Data collection:
Population 1: Rares that drop from boss kills with 0% mf and 0 NV stacks
Population 2: Rares that drop from boss kills with 234% mf and 0 NV stacks
Sample 1:
200 runs of Warrior's Rest Inferno with 0% mf and 0 NV stacks while only picking up drops that came specifically from the boss kill Data:
Rares found = 71
4 affix rares found = 41
5 affix rares found = 26
6 affix rares found = 4
Sample 2:
200 runs of Warrior's Rest Inferno with 234% mf and 0 NV stacks while only picking up drops that came specifically from the boss kill Data:
Rares found = 214
4 affix rares found = 131
5 affix rares found = 68
6 affix rares found = 15
Affix count was determined using the recent exploit found to view the item chat data again.
Excel Spreadsheet:
http://www.mediafire.com/view/?4cjjbij91xafua1Noteworty observations:
The boss kill never dropped damaged (grey) equipment (even with 0% mf). This could suggest that the final elite kill drops from a separate drop table.
The boss kill always dropped 3 or 4 equippable items (average of 3.5075 drops per boss kill)
There wasn't a single 3 affix item that dropped. Do they exist besides from crafts?
Statistical Test to compare two population proportions:
H0:p2 = p1
HA:p2 > p1
Where p1 is the true population proportion of 5 or 6 affix rares from boss kills with 0% mf and 0 NV stacks and p2 is the true population proportion of 5 or 6 affix rares from boss kills with 234% mf and 0 NV stacks
alpha = .1
n1 = 71
n2 = 214
p1hat = (26+4)/71 = .4225
p2hat = (68+15)/214 = .3879
phat = (26+4+68+15)/(71+214) = .3965
Test Statistic:
Z = (p2hat - p1hat)/sqrt(phat(1 - phat)(1/n1 + 1/n2))
Z = (.3879 - .4225)/sqrt(.3965.6035(1/71 + 1/214))
Z = -0.5165
Z~N(0,1)
p-value = P(Z > -0.5165) = .6972
Our p-value >> alpha, so there is not sufficient sample evidence to support the claim that p2 > p1. That is, there is not sufficient sample evidence to support the claim that magic find (when applied to drops from bosses) increases the proportion of 5+ affix rares found.
Other notes:
The same holds true for magic items.
If we assume that the proportion of 4, 5 and 6 rares are equal regardless of mf and the prportion of 1 and 2 magic items are equal regardless of mf then we can lump the data together and obtain the following 95% simultaneous bonferroni confidence intervals for affix proportions from boss kills (calculations in data spreadsheet above):
Magic items:
1 affix (0.0696, 0.1234)
2 affix (0.8766, 0.9304)
Rare items:
4 affix (0.5361, 0.6709)
5 affix (0.2650, 0.3946)
6 affix (0.0323, 0.1010)
Still todo:
It's still possible that mf affects affix count from nv guaranteed rares or from normal mob drops. I will be collecting data on that in the future.
Please let me know if I'm not thinking of something or if you see any erros in my calculations/thought process. Also, if anyone else has any clean boss kill data or wants to gather some, that would help in shrinking those affix count confidence intervals and helping determine the actual values of those proportions.
EDIT:
First of all I want to say thank you to everyone that has commented and brought up things that I had not considered.
I'd like to collect more data to get a more definitive understanding of exactly how mf affects drops and this conversation seems like a good starting point.
I wish I could change my title at this point, but it is what it is. If a moderator can change it to "A discussion of magic find statistics" I'd appreciate it.
I did a few more calculations and my data can only really say that, as a rough estimate, 234% mf leads to an increase of 8% or less (from ~43% to ~51%) for 5+ affix rares (keeping p-value below .05). That is using the 0% mf proportions assumed in Stasis5001's post.
There is not nearly enough data here to rule out the situation described by Stasis5001.
I know I can't collect enough data myself and if anyone would be willing to do runs in the way I did for my data collection to help get a larger data set it would be beneficial to everyone. If you collect any data, please message it to me here on Reddit and I will compile it into a google spreadsheet for everyone to see (If I get enough data I'll post it in a new topic).
Here is an Auto Hot Key script I used to quickly copy the item chat data to notepad:
http://www.mediafire.com/?9zbtfi07jxmhtlgThe script is written for playing in windowed (fullscreen) mode with an open text file named "Items found.csv" as the first window selected on alt-tab. Simply hover over the item and press ctrl-q to copy the item data to the csv file. If your computer is slow you may need to increase the sleep time. Whenever you're done, just use find/replace from : to , and you have a csv file.