These forums have been archived and are now read-only.

The new forums are live and can be found at https://forums.eveonline.com/

Player Features and Ideas Discussion

 
  • Topic is locked indefinitely.
 

What are the odds?

Author
CzyckLeiut
24th Imperial Crusade
Amarr Empire
#1 - 2015-05-11 08:18:52 UTC  |  Edited by: CzyckLeiut
When it comes to manufacturing, the odds appear to be off. For instance I just ran 10 attempts at generating T2 BPs off one multirun BP copy: it cost me 11.8 M to roll those dice. My 'success probability per run' was 69.7% {skills + Party Decryptor}. I'm going to go out on a limb and say that on average that should have worked 69.7% of the time, about 7 out of 10. A little bit up, a little bit down maybe, but mostly 7 out of 10. 4 out of 10 was what I got. This is not unusual. In my experience this anomaly is common, this aberration is average.

There doesn't appear to be much more I can do about these numbers either. I've got the relevant skills raised to 4 out of 5, sometimes 5 out of 5, the difference supposedly affecting things by something like 1%... so if I max the skills (that take something like a month each) I'll have 70.7% or so Which, given the performance of the odds thus far, should amount to f'all. That is discouraging.

It feels distinctly like the skills required to raise my odds - at all - are time consuming but largely ineffective and my overall odds are not rational at all. 50% should work half the time. It doesn't. 46% rarely works and 51% frequently works (in a single run) but multiple runs, regardless of my odds it would seem, frequently fail.

I suggest this crapshoot at least acquire the predictability of a literal crapshoot. 70% odds should work, on average, 70% of the time, events on either side of the bell curve should occur with the frequency of a bell curve... the greater the deviation from the average the less likely it is to occur.

I don't know what variables are actually at work but I find that the advertized odds and the actual performance of the game are in pronounced disagreement.
thowlimer
Roprocor Ltd
#2 - 2015-05-11 08:50:41 UTC
The percentages involved are correct, its just that you have a very wide distribution so you need
a large(1000+ at least) sample size for it to start converging on those numbers.
Lugh Crow-Slave
#3 - 2015-05-11 10:29:04 UTC
some one doesn't understand how chance works......


just because the odds are 7/10 does not mean you will get 7/10 hell you could get 0/10 or 10/10


(i sure hope op doesn't gamble)
FireFrenzy
Cynosural Samurai
#4 - 2015-05-11 14:43:50 UTC
Thats not how probabilities work broseph....

technically your odds werent 10*n=10n blueprints with a rounding error, its 10 packages of 7/10 chance... Find some d10 (or a random number generator) and roll them and count the number of times you get 7 positives on your "3 or higher is good enough" rolls...

Eventually it'll balance out but with random events occuring in clusters (and both dice and computer randomisations not being actually random) you might not see that result for the next couple of hundered rolls...
Samillian
Angry Mustellid
#5 - 2015-05-11 15:20:44 UTC
You get good and bad runs but that is the way the cookie crumbles when your dealing with a chance of success.

NBSI shall be the whole of the Law

Paranoid Loyd
#6 - 2015-05-11 16:04:06 UTC
mmm, cookies

"There is only one authority in this game, and that my friend is violence. The supreme authority upon which all other authority is derived." ISD Max Trix

Fix the Prospect!

CzyckLeiut
24th Imperial Crusade
Amarr Empire
#7 - 2015-05-12 06:09:22 UTC  |  Edited by: CzyckLeiut
Lugh Crow-Slave wrote:
some one doesn't understand how chance works......


just because the odds are 7/10 does not mean you will get 7/10 hell you could get 0/10 or 10/10


(i sure hope op doesn't gamble)


I do gamble occasionally. But not too often since its stupid to play a game where the house always wins over the long haul. In like fashion, 7/10 odds DO mean that - over the long haul, or on average - you will succeed 70% of the time. Perhaps someone should explain this to you.

The problem with the game - I'm asserting - is that the odds are massively off, in both directions. Some factor is at work, or some calculation is off, such that the odds aren't reflected in reality.
Teckos Pech
Hogyoku
Goonswarm Federation
#8 - 2015-05-12 06:43:56 UTC
CzyckLeiut wrote:
When it comes to manufacturing, the odds appear to be off. For instance I just ran 10 attempts at generating T2 BPs off one multirun BP copy: it cost me 11.8 M to roll those dice. My 'success probability per run' was 69.7% {skills + Party Decryptor}. I'm going to go out on a limb and say that on average that should have worked 69.7% of the time, about 7 out of 10. A little bit up, a little bit down maybe, but mostly 7 out of 10. 4 out of 10 was what I got. This is not unusual. In my experience this anomaly is common, this aberration is average.

There doesn't appear to be much more I can do about these numbers either. I've got the relevant skills raised to 4 out of 5, sometimes 5 out of 5, the difference supposedly affecting things by something like 1%... so if I max the skills (that take something like a month each) I'll have 70.7% or so Which, given the performance of the odds thus far, should amount to f'all. That is discouraging.

