Formula for comparing +haste to +accuracy

[Spider-Dan]

I'm trying to make a spreadsheet that gives me an easy way to determine the following:

At x% haste and y% hitrate, haste+z% is equivalent to Accuracy+n in net hits landed over time (user fills in x+y+z and n is autopopulated)

However, I don't have the mathematical chops to work this out. Any suggestions?

[FailureMidgard]

With y, n as hitrate as a decimal, and x, z as haste as a decimal:

y / (1 - z) = n / (1 - x) is the equilibrium where the 2 are equal.

If you want to solve for n, you have

n = y (1 - x) / (1 - z)

The number you asked for, is 200*(n - y), which would be the delta accuracy needed on gear.

So, if you have 50% haste, 90% hitrate, and you want to figure out the value of 1% haste more:

n = .9 (.5) / (.49) = .9183
200*(.9183-.9) = 3.66 accuracy.

Really though, I think since the addition of Pizza, this is an unnecessary calculation, as you can cap accuracy on virtually any job without sacrificing too much attack, and can go pure haste on gear.

[Spider-Dan]

So, if you have 50% haste, 90% hitrate, and you want to figure out the value of 1% haste more:

n = .9 (.5) / (.49) = .9183
200*(.9183-.9) = 3.66 accuracy.

I presume that's hitrate+3.66, or ~ACC+7, right?

As for pizza, it's not always practical (colibri, ~30min events).

[Kirschy]

I've yet to see conclusive evidence on accuracy to hit rate%. People throw around 2 accuracy = 1% hit rate, but it's only a rough estimate in some cases. There are plenty of tests one could run to contradict the ratio.

The only accuracy tests I'm aware of were Nagamaki's tests from 2+ years ago, and scattered info from translated JP sources. In Naga's case, they were only done on one set of mobs and doesn't account for other factors.

While knowing Haste% vs Hit Rate% would be informative, I think more research into the Acc->Hit Rate% should come first.

If I'm wrong, and this has been proven somewhere, I'll gladly retract my statements and admit my ignorance.

[Mojo]

I'm in concordance with Kirshy on this one. I'm sure that there is at least some kind of level correction function associated with it as the amount of accuracy required to cap your hit rate at lower levels would be unreasonable.

[Kirschy]

There's a level correction function listed on Wiki. But it lists no sources for the formula. When I messed around with accuracy around 50%, the formula and level correction was far off from the estimated amount. I would just like to see more data/experiments showing the validity of Acc->Hit Rate relations.

[Suiram]

Ignore this post, everything I wrote was already mentioned.

[ringthree]

I presume that's hitrate+3.66, or ~ACC+7, right?

As for pizza, it's not always practical (colibri, ~30min events).

Never gear for Colibri. They are the bane of the game.

[Codernaut]

There's a level correction function listed on Wiki. But it lists no sources for the formula. When I messed around with accuracy around 50%, the formula and level correction was far off from the estimated amount. I would just like to see more data/experiments showing the validity of Acc->Hit Rate relations.

I don't have any other sources for accuracy/evasion testing, but the magic accuracy testing which came many months after Naga's post (and included THOUSANDS of casts on sky elementals) showed that relationships between magic accuracy, int/mnd, and magic skill were VERY similar to the ones between melee accuracy, dex, and weapon skill that Naga reported.

I know this doesn't prove the point, but it points in that direction.

I've been watching my parses since those original results and my numbers seem consistent with what he found. I wouldn't mind seeing/doing more tests.


http://robonosto.blogspot.com/2008/1...ist-rates.html

[FailureMidgard]

I presume that's hitrate+3.66, or ~ACC+7, right?

As for pizza, it's not always practical (colibri, ~30min events).

No, that's the accuracy on gear assuming 2acc=1hitrate. The first # is the decimal of the needed hitrate (n). n-y is the delta change on hitrate. Multiplying by 100 is just making it into a percentage number rather than a decimal.

As for the 2acc=1hitrate, /shrug. If someone wants to test it, go for it. The most recent thing I've seen of someone trying to disprove that was a single parse at MJSP, which is a pretty awful way to try to disprove something about accuracy.

[Kirschy]

Unless I'm mistaken, even the Magic Accuracy testing was subject to a lot of debate. A JP blogger posted data that resulted from numerous castings on elementals. Another blogger then took the data and made a bunch of general observations.

While the information was increadibly useful in beginning to understand how mg acc/int/elemental skill/HQ Staves interact, it left a lot open to debate.

Only one monster type @ level 78 was used and for each batch of testing, only 5 Elemental Skill values were used. HQ Staves could be a static +Mg Accuracy or directly added to Mg Acc land rate. 5 values of Elemental Skill is not enough to show if the function is linear with only 1 inflection point. Even dINT was only tested at a few values and could change a lot.

I'm not trying to belittle the experiments or the data that resulted. I think the information is increadibly useful and gigantic step in the right direction toward understanding how our equipment works. I just think it's a bad idea to take what was learned, fit it for all situations, and follow it by faith.

