Damage calculations

[Nazrious]

Don't take this personally but your math makes no sense to me atm. Can you elaborate?

There is no difference between

"PCCfight> Gorgetfight> PCCFigt> Gorgetfight> PCCFigt> Gorgetfight> PCCFigt"

and

"200 fights in PCC then 200 fights in say a gorget or torque"

in data collection.

Actually there is a difference in that alternating it makes cirucmstances much more similar for both sets of data. less of a chance for whms or any one inparty not doing something from fight to fight then there is in 200 strait fights then another 200 strait fights. time of day changes people get tired etc etc. By alternating fights you get more similar cirumstances.

as for my staments after this never said they couldn't do the same had they not used the above method was just goign thruogh the steps. Also of course a sample of 200 is tiny, wasn't saying it was large. However unless you have some dedicated testers I don't see many test running for longer then 400 fights.

it is also much better to alternate in the method I stated when you have testing done over several days rather then just doing say 3000 fights over a few days in gear set A then doing 3000 fights over another few days in gear set B.

doing 6000 fights over a week or so while alternating gear from fight to fight will allow for variables to have a lessened affect on the outcome.

~t~

[AoshiZ]

Actually there is a difference in that alternating it makes cirucmstances much more similar for both sets of data. less of a chance for whms or any one inparty not doing something from fight to fight then there is in 200 strait fights then another 200 strait fights. time of day changes people get tired etc etc. By alternating fights you get more similar cirumstances.

~t~

Ok, so you mean like consistency of the enviromental variables such as debuffs and buffs from the other players

[Nazrious]

Yep. exactly what i was getting at.

~r~

[Aymee]

He also claims that a better judge of which piece of gear gives better damage output is by parsing, while I say that there are just too many variables going on in a parse to give as accurate a readout as paper calculations. What do you all think on this topic?

I am a total parsing addict. Numbers are good. Numbers teach you things. Since you can't mentally process everything that's going on, this is the only real way the vast majority of people are going to be able to judge their performance and improvement over time. Even someone totally new to parsing can rather easily calculate out what their results are like on, say, the classic meat vs sushi debate. If you're too lazy even for that, you can just look at the chart and pick out where you'd like to improve--accuracy and dmg/hit are the two most obvious things you can directly try to increase... avg#WS/mob (a judge of TP build, kinda) is good, as is avgWS assuming you use the same thing all the time as most do. I started with just those few basic things, and over time I've been able to tweak out the numbers more to my liking. This takes potentially a while lot of data and a lot of PTs/time.

Parsing is subject to a billion errors, of course. It can't be used for anything entirely scientific, though with repetition and comparison you can get a good feel for a lot of general statistics, which is one way we can keep up with pDif and modifiers and whatnot. For the most part, if you're meritting out as the same job in basically the same gear using basically the same food, etc., Results after ...eh, 500 mobs or so... I'd say are accurate within 2 standard deviations. Go save it and look at it later.

Common parse issues:
--People who think they know how to parse, and fail. Like, the kind of people that like to give results based on dmg%/mob (or dmg/mob for that matter) instead of dmg% total. Also, people who give results based on just a portion of the PT length, or off of a limited number of mobs (anything under 100 mobs I'd say is junk data).
--Parsewhoring makes me cry. An example of this would be a SAM using gekko on a bird with 1% hp, doing full dmg on a 1-hit WS, whereas if someone popped a rampage they'd only get the first hit dmg. More people do this than think they do, even if it is accidental.
--Adds: Any time that not everyone in the PT is attacking the same mob, it ruins consistency in results. Additionally, depending on the parser/system, some data can get screwed up and/or lost when multiple things are being dealt with at the same time.
--'Wyvern Fallacy': Mob pets (raptors, puppets, wyverns) totally ruin the dmg%/mob to where it is essentially invalid. Quick example--say 3 melee attack a DRG mamool. One goes just for the wyvern. The other two have TP and unload on the mamool, killing it in the same amount of time it takes the wyvern to be killed with normal hits. DMG(%)/mob in parsers is based only on the number of mobs a person is recorded as touching. The two that attacked the mamool get 50%dmg/mob recorded for 1 mob attacked, and the 3rd gets 100% for 1 mob. Some people abuse stuff like this and it's basically an extension of parsewhoring in a lot of cases. If you think it doesn't add up, the 100%/mob statistic plugged in for a theoretical 10% of pulls being DRGs ruins the %/mob 10%. It's useless data.
--Duplicate Mob Fallacy: If you're doing exclusively colibri, or multiples of puks or w/e in a row, it's quite possible that the 2nd/3rd+ mob is going to attack or be attacked before the first mob was killed entirely. Hell I do this--if mob1 is at 10% and mob2 is eating a brd, I'll switch to the next mob and dump TP if I have it. The parser doesn't know which mob is separate, as it just looks at the name ('Greater Colibri', etc.). As a result, the parser ends up with wildly different amounts of HP for each mob... thus dmg and %/mob is thrown off. The 'polite' alternatives mess with parsing for the good of the PT as well--I might mistral a mob at 5% if we're 2 seconds away from losing chain when I could have done 3x the dmg on the next mob instead.
--BNS (big number syndrome): Never trust people that say they always have an average WS of whatever or a constant accuracy of whatever, even if they say they can give you proof. What mob you're exping on makes huge differences. Or... say... Soboro SAMs that have parse data saved of beating up ridaberk WARs 2:1 from the week that the equations were literally broken. Don't trust stuff like this.