It feels distinctly like the skills required to raise my odds - at all - are time consuming but largely ineffective and my overall odds are not rational at all. 50% should work half the time. It doesn't. 46% rarely works and 51% frequently works (in a single run) but multiple runs, regardless of my odds it would seem, frequently fail.

I suggest this crapshoot at least acquire the predictability of a literal crapshoot. 70% odds should work, on average, 70% of the time, events on either side of the bell curve should occur with the frequency of a bell curve... the greater the deviation from the average the less likely it is to occur.

I don't know what variables are actually at work but I find that the advertized odds and the actual performance of the game are in pronounced disagreement.


Good God...haven't seen one of these posts in a loooonnnngggg time. Lets see, you have a 0.697 chance of success, which implies a 0.303 chance of failure. The chance of failure (and success) is i.i.d. In fact, the underlying distribution we are looking at is a binomial distribution. And sadly your chances of getting 4 or less successes is 0.0496...or about 1 in 20.

That is if we took say, 2,000 inventors each one installing jobs just like you did, then we'd expect to see about 100 inventors with 4 or less successes.

In other words...you just got unlucky.

"The curious task of economics is to demonstrate to men how little they really know about what they imagine they can design."--Friedrich August von Hayek

8 Golden Rules for EVE Online

CzyckLeiut
24th Imperial Crusade
Amarr Empire
#9 - 2015-05-12 08:31:04 UTC
FireFrenzy wrote:
Thats not how probabilities work broseph....

technically your odds werent 10*n=10n blueprints with a rounding error, its 10 packages of 7/10 chance... Find some d10 (or a random number generator) and roll them and count the number of times you get 7 positives on your "3 or higher is good enough" rolls...

Eventually it'll balance out but with random events occuring in clusters (and both dice and computer randomisations not being actually random) you might not see that result for the next couple of hundered rolls...



I understand that this is what's supposed to happen. It is my assertion that it is not.
Samillian
Angry Mustellid
#10 - 2015-05-12 08:39:55 UTC
You may be exceptionally unlucky then.

I kept records over my first six months and while I had some serious peaks and troughs over all I came out slightly over the expected success rate for that period and everything I've seen the last couple of years supports that.

NBSI shall be the whole of the Law

Null Infinity
Imperial Academy
Amarr Empire
#11 - 2015-05-12 08:59:28 UTC
thowlimer wrote:
The percentages involved are correct, its just that you have a very wide distribution so you need
a large(1000+ at least) sample size for it to start converging on those numbers.


QFT. I believe it is a subject of staistic and CCP cares about proper statistic for sure.

New mining menthods: interactive mining and comet mining

Teckos Pech
Hogyoku
Goonswarm Federation
#12 - 2015-05-12 20:03:58 UTC  |  Edited by: Teckos Pech
CzyckLeiut wrote:
FireFrenzy wrote:
Thats not how probabilities work broseph....

technically your odds werent 10*n=10n blueprints with a rounding error, its 10 packages of 7/10 chance... Find some d10 (or a random number generator) and roll them and count the number of times you get 7 positives on your "3 or higher is good enough" rolls...

Eventually it'll balance out but with random events occuring in clusters (and both dice and computer randomisations not being actually random) you might not see that result for the next couple of hundered rolls...



I understand that this is what's supposed to happen. It is my assertion that it is not.


And you don't have sufficient data to support your assertion.

Look if this were true there would be lots, lots more people here posting. If the probability shown to you is 0.7, but in reality it is working at 0.4 then EVERYONE would see this. I've done thousands of invention jobs and I usually will buy data cores in bulk. I then grind through them via invention and look at the finished products. If I had 200 datacores, and a 50% chance of success I should end up with 1,000 t2 items. I am invariable "close" to that number.

There are hundreds if not thousands of other players out there doing invention. If there was a systematic error in CCP's psuedo random number generator used for invention....we'd hear about it. Why? Because people who do invention would be noticing (using the numbers I provided above) that they were getting approximately 800 items vs. 1,000 and over multiple items (yes, I don't invent just 1 item, I have done up to 4-5 of them at a time...meaning that at the end I should be damn close to 4,000 or 5,000 items).

Doing 10 inventions and getting 4 successes is not a large enough sample to make a determination that there is a problem anyways. As I noted, the probability of getting that result is 5%. Results that happen with a 0.05 probability happen quite often. When explaining spurious correlation to people, I generate 1,000 variables each with 100 observations, I then run a regression on each of those 1,000 variables against a simple linear trend. Invariably about 50 regressions pop up with various goodness of fit statistics that look "good". High R-square, nice t-test, etc. Problem is all the variables are complete and utter bull****. I know, because they were created with a Mersenne-Twister pseudo-random number generator. They mean literally nothing and those correlations are just dumb luck.

"The curious task of economics is to demonstrate to men how little they really know about what they imagine they can design."--Friedrich August von Hayek

8 Golden Rules for EVE Online