2011-09-15, 07:02 PM
Since noone else is posting any numbers, I've used 6 regular EEs on 1st slots and 5 worked. (bellocce, morningstar, pinaka (sploded), rsb, top, belt)
The chance of 7 of 9 80%s failing is 0.03% which means it will happen to a lot of people given that Maple's population is in the millions. In particular it's about 1 in 3400, which is less than the population of 160+ on bera alone. And with the risks people took (thinking of a guy who put 5+ stars onto a 140 attack Ski) it's no surprise some came out badly.
The only theory I have for this is that the RNG is seeded with your IGN, and gives good ensemble performance but behaves badly at specifics. That's hard to prove, but fits the facts, and avoids the claim that they screwed up one single scroll somehow.
If you want to test that, it's gonna be expensive, cause you want to run statistically significant numbers on a variety of characters. I would go with 95% confidence (pretty standard) for a +-20% range just due to the alleged large difference in probabilities. That gives Z = 1.9599, E = 0.2, n = Z^2/(4E^2) = 24 tests. Round it to 25 and it'll give you nice easy numbers to work with.
So then the sample program will look like
[table=]IGN|Successes|% success
nameo|20|80%
alex|6|24%
Seven|16|64%
...|...[/table]
If someone does wind up in the range where it's unlikely to be chance (<60% success) then you could expand testing on that character to see if the rate continues through another test set.
You can run this with other scrolls if you want, it shouldn't just apply to EEs. Aiming for scrolls nearest 50% is probably best cause then you can find anomalies either way. Eg. My Paladin has used 250 60% scrolls logged, 167 of them worked = 66.8%. For a Z test of 95%, that says there's a 95% chance they work something other than 60% for me (60.6 ~ 73% is the 95% confidence interval). My Paladin used 132 10%s logged, 18 worked = 13.6%. That's still within the confidence interval of (5.1%~22.1%). As you can see, even a fairly major increase in success rate from 10%s is really not significant.
Nowadays there's also the % scrolling bonus you get from Dilligence to consider, but statistical tests still work.
The chance of 7 of 9 80%s failing is 0.03% which means it will happen to a lot of people given that Maple's population is in the millions. In particular it's about 1 in 3400, which is less than the population of 160+ on bera alone. And with the risks people took (thinking of a guy who put 5+ stars onto a 140 attack Ski) it's no surprise some came out badly.
The only theory I have for this is that the RNG is seeded with your IGN, and gives good ensemble performance but behaves badly at specifics. That's hard to prove, but fits the facts, and avoids the claim that they screwed up one single scroll somehow.
If you want to test that, it's gonna be expensive, cause you want to run statistically significant numbers on a variety of characters. I would go with 95% confidence (pretty standard) for a +-20% range just due to the alleged large difference in probabilities. That gives Z = 1.9599, E = 0.2, n = Z^2/(4E^2) = 24 tests. Round it to 25 and it'll give you nice easy numbers to work with.
So then the sample program will look like
[table=]IGN|Successes|% success
nameo|20|80%
alex|6|24%
Seven|16|64%
...|...[/table]
If someone does wind up in the range where it's unlikely to be chance (<60% success) then you could expand testing on that character to see if the rate continues through another test set.
You can run this with other scrolls if you want, it shouldn't just apply to EEs. Aiming for scrolls nearest 50% is probably best cause then you can find anomalies either way. Eg. My Paladin has used 250 60% scrolls logged, 167 of them worked = 66.8%. For a Z test of 95%, that says there's a 95% chance they work something other than 60% for me (60.6 ~ 73% is the 95% confidence interval). My Paladin used 132 10%s logged, 18 worked = 13.6%. That's still within the confidence interval of (5.1%~22.1%). As you can see, even a fairly major increase in success rate from 10%s is really not significant.
Nowadays there's also the % scrolling bonus you get from Dilligence to consider, but statistical tests still work.