Bear stat weights - Mr. Robot wants your input

[Zoopercat]

Warning, math and theorycrafting below...

So it turns out that coming up with a good set of default bear stat weights is really hard! About the hardest of all the classes. The Mr. Robot team has been researching all week to get a good set for bears, and just released an update last night. Yellowsix posted on the Mr. Robot forums this morning, outlining the logic behind the most recent set of stat weights. I'm looking for a good discussion around these new weights to see what the bear community thinks of them.

Edit: removed post duplicate to point to the original one over to Mr. Robot's forums so we can have a discussion over there that crosses multiple communities (MMO-Champion, EJ, etc) and update math and we continue to theorycraft. http://forums.askmrrobot.com/index.p...c=937.msg10523

[Themoonstar]

- In general, as you get more stats, savage defense gets better. As you get more stats, dodge gets worse.

lolwut? how does dodge get worst? Im not very smart with my bear but Im pretty sure no matter what taking 0 damage is better than absorbing only some.

[Ciah]

lolwut? how does dodge get worst? Im not very smart with my bear but Im pretty sure no matter what taking 0 damage is better than absorbing only some.

This is a common misconception.
Sometimes taking MORE damage, but more CONSISTENTLY is actually better. Saves your healers mana by reducing their need to use their "clutch" expensive heals. And really, reducing healer mana usage is the goal in the end, at least once you have the living through the encounter part down.

This occurs with the very small gains in dodge% are outweighed by larger gains in consistency by reforging to other things. Value scaling issues and all that.

[Arielle]

Ohhhhhhhh boy. Where do I even start.

- Mr. Robot really doesn't like reforging every item to dodge as a bear tank. Sure, at some gear levels, it'll get you an extra 2-4% dodge, never much more. BUT, reforging your DPS stats to dodge DEVALUES all of your other stats: reducing crit, hit, expertise, mastery, and haste makes all of your: attack power, agility, hit, expertise, crit, mastery, and haste less valuable. And that dodge suffers from DR as well.

While true, the mitigation gained form Dodge in the form of both Avoidance and SD uptime is greater than any possible loss from reforging away from DPS stats. Additionally, nobody cares about Haste.

- In general, as you get more stats, savage defense gets better. As you get more stats, dodge gets worse.

I believe you meant to say "Dodge Rating", not "Dodge".

- Lastly, the utility of using DPS stats to double as avoidance stats (crit and mastery are the best choices), you turn into the awesomest hybrid in the game -- the Ninja BearCat of Death. Would you sacrifice 2% dodge to become a Ninja BearCat of Death? Mr. Robot would.

Then Mister Robot clearly doesn't value maximizing his mitigation as much as he should.

and has no direct relationship to defensive stats, like armor or dodge

It does have a direct relationship to Dodge. Say you proc an SD shield. That SD shield will be eaten by the next incoming hit. If you dodge said hit, it will then be eaten by the one after that, thereby allowing you to mitigate 2 hits for the price of one shield.

2. Savage defense scales with: attack power, mastery, crit, expertise, hit, haste, agility, strength, and technically stamina, but the latter has been ignored. Yeah, that's a lot of stats.

You can't ignore Vengeance when attempting to produce an accurate model of Savage Defense. Sorry.

1. At some point (when you have enough crit, mastery, and attack power, for the most part), mastery will become better than dodge. Unless my math is totally broken. This point is obtainable with current gear.

Your math is totally broken, and it is not attainable with current gear.

2. Despite common belief, armor is not that much better for bear druids than any other tank class. Sure, putting on a higher item level tank item will generally be better because it has more armor, but it'll probably be better anyway because it has more of everything else. Armor on enchants, trinkets, and trinket procs do not get any of the big bear armor multipliers, and there is nothing magical that makes armor itself give bears more damage reduction than any other tank.

/facepalm. Yes, yes it is. Our armor modifiers make our total armor value massively exceed that of other tanks. Armor was our most valuable mitigation stat prior to 4.0.6, and remains so afterwards.

Bonus armor has gone down in value, if anything.

This is the meat of the derivation of our stat weights for bears. To equate mastery to the other tanking stats, you need to know how hard you are getting hit. For normal-mode raid content, let's assume about 80k melee hits (before reductions). After bear form damage reduction and ~50k armor, that's a ~28k hit.

Except, as previously mentioned, you didn't include Bearform reduction. You only included Demo Roar. Also 50k armor is wayyyyyyy over shooting it.

If you assume around 25,000 AP with vengeance stacked up in a raid, the base absorb is:
.35 * 25,000 = 8750

You are massively understating the amount of AP we get from Vengeance. And I mean MASSIVELY.

