WibbleNZ

A percentage (such as miss rate) is most definitely a proportion, not a mean.

 

sylvestro

Plain and simple to the lamen, at 360 weapon skill (typ rogue) getting belt of one hundred deaths (25 weapon skill) assuming maxed hit and at 360 skill, is there a point where i can add AP now that I have the additional weapon skill from the belt?

IE putting in some 20 ap gems where ive got hit ones to maximize AP without losing hit.

 

Karmon

The raw data we gather is a random series of 0 and 1. A zero represents a HIT, a one represents a MISS.
This is a coinflip experiment with unknown probatility p for "Heads". We try to find this probability.

From theory we know that repeated execution of a coinflip experiment yields a binominal distribution with p as parameter and that a binominal distribution has this p as its mean.
We also know that the empirical mean (ie number of ones/total) is a maximum likelihood estimator for the true p (i.e a "best guess").

There are formulas to give a confidence interval for binomal distributed data but they are tedious.
So we approximate this binominal distribution with a normal distribution (justified by the central limit theorem and the big sample size).
For the mean of this normal distribution we now can give a confidence interval based on the
empirical mean and standard deviation.
And this in turn gives us the probability p for a MISS; usual expressed as a percentage and not as a number between 0 and 1.

So nothing to do with proportions here.

 

◊ songster

Um, the proportion is p. That's what p means in a two-outcome experiment. Yes, you can approximate it as a normal, or you can use the formulas which directly give a confidence interval for proportions. They should be extremely close to each other anyway.

 

Karmon

Sorry then for the confuion here. English is not my primary language and I certainly lack a good proportion (pun intended) of mathmatical vocabulary.

So your saying that this test for proportions WibbleNZ uses is that what I fomulated as "confidence interval for binominal distribution"?

 

Ryjaek

In reality, you can use either test to find the confidence interval. The way you described the data, Karmon, will allow you to handle it as a mean, but it is much simpler to handle it as a proportion. You can consider the proportion to be a special case of a mean where instead of a range of possible values are presented, you are limited to the 0/1 or true/false mentality you talked about. The main reason why the proportion is easier to handle is because the standard deviation is √(p*q/n) instead of √(Σ(x-y)^2/(n-1)), where y=sample mean (no x-bar :P). That's why the listed site doesn't require the stddev for the proportion but it does for mean.

 

Olgas

All I can say is, thank goodness for you math people.

 

sp00n

I'm pretty sure I couldn't have followed this discussion even if it was written in German...

 

Karmon

"And now for something completely different"

I managed to have a couple of hits agains Leotheras dualwielding with 335 hitrating and 353 weapon skill.
335 hitrating should be enough to never miss against lvl 73 boss if Spoon/Karmon theory is correct.
Sadly I had one miss in 254 autoattacks. So that refutes this theory unfortunately.

I will try to get more hitrating tomorrow.

 

Karmon

Nevermind.. not well thought out. Must not post at 3:00 in the morning

 

Gruntle

I'm sorry, but you must have misread my post, I was not talking about the Spoon/Karmon formula, I was discussing Olga's modified wowwiki theory:

This formula gives 25 for def-ws = 10 but gives 26.7 if you let the difference go towards 10 from the positive. A clear jump, the formula is not continuous.

edit: not saying that the formula is wrong, as I wrote earlier it just feels a bit odd.

 

Noktelius

Oups, so I agree with you : A change in the formula around Def - Skill = 10 is weird enought for me, but a jump would be very difficult to understand.

Could someone sum up the facts ? For example :

Def - Skill = 15
MR > 27.95% : At least one miss with Skill=350 & HR=362 & Precision 5/5 (Message 318)
MR < 28.02% : No miss with Skill=350 & HR=363 & Precision 5/5 (Message 318)

Def - Skill = 12
MR > 26,24% : At least one miss with Skill=353 & HR=335 & Precision 5/5 (Message 334)

Def - Skill = 10
MR= 25% ???

