Results may vary

[Xenephex]

The following is not a rant or aimed at anyone, because I honestly believe that this is simply a matter of people not having enough information.

Of all the posts I've seen on this board about long runs without a skill-up and failure rates, etc. I have yet to see any evidence that there is anything 'wrong' with the RNG as it is implemented in EQ.

What I do see is a ton of evidence that people simply do not understand how random numbers and statistics work.

You went 200 attempts (let alone 50 or 100) in a row without a skillup? I'm sorry. You have every right to say AAARGGH as loudly as you want, but not to claim that there is something wrong with the RNG.

You got 3 skill-ups in a row? Yay! But it doesn't prove that the RNG is streaky.

You got twice as many skill-ups on one set of 200 attempts as on the next? Doesn't indicate that there's anything wrong.

All of the above are entirely normal and to be expected.

College statistics are 30 years in my past, but I wonder if someone for whom they are more recent might be willing to take the time to post some hard statistical information as it relates to EQ, the RNG and skill-ups and/or successes.

e.g Standard Deviation, Statistically significant samples, etc.

Specifically, take a situation where there is a 1 in 30 chance of something occurring (skill-up). What are the actual chances of going 200 attempts in a row without a skill-up (very good, actually). What are the chances of getting 2 (or 3) skill-ups in a row? How many is a statiscally significant sample? (not sure, but I know it's not 50). Stuff like that.

I know there are lots of hard-core math people in this game and specifically on these boards. I hope one of you would be willing to take the time to write up some solid information on this subject and that it might end up stickied on the board (or part of the general FAQ).

[Iannyen]

How is random normal?

[Tudani]

There was a great post a while back. Might wanna look for it.

[Qaladar Bragollach]

Re: Results may vary

Originally posted by Xenephex

Specifically, take a situation where there is a 1 in 30 chance of something occurring (skill-up). What are the actual chances of going 200 attempts in a row without a skill-up (very good, actually).

(29/30)^200 = 0.114%

[Itek]

Re: Re: Results may vary

Originally posted by Qaladar Bragollach
(29/30)^200 = 0.114%

Yes.

1 time in 1000 sets of 200 attempts someone will get no skill ups in 200 attempts with an average of one skill up in 30.

Remind me again how many skill ups it takes to go from 180 to 250 at an average of 30 attempts per skill up?

(2100)

Remind me again how many skills can be taken to 250?

(7)

Which means how many "sets" of 200 attempts per person in the 1750 club on average?

(73.5)

So, if something is 0.114% likely how many people on average in the 1750 club would be expected to experience it?

(1 in 13)

If something is 1 in a Thousand likelihood how many people should experience it if one presumes 200,000 active EQ accounts with each having 20 chances at the 1 in a Thousand result? (Presume each active account ONLY GM's one tradeskill with average results.)

(4000)

1 in 50 people that play this game can reasonably be expected to experience a HIDEIOUS run of 200 attempts with no skill ups on something the average person gets one skill point in 30 attempts on.

People tend to report their BAD outcomes with much greater frequency, not to mention VOLUME, than their mediocre ones.

People also tend to be very bad at 1) math, 2) probability expectation, and 3) detecting trends and patterns.

1 in a Thousand. That's low. But given that you should have 20 sets of 200 that raises it to 2%. One in 50. Take a pack of cards. Remove three of the 2's and place in one joker. Shuffle. Now draw one card. If you get the joker that's the same probability of drawing a run of 200 attempts with no skill ups if the average was 30 attempts per skill up.

Even if EQ went with the "default" /rand function built into the library of every processor in existence it would be close enough to "real randomness" that it would take serious and rigorous statistical analysis to show it deviated substantially from mathematical expectation.

Now, math aside.

Is there a single thing that the EQ player community can do to change the /rand function used by EQ?

No.

The field of play is the field of play and will most likely not change visibly or substantially regardless of people posting their results.

The three laws of thermodynamics can be stated in lay terms as such:

Law 1) You can't get something for nothing. (no perpetual motion machines)

Law 2) You can't break even. (friction, look into it)

Law 3) You can't change Laws 1 or 2 in any closed system. (the universe is a closed system, as is the biosphere of earth for all intents)

EQ is a closed system. One for which "the laws of physics" rarely change. Changing the way random numbers are generated for EQ would be harder than changing the motion of the sun in the sky. (One could easily change the direction of travel for the sun, most likely by changing a positive positional value loop to a negative one or vice versa, while changing /rand would require a substantial re-write of very complex and deeply embedded code.)

[Poohjure]

Brilliantly Done Itek and Qaladar=)

/em golf claps

<-- Statistics Major =P

[Xenephex]

Great - thanks for the replies, but let me throw another spaniard into the works.

I assume that the percentage given - 0.114%- is for a single set of 200 attempts. However, if you do a run of 1,000 attempts, you are not, for statistical purposes, doing 5 runs of 200 attempts; you are doing more like 800 runs of 200 attempts (1 to 200 : 2 to 201, etc.) In other words, the streak of 200 non skill-ups can occur in any sub-set of those combines.

I don't think I stated that very well, but I hope you understand what I mean.

I believe this gives a significantly higher probability of runs like that ocurring.

[Xenephex]

See above post first.

I wanted to add: I remember enough about statistics to know that having 800 sub-strings of 200 combines within a string of 1,000 combines does not mean that you can simply multiply the chance for occurrence in a single string of 200 combines by 800. That would give you any 80% chance of having such a streak, and I know that is not correct.