- low-ball a little (maybe not enough rage, some attacks can't trigger SD): 27 melee, 53 specials per minute = 80 total

The only attack that can't trigger SD is FFF. And you shouldn't be using that single target anyway. So the 53 number is wrong. Just use 60.

- assume we have no hit/exp, 8+6.5+15 = 29.5% chance to miss.

This is incorrect. Parry is only 14%, not 15%. That makes most of the rest of your math wrong.

Also, you might want to consider comparing your spreadsheet to Tangedyn's to find any other errors I've missed - http://theincbear.com/bear-mitigation-spreadsheet/

[HarleyM]

Not to mention gemming stam, as much as I despise it, is sometimes necessary. I said it before and I will say it again: you cannot use a strict stat weight system and still get a good model. At least not without instituting more soft cap changeover values. and even then it won't be as accurate as more advanced models that already exist. You are going to run into issues getting help for one main reason: the people that care enough about gearing and are capable of doing the math don't want an inaccurate model out there. When good programs already exist like the aforementioned spreadsheet and rawr, why make it more attractive for people to get the wrong answer rather than pointing them to better tools and helping develop those?

[Yellowfive]

Hi all -- I'm the original author of that post over at the Mr. Robot forums, and also the main developer for the website, as well as all of the tanking stat weights. I realize that the math is definitely rough around the edges, and I am looking for some constructive feedback to improve our site!

Before getting started, keep this in mind throughout: the purpose here is to choose gear, and thus the most important thing is the relative value of stats, not the absolute value. Also, the relative value of stats changes for a bear by quite a bit as your gear improves, so an approximation is necessary.

The overall goal was to find a decent relationship between: attack power, crit, expertise, hit, mastery, and haste. Then, to determine how these all relate to dodge, as they all indirectly provide damage avoidance through SD like dodge. The rough theory was that over obtainable levels of gear, the relative value of stats like mastery, AP, and crit won't change too much, but that their relative value to dodge would change quite a bit. Thus, I focused mainly on relating all of the stats to dodge.

A few clarifications:

1. When I say that "dodge gets worse as you get more stats," I mean the following: as your gear improves, at the least, you get more agility, and so your dodge from gear goes up quite a bit. As you get more dodge on your gear, diminishing returns reduces how much actual dodge you get. For example (using fake numbers to avoid complex math at the moment), if you have a 30% chance to dodge, and you add 176 dodge rating to your gear, you do NOT gain 1% chance to dodge. You gain more like 0.8% chance to dodge. So at that gear level, adding more dodge rating on gear is only 80% as valuable as if you were naked and equipped your very first piece of tanking gear.

2. When I say that "Savage Defense gets better as you get more stats," I mean this: if you have more offensive stats (of any kind), your shields are bigger and you get them more often. Pretty simple.


I'm going to respond to a few of Katarn's criticisms here:

It [savage defense] does have a direct relationship to Dodge. Say you proc an SD shield. That SD shield will be eaten by the next incoming hit. If you dodge said hit, it will then be eaten by the one after that, thereby allowing you to mitigate 2 hits for the price of one shield.

That is a very odd and confusing way to say things. I think that what Katarn is trying to get at, is if you avoid more attacks, it effectively increases the average time between hits that actually do damage to you, and thus one could say that "as my avoidance level increases, my chance of absorbing an attack with savage defense increases."

That is a very good point, one that I did not account for. As far as stat weights go, as a player's total avoidance rate (dodge+miss) improves, the value of savage defense also improves. Note that this would only make savage defense better than I am currently estimating, not worse. I'll look into it.

While this is a good point, Katarn was also misunderstanding my original statement. What I was really getting at was that both the frequency and amount of a savage defense absorb are mainly dependent on your own offensive stats, rather than how hard you are getting hit. e.g. a block is 30% of the incoming damage, an SD shield is a percentage of your own attack power. There is no direct relationship between how much AP a tank has and how hard a boss hits -- this complicates things, and requires that we make extra assumptions to come up with meaningful relationships between offensive stats and defensive stats for bears.


With regards to armor not being that much more valuable for a bear:

/facepalm. Yes, yes it is. Our armor modifiers make our total armor value massively exceed that of other tanks. Armor was our most valuable mitigation stat prior to 4.0.6, and remains so afterwards.

Bonus armor has gone down in value, if anything.

