250 dex cap and parry - Test results!

[Zoia]

When i read the Grab Bag, i wasn't sure if they were mixing up 250 cap or 450 cap or what not.
The Camelot Herald
I decided to do a little test myself to find out. I know dex and the effect it has on parry has been tested before, but i can't remember seeing a test with 250 dex vs higher dex.

In these tests i was hitting my valk(without a shield equipped) with my bot. With my bot's crappy weaponskill, it shouldn't be hard for me to come close to whatever potentional parry i had or reach the 50% cap if possible.
I had 4+6 parry and no MoParry in both tests.

In the first test i had 252 dex.
http://img166.imageshack.us/img166/7708/252dexgz8.jpg

In the 2nd test i had 361 dex(max dex cap and aug dex 4).
http://img166.imageshack.us/img166/4118/361dexrv1.jpg

I actually had 0.2% lower parryrate with 361 dex, but some variance is to be expected with such a "small" test.

Looks like my zerker don't need any dex in it's template after all. With no dex at all, i should get exactly 250 dex with buffs. =)


Just as a comparison, Charplan said i would get a 30,1% parry rate with 252 dex and 36,05% with 361 dex. I guess my bot is stronger than i thought.


Edit: The 250 cap should apply to evade and blocking as well, according to the grab bag.

 

[aika]

nice test thanks for sharing the info Zoia!

 

[Tuthmes]

I count 6 extra parry's with le high dex though? Anyways i've always bin curious about the block and parry cap so cheers for testing.

 

[Zoia]

I count 6 extra parry's with le high dex though? Anyways i've always bin curious about the block and parry cap so cheers for testing.

That's right, but the last test had 43 more hits/swings. Look at the percentage.

 

[Tuthmes]

That's right, but the last test had 43 more hits/swings. Look at the percentage.

Bit dodgy on the % tbh. You (could) have bin extreemly (un)lucky in those 43swings. So 6 parry's in 43 attacks = 13.95%.

Not even counting higher ws vs dex etc.

But, still an interresting test. Conclusion still would be that there's a cap @ 250, but then again

 

[anioal]

nice grab bag and nice tests, thanks zoia.

PS. thnx for lettin' me bard to do his assassin missions

 

[Tuthmes]

The way i read it, evade is calculated as (qui+dex)/2 with a cap at 250, no matter where you put it in (100qui + 400dex for example).

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).

Just not sure how you count the + skill and the dex for parry/blocking. 250dex would make me cap and adding + in skill would raise my cap perhaps?

 

[Zoia]

Bit dodgy on the % tbh. You (could) have bin extreemly (un)lucky in those 43swings. So 6 parry's in 43 attacks = 13.95%.

Not even counting higher ws vs dex etc.

But, still an interresting test. Conclusion still would be that there's a cap @ 250, but then again

I think you misunderstood my test a little here.
Yes, i could have been lucky IF those 6 parries had been during those last 43 swings, but they didn't. They happened during the 1132 swings.
You wont get an accurate test from only 1000 swings. You need 10.000 or more for that.
It was 1089 swing with 206 parries and 1132 swing with 212 parries. That's how you have to look at it.

 

[Zoia]

nice grab bag and nice tests, thanks zoia.

PS. thnx for lettin' me bard to do his assassin missions

Aaah, so it was your bard... Grats on RR2! I saw you were a cosantoir on the 3rd run.

 

[anioal]

Aaah, so it was your bard... Grats on RR2! I saw you were a cosantoir on the 3rd run.

ye, strong rp farmage

 

[kirennia]

I thought block rate was effected over 250? In fact I thought the only cap for it was percentage based, going on +skill, mob, dex etc etc If this is true, different template could be incoming...

 

[Tuthmes]

I think you misunderstood my test a little here.
Yes, i could have been lucky IF those 6 parries had been during those last 43 swings, but they didn't. They happened during the 1132 swings.
You wont get an accurate test from only 1000 swings. You need 10.000 or more for that.
It was 1089 swing with 206 parries and 1132 swing with 212 parries. That's how you have to look at it.

No, you cant. Because the test is not equal. You could have 43 or no parry's at all in those extra swings for example. That they happend in those 1132 swings is the only fact you know for sure.

 

[anioal]

hmm, the test was done using a low ws toon... maybe having over 250 in the governing stat will help versus high weaponskill opponents... else i have no explanation why 400dex scouts block that feckin much.

 

[Tuthmes]

hmm, the test was done using a low ws toon... maybe having over 250 in the governing stat will help versus high weaponskill opponents... else i have no explanation why 400dex scouts block that feckin much.

I find it strange aswell. But if you read the grab bag stuff Zoia posted, it shouldnt matter.

 

[Zoia]

No, you cant. Because the test is not equal. You could have 43 or no parry's at all in those extra swings for example. That they happend in those 1132 swings is the only fact you know for sure.

