Magic Accuracy and INT

[Vodou]

I'm more or less confused at the allusion to "scientific method" in some posts.

The term has zero to do with proving to others that you did something by taking a video, screen shots or bring along a pal to watch. Providing methodology and data SUFFICIENT to be repeated is part of the method, not providing proof to a skeptical audience that you carried out the tests. Whether he used a bot program, did it personally or had others do it, is irrelevant to whether there is sufficient information to repeat it. (Unless people think a bot program by it's nature has more INT or Macc than a manually controlled player.)

I'm not sure what part of the OP was so unclear that an objective bystander could not repeat the test. Some people think the sheer number makes it questionable, but that is a personal issue IMO. It seems to me, that anyone that wants to repeat this can do so from the OP. If not then, that is a valid question.

[Tagus]

I'm more or less confused at the allusion to "scientific method" in some posts.

The term has zero to do with proving to others that you did something by taking a video, screen shots or bring along a pal to watch. Providing methodology and data SUFFICIENT to be repeated is part of the method, not providing proof to a skeptical audience that you carried out the tests. Whether he used a bot program, did it personally or had others do it, is irrelevant to whether there is sufficient information to repeat it. (Unless people think a bot program by it's nature has more INT or Macc than a manually controlled player.)

I'm not sure what part of the OP was so unclear that an objective bystander could not repeat the test. Some people think the sheer number makes it questionable, but that is a personal issue IMO. It seems to me, that anyone that wants to repeat this can do so from the OP. If not then, that is a valid question.

One of the purposes of the scientific method in real life is to provide repeatability. In a closed pseudo-environment like ffxi, there is nothing preventing this test from being repeated. It's just unfortunate that this is the only way to verify his results because he is unwilling to talk more about his methodology.

[Vodou]

One of the purposes of the scientific method in real life is to provide repeatability. In a closed pseudo-environment like ffxi, there is nothing preventing this test from being repeated. It's just unfortunate that this is the only way to verify his results because he is unwilling to talk more about his methodology.

Yes, ironically, you say its unfortunate, but...

...there is nothing preventing this test from being repeated...this is the only way to verify his results...

This is the whole point. Verification by a third party IS how its done.

When I read the OP, it was with the mind set: this was his theory, this was his method, these were his results, these were his conclusions, now go test and verify them.

Not: Here are the results, go change your formulae to agree with them.

[Yugl]

One of the purposes of the scientific method in real life is to provide repeatability. In a closed pseudo-environment like ffxi, there is nothing preventing this test from being repeated. It's just unfortunate that this is the only way to verify his results because he is unwilling to talk more about his methodology.

1. What does him explaining his methodology have to do with media? You can use words to explain how you did something (Albeit, it may not be as effective without pictures, but in this case, it's self-explanatory).

2. The degree to which his numbers turned out is one issue, but criticizing his explanation of the methodology is different. Everyone is asking Pchan to explain in more detail about his modus operandi, but I doubt many of you bother to read, in detail, what he wrote.

All you need to do is equip a character similar to his or a character with identical related stats (INT/Skill) and cast bind on Hmpede of the same level as the ones he used. Whether he used or a bot is irrelevant for repetition procedures. Whether the value of Hmpede is correct or not can be adjusted for regardless of whether it is true or not. The degree to which his numbers were precise can be reaffirmed or denied by a repetition of his experiment. The time allocation is covered given the period over which he conducted the trial. Videos, pictures, parse tables, and other associated media are unnecessary and do not prove legitimacy of the experiment.

Finally, while many of you wave the "explain" banner, you do not bother to do explaining yourself. If the current information is not enough, what exactly do you need to reconstruct the experiment?

[Taruking]

We do. If you resist a lot, your magic accuracy is not high enough, that IS how it works. Otherwise, why would SE even let you land threanodies at all ? You don't seem to understand the concept. Your hit rate is necessarily between 5% and 95% (except immunities / change of "cap" values which seem to happen with resist - traits). 5% means you have -90 magic accuracy than the mob or less, 95% means you have 45 more magic accuracy than their magic evasion, or more. Simple. Let's say the max magic accuracy a RDM can get is 400. It's easy for SE to gimp your ability to slow the mob while not making it immune : give him 500 magic evasion. It's easy for players to resist monsters spells : up your own magic evasion with (-> fire resists for instance). You are not happy that Tiamat systematically hurts you with firaga III ? All you need is 90 more fire resist than its magic accuracy. Game mechanics apply to HnMs and we do understand it.