Once again, Katarn is being a little obtuse. Yes, bears end up with more armor than other tanks (my druid has about 48k armor in good pre-raid/entry T11 gear, whereas a warrior or paladin would be about 10k less or so). But there are only a couple ways to get more armor: the armor straight on your items, "bonus" armor (e.g. enchants, trinket procs), and buffs (devo aura). The armor straight on e.g. a piece of leather is pretty irrelevant for choosing gear. You just want a stat weight that will favor picking higher ilvl gear (e.g. more armor) pretty heavily, regardless of which stats are on it. So as far as actual gear choices that concern armor, we have: enchants, armor meta, trinkets, trinket procs. NONE of those things get our bear multipliers (edit: except armor meta, but that's a no-brainer to get anyway). Thus, they are just as good for bears as any other tank class: all tanks have the exact same damage reduction formula for armor. Katarn is very confused here.


And a few other random things:
1. I used 25,000 AP as a low-end estimate. The actual stat weights are calculated at 30,000 AP, which is pretty reasonable for a fully-buffed ilvl350ish tank with 150k hp, max vengeance.
2. I'll change parry from 15% to 14% -- recall that this math is an approximation to get relative stat weights: this change will have little or no impact on the outcome.
3. You are right, I did not include both demo roar and bear form reduction when finding the final damage. Note that this doesn't matter much though: I'm just taking a guess at how hard bosses hit anyway, and I'm sure that different bosses hit harder or not as hard. Also note that by excluding demo roar, it is causing me to underestimate savage defense (SD is less valuable relative to dodge if a boss hits harder). Note a trend here? I am being conservative because I wanted to ease people into the idea that mastery is not as undervalued a stat for bear tanks as they think.
4. Yeah, I used a low-end estimate for number of special attacks per minute. I did this because there are plenty of situations where a tank needs to do other things, maybe use an interrupt, a taunt, move around, whatever. Note that once again this will cause my estimated for SD to be conservative.


I will take your suggestions and do some updates: if I do, savage defense will come out to be a little better than I am currently estimating. I will also try to put together a simple example of a set of player stats where adding 100 mastery rating is worth more than adding 100 dodge rating, and yes, it is within the realm of obtainable gear.

Hopefully that clarifies things a bit -- I know this is a bit rough, but I think that it comes up with decent relative stat weights. Feedback is always appreciated, though Mr. Robot would prefer a little less scorn than Katarn is bringing

---------- Post added 2011-02-17 at 01:23 PM ----------

HarleyM -- I agree that in some cases you will want more total health to deal with a special circumstance. But for most cases in cataclysm, it's not necessary to stack hp. 150k hp is more than sufficient for most normal mode fights.

That said, it's really easy to bump the stat weight on stamina up on Mr. Robot -- then it'll gem/enchant your gear for stamina, and probably recommend some stamina trinkets -- BAM, problem solved in 15 seconds!

Also, you say the following HarleyM, (paraphrasing) "Why bother trying to make Mr. Robot good for bears, when there are spreadsheets and other complex programs that are already good for bears?" The answer is simple: we want a good, accurate tool for choosing bear gear that is easy and fast to use. All other existing tools are seriously lacking in the user experience department.

Anyway -- we would love feedback, and are always open to suggestions for improvement. I'll be monitoring things closely, and tweaking the bear portion of our website to hopefully make it into a tool that all feral druids will feel confident using.

[Mystile]

The only attack that can't trigger SD is FFF. And you shouldn't be using that single target anyway. So the 53 number is wrong. Just use 60.

Faerie Fire (Feral) and Demoralizing Roar, sir. Other than that, I agree with most of your points. There is far too much assumption in these values and you cannot ignore stamina:ap gains for savage defense. I notice quite a large difference between a max vengeance savage defense for 28k compared to the typical 17k.

[Arielle]

1. I used 25,000 AP as a low-end estimate. The actual stat weights are calculated at 30,000 AP, which is pretty reasonable for a fully-buffed ilvl350ish tank with 150k hp, max vengeance.

Incorrect. Said tank will end up with ((39533 * 0.1) + (150,000 / 14)) * 1.1 * 1.25 = 20167 AP from a full Vengeance stack (for a night elf). And since I am over 10k AP in Bear form with 0 buffs and no trinket stacks, it's pretty safe to assume that this number is too low.

3. You are right, I did not include both demo roar and bear form reduction when finding the final damage. Note that this doesn't matter much though: I'm just taking a guess at how hard bosses hit anyway, and I'm sure that different bosses hit harder or not as hard. Also note that by excluding demo roar, it is causing me to underestimate savage defense (SD is less valuable relative to dodge if a boss hits harder). Note a trend here? I am being conservative because I wanted to ease people into the idea that mastery is not as undervalued a stat for bear tanks as they think.