Biggest issues that no one ever seems to pay attention to and will just declare themselves parse winner instead:
--What camp? MJSP lets you use food, but has mobs that throw off accuracy, and puks are overly weak to some things more than others. On colibri, some people will spam [cheap] food anyway, knowing it'll get taken. Also, piercing bonus can throw things off. Imps/mire crap... various status effects will hurt the output of some people.
--Tanking? There's no way to compensate for a NIN-WAR-PLD-MNK that's voking and spending time casting versus say, a SAM.



I'd say the opinion of the OP of 90% or so accurate in any given situation is pretty good. Even stuff like moon phase can mess with results in a small way... you can never account for anything. 99% is ridiculous--we will never be able to get to 99% on anything whatsoever, past obvious dmg caps and using that data to plug in. Look at it this way--practical cap on acc is 95% even if you're 'perfect', so why necessarily think you'll be able to be substantially more accurate than that?



What's the closest to an actually 'fair' parsing situation? After a lot of thinking, this is what I come up with: Main deco camp, no food, no brd or cor, blinkers cast only between mobs and Ni during mobs if desired (time wasted = roughly the length of a whirl stun on a non-blinker so I'll call it close enough to even), with a dedicated person tanking. No weird bonuses or penalties to worry about, no pathos randomly afflicted to anyone that could mess with their numbers. If you want any sort of 'real' comparison between you and another person, this is the only way to do it (imo). Yes, more 'normal' exping will give decent results that can average out to being practically the same, but this is a showdown that doesn't play favorites to anyone.

[Klutz]

Damn.. /walloftext. >_>

[Woozie]

Our approach at understanding the damage calculation is comparable to a physicist approach at understanding nature.

Probably not a great example seeing how no single physicist has been able to accurately and completely explain nature. Every working theory has major gaps, and no one can agree on a universal understanding of nature. It is one of the least understood and most controversial topic in science. The only thing everyone can agree upon is we know the "physical laws" we operate under today are wrong. But no one has yet to find the right answer.

I actually think its a great example.

Since damage calculation's exact inner workings are based on our frame of reference and the data we are able to collect thus far. Similar to how we have formulas for gravity such as x = 1/2gt^2 that give us great results for our reference frame. Even though there are possibilites with darkmatter and newtons grav theory being wrong and all that.

AoshiZ's answer is exactly what I was going to say. The reason I used physics as an example is because although the laws of physics we use may not be the way nature intended, they still work in their domain of applicability and can still be regarded as "correct" because they DO give the correct answers.

Our damage formula may not look exactly like SE wrote them, but they still give us the correct answers.

[Aymee]

Damn.. /walloftext. >_>

Uh... sorry about that. I just ramble. A lot. That was like 2+ years of annoying people not understanding wtf a parser actually spits out in data coming out in a post--no, it's not just whatever number people have the highest of so they claim they can 'win'. I'd go insane with numbers and go and calculate margins of error depending on # mobs or something, but thank god I forgot how to do that kind of stuff.



