Comps and counters in 2v2 arena

The strong compositional character of 2v2 arena makes for many heated arguments around comp strengths and counters. Due to the lack of large-scale arena data, such discussions are necessarily anecdotal and often based on perceptions of strength as much as actual strength. Recently, we've been able to gather a fair amount of matchup data from 2v2. This now allows us to develop an overview of the comps and counters in 2v2 in Season 1 of the Shadowlands expansion.

Comp overview

This is a view of the data that can be found on the 2v2 comps page. This page provides a regularly updated view of the most played 2v2 comps for each spec. At the time of writing, our dataset contains 135 000 unique 2v2 games.

Rank Comp Number of matches % of matches
Fire MageFire Mage
Subtlety RogueSubtlety Rogue
3030 10.5
Arms WarriorArms Warrior
Holy PaladinHoly Paladin
2223 7.7
Holy PaladinHoly Paladin
Windwalker MonkWindwalker Monk
1632 5.7
Discipline PriestDiscipline Priest
Windwalker MonkWindwalker Monk
1503 5.2
Arms WarriorArms Warrior
Restoration ShamanRestoration Shaman
1477 5.1
Discipline PriestDiscipline Priest
Subtlety RogueSubtlety Rogue
1191 4.1
Discipline PriestDiscipline Priest
Retribution PaladinRetribution Paladin
943 3.3

Fire Mage/Subtlety Rogue is the most common comp, with both classes enjoying huge burst in their respective burst windows. Mage/Rogue has been a staple of the arena since its inception, and so it is no surprise that they are the most popular comp on the ladder.

Both Arms Warriors and Windwalker Monks make up one part of a number of the most popular comps, with Holy Paladins, Discipline Priests, and, to a lesser extent, Restoration Shamans providing the heals.

A note on comp strength and the definition of a counter

Comp strength shares many of the properties of spec strength: it's not a trivial thing to calculate, and you rely on very good data to do so. A major complication is that popularity and strength can easily get conflated. A comp that is played by a tonne of people may not actually be that strong (for whatever reason). With specs we see this phenomenon very clearly with Marksmanship Hunters: MM Hunter is one of the most popular specs in PvP at the moment, but they are also severely underrepresented on the leaderboards across all brackets. On the other hand, Elemental Shaman is a fairly unpopular spec, but is by all accounts one of the strongest casters in the game at the moment. In the present article, we focus on counters rather than comp strength; this is because calculating a counter is far more tractable than calculating comp strength.

So how do we define a counter? Intuitively, this is just the win rate of one comp against another. For example, if two teams have the same rating, and Comp A beats Comp B 64% of the time, Comp A can be said to be a counter to Comp B. There are, of course, degrees to this, and an alternative way of expressing this is in terms of the rating advantage. If Comp A beats Comp B 64% of the time, we can equivalently say that Comp A has a rating advantage of 100. Having a rating advantage of 100 means that Comp A at 1600 rating would expect to beat Comp B of 1700 rating 50% of the time. The nice thing about the Elo rating system is that this is true for all rating values (in fact, this is how rating is defined mathematically). Note that is still possible for a counter to be stronger at a certain rating. When I have looked at this previously, I have checked the top counters for certain matchups above and below 1800 rating. These remain largely consistent, and so do the magnitude of the counters. That being said, approximately 2/3 of the games analysed here are in the 1400-1900 range (with an average rating of around 1650). Thus, this analysis likely applies to the majority of WoW players, but may not hold true at the upper echelons of the 2v2 ladder.

Now that we've stepped through some of the groundwork, let's have a look at what actually counters what in 2v2. For these analyses, I only consider matchups where we have at least 100 games.

Fire Mage/Subtlety Rogue

The most obvious set of compositions that Fire Mage/Subtlety Rogue counters is Windwalker Monk/Healer. This is consistent across all WW/Healer comps, with each comp suffering a 50-100 rating disadvantage against Fire Mage/Sub Rogue. This means that a 1500-rated Fire Mage/Sub Rogue would have a 50% win chance against a 1600-rated Windwalker/Resto Shaman. Windwalker Monks have some of the highest single-target damage in the game, with frequently occurring burst windows, and very high mobility. A weakness of Windwalker Monks, however, is that they are relatively squishy, and particularly vulnerable to getting blown up in a stun. Of course, this is exactly the modus operandi of Mage/Rogue.