In the same trend : clearly it's easy for SE to make puddings very resistant to silence, flayers very resistant to poison. This is not "special" mechanics.

You still don't get it do you?

You are making assumption again, unless you actually run tests on HNMs you can't be 100% sure the same mechanism also applies to them. The same reason why scientist can't just conduct new drug test on guinea pigs and put them through the market straight away. You can probably argue FFXI is just a game and it's unlikely SE would have created dual standard but it doesn't completely rule out the possibility either.

I didn't really bother to check the details of your methods because they are completely irrelevant. If you are going to spend hours running those tests why waste your time on subjects that have no impact on the current gameplay? Many statistically valid research papers never make it to publication because of the same reason. Pick a good topic before you start your research.

[Callisto]

You still don't get it do you?

You are making assumption again, unless you actually run tests on HNMs you can't be 100% sure the same mechanism also applies to them. The same reason why scientist can't just conduct new drug test on guinea pigs and put them through the market straight away. You can probably argue FFXI is just a game and it's unlikely SE would have created dual standard but it doesn't completely rule out the possibility either.

I didn't really bother to check the details of your methods because they are completely irrelevant. If you are going to spend hours running those tests why waste your time on subjects that have no impact on the current gameplay? Many statistically valid research papers never make it to publication because of the same reason. Pick a good topic before you start your research.

Your analogies are kinda retarded here, but you need to be practical as well. Are you going to ask someone to hold Khim still for 3 hours so you can spam Paralyze on him?

A direct increase to magic evasion on a spell-by-spell basis is actually a pretty solid theory even though it's difficult to back up. People who have dicked around with Death Blossom can provide a fair amount of anecdotal evidence that this can have a direct influence on landing spells that mobs were previously thought immune to just because they were really fucking resistant(i.e. I about shit myself the first time I landed a plain Silence on Suzaku). I know that's nowhere near as useful as actual methodical testing, but it's not a theory that's so far out there to be dismissed entirely.

If his methods/data/shitty english are good enough that we can find a bar to easily measure MEva values and confirm that they can differ noticeably between different spells on the same mobs, that'd be a pretty solid start, even if it is so late in the game that it likely won't lead to anything fruitful.

[Kaeko]

You still don't get it do you?

You are making assumption again, unless you actually run tests on HNMs you can't be 100% sure the same mechanism also applies to them. The same reason why scientist can't just conduct new drug test on guinea pigs and put them through the market straight away. You can probably argue FFXI is just a game and it's unlikely SE would have created dual standard but it doesn't completely rule out the possibility either.

I didn't really bother to check the details of your methods because they are completely irrelevant. If you are going to spend hours running those tests why waste your time on subjects that have no impact on the current gameplay? Many statistically valid research papers never make it to publication because of the same reason. Pick a good topic before you start your research.

I will actually defend Pchan on this issue because I've had to defend my own research against the exact same argument. Every value related to enmity is based 99% of lvl-1 Mandragoras. Everyone uses these values (mostly for HNM fights) but the mob they were tested with is not significant. While I did do a level correction test, I never went out and tested formally on any HNMs.

It comes down to this: it's a leap of faith. You either buy it or you don't. No one is going to conduct a trial of meaningful sample size on Tiamat. It's unreasonable and you will be GM'd, or kicked from your LS. It is reasonable to most people to make this leap of faith. You may be in that minority and don't - in which case that's your choice and you can ignore the data and do your own thing. No experiment is perfect. There are ALWAYS assumptions and conditions to which the experiment crumbles if the assumption is false.

Comparing this to medical research is taking it too far. Medicine and the human body are infinitely more complex than anything this game can provide. It also has lives at stake, which can't be compared to what we're talking about. Not all "studies" are double-blinded gold standard trials - even in medicine. All you have to do is clearly state your assumptions, and importantly, your limitations, and the "study" is useful. It may not be a smoking gun, but it will always still be useful.

[pchan]

People who have dicked around with Death Blossom can provide a fair amount of anecdotal evidence that this can have a direct influence on landing spells that mobs were previously thought immune to just because they were really fucking resistant

I've read this bullshit somewhere. When i did a weak sample sized test on death blossom it gave 5-10 mac increase only, I don't see which NMs you can enfeeble with DB that you can't w/o.

(i.e. I about shit myself the first time I landed a plain Silence on Suzaku).

Try going melee/whm and spam silence on him, you will eventually land it. If it doesn't have a different lower cap of landing rate, then you should be expecting 9.75% landing rate (5% full duration and 4.75% half duration).

