Thrugadin? Cursed or RNG Jacking Me Around?

[waldora]

SOE has alredy stated that the RNG, while being streaky at time, is random over the long haul.

Consider the massive amounts of random events EQ does in a single day:

Every Spell cast has to be checked for fizzle, against skill, specialization, skill up, resists, to actually determine its effectivness.

Every melee event has to be checked also, comparing skill, to AC, avoidence, skill, to determine how many HP to remove.

Characterd doing tradeskills, and /randoms.

This is for every character every second.

The NPC's path have a randomness built in, what will spawn next, will it be a named? What loot will spawn with it, etc.

Then consider the randomness of EACH character performing their own mental calculation to defeat the randomness of the RNG. Every person will decide to do something different based on the results of the random events above.

Something as simple as replying to a tell, l before your next combine, changes the current state of the RNG.

Stopping to buff someone before your next combine, may be the difference between a success and a fail (and it could go either way)

With the RNG being streaky, you do have a chance, if you detect it early, to change YOUR actions for a more desirable result.

Bottom line, its random in the VERY long term, so every period of faliures, time without skill ups, low /random results, will, eventually be balanced by a period of unusually high successes, skill ups, and high /random results.

WHEN this will happen ... is random.

[Boleslav]

Holy digging up an old thread Batman!

That said, if you want to go through the process of running a bunch of /random 0 100, here are some ways you could analyze that data without a degree in statistics.

Expectation: I would expect that the results of /random 0 100 should be randomly distributed between 0 and 100, with an equal chance to get any value. I would also expect each value to be independent of the previous value.

This means that the average (mean) of the values should be 50. An average far from 50 would show a large-scale shift from expected value.

Set up a second column. Have this column be the absolute value of the difference between two consecutive values. So the first cell in the column would have the first random number minus the second random number. Continue that until you get to the end. (The last random number will not have a partner, discard it for this bit)

The mean of the second column should be around 33 (exact expected value would be 32.835). This would reflect an even distribution of differences based on an expected value of 50 on the first reading and an independent second reading. Values far from 33 would reflect some correlation between one value and the next.

For example, I set up a sheet in Excel to do this, and used Excel's random number generator to give me a sample of 100 readings between 0 and 100. I repeated it with 1000 readings. Here are the results.

Sample of 100: average = 50.59, average of 2nd column = 32.77

Sample of 1000: average = 50.128, average of 2nd column = 33.157

Try this with data from the EQ generator. If you don't want to do the analysis, post and excel file with the values and I would be happy to crunch them for you.

I'm willing to accept Excel's random number generator as pretty random. Doing several runs of 1000, it gave averages as low as 48.14 and as high as 51.8. So (again without digging out big test statistics) if your examples in a run of 100 are within 5 of the expected value I would say the test does not support the generator being non-random. In a sample of 1000, cut that in half.

If anyone neeeeeds a big statistical analysis that can be arranged as well.

Boleslav (math geek) Forgehammer
Paladin of Brell in his 67th Campaign
E'ci – Sacred Destiny

[Nolrog]

There's only one thing that could possibly explain this.

The RNG is sentient, it is evil, and it doesn't like you.

[Trellium]

But it may like someone else, and the choice it makes is ... random. Lol

[Sukrasisx]

This means that the average (mean) of the values should be 50. An average far from 50 would show a large-scale shift from expected value.

Set up a second column. Have this column be the absolute value of the difference between two consecutive values. So the first cell in the column would have the first random number minus the second random number. Continue that until you get to the end. (The last random number will not have a partner, discard it for this bit)

The mean of the second column should be around 33 (exact expected value would be 32.835). This would reflect an even distribution of differences based on an expected value of 50 on the first reading and an independent second reading. Values far from 33 would reflect some correlation between one value and the next.

I ran my experiment and edited my earlier post to include the results. Your suggestions are also good for quick checks. On the 42155 samples I took, the average was 50.596, which is very close since I was using /ran 1 100 (not 0 100) so the expected average would be 50.5. The standard deviation was 28.831, which is again very close to the expected s.d. of (I think) 28.866.

I looked at the difference between pairs of samples, but I only looked at non-overlapping pairs (first vs second, third vs fourth, etc.) since I'd be worried that including overlaps (such as second vs third) would mean the differences are no longer independent tests. The expected average difference for 1-100 is 33.33, and what I found was 33.18.

Of course, it's possible that all this means is that I ran my test during a period when the RNG was mostly "normal", and there could still be long intervals when it's consistently high or low. But I'd be pretty impressed if SOE uses a RNG that can be consistently high or low for periods of hours at a time.