250 dex cap and parry - Test results!

[Inso]

Respecced to Aug Dex5, Mastery of Blocking 5 and ran some tests versus someone that styles.

Attacker is a 2168 WS warrior using Draw Out. Using ~1000 hits as base for the data.


268 dex, 50+14 shield
2,8% evade
21,1% parry
69,2% block

368 dex, 50+14 shield
3,6% evade (qui increased by specbuffs)
26,1% parry
79,2% block

Seems that it does help versus styled hits. Though it might also be the MoB5 giving strange results.

Can't edit my posts for some reason, remaking it here since it seems I can't read properly at the moment...


Respecced to Aug Dex5, Mastery of Blocking 5 and ran some tests versus someone that styles.

Attacker is a 2168 WS warrior using Draw Out. Using ~1000 hits as base for the data.


268 dex, 50+14 shield
1,2% miss
22,4% hit
2,8% evade
20,5% parry
53,1% block

368 dex, 50+14 shield
0,9% miss
13,9% hit
3,6% evade (qui increased by specbuffs)
25,1% parry
56,5% block

 

[Tuthmes]

So the statistics are now saying block and parry rate go up with more dex, or are the numbers wrong this time?

 

[Inso]

Versus styled hits there was a small increase.

 

[Kinag]

I don't get this (not that I'm a genius at statistics anyway).

Shouldn't your blockrate increase with MoB5, but instead it seems to go from 79,2% to 56,5%.

Am I missing something here or does MoB seem to be ineffective?

Sorry if I missed a 'obviously' clear point, cause as I said, I don't understand Statistics much

 

[SethNaket]

So the statistics are now saying block and parry rate go up with more dex, or are the numbers wrong this time?

Depends on how certain you want to be. The following is the standard way to analyze a double test, starting with parry rate (because the thread started about parry).


The starting assumption is that there is no difference, so you form the "null hypothesis" and the "alternate hypothesis":
H0 : pr1 = pr2 (pr=parry rate)
Ha : pr1 != pr2

You also have to decide on a level of significance, or "how certain" you want to be. Usually the choice is 95% certain (aka two standard deviations rule). Since one assumes the rates are the same, this translates in to testing if there's less than 5% chance they are the same, or a probability of 0.05.

Calculate the proportions for each test (already done):
p1 = 0.205
p2 = 0.251

Calculate the difference in proportions:
pd = p2-p1 = 0.046

Calculate the weighted average proportions (here I assume both tests are 1000 hits so n1=n2=1000 so this is just the average of the two rates):
p = (n1p1+n2p2)/(n1+n2) = (p1+p2)/2 = 0.228

Calculate the standard deviation of the difference (the formulas I showed you earlier) again assuming n1=n2:
sd = sqrt( p(1-p)/n1 + p(1-p)/n2 ) = sqrt( 2* p(1-p)/n ) = sqrt( 2*0.228*0.772/1000 ) = 0.0188

Calculate the "Z value" for the test:
z = pd/sd = 0.046/0.0188 = 2.447

Finally look up the Z value in a table to see what probability it corresponds to, and compare with the chosen significance above (0.05)

z=2.45 corresponds to approximately 0.015 whichs is less than our chosen limit 0.05. In other words, there's less than 5% chance they're the same, so we reject the null hypothetis and accept the alternative. The tests show that under 95% certainty there's a real difference in parry rates.

You can shortcut and only look at the z value => propability conversion and say that we're 98.5% certain they parry rates are really different. Another shortcut is to simply know from experience that 0.05 probability corresponds to z=1.96 and simply state that 2.45>1.96 so we reject the null hypothesis.


However he also tested shield, and repeating the same process there gives (without writing everything again)

z = (0.565-0.531)/sqrt(2*0.548*0.452/1000) = 1.53

1.53 is less than 1.96 so we can't with 95% certainty reject the null hypothesis for shield from this test.


So to sum it up, with a confidence of 95% the test shows that there is a significant difference in parry, but it doesn't show a significant difference in blocks. With reservations for typo's along the way, only used the windows calculator to do this, and the assumption that both tests are 1000 hits (wasn't stated explicitly if BOTH were 1000 or total was 1000).

 

[Gidget]

i think what he was saying was : i made a mistake reading /posting the results of my tests, but couldn't edit my post, so here the correct results from the same test.

something else i was curious about... from what i remember (no clue if it's correct tho) is that the engine will first calculate parry, and therefore fuck up your blocks if you parry. i don't have an open account so i can't try myself, but can't you prevent your character from parrying when unequiping your weapon? also have i no idea if this influences your testing results really...

