The Curious Case of the Frost Buff

So if you have read the patch notes concerning mages, you’ll notice a few things.  First, arcane mages are getting nerfed to the ground, with an overall damage reduction of 13.6 to 22% for relevant spells.  Fire mages aren’t getting any specific nerf or buff, although the switch of Scorch to a fire-only spell will operate as a buff as it gives fire mages an extra talent to play with.  (I think it would be very difficult to predict what size this buff would be, but I think it’s significant.)  Frost gets an obvious and major buff, with Frostbolt getting a 32% increase in damage and Fingers of Frost getting a 25% greater proc chance. (The patch notes say that Frostbolt gets a 52% damage increase but loses the ramping buff; according to Ghostcrawler this has been changed.)

To me this is a curious state of affairs. According to Noxxic’s “realistic” DPS rankings, Frost Mages are the number two spec in the game, with only Arcane Mages doing bearer.  (Fire is all the way down at number 21.) Arcane’s nerf is understandable, but why would you buff the number 2 spec in the game so severely?

Well, it turns out that in practice rather than theory, Frost is not doing all that well.  If you look at Raidbots, which I understand to be viewing World of Logs parses, you’ll find a few interesting statistics about mages.  I’m using Feng the Accursed on Normal which seems to be as good a fight as any, but I haven’t noticed significant differences on other fights I’ve looked at:

• Arcane mages account for the second most samples of all DPS.  Their median dps is 65.4k.  The highest DPS recorded for an arcane mage is 151k.
• Fire mages account for slightly less than half of arcane mage parses.  They have a median dps of 62.6k.  The highest DPS recorded for a fire mage is 128.6k.
• Frost mages account for slightly more than half of fire mage parses. They have a median dps of 60.3k. Their highest DPS is 115k.

I chose Feng because of all the fights it seems to be the most “Patchwerky” in style, in the sense that it best illustrates a class’s single-target abilities that are typical in fights.  Some fights differ: fire mages and frost mages both do better on Stone Guardians, for instance, but most fights follow that particular ranking.

A few notes.  Blizzard’s supposed aim of getting people to play the mage spec they want has apparently failed; roughly 60% of mages are arcane mages (and I refuse to believe that there’s such a significant preference for a drinky-bird spec).  But more significantly, practice seems to differ from Noxxic’s theory in a significant way.

Let’s turn from the listing of all parses to the listing of the top 100 parses for given specs.  These should, in theory, weed out the chaff and consider only people who are really good at their spec, the kind of people who, like me, play a spec because they enjoy the play style and have taken the time to research it and get it to perform at its best.  (To be clear: these people are a lot better than me at it.)  Here’s what we get:

• Arcane mage dps ranges from 104k to 151k, with a median of 112k.
• Fire mage dps ranges from 84.6k to 129k, with a median of 89.5k.
• Frost mage dps ranges from 77.1k to 115k, with a median of 83.8k.

Those are some pretty stark differences.  The best frost mage parse on Feng ever recorded would barely scratch the top 50 arcane mage parses. These are people who should, in theory, be able to get the most out of their classes.  I’m sympathetic with the claim that perhaps an average frost mage might be more familiar with PvP than PvE and thus frost mage results would be suppressed on the whole.  But the top frost mage in the world?

Let’s be clear about this: the theoretical dps figures described by Noxxic for frost mages do not align with the observed dps figures listed by Raidbots.  There are two possibilities for this:

1. Frost mages suck and are incapable of playing to the classes true ability as determined by Noxxic.
2. Noxxic’s simulators suck, and frost mages’ true capabilities are more in line with their observed performance.

Let’s consider what I think the best argument in favor of number one would be.  Frost mages may be listed as the number 2 dps spec in the game by Noxxic, but they’re still behind arcane mages. Thus, the kinds of hard core world-best raiders are going to choose arcane over frost. The stark difference in the top spots is because arcane includes the top mages in the world, while frost is limited to people who are on a decidedly lower tier of raiding.  The overall difference is similar as well; most people whose raid leaders use World of Logs will min-max and play the best theoretical spec, and the difference observed between arcane and frost is a self-selection bias.

This argument to me plausibly explains the difference between arcane and frost.  The overall stats show that frost mages, on the whole, do about 8% less dps than arcane, which really is not that huge a difference from the 3.2% Noxxic would predict.

The problem with this theory is that it does not explain frost’s poor performance relative to fire.  Fire mages, according to Noxxic, should do a bit over 15% less dps than frost mages. Yet according to Raidbots they do almost 4% more dps than frost mages on Feng, a fight that doesn’t seem to benefit one spec over the other.  (Personally, I don’t enjoy Feng and I don’t think fire mages should do well at it at all.) And unlike arcane, there’s no real plausible self-selection bias here. There’s no reason to think that raiders who push their dps above other considerations would chose fire over frost, as they would eschew both and go arcane.

This leads me to reject the first premise. I don’t think it’s reasonable to claim that frost mages suck, because a 20% theoretical swing just seems too unlikely. I also think that if frost mages really had the theoretical capability of being within 3% of arcane, we’d see some excellent frost mage somewhere put up a top 10 ranking for mages. But we haven’t, and that suggests that the simulations aren’t quite getting it with respect to frost (and certainly with respect to fire).

Blizzard’s actions support this view. There’s a cynical part of me that entertains the notion that frost is getting a huge buff because they’d like to get a balance between the mage specs. (This conflicts with the equally cynical part of me that entertains the notion that Blizzard has an unhealthy love for arcane.)  For whatever it’s worth, it seems highly unlikely that Blizzard would give what they believed to be the second best dps spec in the game the huge buff that frost mages are getting in 5.2.

That brings into question something else. Assuming Blizzard is acting in good faith here, Noxxic’s simulated spec rankings are highly unlikely to align with Blizzard’s understanding of the capabilities of the class. If what Noxxic says is the second best spec in the game is actually such a poor class that Blizzard believes it necessary to give a 32% damage boost to its primary spell, what does that say about Noxxic’s methodology? Why should anyone trust anything their simulations say?

I’d really like to know why they’ve gotten frost so wrong. (For what it’s worth, they’ve also gotten fire wrong; it’s performed worse than arcane but not as much worse as Noxxic suggests it should.)  But that’s the sort of thing I think I’ll have to leave to a frost mage.  I have suspicions about why their simulations don’t work well for fire mages, but that’s a post for another time.

(Thanks to @_Rades for the conversation prompting this post.)