Even if you only wanted to include the Bearform reduction, it's 12%, not 10%. Moreover, I can't even find a raid boss that only hits me for 28k even after Demo Roar and Bearform reduction. Your initial hit needs to go up to 120,000 unmitigated. Bare minimum. 130,000 -> 140,000 would be more accurate.

4. Yeah, I used a low-end estimate for number of special attacks per minute. I did this because there are plenty of situations where a tank needs to do other things, maybe use an interrupt, a taunt, move around, whatever. Note that once again this will cause my estimated for SD to be conservative.

Interrupt is off the GCD. If you're taunting the boss isn't attacking you. You can attack while moving.

And finally, you cannot produce stat-weights in a vacuum. They will always change relative to whatever gear you are currently wearing.

[Aukas]

I give this a HUGE thumbs down. Inaccurate crap.

[Arielle]

I give this a HUGE thumbs down. Inaccurate crap.

There's nothing wrong with trying to produce a model that helps players select the right pieces of gear, as long as you do it correctly. I'm just trying to help him correct some of the inaccuracies in his model.

[Aukas]

Probably shouldn't have read the initial post the just scrolled down and added my 2 cents without reading anything else ahaha

[Mithfin]

2. Savage defense scales with: attack power, mastery, crit, expertise, hit, haste, agility, strength, and technically stamina, but the latter has been ignored. Yeah, that's a lot of stats.

Eh? How exactly SD scales with hit and exp in the world of one-roll system? Enlighten me please, will you?

[Mystile]

Eh? How exactly SD scales with hit and exp in the world of one-roll system? Enlighten me please, will you?

Probably because it's a two roll system.

[Yellowfive]

Well, I got what I was looking for from this thread -- some links to a few spreadsheets that look pretty solid. I'll use those to tweak the numbers used on the Mr. Robot site, and see how things come out. Thanks for the links.

I'm almost done converting this into a simulator rather than screwing around with a spreadsheet. I'll update all the numbers for vengeance and blah blah based on the referenced spreadsheets, then run a few million combinations of stats through it. That should definitively tell me how valuable e.g. 100 mastery is in terms of damage reduction at any available combination of stats -- same idea for any other stat.

From there, it's a game of coming up with a set of stat weights that works reasonably well across the spectrum of available gear in the 346-372 level. That is always going to be a minor approximation, but my intuition is that there is a set of weights that will work well for all bears across that spectrum. We'll see!

[Mithfin]

Hm. Well, ok. It's one-roll for white hits and two-roll for yellow. Thanks for pointing this out.

[HarleyM]

1. When I say that "dodge gets worse as you get more stats," I mean the following: as your gear improves, at the least, you get more agility, and so your dodge from gear goes up quite a bit. As you get more dodge on your gear, diminishing returns reduces how much actual dodge you get. For example (using fake numbers to avoid complex math at the moment), if you have a 30% chance to dodge, and you add 176 dodge rating to your gear, you do NOT gain 1% chance to dodge. You gain more like 0.8% chance to dodge. So at that gear level, adding more dodge rating on gear is only 80% as valuable as if you were naked and equipped your very first piece of tanking gear.

Check your math. Using your numbers, if you started with a 30% chance to dodge, and you increase your dodge by .8%, you are eliminating .8/70 = 1.14% of the incoming damage. Therefore, dodge at that point is BETTER than when you were standing naked, by about 14%.

HarleyM -- I agree that in some cases you will want more total health to deal with a special circumstance. But for most cases in cataclysm, it's not necessary to stack hp. 150k hp is more than sufficient for most normal mode fights.

That said, it's really easy to bump the stat weight on stamina up on Mr. Robot -- then it'll gem/enchant your gear for stamina, and probably recommend some stamina trinkets -- BAM, problem solved in 15 seconds!

Also, you say the following HarleyM, (paraphrasing) "Why bother trying to make Mr. Robot good for bears, when there are spreadsheets and other complex programs that are already good for bears?" The answer is simple: we want a good, accurate tool for choosing bear gear that is easy and fast to use. All other existing tools are seriously lacking in the user experience department.

I realize this is a long thing to quote, but they go together to make my point. I agree that Mr. Robot is easier and faster than the other tools out there. That is what makes it so dangerous. Right now, it is NOT a good tool from an accuracy perspective. The reason I have no desire to help it succeed is because it is married to the stat weight idea, which is fundamentally flawed. You say I can bump up the stat weight of stamina and BOOM, problem solved, but it isn't, because stamina isn't ALWAYS better than any other stat. You need it to eliminate the chance of being killed before it is possible for healers to kill you, then it is less valuable up until the point where the healer would have to be dead to not heal you in time. You make some very valid points, but you seem satisfied with a product that is very unfinished, and that bothers me. Especially since so many people will say 'ooh math, he knows what he is talking about.' If you provided a disclaimer it would make me much happier.

