CCP Greyscale
   |
Posted - 2008.02.16 21:55:00 -
[1]
Posting with my thread bars off because I am posting as someone who does know basic statistics but does not know anything about the implementational details of EVE's RNG, and hoping that someone from Software doesn't leave a nasty present on my desk on Monday...
In general terms and referring to life in general, if you claim any kind of statistical irregularity and want people to treat it as a serious issue you usually need to demonstrate that there is a statistical irregularity, not a gut feeling. hy? Because human gut feeling sucks at statistics. We're naturally great at spotting patterns. That makes us naturally terrible at stats, because it makes us get selective, and the hard part of basic stats is avoiding selective data.
For example, if you wanted to posit an irregularity in a game's RNG you'd want to show that the RNG as a whole has a flaw. Granted, if you don't know the implementational details this becomes a trickier exercise than it would otherwise be, but it's still possible to do some stats, think about the results carefully, and get a few pointers at least as to what's going on.
For example, say I did an activity once a day for six months, and I got a load of negative results that I was unhappy with. My natural human tendency is to conclude that something's wrong with whatever's generating my results, which I assume is the RNG, and if I'm a community-minded person my follow-up action is to post on the forums about it.
This raises two issues. The first is "am I representitive"? Does the fact that I personally have seen an unexpected and unlikely string of results mean something is wrong, or that I'm just a victim of statistics? In order to work this out I need to find out how many likely strings of results are being generated, to give me a baseline to find out how unlikely my unlikely set of results actually are. If I do the math on my set and realise there's only a one in a thousand chance of them happening, say, and then I realise there's 999 other people doing the same thing who didn't get my one in a thousand set of results, well... everything's working as expected.
Secondly, my conclusion that the RNG has generated an "unlikely streak" relies on the premiss that my set of results are both "unlikely" (see above) and also a "streak". If I'm running it once a day, in order for my results to be a "streak" for the RNG I must assume that the server is running a special RNG especially for me when I do that one activity, which generates one result a day. Not only is that RNG separate from everyone else doing the same activity, it's also separate from everything else I do requiring a random number. This may of course actually be how it's implemented, but do I find this likely? (Har har.)
I guess what I'm saying here is I have no idea whether the RNG's playing up, but I've yet to see any good statistical evidence that anyone else outside our Software department knows either  I'm not saying "be quiet", I'm not saying "if you don't have conclusive proof, be quiet", I'm just saying "I don't see conclusive proof". If someone can show me actual statistical anomalies in the system as a whole I'll send extremely long emails to my boss until he does something to shut me up 
Again, posting with my thread bars off because I am posting as someone who does know basic statistics but does not know anything about the implementational details of EVE's RNG
|