Def - Skill = 5
MR > 24.46% At least one miss with Skill=360 & HR=307 & Precision 5/5
MR < 24.54% No miss with Skill=360 & HR=308 & Precision 5/5

Def - Skill = 0
MR = 24% ???

 

Dontmindme

I know that for weapon skill difference 10 or less, there have been multiple tests over at least 3 threads (this and 2 Rogue spreadsheet discussion threads) that prove the 0.1 miss per weapon skill ratio, at least to my satisfaction.

So we are really down to what happens at 11, 12, 13, 14 and 15 weapon skill difference as this is where added miss chances seem to have crept in.

 

Kallistae

Not to derail the topic, but has anyone done any testing on how +weapon skill affects Shield Slam/Shield Bash? Given there's no +shield skill, I'm curious as to whether your mainhand skill affects those two abilities.

I believe this was brought up earlier but I never did see a conclusive response.

 

Amorpheus

These skills depend on weapon proficiency just like other attacks. Somebody once brought up the example of shield slam becoming severely hampered while disarmed with low Unarmed skill.

"You are better than I am," Inigo admitted.
"So it seems. But if that is true, then why are you smiling?"
"Because," Inigo answered, "I know something you don't know."
"And what is that?" asked the man in black.
"I'm not left-handed."

 

Olgas

Unfortunately, I just haven't been able to raid too much lately. Real life has gotten in the way.

However, tonight, I did manage to get another Gruul's run in. And again I was equipped with exactly 142 hit gear (9.00%), just like the last time. This time I had a total of 1427 shots against bosses with ZERO misses.

So, combining this run with my last one, I now have had 2,323 shots with ZERO misses.

Really is looking like 9.00% is our magic number. I'm going to do at least one more run with 142 hit. So, my sample size will be at least 3000 shots. Then, I'm going to drop to 141 hit and see if I can get a miss.

Also, I was paying close attention to my pet's miss rate this time again.

My pet had 2,734 attacks with 142 misses (5.194%). Adding that to my pet's +4% Hit from the Animal Handler talent yields an experienced miss rate of 9.194% (and 9.00% is clearly within the confidence range of this result).

 

thelastrace

From my experience, having the Belt from Vashj and all the rogue talents for hit and weapon skill you only need a 299 hit rating to be capped. I have a few misses having 298 hit rating and have never seen a miss at 299. I am now over that cap at 304 until I find the needed gear to level out the hit cap for me again.

 

Olgas

This would be consist with the working theory.

With that belt, you have a Weapon Skill level of 366. So your Weapon Skill actually exceeds the Defense Skill of a Level 73 boss by 1.

So your base miss rate against bosses should therefore be 23.9% (24% base - 1(.1%)).

With 298 hit rating and Precision, your +Hit will be 23.89756%, just enough to see the very rare miss.

With 299 hit rating and Precision, your +Hit will be 23.96098%, enough to never miss.

 

sp00n

Damnit, I have to do these testings again. Reason: stupidity.
Obviously equipping exactly 45 hit rating was even more difficult than I thought and I equipped only 44 hit rating for these tests. Which of course I didn't notice (WHY THE HELL NOT?) until yesterday I equipped the wardrobe outfit for that test again. And although I tried to convince myself that it was 45 hit rating when I did the tests, I couldn't find a reason why it suddenly should have changed.

Well, again some PvP action then.

 

sp00n

A 'few' Alterac matches later.

I have gathered around 1150 attacks against Drek'Thar and Captain Galvangar (inlcuding special attacks) and have encountered 2 misses (both against Drek btw).

It is a huge difference percent wise to 44 hit rating, but encountering any miss still invalidates the 0.6% theory.

2/1150= 0.00174%, but I have no clue how to calculate these standard deviations.

 

Olgas

Hmm.... very interesting. It does seem to suggest that the .6% theory can't be right.

I guess we next need to figure out a test to see if the modified Wowwiki theory holds up. But I'm having a hard time conceiving of a good test. Any suggestions?