If his methods/data/shitty english are good enough that we can find a bar to easily measure MEva values and confirm that they can differ noticeably between different spells on the same mobs, that'd be a pretty solid start, even if it is so late in the game that it likely won't lead to anything fruitful.

Do you play the game ? Can't you give me a mob that is resistant to X spell and not resistant to Y ? This is exactly what having different magic evasions means. What do you think barblind and barwater do?


Try sleeping an imp then binding it. Try poisoning a soulflayer then silenceing it. Try silencing a pudding then nuking with fire. And no imps are not immune to sleep, puddings are not immune to silence, flayers are not immune to poison.

[Taruking]

Your analogies are kinda retarded here, but you need to be practical as well. Are you going to ask someone to hold Khim still for 3 hours so you can spam Paralyze on him?

They are not analogies, anyone who does research regularly should understand those basic concepts. Choosing a useful topic and the right test object are the 2 most important criteria in determining wheter a research is worth publishing.

Obviously holding Khim for 3 hours is not really practical (probably impossible because other LS is likely to call GM on you). One possible way of dealing with situation like this is to collect data on several occasions and pool the results to increase the statistical power in studies with small sample sizes. It's called meta-analysis.

A direct increase to magic evasion on a spell-by-spell basis is actually a pretty solid theory even though it's difficult to back up. People who have dicked around with Death Blossom can provide a fair amount of anecdotal evidence that this can have a direct influence on landing spells that mobs were previously thought immune to just because they were really fucking resistant(i.e. I about shit myself the first time I landed a plain Silence on Suzaku). I know that's nowhere near as useful as actual methodical testing, but it's not a theory that's so far out there to be dismissed entirely

Yes and it will always remain a theory until someone bother to test it.

[Callisto]

@ pchan: Relax there Tiger, I was actually agreeing with you.

[BerenTebogo]

@ pchan: Relax there Tiger, I was actually agreeing with you.

lol'd

[FailureMidgard]

I went out and did a more extensive test, and for what it's worth, every data point I got that conflicted works if it's assumed that for dINT < 0, the value of m is .5 [ in D = V + (dINT * M) ]. My apologies if this was posted somewhere, I found no such information on either wiki/etc.

There are other solutions that work as well I believe, but with the other assumed information (I have no idea what testing went into coming up with these numbers), I'd say 58 may be the correct int value for a lvl 75 goldfish.

[Kaeko]

I'll attempt to continue posting in this topic, trying to be as unbiased and drama free as possible. If I get responses that end up getting the OP flamed, I will stop. Round 2, let's hope it goes better.

I'll preface my criticism but AGAIN stating that I am not trying to turn these into personal attacks. If said OP wants to take them that way, then I am sorry; however, any "new findings" need to be bombarded with criticism to see if it passes muster. You don't get a free pass - especially if you are attempting to establish something new and against the "norm". The entire point of posting is to defend your conclusions.

*********

(1) Regarding "Faking Data"

There is no way for me to definitively prove he faked it. There is no way for him to prove it's real since as mentioned before - super easy to shop, etc. Considering this, and the fact that BG is a popularity contest, I'm going to avoid talking about faked data again (though I still firmly believe my position); instead I'll word it a bit more nicely as "questions regarding your methodology". How about that?

(2) The 1-(1-p)^2 Conversion

Lodeguy never specifically mentions that this geometric distribution holds true for 1/4th and 1/8th resists. The first mention of this comes from Robonosto here. In it, he mentions that the data "seems to fit" and that it is the most logical way to program this phenomenon; however, it is never statistically been shown to be true. Until it is, this is something we must keep in mind as all the percentages used in the "raw data" are based on this formula which "seems to fit", but lacks statistical proof. Again, it does not negate the data set - it is just something to keep in mind.

(3) The BLM75/RNG1 Gear Set

Which gear set did you use for these tests (Tests1-5)? Very simple question. You posted your PUP/RDM set. All I am asking is to post the BLM one. No ill-will, no secret backhand attempt to make you look bad. Just want the gear set.

(4) Test 1 (your control)

Your "control" test is Test1, where you hit 49.99%, the exact established inflection point of MACC. This alone could have been luck; however, I believe this was derived. I say this because if we look at your base stats (246skill, 0MACC, 55INT), we'll notice that the INT you used is abnormally low, even for an ELV lvl1 sub. Using the FFXI stat calculator, and plugging in 75 ELV BLM75/RNG1, we get base INT of 58 non-merited.