So for a test like this to be valid for you, both tests have the have the same number of swings?
The whole point of doing a 1000+ swing test is to get a number that is as close to the real thing as possible! Whether it's 1000 or 10050 wont matter at all!
Even with two 10.000 swing tests you might get a 0.1% variance. It wouldn't matter if one was 10.150 and the other 10.380. It's the average result that counts.
A 1000 swing test will give a little more variance than a 10.000 test, obviously, but it's still accurate enough to give you an idea of how things are.
In this case, 109 dex should have made a difference, but it didn't.
The fact that my results only had a 0.2% variance speaks for it self. It's simple, basic math, ffs.

 

[Zoia]

hmm, the test was done using a low ws toon... maybe having over 250 in the governing stat will help versus high weaponskill opponents... else i have no explanation why 400dex scouts block that feckin much.

The WS of my bot only helps confirm it. My bot has shit WS and should have little effect on my parry rate. If anything, i should parry more against my bot than i did in this test.

 

[Tuthmes]

So for a test like this to be valid for you, both tests have the have the same number of swings?

To keep it short, yes and no. Much more use would be 10 or 100 test's of 1000 swings, but 43 extra swings can put the % number off quite a bit.

The whole point of doing a 1000+ swing test is to get a number that is as close to the real thing as possible! Whether it's 1000 or 10050 wont matter at all!

It does, especially if your using unequal numbers to compare

Even with two 10.000 swing tests you might get a 0.1% variance. It wouldn't matter if one was 10.150 and the other 10.380. It's the average result that counts.
A 1000 swing test will give a little more variance than a 10.000 test, obviously, but it's still accurate enough to give you an idea of how things are.
In this case, 109 dex should have made a difference, but it didn't.
The fact that my results only had a 0.2% variance speaks for it self. It's simple, basic math, ffs.

Can only agree with you on these points, especially that 109dex should have made a (huge) difference.

 

[Flintlock]

How does this parry work now? You fight mobs and some of the 2 handed guys parry and then whack you for enormous amounts. Really hate taking such mobs on. So parry sux against multiple attackers in RVR? Thought they altered it somehow? Against one though it is still very effective. And it also works off dex only?

 

[Zoia]

You haven't done many tests like this yourself, have you Tuthmes?
Once you get past 600-700, you wont see any huge differences in numbers. Maybe 1-2% or so, but that will get even less the larger your test becomes.
Once you hit 1000+, you should be within +/-1% variance.

Again, it matters very little if the tests are 1000 vs 1100. If it was a 200 test vs a 300 test, then it would matter.

 

[SethNaket]

Increasing the number of attacks in a test increases the accuracy of the result. The second test has a higher accuracy (although the difference is miniscule) than the first one. To claim that you need to lower the accuracy of the second test to be able to compare it to the first one is nonsense. The entire point of statistical analysis is that you can take random samples with random sizes and compare them with varying degrees of confidence. Removing 43 swings from Zoia's second test would make the comparison LESS ACCURATE, nothing else.

This can be seen directly from the formula for standard deviation of the difference between means, provided the samples are independant:

σd = sqrt( σ1^2 / n1 + σ2^2 / n2 )

Increasing EITHER sample size decreases the standard deviation of the difference between the two means, meaning that the calculated difference will be more accurate (in a statistical sense, closer confidence). In this case Zoia didn't test for means but for proportions (% of hits parried) but the formula for difference between proportions is identical.

 

[Frozensolid]

Intresting, never knew this really...
Always thought there wasnt a cap on parry/evade/block statswise...

<starts wondering if his high dex, aside from castspeed, has any use on the poor vw>

 

[Reno]

I'm not sure if I can fully agree with you Zoia.
You really need to do a similar test with a toon with higher WS to see if the results are similar.

Say I have a composite parry score of 10 and a normal dex score of say 150 without any dex buffs.
I now do a tests vs 2 chars, 1 with WS X and 1 with Y, X being really low and Y being a medium score.
With no dex applied X is parried 10% and Y is parried 5%.
I add a few dex buffs and I redo the test, and X still is parried 10% but Y is now also parried 10%. What have you proven then?

You have tested and proven this for 1 set of variables, but you need to test the other variables ( WS and maybe even different parry scores vs various WS's) as well to say with a 100% degree of confidence that the Grab bag is right for a change.

Danita

 

[Kagato]

Curious, though I do have a few doubts.

On the other hand with Parry already having such a very low hard cap at 50% It's not as if it's really going to make a huge differance, Either save a few SC points or save a couple of RA's on masteries. Either way you will easily hit the cap if you really want to.

 

[Inso]

Did a small test on my own. Had a warrior attacking my warrior. Used a smaller datasample though, if anyone wanna wack me for 3 hours while I AFK, pm me.

