G
granny
Guest
(reposted from Seven forums)
Someone started a rumour that the chance of getting 99's had been lowered in 1.62 so I've just checked it out and it hasn't. When I spellcraft I keep a record of how many retries are needed for each gem (all gems have to be 99% qua for overcharging) in Leladia's calculator since final price is based on number of retries.
Here's the results from 68 crafts in 1.60 and 32 in 1.62:
1.60: Average (mean) retries for a 99% craft = 7.09 (+/- 6.38) ( n=68 )
1.62: Average (mean) retries for a 99% craft = 6.22 (+/- 5.88) ( n=32 )
As you can see it actually looks as is the average retries is lower in 1.62 than 1.60 but the standard deviations for both data are fairly large so is there a real difference here? Doing a simple t-test on these gives a significance of 0.52 ie, no significant difference. For both sets the mode (ie. most frequently occurring number of retries) is 1. Pretty good. The median (central value - a good measure of frequency for this kind of distribution) is 5 for 1.60 and 4 for 1.62 - again unlikely to be significantly different due to the spread of the data.
Here's the frequency distribution curve for the 2 data:
This kind of frequency distribution is pretty clearly what's known as a Poisson distribution which is what makes crafting seem so frustrating sometimes - most frequently the retries needed for 99's are fairly low but occasionally you get some that take a *lot* of retries. The highest I had in those 100 crafts was 32 retries. If you were trying to make someone a 99% chain hauberk and it took 32 retries you'd be ticked off - but it's entirely possible that it can happen.
These statistics conclusively show that crafting rates haven't been nerfed in the last patch but I do think it's invaluable to have an understanding of the nature of random numbers and frequency statistics when crafting
Just occurred to me to try another test too. We're told that the chance of getting a 99 is 1 in 5, or 20%. So if we do 1-tailed t-tests on those 2 sets of data comparing each to a value of 5 are they significantly different?
1.60: p = 0.46
1.62: p = 0.51
I.e. neither of these data are significantly different from the number 5. I.e. the chance of getting a 99 craft is indistinguishable from 1 in 5 or 20%.
It may not *feel* like 1 in 5 but the maths shows that that is exactly what it is.
(nb. there is one caveat I would like to add to these data and that is that for a Poisson distribution n values of <100 could be considered low. It'd be nice to extend these numbers greatly so if anyone feels like adding to them feel free to contact me )
Someone started a rumour that the chance of getting 99's had been lowered in 1.62 so I've just checked it out and it hasn't. When I spellcraft I keep a record of how many retries are needed for each gem (all gems have to be 99% qua for overcharging) in Leladia's calculator since final price is based on number of retries.
Here's the results from 68 crafts in 1.60 and 32 in 1.62:
1.60: Average (mean) retries for a 99% craft = 7.09 (+/- 6.38) ( n=68 )
1.62: Average (mean) retries for a 99% craft = 6.22 (+/- 5.88) ( n=32 )
As you can see it actually looks as is the average retries is lower in 1.62 than 1.60 but the standard deviations for both data are fairly large so is there a real difference here? Doing a simple t-test on these gives a significance of 0.52 ie, no significant difference. For both sets the mode (ie. most frequently occurring number of retries) is 1. Pretty good. The median (central value - a good measure of frequency for this kind of distribution) is 5 for 1.60 and 4 for 1.62 - again unlikely to be significantly different due to the spread of the data.
Here's the frequency distribution curve for the 2 data:
This kind of frequency distribution is pretty clearly what's known as a Poisson distribution which is what makes crafting seem so frustrating sometimes - most frequently the retries needed for 99's are fairly low but occasionally you get some that take a *lot* of retries. The highest I had in those 100 crafts was 32 retries. If you were trying to make someone a 99% chain hauberk and it took 32 retries you'd be ticked off - but it's entirely possible that it can happen.
These statistics conclusively show that crafting rates haven't been nerfed in the last patch but I do think it's invaluable to have an understanding of the nature of random numbers and frequency statistics when crafting
Just occurred to me to try another test too. We're told that the chance of getting a 99 is 1 in 5, or 20%. So if we do 1-tailed t-tests on those 2 sets of data comparing each to a value of 5 are they significantly different?
1.60: p = 0.46
1.62: p = 0.51
I.e. neither of these data are significantly different from the number 5. I.e. the chance of getting a 99 craft is indistinguishable from 1 in 5 or 20%.
It may not *feel* like 1 in 5 but the maths shows that that is exactly what it is.
(nb. there is one caveat I would like to add to these data and that is that for a Poisson distribution n values of <100 could be considered low. It'd be nice to extend these numbers greatly so if anyone feels like adding to them feel free to contact me )