In statistics, there is nothing like finding a simple pattern that explains a complex universe. A long time ago we introduced PTK (Percentage of Teams’ Kills) as a simple metric, which by the way acts as a great “control-stat”. To research and conceptualize different playstyles I developed the PTK-Model. And this quickly yielded some interesting results, which made me question the West’s approach to Overwatch. It is not DPS that carry your team, no, Tanks and Supports win Championships.
Now that the first Stage of Overwatch League has come to its end it is time to look at the PTK-Model once again. And this time I want to focus on something that I have been preaching in private circles for a long time: Flex-Support is the most impactful role in Overwatch!
The Divine Correlation
It is rare to find strong correlations in statistics, especially in a complex environment. Overwatch is a complex game. And yet, the role of Flex-Support provides a truly divine correlation.
When I talk about Flex-Supports, I am speaking of the non-Lucio/Mercy Supports. The players that occasionally flex to Sombra or Soldier, the Zenyattas, Anas, and Moiras of this world.
When looking at the PTK-Model for Overwatch League you can quickly identify one thing: Higher support PTK means a better placement. On the right, you can see how an ordered support PTK frighteningly matches the standings. Two different map-outcomes on the last day of stage 1 and the first three positions would have been the same. While it is not perfect and has its outliers (Houston, what are you doing down there -.-) if you actually do the math it is the best predictor for standings (out of all PTKs by role).
This is not entirely new, as explored in the article referenced above. What is new, though, is that this meta only really sees one support player fragging, Mercy keeps violence to a minimum. And if we split up all the roles for a more detailed look the impact Flex-Supports have on the game only gets more apparent.
But before we do that, let’s quickly define what I mean with “impact”. Because obviously, DPS players get the most kills, no matter which team we examine. Impact, here means the potential a role has to win a game (or the tournament). An understandable metaphor would be the following: Let’s say a team with perfectly average players has a skill level of 100. And let’s say bringing in a beast DPS player, a top 3 player in the world in his role, would increase your skill level to 150. Then the observed phenomenon would mean, that bringing in a top 3 flex-support to an average team would raise its skill level to 170.
So again: Flex-Support is the role which has the biggest potential to improve your team. JJonak, Bdosin, and Neko (who doesn’t get enough love) are the big names of OWL Stage 1. And funnily enough in the final JJonak underperformed while Bdosin was better than usual. In the game NY and LDN played earlier that day it was the other way around. It was actually true for a lot of games that the team which won was the team where the Flex-Support had a higher PTK, as you can see in the graphic below.
Impact by Role-Subcategory
A detail-level deeper we split up each of the 3 roles into two subcategories. Support becomes Flex-Support (Ana, Zen, Moira) and Main-Support (Lucio, Mercy). Tank gets split up into Off-Tank (D.Va, Zarya, Roadhog) and Main-Tank (Rein, Winston, Orisa). And we can simply label DPS as either hitscan or projectile.
Doing that gives a clearer view of what’s going on. A simple means to measure impact is to look at each match and check for each role if the winner was the team that had a higher PTK in that role. From the 62 matches of stage 1, in 38 (61%) of them, the winner had a higher Flex-Support PTK. Hitscan meanwhile sits at 39%, basically meaning if your DPS get a lot of your teams’ kills you lose.
Now, this is by no means the best metric to measure the impact of a role, but it is easy to grasp. A more accurate measurement can be done when looking at the Pearson Correlation Coefficient of role PTK and fight win rate by match. At OWL Flex-Support PTK has a correlation of r=.24 which is insane, considering that PTK as a whole, combining all roles, has no correlation to winning at all. The reason for this is that every team always has a PTK of 100% in total, so getting to .24 when examining a single role is crazy. For stats junkies, I will go into more details at the end of this article.
If we look at the PCC of role PTK and FWin% by team (for stage 1), we can see a correlation of r=.65. With this method only having 12 data-points, it is certainly questionable. None the less, on this level Flex-Support PTK is a better predictor of success than deaths/10min! Yes, this indicates that it is more important for your Flex-Supports to get kills than it is for your team not to die.
Again, take that last statement with a grain of salt. On a per match basis it is definitely not true, it’s just a little fun-fact.
In my dream world, I would now be able to say that Flex-Supports are always the most impactful players in Overwatch, no matter the meta. Unfortunately what I started saying about a year ago is not always true. Sometimes main supports were more important, sometimes main tanks, buffs and nerfs had strong effects.
Over the whole lifespan of Overwatch Flex-Support has seen the most time at the top, though. But in the end, there is only one fact that persisted throughout all metas: If you do not have bloodthirsty Supports and Tanks you will lose! DPS alone cannot carry your team – most of the time, there are outliers of course.
There was a reason why Lunatic-Hai with Esca, who was not known for crazy plays and getting tons of frags, was so dominant. The PTK-Model gives clear directions on how teams should be built. You shall not have more than one aggressive DPS player and having less than 2 bloodthirsty, play-making, Supports/Tanks means trouble. But this is not the place to go into details about this, I might write about it another time.
All said and done, the one rule that stands above all is Tank and Support players win championships.
For everyone who likes to look at numbers here are some more, without many explanations.
Kills and Deaths have been more important than ever before. All-time in Overwatch kills/10 correlate with r=.817 to fight win rate, deaths/10 with r=.804. In OWL kills/10 are at r=.875 and deaths/10 at r=.87. This means that either, teams have become drastically better at getting kills at the perfect time and snowballing quickly. Or, which I think is the case, that this meta was the least strategic meta we ever had. In my opinion, playing dive makes it easier to coordinate, easier to get chain-kills, easier to snowball the fight.
If we go back to slower Tank-compositions next stage kills should once again become less important, because not being out of position and coordinating well would be the focus of those comps. As opposed to shouting “Monkey, monkey, monkey” in voice chat and jumping in at the same time 😉
Because it is kinda cool to look at here are the Fight Win + Flex-Support PTK data points for each match in OWL. Blue dots means home team, red dots are for the away teams.
I put in the graphic above to amphasize how steep the trendline really is, here is a closer view of it.
And now lets look at the chart for Hitscan PTK:
Here are the correlation coefficients for all subcategories and kills/10 and deaths/10 to fight win rate.
All-time, Hitscan has a correlation of r=-.083 Projectile is at r=-.072.