 

sp00n

I've regemmed some of my gear to be able to dual wield and I've encountered a miss with 347 hit rating.
This means that one point of weapon skill is worth less than (28-5-(347/(10*82/52)))/2= 0.49756% tohit.
For the modified wowwiki formula I'd need (4-(2*(0.2+(2/30)))+19)*(10*82/52)= 354.28 hit rating to not miss (0.26667 transformed to 0.2+(2/30)).

I'm going to do some tests with 354 hit rating and see if I encounter a miss.


And looking back to my "I missed a mind controlled warrior" posting, he had 511 defense vs. 360 weapon skill, meaning 151 additional defense (and I was having 297 hit rating and missed a special attack).
(297/(10*82/52))/151= 0.12473

So currently 1 weapon skill is worth something between 0.125% and 0.498%.


If there are some more people interested in finding this out, maybe we could meet on the (European) PTR server and take an approach at Anachromos or some other level 73 mob.

 

Olgas

Great thinking Sp00n. I think this test will tell us a lot. I can't wait to hear the results.

 

0nighthawk0

So...after following this discussion (and i know its not finished), what have we detemrined so far about the value of weapon skill and how it affects things vs Level 70 mobs, 71,72 and 73/bosses?

Understand testing is stil lgoing on, but after reading this thread and getting my head overloaded with figures a summary would be great

 

Matari

I would like to propose a new model for the effects of increased weapon skill versus "Boss" level mobs. Let me state that I am fully aware that the is based purely off of speculation, meaning that I have done no testing of my own. The motivation then, is to propose a model which maximizes the consistency of the various odd reports that we have gotten from Blizzard about increasing weapon skill. The following quotes are the basis of this new model:

[From WoW patch notes]
[From WoW-Eu Forum CM]
[From WoWWiki, no references given]
My assumption is that the 2nd two are correct and that in the first, there is a typo in that the word "rating" does not belong. Assuming this, the 3 quotes are completely consistent with each other. Let us examine what they say.

Against a "Boss" level mob, the player begins with a 15 point deficit. Define the "Base Chance" as the player's chance to miss, be dodged, be parried and to crit with +0 weapon skill. According to the WoWWiki quote, the first 5 points of skill added will result in percentage increases at a rate of 2*R. After 5 points, percentages will increase at a rate of R. This mimics the behavior mentioned in both quotes two and three. At +4 skill rating, players percentages will be increased by 8*R, at +5 10*R and at +6 11*R. We can now use the specific values in quote two to find out what R should be for the different swing outcomes.

to Miss: Decreases at R = 0.1
be Dodged: Decreases at R = 0.05
be Parried: Decreases at R = 0.3
to Crit: Increases at R = 0.1

This is likely the source of quote 1, which states that each point of weapon skill grants 0.1 of *something* related to crit.

In this model, let's see what the talent weapon expertise does. It grants +10 skill for two points, so the first 5 give 10*R percentage points, while the second 5 give 5*R, so the total increase is 15*R. This means that for +10 Skill, the player gains:

+1.5% to Hit
-0.75% to be Dodged
-4.5% to be Parried
+1.5% to Crit

Which seems like a very good bonus indeed. How does this compare with other talents that give +Hit% and +Crit%? One immediately knows that such talents give 1% per point. Given that weapon expertise is much higher in the tree, and has a prerequisite, an increase of 3% for two talent points is not totally out of line.

One final comment about the differing values of to be Dodged and to be Parried. I propose that these values are what they are to give the advantage to the player. Thinking about dodge, Warriors gain the ability to overpower if their attack is dodged, so being dodged is more valuable that simply missing. Thus it is harder to reduce your to be dodged with +skill. Thinking about parry, when a mob parries an attack, it gains an attack speed advantage, so to give a better advantage to the player, it is easy to decrease your chance to be parried with +skill. Also consider that in the vast majority of "Boss" level situation s, players which have +skill items are likely to be attacking from behind, where the to be Dodged and to be Parried portions give them no benefit.