.org generates their numbers.

They answer that in their FAQs, I believe. They get their numbers from atmospheric noise.

[a.sub]

sig

[kiddicus maximus]

Without delving too deeply into the mathematical, I'll say only this:

When your opponent rolls two 6's, stop. Give random.org a chance to recalibrate its randomness. Sometimes I'll even close out the browser and start over...

Probably more superstition than actuality, but it seems to work for me.

[Timminz]

When your opponent rolls two 6's, stop. Give random.org a chance to recalibrate its randomness.

That might be logical, if each roll were pulled directly from random.org, but (unless things have changed recently) CC buys a large file of random numbers, and draws each line of rolls from there. Also, using standard attack (as opposed to auto), you're not drawing sequentially from the file anyway, since dozens (or more) other users will be drawing lines from the file in between your attacks.

[xelabale]

If I lose both dice I close my eyes for the next roll. Works without fail. You guys can thank me later.

[Feanor79]

Well, I will ponder quietly in the corner, being proven wrong. I e-mailed Mads Haahd. He works in the same state as I do at a rival University, so he was kind enough to respond.

>On Wed, 15 Apr 2009, Timothy wrote:

> Professor Mads Haahr,
>
> I know you are a very busy man and I hate to bother you about such a tiny problem. A few months ago I joined a site where people can play Risk like games online (http://www.conquerclub.com). On the site forums the administrators say they use your website or program (not sure, they are very quiet) to generate the dice rolls (typical 6 sided dice). I have, along with other members, notice a lot of streakiness in the rolls. We will win 10 to 20 battles in a roll and then lose like 10 to 20 times in a roll. The site has done a lot of statistics and observed that every number is evenly represented over time. However, I don't think that they have been able to perform a time based analysis. Have your or your institue performed a repeated measures analysis to determine if their is an element of time that may be baising the number generators.
>
> I know that it is probable I am daft in my head, however I do have a scientific background. I am a PhD student at UGA and have a masters from University of Wisconisn. I have taken numerous graduate level stat courses, though never on random numbers or modeling, always on data analysis and experimental design.
>
> Well, thanks for your time, and I would appreciate any response to quell my fears (either their is a time based bias with streaks or I am not a great Risk player!).

He responded:

Hello Timothy,

Thank you for your email. There are many people using numbers from RANDOM.ORG, and I am not aware of what the good people at Conquer Club are doing with them.

You will probably have seen my real-time statistics and in-depth analysis pages here:

http://www.random.org/statistics/
http://www.random.org/analysis/

These tests are concerned with the raw random bit stream and not with numbers that have been scaled to fit particular intervals, such as [1,6]. However, you should of course observe similar characteristics for scaled numbers, such as uniformity of distribution and serial independence, as you would for a raw bit stream that has passed the tests.

It sounds to me like the streakiness that you describe should show up as correlation and therefore be picked up by a serial dependency test. Although a streaky sequence might be able to pass some of the standard (e.g., frequency) tests, I would expect it to perform poorly across a good range of other tests, such as Information Entropy, the Spectral Test and the Serial Test from the NIST suite. Many of the tests in the NIST suite would fail if the numbers in your stream appear correlated.

As you will have seen from my pages, the random bit streams from RANDOM.ORG have passed the NIST suite. If Conquer Club are getting a uniform distribution, I think it is safe to assume their scaling is correct, but there could be other things worth considering, such as whether they are using a fixed pool of numbers or are constantly renewing their dice rolls. For example, one risk of repeatedly using the same pool (and I should stress I have no idea whether Conquer Club do this or not) is that if the pool is small and by chance happened to contain streaky sequences, then you might see that same streakiness occur several times across games.

Best regards,
Mads
--
Dr. Mads Haahr <Mads.Haahr@cs.tcd.ie> | Lecturer in Computer Science
Department of Computer Science | Phone: +353 1 896 1540
Trinity College Dublin | Web: http://www.cs.tcd.ie/

[oVo]

That's interesting... I've always felt the defensive dice are stronger than people give them credit for and particularly when it's 3 dice rolled against 2. Is there a way of factoring in the defender winning all draws?

With 3 vs 2... or 18 dice faces against 12 defenders who prevail over all ties, is there a legitimate way to reliably predict the odds of coming out on top? I've been surprised by the number of times all 5 die come up the same and it always makes me wonder what the odds of "quad-snake eyes" actually is.

Another curiosity is exactly how many dice rolls occur on the average CC day?

[aineolach]

weakest test. For example: 1,1,1,1,2,2,2,2,3,3,3,3,4,4,4,4,5,5,5,5,6,6,6,6 will give you even final averages, but noone would argue this is random.

If you saw those numbers somewhere you would assume they weren't random... but they very well could be.

Maybe CC uses only a small set of numbers that get used up every two days (hence fits in with your odd/even hypothesis).