Attacker is a 1963 WS warrior. (Cheers Obblado <3 )

165 dex, 50+16 shield, 31+13 parry, evade 1 Mastery of Blocking 1, Mastery of Parry 1
42,7% hits
1,3% evaded
19,8% parried
32,7% blocked

258 dex, 50+16 shield, 31+13 parry, evade 1 Mastery of Blocking 1, Mastery of Parry 1
29% hits
3,3% evaded (specbuffed, both dex and qui increased)
29% parried
46,3% blocked

320 dex, 50+16 shield, 31+13 parry, evade 1 Mastery of Blocking 1, Mastery of Parry 1
33,9% hits
3,5% evaded
25,9% parried
45,4% blocked

I used a datasample of 300-500 hits, so that's where the odd numbers come from. It all points to 250 being the cap though, IMO. Looks like the test is correct.

 

[aika]

I think it matters if the hits are styled or not as well.
i.e the cap is maybe true for unstyled hits, but might be higher for styled.
And also Higher WS helps with blocking for yourself (doesnt affect guard)

 

[Tuthmes]

Increasing the number of attacks in a test increases the accuracy of the result. The second test has a higher accuracy (although the difference is miniscule) than the first one. To claim that you need to lower the accuracy of the second test to be able to compare it to the first one is nonsense. The entire point of statistical analysis is that you can take random samples with random sizes and compare them with varying degrees of confidence. Removing 43 swings from Zoia's second test would make the comparison LESS ACCURATE, nothing else.

This can be seen directly from the formula for standard deviation of the difference between means, provided the samples are independant:

σd = sqrt( σ1^2 / n1 + σ2^2 / n2 )

Increasing EITHER sample size decreases the standard deviation of the difference between the two means, meaning that the calculated difference will be more accurate (in a statistical sense, closer confidence). In this case Zoia didn't test for means but for proportions (% of hits parried) but the formula for difference between proportions is identical.

You roll a dice, you got 1/6 of a chance to roll 1(or any number). You roll a dice 100 times and the averidge number will be ~3.5 .
Shamefully we're not working with dice here. I'm not saying the test is wrong at all and i've said aswell:

Can only agree with you on these points, especially that 109dex should have made a (huge) difference.

But using statistic's you should know you can't compare 2 different numbers. The difference with counting those 43 extra swings can be 3.7%. Which imo is quite a lot.

So once again. Yes increasing the numbers makes testing more accurate, but this doesnt allow you to compare 2 different numbers, especially when working with lower numbers.

Once again, yes i agree the test is fine (to some extent). Because one would think with the higher dex you would have a much higher parry rate on averidge.

If you read the grab bag you can only conclude it does cap at 250.

If you look at Caeli's test you can only come to the same conclusion.

Ie. back to Loki.

 

[SethNaket]

Shamefully we're not working with dice here. I'm not saying the test is wrong at all and i've said aswell:

As I said, the formula for proportions, which is what Zoia is testing is basically identical to the formula for means. You just replace σ1 and σ2 with p*(1-p) where p is the weighted average of the two proportions (the parry rates). The effect of the sample sizes n1,n2 are the same wether you roll dice or hit a valk with your shaman.

But using statistic's you should know you can't compare 2 different numbers. The difference with counting those 43 extra swings can be 3.7%. Which imo is quite a lot.

You CAN compare tests with different sizes and that's the whole point of statistics. Here's a little homework for you that I found on a simple google search on proportions (you can find many examples like it on testing differences in your statistics textbook). As a bonus, it's in pink!

PinkMonkey.com Statistics Study Guide 8.15 Test for difference between proportions

 

[liloe]

Oh nice, less dex in my temp....oh wait, I'm pierce spec

Tbh that sucks for elven BM's cause the naturally high dex would favour more defense.....but it doesn't

 

[aika]

elven bm is like avalonian paladin!

 

[Tuthmes]

The effect of the sample sizes n1,n2 are the same wether you roll dice or hit a valk with your shaman.

Nope, the dice and the valk are not the same. Just look at it as if you where throwing with a corrupt dice. You do not know if the valk or Mythic's coding is perfect (so to speak).

I don't know who taught you about statistics but whoever it was didn't know what he was doing. You CAN compare tests with different sizes and that's the whole point of statistics. Here's a little homework for you that I found on a simple google search on proportions (you can find many examples like it on testing differences in your statistics textbook, if you even have one).

Once again you can and you can't. The lower the number of your test, the more inaccurate it gets. Comparing 2 of those with eachother is even more inaccurate and even more inaccurate is comparing test with different numbers.

If i'd roll the dice the averidge outcome on 10 rolls should be 3.5, but dont be surprised if someone has 5 as an averidge. If someone else rolled 10 times and hade exactly a 3.5 outcome and rolled 2x 1 after that his averidge is lower then usual.