How do you rate players in Overwatch? Put another way, how do you tell who has the most impact in a match? Finding the answer is a challenge that is nearly impossible. After all, how do you develop a reliable way to compare apples to oranges in a world where an apple can swap to a banana and an orange can swap to a cantaloupe at a moment’s notice?
In the first season of the Overwatch League, we solved half of this problem by normalizing stat accumulation as a rate over time—“per 10 minutes” statistics—which equated roughly to the average of a map’s duration. Thus, when looking at player-hero pairings, we were always at least comparing apples to apples.
But that’s not good enough in the long run. Stat aggregation across an entire map or match has less analytic power compared to moments of greater significance: teamfights. Teamfights allow us to focus directly on the most important time spans within each map. By knowing when teamfights begin and end, we could remove the fluff and build more valuable aggregations.
However, the challenge of comparing an apple to an orange (e.g. a Winston to a Tracer) still remains. Over the last few months, inspired by several systems ranging from fantasy esports to internal balancing metrics, I created “Impact Points”—an abstract method of measuring each hero’s contribution to a teamfight, where each hero is balanced via a two-tiered process to produce the same amount of points over time.
With Impact Points as the counting statistic, Player Impact Rating (PIR) is therefore the rate at which a player accumulates these hero-agnostic points, and ultimately the best comparator across a variety of heroes and over unequal periods of time.
Impact Points are produced by applying individual weights to a suite of statistics that heroes generate in teamfights. Overwatch is an objective-based game, but final blows and deaths are the currency that teams spend to achieve those objectives, so the stats I’ve used are centered around final blows and the things teams do to accumulate or prevent them. The list includes—but is not limited to—raw damage and healing, crowd-control abilities like hacks or hooks, and even teamfight-winning plays like ult negation.
These stats are rolled up by each hero and balanced as close as possible to an equilibrium of 100 impact points per minute (ppm) of teamfight time played, as an average across the last five patches played in the Overwatch League. This ensures that this “first-tier” average comprises four stages and playoffs.
However, this first tier of balancing is imperfect. Some heroes, by virtue of their design or role, tend to accumulate valuable teamfight statistics at a rate that slightly exceeds their peers. This is further exacerbated by patch differences each stage. For example, if I applied just the first tier of balancing to Stage 1 of the 2019 season, Zarya players’ impact would be off the charts due to how much more damage the hero can deal in the triple-tank, triple-support meta compared to the first tier’s predicted output.
Player Impact Rating has to be meta-proof, but also accurate in showing the strength of individual players within a specific patch. To address this shortcoming, the first tier of the system serves primarily to provide a general idea of the historic strength of a hero. When it comes time to roll a new patch into the aggregate, the stat weighting generated by the first tier is applied to each individual statistic, but each individual hero is then also weighted individually to bring the entire Overwatch cast into exact 100 ppm equilibrium for that current patch. So if you are an Overwatch League player performing at exactly the league-average level, you should expect to generate 100 Impact Points per minute, regardless of what hero or patch you’re playing on. This hero- and patch-specific weighting comprises the second tier of the Player Impact Rating system.
Let’s try summing up both tiers: Player Impact Rating is a hero-agnostic system for comparing Overwatch League players, balanced by long-term stat weighting and hero strength during a given patch.
Note: to avoid outliers generated by small sample sizes, a player must have played in 60% of the minimum possible maps within the time range being measured in order to qualify for Player Impact Rating.
And now for the fun part—who were the top 30 most impactful players in the 2018 season of the Overwatch League, according to Player Impact Rating?
|Player||Team||2018 Player Impact Rating|
|Jjonak||New York Excelsior||127.79|
|Fissure||London Spitfire/Los Angeles Gladiators||118.28|
|Space||Los Angeles Valiant||115.21|
|Fate||Los Angeles Valiant||114.94|
|Mano||New York Excelsior||112.28|
|Saebyeolbe||New York Excelsior||110.76|
|Meko||New York Excelsior||110.34|
|Kariv||Los Angeles Valiant||108.80|
|Danteh||San Francisco Shock||107.70|
|Shaz||Los Angeles Gladiators||107.53|
|ArK||New York Excelsior||106.15|
|Agilities||Los Angeles Valiant||105.74|
|BigGoose||Los Angeles Gladiators||104.78|
By and large, the rankings match the eye test and then some. Is anyone surprised that Seong-Hyun “Jjonak” Bang had the top rating, or that Jae-Hui “Gesture” Hong and Jun-Young “Profit” Park were very close behind? Or that Philadelphia Fusion duo Jae-Hyeok “Carpe” Lee and Josue “Eqo” Corona were the highest ranking damage pair? Or that London, who ultimately won the inaugural season championship, had four players in the top ten?
Player Impact Rating even provides a way to track the ups and downs that players and teams experienced during the season:
A picture is worth a thousand words, and here’s about 60 from this one: The London Spitfire kicked off the inaugural season on a high note, but an injury to Ji-Hyeok “Birdring” Kim nearly torpedoed their playoff hopes in Stage 3. However, he rallied in Stage 4 to help London limp into the playoffs. Once there, London ripped off an incredible playoff run led by outstanding individual performances by eventual Grand Finals MVP Profit.