 

[SethNaket]

something else i was curious about... from what i remember (no clue if it's correct tho) is that the engine will first calculate parry, and therefore fuck up your blocks if you parry. i don't have an open account so i can't try myself, but can't you prevent your character from parrying when unequiping your weapon? also have i no idea if this influences your testing results really...

Yea, if you want to be really thorough you need to account for the order they're checked. I don't even remember which order it is so I didn't compensate for it above, but it could increase the difference of whichever comes after the other (so if parry comes before blocks then the difference in blocks might be significant afterall).

 

[Zoia]

I believe it's evade, parry, block and then misses. Same order as in the screenies of my test from the parser.

Caeli, it would be great if you could post exactely how many swings you had in each test and how many parries, etc, you had.

 

[Rigga Mortice]

Yea, if you want to be really thorough you need to account for the order they're checked. I don't even remember which order it is so I didn't compensate for it above, but it could increase the difference of whichever comes after the other (so if parry comes before blocks then the difference in blocks might be significant afterall).

The order of combat resolution is Evade, Parry, Block, Guard, Hit/Miss, Bladeturn, so only hits that aren't parried or evaded should counted in block calculations.

Edit: nerf, Zoia beat me to it

 

[SethNaket]

I believe it's evade, parry, block and then misses. Same order as in the screenies of my test from the parser.

Ok, in that case (still assuming 1000 hits each), the z-values become:

Evade: z=1.00 < 1.96, H0 stands -> not significant
Parry: z=2.565 > 1.96, reject H0 -> significant difference
Block: z=4.391 > 1.96, reject H0 -> significant difference

Edit: Just a quick note, z=4.391 is huge, means 99.98% significance

 

[SethNaket]

As a comparison, Zoia's original test show z=0.114, or about 91% chance the rates are really equal. (nerf 10min edit limit!)

 

[Zoia]

Can't edit my posts for some reason, remaking it here since it seems I can't read properly at the moment...


Respecced to Aug Dex5, Mastery of Blocking 5 and ran some tests versus someone that styles.

Attacker is a 2168 WS warrior using Draw Out. Using ~1000 hits as base for the data.


268 dex, 50+14 shield
1,2% miss
22,4% hit
2,8% evade
20,5% parry
53,1% block

368 dex, 50+14 shield
0,9% miss
13,9% hit
3,6% evade (qui increased by specbuffs)
25,1% parry
56,5% block

This test is flawed though. The combined percentage in these two tests is 100%, so it looks like you didn't use a parser.

If you have a 1000 swing test, then get 10 evades, 200 parries and 500 blocks, that doesn't mean you have 1% chance to evade, 20% chance to parry and 50% chance to block.

It would be 1% evade, but then you'd be left with 990 hits when calculating parry. 200/990=0.2020 or 20,20% parry.
You would then have 790 hits left that might get blocked. Out of those, 500 got blocked. 500/790=0.6329 or 63,29% blocks.
Then you will calculate misses.

Because your dex/quick buff increased your evade rate significantly and evade is calculated first, this affected your results when you calculated it the way you did.
You may actually have an even higher parry/block rate than what you got, but post the numbers and we'll see.

 

[Zoia]

Respecced to Aug Dex5, Mastery of Blocking 5 and ran some tests versus someone that styles.

Attacker is a 2168 WS warrior using Draw Out. Using ~1000 hits as base for the data.


268 dex, 50+14 shield
2,8% evade
21,1% parry
69,2% block

368 dex, 50+14 shield
3,6% evade (qui increased by specbuffs)
26,1% parry
79,2% block

Seems that it does help versus styled hits. Though it might also be the MoB5 giving strange results.

I went over your numbers again, and it seems like these are the correct ones. Kinda weird you got such a high increase in parry and block, tbh.
/ponder

 

[Tuthmes]

Depends on how certain you want to be. The following is the standard way to analyze a double test, starting with parry rate (because the thread started about parry).

Ohh now certainty matters all of a sudden, seriously..... Like i said before, you can use numbers and statistics all you want and asses your certainty, you just don't know (exactly) how Mythic coded everything.

With reservations for typo's along the way, only used the windows calculator to do this, and the assumption that both tests are 1000 hits (wasn't stated explicitly if BOTH were 1000 or total was 1000).

Why do you want to assume both test are 1000hits, hmm? According to your earlier posts it doesnt matter what size the samples are.

And you can set your Windows Calc to sientific btw.

Anyways enough with the bs'ing around, bit more OT now.

The part in the grab bag suggests there is a cap @ 250 (ill quote it again) for every skill.