The strongest counters to Mage/Rogue are Holy Paladin/Marksmanship Hunter, Discipline Priest/Survival Hunter, and Discipline Priest/Frost Death Knight. In fact, these three comps are three of the four strongest counters in the game, with each comp having an estimated 150 rating advantage on Fire Mage/Subtlety Rogue. Frost DKs provide a very strong toolkit to deal with Mage/Rogue. Icebound Fortitude gets you out of stuns, Anti-Magic Shell and Zone allow you to effectively deal with Combustion, Wraithwalk ensures that you can chase the Rogue down to prevent resets, and Chains of Ice keeps everyone at geriatric running speeds.

Arms Warrior/Restoration Shaman

A pretty consistent results across all Arms Warrior comps is that they counter Subtlety Rogue/Healer fairly hard. Arms Warrior/Restoration Shaman is no different. Rogue/Healer is not a particularly commonly played comp (which is not surprising given how popular Mage/Rogue is), and so the only comp that meets our 100 game threshold is Discipline Priest/Subtlety Rogue. For this matchup, there is a very substantial estimated 108 rating advantage in favour of the Arms Warrior/Restoration Shaman (with the 95% confidence interval being between 80 and 137 rating). Arms Warriors also generally play very well into Shadow Priests across most Arms/Healer comps.

Arms Warrior/Healer provides a fairly even matchup against Mage/Rogue at the rating under analysis, with usually only a small (but statistically significant) advantage in favour of the Arms team (in the case of Arms/Resto Shaman, it's 32 rating; for Arms/Holy Paladin, it's 25 rating).

The strongest and most consistent counter to Arms Warrior comps is a Frost Death Knight. Against Arms/Resto Shaman, Frost DK/Holy Pala has an estimated 80 rating advantage, while for Arms/Holy Paladin this is reduced to 53 rating. Playing into a Frost DK as an Arms Warrior can feel like a Catch-22: if you try to run at the healer, you get perma-slowed with Chains of Ice. If you go on the DK, you and your healer get cleaved hard by Chill Streak. The advantage of Frost DK against Arms Warriors is something that is observed across all Arms/Healer comps.

How well an Arms Warrior does against a Windwalker Monk depends entirely on the healer both DPS play with. The weakest healer for both DPS in a Windwalker/Arms faceoff is a Discipline Priest: whoever doesn't have a Discipline Priest generally comes out on top (statistically speaking). Arms Warrior/Restoration Shaman does pretty poorly against both Windwalker/Holy Paladin (-60 rating) and Windwalker/Restoration Druid (-70 rating).

If you'd like to see a more in-depth analysis of the Arms/Healer variety, you might be interested in this collaboration video I did with Warroid.

Holy Paladin/Windwalker Monk

Holy Paladin/Windwalker Monk reliably counters Arms/Healer when that healer is not another Holy Paladin. It also does very well against SP/Sub, although this is a far less popular comp in 2v2. A general observation across the different Windwalker/Healer comps is that Windwalkers are incredibly performant against Discipline Priests, especially when paired with another Discipline Priest (not shown). This can also be seen in data for this comp by observing that Holy Paladin/Windwalker soft counters Discipline Priest/Windwalker.

Holy Paladin/Windwalker experiences few natural predators, but the ones that are significant are Fire Mage/Sub Rogue, Frost DK/Holy Paladin (marginally), and the strongest counter to this comp: Discipline Priest/Retribution Paladin.

Discipline Priest/Retribution Paladin

Disc/Ret does incredibly well against Holy Paladin/Windwalker Monk. This is likely due to the enormous support toolkit offered by Retribution Paladins. Blessing of Protection, Freedom, and Sanctuary all provide ways to effectively stop Windwalker goes. Blessing of Protection is particularly useful given that Holy Paladin/Windwalker will have no spammable purge. Indeed, against Restoration Shaman/Windwalker Monk (not shown), there is no statistically significant advantage in favour of the Discipline Priest/Retribution Paladin, and against Windwalker/Discipline Priest, there is even a 52 rating advantage in favour of the Windwalker team.