This value is VERY nice for 2 reasons - (1) it's the base INT of the mob you are using, making dINT=0, a nice control value, and (2) it's your naked INT. However, you chose to purposely lower your INT by 3 to 55 (or more if you are merited in INT). Why did you do this? Coincidence or did you "guess" or "derive" the INT value that would grant you this ~50% "hit rate" and adjust your gear accordingly? If the latter is true, then test1 was not the actual first test conducted. In which case, which is?

Again, whatever your answer is, it will not negate your data. All I want is the THOUGHT PROCESS you followed. Were you lucked into finding the inflection point, or did you somehow derive it from a separate test and manually adjust your INT by -3?

I'll just start there. Easy questions imo. If you choose not to respond to me meaningfully, I will respect that and leave this thread.

[BerenTebogo]

I'll attempt to continue posting in this topic, trying to be as unbiased and drama free as possible. If I get responses that end up getting the OP flamed, I will stop. Round 2, let's hope it goes better.

You could post pictures of Pchan having a three-some with Jenna Haze and Kayden Kross and he'd still get flamed on improper technique. There's only so much you can do.

[Callisto]

That'd make for a pretty sweet thread at least.

[BerenTebogo]

Agreed. Let's /r/ some media on that.

[pchan]

That's why I only test INT with dint=0 (obviously not doable with too low INT). The formulae may be inaccuracte for negative dint, or maybe it's just due to being too low on dint ; we know that weird things will happen for high dint, "high" depending on the spell, as far as damage go. It's not surprising if you look at stone I for instance, because the formula

floor( (10+dint)*1.24 )

can bring negative values for dint < -10.

[Yugl]

I might have missed it, but why did you decide to start with dINT=(-3)? Did you just happen to have gear on at the time that reduced your INT or was there a reason for you picking that?

[FailureMidgard]

first tests are done on BL75/RNG1

Test1:skill=246,mac=0,int=55:1865/7457=49.99%@[49.02,50.98]
Test2:skill=246,mac=0,int=98:287/7418=80.33%@[79.24,81.48]
Test3:skill=246,mac=0,int=68:711/5054=62.49%@[61.24,63.79]
Test4:skill=246,mac=0,int=73:482/4896=68.62%@[67.32,69.98]
Test5:skill=246,mac=0,int=78:462/5639=71.38%@[70.15,72.66]

last tests are done on PUP75/RDM37

Test6:skill=142,mac=8,int=73:3024/3829=11.13%@[10.41,11.86]
Test7:skill=142,mac=8,int=67:3605/4265=8.06%@[7.47,8.65]
Test8:skill=142,mac=8,int=81:4709/6348=13.87%@[13.25,14.5]

interpretation

Tests 6,7 and 8 confirm that this relation still happens under 50% hit rate, which I've seen contested on various forums (by idiots).

Forgive me if I'm missing something.
Going from Test 7 to Test 6, you gain 6 int, and only gain ~3% macc while under dINT of 15 when gaining 6 int. In fact, if you take the lowest part of your interval in 7, and highest part in 6, a gain of 6% accuracy is not even possible within your confidence intervals. Compare this to test 3>4 where you saw a gain of 6% accuracy with +5 int. May I ask why you are calling people idiots for saying that the relationship doesn't happen under 50% hit rate? It looks to me like there's no way 1int=1macc at that low, unless your test result in 8% was floored, but if that's the case, you really didn't prove anything at all in that test...
Past that, the difference going from Test 6 to Test 8 (where you definitely are not at the floor) where you have dINT>15 is not really giving 1macc per 2 int. Looking at the confidence intervals, you would have to have results at the extreme ends to match it.

Finally, did I miss some testing that shows dINT being the sole cause of any inflection point, and not some more complex formula including magic accuracy/land rate as a whole? Because I don't see the same conclusion from your data that you are even getting at...though I may be missing something, and I'll admit I'm not well versed in the formulas for offensive magic in this game.

[rog]

Forgive me if I'm missing something.
Going from Test 7 to Test 6, you gain 6 int, and only gain ~3% macc while under dINT of 15 when gaining 6 int. In fact, if you take the lowest part of your interval in 7, and highest part in 6, a gain of 6% accuracy is not even possible within your confidence intervals. Compare this to test 3>4 where you saw a gain of 6% accuracy with +5 int. May I ask why you are calling people idiots for saying that the relationship doesn't happen under 50% hit rate? It looks to me like there's no way 1int=1macc at that low, unless your test result in 8% was floored, but if that's the case, you really didn't prove anything at all in that test...

I thought you just need 2 macc to raise land rate by 1% when under 50% total land rate? 3% land rate = 6 macc, unless i'm missing somehting.