London’s entire season, as told by a single metric.
The same can be done to show the emergence of two of the league’s young stars, San Francisco Shock duo Jay “Sinatraa” Won and Matthew “Super” DeLisi:
Upon reaching eligibility, their impact steadily rose over time, peaking during the Shock’s golden stage before dropping back to earth in Stage 3.
How about the ascension of the Shanghai Dragons to the team’s first stage playoff win?
Min-Sung “Diem” Bae and Jin-Hyeok “Dding” Yang have continued to improve over time this season, but perhaps it was Young-Jin “Youngjin” Jin receiving the green light to use his signature Doomfist that helped put Shanghai over the edge in the playoffs!
While Player Impact Rating is an exciting new way to compare players, there are some caveats to keep in mind. The statistic—like most in Overwatch—is still somewhat dependent on team performance. If your team is winning more teamfights, your raw output will most often exceed the teams that you’ve defeated. Some dependency on team performance is acceptable, since Overwatch is one of the most team-oriented esports around, but team strength alone cannot fully explain a 20% deviation from league average. Additionally, the best teams don’t become the best teams by having below-average players.
Even within teams, there is a decent amount of shuffling in Player Impact Rating throughout each stage. Take the NYXL last season, for example:
Here we have four NYXL players last season who—besides a shared nosedive in the playoffs—had very different paths through the season. Jjonak and his bodyguard, Tae-Hong “Meko” Kim, maintained fairly consistent teamfight impacts, even if Jjonak’s was ridiculously high. Yeon-Jun “Ark” Hong, on the other hand, was constant, but only in his decline in PIR over the season. Jong-Ryeol “Saebyeolbe” Park, the charismatic but streaky captain, was all over the place. So even if you’re worried by how much team performance affects players on different teams, it can still be useful for comparing players within a single roster.
Finally, the moment of truth: with three stages completed and 2019 MVP discussions reaching fever pitch levels, who is leading this season’s PIR race?
|Player||Team||2019 Player Impact Rating|
|Sinatraa||San Francisco Shock||117.94|
|Anamo||New York Excelsior||117.78|
|Jjonak||New York Excelsior||116.9|
|Void||Los Angeles Gladiators||116.12|
|Super||San Francisco Shock||114.36|
|Mano||New York Excelsior||114.27|
|Nenne||New York Excelsior||112.29|
|Viol2t||San Francisco Shock||111.11|
|Rascal||San Francisco Shock||110.47|
|Libero||New York Excelsior||109.36|
That’s quite a list! Some players recently cracked the top 30 after the meta shakeup in stage 3—Diem, Youngjin, and many Spark players for example—but the upper echelons of Player Impact Rating are reserved for teams with continuous, overwhelming success. To that point, it doesn’t surprise me in the slightest that three Vancouver players own the top three spots. Despite their playoff losses in Stages 2 and 3, no team—not even the 2018 NYXL—has matched the pace the Titans are setting. At their peak after Stage 3, the NYXL’s regular-season map win rate was just over 75%. Through three stages, Vancouver’s is over 81%.
As for why, look no further than the man at the top of the league: Hyo-Jong “Haksal” Kim. Even without Player Impact Rating as a guide, I knew he was doing something very special on Brigitte this season. Generally, Brigitte players have fallen into two statistical categories this season. One version is the aggressive, playmaking Brigitte who racks up final blows, hero damage, and Shield Bash kills—setting up their team for picks and combos. The other is the passive, peeling Brigitte whose focus is on protecting their backline and interrupting pushes with clutch stuns—all while maintaining a high rate of healing.
Rather than falling into one of these buckets, Haksal does it all. Over the entire season, he has consistently ranked in the top three for nearly every statistic that Brigitte can accumulate; other teams even tell their players to study his Brigitte play in order to improve.
But this isn’t a story about Brigitte, it’s a story about Haksal. Role lock is looming, and while most of his Player Impact Rating comes from his Brigitte play (89%, to be exact), Haksal is historically best-known for a much different hero: Genji. While the viability of Genji in the coming meta is still up in the air, the 45 minutes of Genji that he has played this season has produced a PIR of 142.58. Something tells me that no matter what damage hero he plays, Haksal is going to do just fine in Stage 4, and that his impact will continue to be felt throughout the league.
Player Impact Rating is not a predictor for wins, but winning teams do tend to produce higher PIR. It measures raw output of “good stuff” and attempts to quantify how much “good stuff” a player like Haksal is producing during a teamfight, relative to the league-average player. The best players tend to do better than their peers in teamfights, no matter what hero they’re playing, and the best players on the best teams perform even better than that.
Player Impact Rating, as with any statistic, is just one broad stroke in the larger discussion of a player’s value. Clutch factor, striking first, and many other intangibles aren’t reflected as heavily in Player Impact Rating, but still matter a lot when it comes to evaluating individual and team talent in the Overwatch League. Still, PIR is a good start, and I’m excited to have a powerful new way to compare players.