Moar tests need to be done on varying levels of different mobs with a wider range of elemental skill before we can be more confidant in applying a general rule/formula to all situations.

[pchan]

I don't have any other sources for accuracy/evasion testing, but the magic accuracy testing which came many months after Naga's post (and included THOUSANDS of casts on sky elementals) showed that relationships between magic accuracy, int/mnd, and magic skill were VERY similar to the ones between melee accuracy, dex, and weapon skill that Naga reported.

All the tests I've done or seen on magic accuracy say the opposite. Sometimes int does a lot sometimes it does nothing, depending on the value of dint. If stacking int would be the key to get no resists, nothing would resist. So it's not similar to melee stats in any way.

To the OP, the relation haste/hitrate would need to take into account the increase in WS frequency given by accuracy and haste ; for this you would need an accurate knowledge of Ws DMG and melee DMG which will vary a lot by jobs.

[Spider-Dan]

OK, thanks for the help. I was able to get it working.

So according to this formula, Ace's is better than w.turban (completely ignoring fSTR) as long as the +accuracy isn't over the cap? Even with double March/WHM haste/Hasso, if you are at 92% accuracy or below, ACC+7 beats haste+1%.

[Spider-Dan]

To the OP, the relation haste/hitrate would need to take into account the increase in WS frequency given by accuracy and haste ; for this you would need an accuracte knowledge of Ws DMG and melee DMG which will vary a lot by jobs.

No, it would not, and you won't.

The point of this comparison is determining which will provide a greater increase to your net hits landed: adding Accuracy+n or Haste+z%. WS frequency/DMG has nothing to do with it.

[Codernaut]

All the tests I've done or seen on magic accuracy say the opposite. Sometimes int does a lot sometimes it does nothing, depending on the value of dint. If stacking int would be the key to get no resists, nothing would resist. So it's not similar to melee stats in any way.

The blog post I referenced tried to show that int doesn't do much below a 50% unresisted landing rate (pre staff/stats). It does much more once past the 50% threshold.

As a career RDM who's tried to nuke/drain/aspir just about every high level mob I've come across (where I would be in the under the 50% unresisted realm), I find these findings reasonable. I wouldn't complain about more test results though .

[pchan]

The blog post I referenced tried to show that int doesn't do much below a 50% unresisted landing rate (pre staff/stats). It does much more once past the 50% threshold.

So does magic accuracy/skill and HQ staff, they have a worse effect at low rate than high rates. Like 15% for HQ staves at low base rate (30% ish) and more than 30% at high rates.

[Indalecia]

Midgar math weenies unite.

Also, I wanna see an algorithm to explain Zylens drop rate when he's up for loot. It's not natural.

[Codernaut]

So does magic accuracy/skill and HQ staff, they have a worse effect at low rate than high rates. Like 15% for HQ staves at low base rate (30% ish) and more than 30% at high rates.

That's what the results show. Here's Kanican's (he's done a bunch of great testing of other things on his own) summary and thoughts from it: Kanican - Scholar Guide (Part IV)

P.S. Zylen sucks

[FailureMidgard]

To the OP, the relation haste/hitrate would need to take into account the increase in WS frequency given by accuracy and haste ; for this you would need an accurate knowledge of Ws DMG and melee DMG which will vary a lot by jobs.

As usual, you are wrong. Neither accuracy nor haste does anything except net you more landed hits in a timeframe.

OK, thanks for the help. I was able to get it working.

So according to this formula, Ace's is better than w.turban (completely ignoring fSTR) as long as the +accuracy isn't over the cap? Even with double March/WHM haste/Hasso, if you are at 92% accuracy or below, ACC+7 beats haste+1%.

That sounds correct. Sadly, as I said, that's irrelevant. Since the addition of Pizza, you should seriously almost always be at accuracy cap - at least as far as normal situations go. If you are parsing 85-90% accuracy, use a Pizza, use WTurban. Basically, the goal should almost always be to get as much haste as you can, then get as much accuracy as you can in slots that you don't lose any haste (including wturban/blitz), then check your accuracy - if not capped, use pizza, if capped, use meat. Check accuracy again, if still not capped, then it's ok to switch out a blitz or turban for an aces/toreadors. Otherwise, almost always, the pizza/wturban/blitz setup will beat out meat/aces/toreador.

Also, realize that at something like MJSP merits, parsing ~88-90% accuracy is capped accuracy, so this formula isn't exactly correct in a situation like that.

[Spider-Dan]

That sounds correct. Sadly, as I said, that's irrelevant. Since the addition of Pizza, you should seriously almost always be at accuracy cap - at least as far as normal situations go.

Well, again, that's for a pretty selective range of "normal." Essentially, you are talking about situations in which:

a) pizza caps your ACC (i.e. not HNM or dyn-xarc)
b) you are not meriting on colibri
c) you don't mind using 3h food for a ~30min activity (limbus, ein)

If you are in a situation where pizza doesn't cap your ACC, it seems like STR+4/ACC+7 in exchange for haste+1% is a pretty good deal (relative to other +accuracy slots you can trade out).