There is a relatively generic cap at 250 in terms of skill use, for example when using parrying, shields, evade, etc. If the skill uses more than one attribute, the cap will be applied to the average of the two (parrying uses just dexterity, whereas evade uses dexterity and quickness, for example).

Then again, the frase; "relative generic cap" makes my eyes bleed aswell.

What "troubles" me is that we atm we should be able to go above 75% blockrate (cause inc patch it gets nerfed down to 75% max). If you take the stats Caeli gave us, you need tobe rr11 with 50+14 shield and mob5 to even hit that cap.

Anyone mind if i'm posting here again btw, or do i need to foad in bad ways?

 

[Inso]

I don't get this (not that I'm a genius at statistics anyway).

Shouldn't your blockrate increase with MoB5, but instead it seems to go from 79,2% to 56,5%.

Am I missing something here or does MoB seem to be ineffective?

Sorry if I missed a 'obviously' clear point, cause as I said, I don't understand Statistics much

Read my post again. I had to repost since FH is wank and wont let me edit/delete posts.

I believe it's evade, parry, block and then misses. Same order as in the screenies of my test from the parser.

Caeli, it would be great if you could post exactely how many swings you had in each test and how many parries, etc, you had.

The first test (yellow basedex, 268 dex, 2148WS attacker using styles) was 1030 attacks, 231 hits, 29 evades, 211 parries, 547 blocks and 12 misses.

The second test (red basedex & specdex, 368 dex, 2148WS attacker using styles) was 983 attacks, 137 hits, 35 evades, 247 parries, 555 blocks and 9 misses.

 

[Inso]

What "troubles" me is that we atm we should be able to go above 75% blockrate (cause inc patch it gets nerfed down to 75% max). If you take the stats Caeli gave us, you need tobe rr11 with 50+14 shield and mob5 to even hit that cap.

Forgetting that Booooomer has kinda high WS? Not all get 2100+ WS... Remember that it will be 75% of the hits that bypasses evade/parry aswell, so the actual number of hits deflected will be alot higher then 75%.

 

[Gidget]

so assuming caeli did a test with 1000 swings, subtract evade and parry and have 713 swings left, 56,5% out of 1000 blocked so 79,24 block, with a slightly higher deviation, due to the order if i did understand that correctly?

if so; his parrys going up from 21 to 26% and his block from 69 to 79% have to be somewhat related to using styles, since he did a test unstyled before, where 60 dex made the block rate go down by 1% due to the deviation (i guess?) and i doubt that specing high moblock makes you benefit from overcaping the dex hard cap...

which proves this wrong, doesn't it?:

Styles only decreases my chance of missing with their to-hit bonus and has nothing to do with penetrating defences. Since misses is calculated after parries, that is irrelevant to what i'm testing here.

also someone said a block cap of 75% will be implemented in an upcoming patch, but isn't now? or why would he possibly get a 79% block rate? there is a small chance for the deviation to be that high, isn't there?
and that vs 2.1k WS, shouldn't you block low ws players 100% then, or what's the cap at the moment?

oh an tuth do whatever you like, just since you were asking; in my opinion your posts are a bit tiring, and in this thread not worth reading any more

 

[MegaMaejter]

Would be nice if someone could do a test where the test enviroment is more controlled.

1) 1000ish swings with 250ish dex ( no MoB ), no weapon equipped to get rid of parry as a variable.

2) 1000ish swings with as much dex as possible ( no MoB ), no weapon equipped.

Then you might wanna do a test that includes MoB, to see if it has any effect on the assumed 250 cap at wich dex no longer increase your blocking.

Basically you wanna test if having MoB increases your block to such an ammount there is reason to think that it raises the 250 "cap" and its own said value.

Exampel to make my point abit clear.

Test 1), 250 dex, no MoB 45% block
Test 2), 400 dex, no MoB, 45% block

Test 3), 250 dex, MoB3 ( 10% added block ) and 55% blocking
Test 3), 400 dex, MoB3 ( 10% added block ) and then you get all of sudden 65%+ Blocking with only alternation from test 1 and 2 added dex combined with added MoB as exponation.

P.S fictional nummbers used to clearify my point.

P.S2 I guess you will need at some point also test styles effect on the block %, by reading abov posts it seems like block% nummbers varied alot when styling and not styling wich is odd since I was under the impression styling only boosted your to hit chance, as in lowering the chance to flat out miss.

 

[Tuthmes]

Forgetting that Booooomer has kinda high WS? Not all get 2100+ WS... Remember that it will be 75% of the hits that bypasses evade/parry aswell, so the actual number of hits deflected will be alot higher then 75%.