So long as you are using stat weights you will have no chance of every bear being confident using it.

[Barawr]

Unless you have a manner in which to code Mr. Robot so that the stat weights are dynamic it will still be horribly inaccurate in comparison to the other tools currently available. Static stats have never and will never work well.

[Deleted]

So you want to make a website for l33t bears?

Here is feedback for you: Nobody cares about theorycrafting for softmode content. Why? Because said content is clearable no matter what you do with your gear. You could enchant spellpower and gem for pure haste and still tank everything including Nefarian fine, as long as you know when to use cooldowns.

My advice:

1- before calling people with real pve experience obtuse because you know how to use excel, get over yourself a little, and listen to Katarn, he has a lot more to offer you than you might think.
2- focus on what matters, which is telling guys like us what is the best set up for tanking massive incoming damage that threatens your life every second global during a fight. As of now, all the maths in the world don't stand in front of "gear for time to death, which means HP, end of story".

[Yellowfive]

Check your math. Using your numbers, if you started with a 30% chance to dodge, and you increase your dodge by .8%, you are eliminating .8/70 = 1.14% of the incoming damage. Therefore, dodge at that point is BETTER than when you were standing naked, by about 14%.

This statement is very wrong. Use some round numbers as an example:

You get hit 100 times in a fight for 100 damage after all is said and done.

If you had no dodge from gear, you would have:
~15% dodge (base 5% dodge, 10% from talents, this does not suffer DR).
You would avoid: 1500 damage

Say you add 10% dodge "from gear" (i.e. subject to diminishing returns -- dodge from agility and doge rating on gear).
After applying the DR formula, your final dodge chance is: 24.46%
You avoid 2446 total damage.
Your 10% dodge from gear has avoided 946 extra damage

Now say you add another 10% dodge from gear for a total of 20% from gear.
After applying the DR formula, your final dodge chance is: 32.50%
You avoid 3250 damage.
Your 20% dodge from gear has avoided 1750 extra damage.

If dodge became MORE valuable, as you went to 20% dodge from gear, you would have had to avoid more than 946*2 extra damage, or 1892. But you didn't -- you only avoided 1750 extra damage. Thus, as you add more dodge, it becomes LESS valuable due to the diminishing returns formula.

Barawr and Nodata:

I am almost done creating a simulator that will quickly and accurately determine the relative value of stats for bear druid damage mitigation at any gear level. Its accuracy will surpass that of any available spreadsheet, because it is very difficult to come up with a formula for this -- instead I'm taking the approach of doing statistical analysis on a large number of trials of a boss actually hitting a bear druid. It dynamically models a boss's attacks, vengeance, savage defense, your relevant offensive attacks, armor, etc. -- with plenty of knobs that you can tweak.

I'll provide a link when it is completed in the next day or two -- it will be integrated into the powerful and simple gear optimization UI of the Mr. Robot site. We are committed to the accuracy of our tool, and we will make whatever changes are necessary to make the tool the best available for bear druids of any gear level -- that's why we asked for feedback in the first place! You guys need to think out of the box a little... a lot of you see "stat-weight based tool" and then you completely shut down and go into reflexive "ur bad" shock. I'm going to show you that our simple yet powerful site can give you the best of both worlds: the ease and speed of stat weights, backed by the accuracy of a dynamic model.

[Arielle]

Its accuracy will surpass that of any available spreadsheet, because it is very difficult to come up with a formula for this

I sincerely doubt this, since Fasc and Tangedyn have been doing this for over 3 months now. But I'll wait for the finished product.

I'll provide a link when it is completed in the next day or two -- it will be integrated into the powerful and simple gear optimization UI of the Mr. Robot site. We are committed to the accuracy of our tool, and we will make whatever changes are necessary to make the tool the best available for bear druids of any gear level -- that's why we asked for feedback in the first place! You guys need to think out of the box a little... a lot of you see "stat-weight based tool" and then you completely shut down and go into reflexive "ur bad" shock. I'm going to show you that our simple yet powerful site can give you the best of both worlds: the ease and speed of stat weights, backed by the accuracy of a dynamic model.

We're more concerened about the "weightings" you showed in your math evaluating gear increases with other constant stats. Which isn't accurate. Those "weights" are only applicable to that one static set of stats showed in the math/spreadsheet you linked. They are completely inaccurate in any other context.