..........Ok, I looked it up anyway. In my last post I said that anything under 100 mobs/samples is junk data. Wikistuffs and some fragments of my memory show that 96 samples are needed for a margin of error +-10% with 90% confidence... which is probably why I picked 100 mobs in the first place when I started doing this. So there you go, for the OP's hope of ~90% accuracy in data calculation, we'd need to test out whatever 96 times. It's quite possible to make some general equations of things with this kind of sample, but obviously room for improvement. On the other hand, for a 99% confidence with a margin of error of +-1% (to theoretically encompass 'perfection'), you'd need a sample size of, uh... 16600 or so. I think. This is why people would rather hack the game code or work off general starter equations. In other words, unless we get an exact formula of things, it's practically impossible to be 99% certain. Your friend is exaggerating things a bit.

So if you're parsing vs your friend in a merit PT... you hit 34% and they 31%, hmm... @ 100 mobs (or ~1hr of PTing), 90% confidence level would be +-8%... that's 8% of your dmg... so we're 90% certain the 'true' range is ~28.5% to ~36.5%, which is a big gap yet. 1000 mobs (multiple PTs of collected data), for 90% confidence we're to +-2.5%... within that range your minimum is larger than your friend's potential max so the data has far more validity. Getting +-5% (which would just about max the highmin = the lowmax) is ~270 samples... length of an average PT. That works out. So from a random theoretical standpoint, anything under ~250 samples and I'd hardly call the data worthwhile.

On the other hand, when I did 43% to 21% for a BLU (this one needs a lot of help) last night, we'd only need +-33% margin of error, or so. 90% confidence is 6 samples! Rawr. 99% confidence is a mere 15 samples, and for the number of mobs we killed... It ends up being about 4 standard deviations off, so we can be 99.994% sure that something is quite awry with this BLU.

Ok I need to stop. My head hurts. Now I want to write this out so I can plug numbers in from my PTs

[Wraith]

Ok, I think some people are misunderstanding what I'm talking about, and assuming I've forgotten important parts of the equation. I'm fully aware that the margin of error in a parse is huge, because there are so many uncontrollable variables going on that it's near impossible to get any accurate data from one in terms of overall production. They're good for small things like, "do I have enough accuracy to use meat," or "does this weapon substantially improve my average WS." These questions don't require much precision to get a satisfactory answer. However, what I'm talking about are minute changes that would be hard to see on a parse, like the example of Myochin kote +1 and chiv chain vs. bandomusha kote and pcc. For those who are misunderstanding what I was talking about, I'll go through the entire process I used.

First off, the stats for Myochin kote +1 and Chivalrous chain (Set A):
10 str, 7 dex, 5 accuracy
We know that for 2hand users, the conversion of dex to acc and str to atk is 4:3
Therefore, the above setup has an accuracy of 10.5.

The stats for Bandomusha kote and Peacock charm (Set B):
22 attack, 10 accuracy

It's clear to see right off the bat that Set A has better accuracy. The question now becomes, which will give the best boost, on average, to a melee hit. Since I was calculating this for my friend, I'll use his stats in the calculation.

Ok, so the equation for a melee hit (taken from Studio gobli and translated by our buddy VZX):

Damage = { DMG + floor[( Your str - mob vit + 4 ) / 4 ]} * [( your atk / mob def ) - LC ]

The stats for a level 82 Greater Colibri are 67 vit and 327 defense.
My friend uses Hagun, has 103 str, and 438 base attack. I accounted for berserk and minuet4 being on him, which would bring his attack to 571. Also, since the colibri is 7 levels higher, the LC would be .35.

Using the above values for the variables in the equation, his melee damage would be 118.675.
Now, to calculate exactly how str compares to atk, I take the above equation, subtract 4 from my friend's str, and see how much attack it takes to have an equal damage as before.

118.675 = ( 75 + floor((99 - 67 + 4 ) / 4 ) * ({[( 438 + x ) * 1.15 + 61 ] / 327 } - .35 )

Solving for x gives a result of 10.2, which divided by 4 gives a result of 1 str ~ 2.55 atk.

This value gives a good conversion between str and atk (for my friend's specific situation) to determine which set helps him the most.