Yep, but still doesnt chance the fact that you need to be way up there to hit the cap ^^.

Also look at it from different char's. Ie. Scouts have no parry for example or chimps/skalds with just parry. Not to mention other specs with lower shield/parry (42shield/20parry for example).

Not too mention you need to test WS and how much effect it has (in general mebbe too!) on defensive stuff aswell. And ofc if WS has a cap too!

 

[SethNaket]

Ohh now certainty matters all of a sudden, seriously..... Like i said before, you can use numbers and statistics all you want and asses your certainty, you just don't know (exactly) how Mythic coded everything.

Excuse me? Did nothing of my explanation of statistics earlier register at all with you or did you get so upset that someone corrected you on a messageboard that you now have to hold a grudge? I demonstrated in my post how you statistically analyze a test, no ifs and buts. That's how it's done and it really doesn't matter if you like it or not.

Why do you want to assume both test are 1000hits, hmm? According to your earlier posts it doesnt matter what size the samples are.

I don't "want" to assume 1000 hits, but Caeli only stated it was based on about 1000 hits. Not 1000 each, 1000 combined, or even exactly 1000, just "~1000 hits". I never stated size doesn't matter, seriously, are you just stupid? If you don't know the answer but still want to analyze the test you have to make assumptions, it's basic scientific methodology.

And you can set your Windows Calc to sientific btw.

You've obviously never used a real scientific calculator.

 

[SethNaket]

The first test (yellow basedex, 268 dex, 2148WS attacker using styles) was 1030 attacks, 231 hits, 29 evades, 211 parries, 547 blocks and 12 misses.

The second test (red basedex & specdex, 368 dex, 2148WS attacker using styles) was 983 attacks, 137 hits, 35 evades, 247 parries, 555 blocks and 9 misses.

Using significance level 0.95 (95% certainty) the difference in rates between the two tests are:
Evade: Not signifcant (z=0.91<1.96)
Parry: Significant (z=2.59>1.96)
Block: Significant (z=4.36>1.96)

 

[Tuthmes]

Excuse me? Did nothing of my explanation of statistics earlier register at all with you or did you get so upset that someone corrected you on a messageboard that you now have to hold a grudge? I demonstrated in my post how you statistically analyze a test, no ifs and buts. That's how it's done and it really doesn't matter if you like it or not.

Corrected? But yeh I do actually hold a grudge. People trying to be smart with statistics, but completely missing the point, pisses me off.
Every 2 year old could conclude with Zoia's first test that there seem(ed) to be a cap @ 250 dex.
Every 2 year old knows that with higher numbers the accuraty goes up, the point i started with, is that there whas a differance in accuraty. Which is something you really want to avoid here.

I never stated size doesn't matter, seriously, are you just stupid? If you don't know the answer but still want to analyze the test you have to make assumptions, it's basic scientific methodology.

See for example:

there's no problem comparing two samples of different sizes using statistics and the sample size affects the accuracy in the exact same way wether you roll dice or try to test parry.

But then again, you prolly don't care at all if the 2 samples givin to you are relative to eachother. Saying size matters or its no problem comparing two samples of different sizes, really is saying 2 different things.
If don't know the answer but still want to analyze the test you make an hypotheses, not assumptions and you rule out anything that can effect your test, in any way possible (and add in a control group). It's basic scientific methodology, really.

You've obviously never used a real scientific calculator.

Nope dont need it <3

 

[Zoia]

Corrected? But yeh I do actually hold a grudge. People trying to be smart with statistics, but completely missing the point, pisses me off.
Every 2 year old could conclude with Zoia's first test that there seem(ed) to be a cap @ 250 dex.
Every 2 year old knows that with higher numbers the accuraty goes up, the point i started with, is that there whas a differance in accuraty. Which is something you really want to avoid here.

And we're trying to tell you that 43 more or less hits in one test wont make any huge difference in accuracy in this case, when both tests have 1000+ hits. It's perfectly fine to compare the two.
You're really starting to piss me off now. No matter what people tell you, you keep arguing and ignoring what we say to you.
You're also trying to make it sound like people are saying different things than what they actually are saying, often by using quotes out of context.


Caeli's test was interresting though, and i'm doing one with my skald now. Maybe someone with higher parry than my valk will produce other resuslts. We'll see.
Either way, i can guarantee you the samples wont be of the exact same size.