I am pretty sure that for a string of 1,000 combines it is more than 5 times the base percentage and less than 800 times it, but I do not know how to do the exact calculation.

I think I may go running for a statistics page.

[Ursawraith]

I couldn't agree more with you. The RNG is just that Random. if you are going to use statistics to prove it's broken it will take more then a sample group of 200 combine 200,000 combines may even be too small a sample. Time and Time again I have seen all these people who post there skill up runs and the average of those reported seems to be about 1 in 25. I know that whan I set up for a skill up run I hope to get that but I know it will usually be more or less depending on the stars.

As a funny aside I have been stuck at 198 brewing for 887 combines now and counting. Do I think the RNG is broken? Heck no, am I unlucky? Yep

[Qaladar Bragollach]

I get what you are saying Xenephex and I am afraid the actual computation of the odds of getting a streak of 200 failures in 1000 combines is beyond me.

However, on the web I found a little java applet for computing hitting streaks.

Hitstreak calculator

Its pretty generic and can be converted to our purposes.

So... in our case we're interested in "hitting" streaks of failures.
Our success rate is 29/30 or .9666667.
Our "games in season" would be 1000... it will only go to 999 but thats close enough for us.
And our streak length(at least) is 200.

That works out to 1 chance in 32.15.
We're 1 combine short but thats slightly better than 3%.

That good enough for you?

[andyhre]

The same website says the chance of 3 failures in a row on something with a 95% chance of success over the course of 200 combines is more than 2%. I think this is one of the "it can't be random - I was making a bagfull of heady kiolas and I got 3 fails in a row" triggers we all see, but, yep, it should happen to about 2 in 100 of us.

And similarly, if you have a 1 in 20 chance of a skill up, there is a 2% chance you'll get three of them in a row over the course of 200 tries.

[Cigarskunk]

The vast majority of the folks complaining about thier combines bogging down are in the hell levels of tradeskills - high 180s to 220s.

You can toss around all of the RL statistics you want, but it is a proven fact that those levels are tradeskill hell levels - from about 188-225 (on average) a person will encounter a significantly higher number of combines needed in order to get a skill up.

Additionally, the RNG is not truelly random - it's definately wieghted.

Some tradeskills have been designed to require more combines on average to skill up then others - pre-nerf pottery is an excellent example - it used to be that one would get a skill up every combine or every other combine on average.

There is also growing evidence that some combines actually give one a better chance of a skill up vs others - fletching is the best example at the moment - people have consistantly found that skilling up on bows produces more success per for the same number of combines vs skilling off of WE cultural arrows.

[Qaladar Bragollach]

All of that may be true Cigarskunk but none of it is evidence or proof that the RNG is not truely random.

It merely indicates that the RNG values are manipulated after the fact.

For example, pre-nerf pottery might have been driven on a skillup ratio based on a random(1,2) or a 50% chance. The range that everquest passes to the RNG could be manipulated in all kinds of ways. Theres probably a stat based manipulation, and a sliding scale manipulation based on the trivial point of the item. Also, bonuses may be added after the RNG call. So for example, you might do a random(1,30) for fletching skill up and then add 5 if you're making a bow and then decide its a skill up if the final result is >=30.

Many, many, many ways the numbers can be cooked.

Still the same RNG, and none of the complaining I have seen has presented any evidence that the RNG is not random.

[Magrath]

To be quite honest, I'd rather play Everquest than spend time evaluating its random number generator, but if anyone is truly interested in doing so, here's a nice overview of how it would be done. Might make a nice science fair project for someone, I suppose. No, Mom, I'm not playing, I'm doing schoolwork.

Edit: Here's another example with more detail. Same tests, though, just more explanation. The autocorrelation tests are probably of particular interest, as these deal with the "streakiness" conjecture.

[Itek]

groan...

1) no 1000 attempts is not = 5 sets of 200 attempts statisticly

2) the difference, for our purposes, is trivially small

3) getting supplies for 200 attempts in a row is feasable

4) getting supplies for 200,000 (or even 1,000) attempts in a row is impractical

5) gonna have to disagree AND agree with Cigarskunk.

a) 180-220 skill is the "hell-levels" and substantially harder than skill ups both before and after

b) this doesn't, however, mean the RNG isn't random but weighted.

What it really means is that the "skill up precent chance" varies by current skill AND Int / Wis.

i.e. With a 255 Int I managed to get Fletching from 1 to 169 in very short time.

With 200 Wis my wife's baking twink got from 135 to 177 very fast. Then started to slow WAY WAY down. (Went from 5 - 10 attempts a skill point to 20+ a skill point.)

Plus some skills go faster than others.
(Brewing, fletching and jewelcraft went very fast for me. Pottery and tailoring a little slower even at lower skill levels.)

But the underlying truth remains. The RNG is embedded technology. Complaining about it, or trying to examine it, really falls outside the scope of things to even put on a "wishlist."

Itek: "The RNG seems broken because I tend to get huge runs of tailoring with no skill up."

NewbieTank: "The RNG seems broken because I miss a lot."

Even if SOE published the numbers we would still get these freaky occurences.

SOE: There is a 0.33 chance of getting a skill up when doing Othmir Fur Caps.
IrateCustomer: Liars, I did 300 Othmir and didn't get a single skill point.

If something is 1 in a Thousand... how many TIMES each day do you figure it happens with 200,000 players?