Myochin kote +1 and Chivalrous chain (Set A):
25.5 "attack points," 10.5 accuracy

Bandomusha kote and Peacock charm (Set B):
22 "attack points," 10 accuracy

So, from these calculations it's clear to see that for my friend's situation, Set A is better.

So this is the basis I used to tell my friend that Set A would be slightly better for him. He then claimed that his parses showed the opposite result. So my original question would be, are my calculations accurate? If I made a mistake somewhere, that could account for his claim, otherwise I think it's pretty clear that AF+1 gloves and chiv chain are better than b.kote and pcc.

[Amele]

7 dex is 5.25 accuracy (not 5.5) which is to say that '25%' of the time (when you have a remainder of 3 dex from other gear) you'll get 6 and the rest of the time you'll get five.

you also state 103 and then use 99 in the formula, so that doesn't add up quite right either. I'll assume 103 is accurate for PCC+bando and the 438 is excluding his bando kotes (that is, using set B he has 460 attack.)

you also didn't account for pdif max and min (cRatio isn't really accurate by itself.) and berserk when up, is +25% attack - I'll calculate both the 'most optimistic maximum' and 'worst case minimum' for both, I'll ignore hasso because I'm not sure if your 103 str reference includes hasso or not. worst case minimum will be his raw attack with no additional buffs (does 460 include food?)

for both case minimums his cRatio doesn't exceed 1.5 so his pdif_min is 1.
case 1 is bando+pcc; case 2 is chiv+myo+1

so you have case 1: max; 75+10 ) * 1.2*(651/327 - 0.35) = ~167. min; 75+10 * (460/327 - 0.35) = ~89.
case 2: max; 75+12 ) * 1.2(632/327 - 0.35) = ~164. min; 75+12 * (445/327 - 0.35) = ~87



so yeah, your approximations are wrong and your friend was right via parsing. (given the real numbers I don't know if I'd still call it negligible like I did earlier.) doing it the way you did isn't accurate, because str adds attack and the calculation you did for the approximation was incorrect (primarily because his attack is actually higher.) and also because (and I admit I'm having trouble following your logic) that I think you solved for the wrong 'x' (you need to do the deltas, and since you have no real unknowns it'd be faster to just do the full equations, as I did above.)

[Wraith]

Ok, you're still not understanding what I did. First of all, the *1.15 for berserk is because you only have berserk active 3/5 of the time. Also, I didn't account for hasso because he told me he usually uses seigan and third eye, and i didn't account for food because he told me he doesn't use food at birds. And you're right about the dex, i did that quickly in my head, lol.

Ok, for what you said about the str, I first calculated his melee damage at 103 str, which is what he normally wears. For the second calculation, I calculated it at 99 str, and the variable x is how much attack it would take to balance the equation to give the same damage as when he had 103 str. Doing this, I can find a direct conversion between str and atk for his situation.

The equations I used are directly from VZX's translation of Studio gobli's information.

[VZX]

I think I did some calculation a while ago how to convert X base damage to y atk

Formula is diff = ?BD/BD - ?atk/atk
if diff > 0, then the increment using base damage is better than the increment using atk.
Conversely, if diff <0 then the increment using atk is better than the increment using base damage.
if diff = 0, then they are equal.

This is all under assumption the increment using atk doesn't push his pDIF to different region

[Amele]

Ok, you're still not understanding what I did. First of all, the *1.15 for berserk is because you only have berserk active 3/5 of the time. Also, I didn't account for hasso because he told me he usually uses seigan and third eye, and i didn't account for food because he told me he doesn't use food at birds. And you're right about the dex, i did that quickly in my head, lol.

Ok, for what you said about the str, I first calculated his melee damage at 103 str, which is what he normally wears. For the second calculation, I calculated it at 99 str, and the variable x is how much attack it would take to balance the equation to give the same damage as when he had 103 str. Doing this, I can find a direct conversion between str and atk for his situation.

The equations I used are directly from VZX's translation of Studio gobli's information.

no, I understand why you used 1.15 and I disagree with the reasoning. you're not doing a true dot calculation to compare subjobs or any other nonsense like that, you're trying to decide which piece is 'better', so look at max and min and consider that you'll be 'high' and 'low' the same amount of time regardless of setup.