 

[Tuthmes]

And we're trying to tell you that 43 more or less hits in one test wont make any huge difference in accuracy in this case, when both tests have 1000+ hits. It's perfectly fine to compare the two.
You're really starting to piss me off now. No matter what people tell you, you keep arguing and ignoring what we say to you.
You're also trying to make it sound like people are saying different things than what they actually are saying, often by using quotes out of context

Fair enough, if you wanne stick your head in the sand and keep testing be my guest.

Ill be off again...

 

[SethNaket]

Every 2 year old knows that with higher numbers the accuraty goes up, the point i started with, is that there whas a differance in accuraty. Which is something you really want to avoid here.

LOL! Stop right there. First of all, your "point" as you put it was that it's not possible to compare samples of different sizes, a statement that is simply FALSE. You didn't say "there's a difference" you said "you can't".
Secondly, only one of us has made any ACTUAL assessment of what the accuracy of these tests really are. I'll give you a hint: NOT YOU!

I never stated sample size doesn't matter, the quote you copied explains exactly what I've been saying all along that you just don't understand. Statistics, which is a METHOD, work the same on any sample size and larger samples give higher accuracy regardless of what you're testing. I'm sorry that you're not able to follow this logic, there's no contradiction in that statement.

Nope dont need it <3

So why try to make witty comback when you don't even know what the difference is? Purposely trying to make yourself look even worse?

 

[Bistrup666]

Because the only thing you can do with statistics is to estimate how accurate a test (of whatever kind) is. A test with a "margin of error" of 2 is more accurate than one with a margin of error of 3. A test with a standard deviation of 3.9 is better than one with 4.0 even if there difference is tiny. Those statements are facts regardless of the sizes of the indivudual samples. Chosing identical sample sizes doesn't magically improve the accuracy as you seem to think. The ONLY thing you can alter that affects it is the sample size n, and increasing it always increases accuracy as I showed you with the formulas above (that you keep ignoring).

If you knews as much about statistics as you seem to think, you would know that the only thing I need to say to refute your claim is to post the formula for std.dev. of the difference between proportions.

So, to recap, comparing 1089 swings and 1123 swings is MORE ACCURATE than comparing 2x1089 swings. ALWAYS.

So true the reason being as Zoia started out with, 1123 gives U less variance than 1089 hence it is more accurate.

The ONLY difference between the 2 tests is the variance, and that is marginally lower for 1123 than for 1089.

 

[Zoia]

Fair enough, if you wanne stick your head in the sand and keep testing be my guest.

Ill be off again...

Eerh, i'm really not the one in this thread sticking my head in the sand. Or should we say holding your hands over your ears?
I have no problem with listening to people if they have something usefull to say. It seems you're just arguing for the sake of arguing now.

 

[Zoia]

Oki poki, new test results are in!

This time i did 3 tests and used my skald with 17+16 parry and MoParry 1. All 3 tests have exactely 1200 attacks. Hopefully that will make Tuthmes happy(you know i love ya, you stubborn bastard). :wub:

152 dex:
4,7% evades
30,7% parries
11,5% misses

251 dex:
5,4% evades
40,5% parries
14,5% misses

345 dex:
6,8% evades
42,5% parries
12,6% misses
Edit: In this last test i had dex/quick buffs, which may explain the higher evade.

There's an obvious increase in parries from 152 to 251 dex, but only a minor increase from 251 to 345. Still some increase though.
I can't really explain why, but maybe 250 is a soft cap and there's diminishing returns after that. Maybe how much you have in parry/shield has an effect as well, who knows.
I still don't believe styling has an effect on parries/blocks/evades(only misses), nor do i think it matters whether i'm using my bot or a class with higher WS in this test.

Perhaps it's worth it to SC in some dex if you're a class with high parry/block skill and not worth it if you're a skald.
I could test it with my high parry zerker as well, but i don't think i will bother. Swinging for hours is boring.

Thanks to Caeli for his tests. Much appreciated!

 

[SethNaket]

Under 0.05 significance level (95%):

Going from 152 to 251 dex the parry change is significant but not the evade change.

251 to 345, neither are significant.

151 to 345 both are significant which basically means there IS an evade increase but we can't from the test say if it's from the first or the second difference, or both.

 

[Zoia]

Under 0.05 significance level (95%):

Going from 152 to 251 dex the parry change is significant but not the evade change.

251 to 345, neither are significant.

151 to 345 both are significant which basically means there IS an evade increase but we can't from the test say if it's from the first or the second difference, or both.

Not with the evade, but definately a difference with parry, when you compare 152-251 vs 251-345?
Edit: Since skalds only have evade 1, which is 5% base evade, it will be hard to tell a difference in a 1200 sample test, if you have, say, +/-1% variance.