Disc/Ret also does very well against both Frost DK/Holy Paladin (78 rating advantage, though note that this is only based on 100 games so the error bars are very large) and Fire Mage/Subtlety Rogue (58 rating advantage). In addition to Windwalker/Discipline Priest, the notable counters to Disc/Ret is Arms/Disc (50 rating advantage), Arms/Resto Druid (61 rating advantage), and Fire Mage/Shadow Priest (87 rating advantage).

Closing thoughts

An interesting point of reflection is the magnitude of the advantage that we generally observe in the arena. A "hard" counter would be a comp that has somewhere in the ballpark of 100 rating advantage on your comp. Interestingly, however, this still gives you a 36% chance of winning the match. There are exceedingly few commonly played matchups that have a rating advantage above 100. Indeed, the strongest counter of all in the current dataset gives a 150 rating advantage in favour of Discipline Priest/Windwalker Monk when playing against Discipline Priest/Havoc Demon Hunter. Most soft counters, on the other hand, are in the 50-100 rating range, which is a fairly modest handicap.

Got any requests for other comps and matchups you'd like to see? Pop by my Discord channel and let me know.

That mean of the rating distribution in this data is 1640, which is significantly higher than for the general ladder. The rating distribution is roughly normally distributed (i.e. a bell curve) with a standard deviation of 280. This means that approximately 95% of all games played are played between 1100 and 2200 MMR.

Estimating rating advantage

The simplest way of estimating the rating advantage is to just calculate the win rate and then convert this to a rating advantage through the Elo formula: $$ P(y=1 | \theta; \beta) = \rho = \frac{1}{1 + 10 ^{\ \theta/\beta}}, $$ where \(\theta\) denotes the rating advantage and \(\beta=400\) is a fixed parameter. Moving terms around we can obtain \(\theta\) by $$ \theta = \log_{10} \left( \frac{1}{\rho} -1 \right) \beta. $$ This is an approach that generally works fine, but is limited by the fact that people do not meet equally rated opponents all the time. While this is true on average (and thus it is an unbiased estimator), we can obtain a more accurate (i.e. lower variance) estimate of \(\theta\) by instead calculating the Elo maximum likelihood of \(\theta\) given the data. This takes into account the fact that sometimes a 1500 rated team meets another team at 1530 rating, and we wouldn't expect to have a 50% win rate.

The Elo likelihood is given as $$ \mathcal{L}(\theta | y, r_A, r_B) = \prod_i (p^{(i)}) ^ {y^{(i)}} (1-p^{(i)})^{1-y^{(i)}}, $$ where $$ p^{(i)} = \frac{1}{1 + 10 ^{\ (r_B^{(i)} - r_A^{(i)} + \theta)/\beta}}. $$ \(r_A\) and \(r_B\) refer to the MMR of Team A and Team B, respectively. In words, we try to find the value of \(\theta\) (i.e. the rating advantage) that can best describe the win outcomes that we have observed given the MMR of Team A and Team B.

The log likelihood is $$ \begin{align*} L\mathcal{L} &= \sum_i \log\left( y^{(i)} p^{(i)} + (1-y^{(i)})(1-p^{(i)})\right)\\ &= \sum_i \log \left ( y^{(i)} \frac{1}{1 + 10^{(r_B^{(i)} - r_A^{(i)} + \theta)/\beta}} + (1-y^{(i)}) (1 - \frac{1}{1 + 10^{(r_B^{(i)} - r_A^{(i)} + \theta)/\beta}})\right) \end{align*} $$ Thus, we obtain \(\theta\) by simply maximising the log likelihood.

For the confidence intervals, I have simply converted the \(\theta\) to a win probability, used the beta-distribution for obtaining confidence intervals on the win probability estimate, and then converted this back to rating advantage. This is simpler than obtaining the confidence intervals on the distribution of \(\theta\) and gives nearly identical results.