I wasn't sure if you had or hadn't counted hasso so I excluded to avoid counting twice, the actual effect on the result is minimal. (hasso actually makes +attack stronger)

what you found was the instantaneous slope of the equation. it's -not- linear w/r to damage. (among other things, you forgot to deduct the effect of losing 3 attack from losing 4 strength)

adding 1 base damage (4 strength) when your base damage is 1 is a bigger deal to your overall damage than adding 1 BD when when your base damage is 100. (in the former case, your damage is doubled, in the latter you gain ~1%) you found a moment but the moment you found is flawed. (in this case, 4str was worth at best ~1.5 damage) and 3 attack is worth ~.85 damage. this is a str:att ratio of 1:1.7, not 2.55. with a 1:1.7 10 str is worth about 17 'attack' not 25. (attack in quotes because this is a poor way to talk about the changes.)

VZX's summary formula bears out the same conclusion. in this case: diff = 2/85 - 15/438 = -0.01 (22 attack > 10 str)

[Max™]

Just a quibble, and it seems a fair spot to ask about this little habit I have.

TP Haste > Acc > Att > Str: UNLESS You can get twice or better of the lesser item.

Kote +1 7 Str (Dex) vs Bando's 22 Att,

7 Str (5ish Att) 7 Dex (5ish Acc) vs 22 Att, if Kote +1 would let you swap haste gear for Acc somewhere, they would win, imo, not on the Str, which is a downgrade to the board of the Bando's 22 Att... but on the 7 Dex, it's acc and +Crit tier aiding. Letting more Acc be dumped elsewhere.

Then you look at Dusk Gloves(Askar) vs kote +1, Dusk has 5 Att, 3% haste, Askar is 4 Acc (5 Dex) 2% haste in this situation, the haste is an upgrade if it doesn't prevent 6 hit, in a lower acc build, or a high dex build which the kote +1 are nice for, the Askar win.

Mind this is all with regards to a 6 hit build using Usu feet ideally.

As a war, Dusk +1 > Askar > Dusk > Are's > Uni +1 with latent > AF2 +1 hands > AF2 > Fourth gauntlets I'd think.

[Amele]

didn't actually see a question, but 2:1 *can* work, although it might be flawed in a few places for common situations.

as a fast example, around 90% accuracy, +2 accuracy (+1% acc) is a dot improvement of about 1.1% ; all examples below assume you do not exceed acc cap as a result of the gear choices in question, and that 90% accuracy is your accuracy without the slot(s) under debate.

1% haste is worth >1.1% when your base haste is in excess of ~11% (this is a trivial amount of haste at 75)

at a more normal amount of ~30% haste (haste spell + 15% in misc. gear/songs) one haste is a dot boost of about 1.4% this is equivalent to about +3 accuracy at 90% (1.6~dot gain)

likewise, 1 acc is worth about +.55% (see above) and to get about a .55% boost to cRatio will take in the area of 2att*current cRatio against a mob with ~300-350 def.

so in cases where your cRatio is already at least 1.5 (the exact cutoff depends on mob, generally 600+ attack vs. say, g. colibri) you'll want 3attack before you switch out 1 accuracy.

so. I think if I wanted a fast and 'sure' answer, I'd use 3:1 instead of 2:1 for haste>acc>att. in the case of strength, 2:1 is too conservative even for 1h users (2str is always a little better than 1 attack, because 2str gives you 1attack+other stuff) but the margin is small enough for 1h that it's an ok measurement. I'd use 4:3 for 2h for similar reasons.

[Max™]

Well, that kind of answered the question that I got distracted in asking, regarding my Haste > Acc > Att > Str "Unless" rule, and yeah 4:3 would apply better for Str now for 2hers due to the benefits towards fStr+Attack.

[Amele]

Well, that kind of answered the question that I got distracted in asking, regarding my Haste > Acc > Att > Str "Unless" rule, and yeah 4:3 would apply better for Str now for 2hers due to the benefits towards fStr+Attack.

it's still conservative, but the more accurate ratio wouldn't be easy to execute as mental math. (and would vary depending on your current attack and current strength) so generally it wouldn't be worth trying to condense into a specific 'rule'.

[Max™]

Well, it's just kind of what I go by when gearing, and why I (and many others) find things like Homam, Haidate, Zerehs, Askar head/hands to be so sexy.