Snowballing is an important part of League of Legends: if you can’t convert a large gold lead into a nexus lead, you will never win the game, and if you can’t convert a small gold lead into a large gold lead, the enemy team will take your lead away. Every team should be constantly trying to increase their gold lead, or dig themselves out of a gold deficit.
However, not every team is equally good at this. Some teams know how to recover from a disastrous early game to a miraculous comeback victory, as TSM did vs Immortals in Game 4 of the NA LCS Summer finals. Some teams seem hopeless, as they either squander a big lead, or dig themselves into a deeper hole every passing minute, such as when Dignitas struggled against CLG in the NA LCS 3rd-place match.
Today we’re introducing a new statistic called Snowball that measures this aspect of League of Legends.
We define Snowball as the following:
Note that whenever a team gains Snowball, the opposite team loses Snowball of equal value. If the gold distribution changes from 52%/48% to 51%/49%, blue team lost 1 Snowball while red team gained 1 Snowball. This rewards teams for increasing their % gold lead, while punishing teams for losing it.
We calculate a team’s Snowball in one game as total of Snowball from each minute, and the team’s overall Snowball as the average of Snowball from each game. While each game length is different and thus contain different number of data points, this places equal importance on each game.
A team achieves positive Snowball if they are increasing the % of the total gold they hold. For an example, let’s look at Game 1 of H2K vs Splyce from Week 1 of EU LCS, specifically from 15:00 to 16:00. At 15:00, Splyce has a gold lead, with 23,443 gold, while H2K has 22,007. At 16:00, this gap widens as H2K goes up to 23,294, but Splyce jumps to 25,156. This means the Snowball for Splyce from 15:00 to 16:00 is:
Conversely, this means H2K’s Snowball at 15:00 is -.00342. Splyce is rewarded for increasing their % gold lead, while H2K is punished for slipping further.
On the other hand, in Game 1 of C9 vs CLG from Week 1 of NA LCS, from 29:00 to 30:00, C9 increases their gold lead from 5,734 to 5,779. However, C9’s Snowball is -.000562, as they didn’t increase their gold lead enough compared to how far ahead they were. CLG is rewarded with .000562 Snowball as they are doing a good job stalling while C9 holds a large lead.
The Snowball coefficients from EU and NA in the 2017 Summer Split look like this:
Click for full size
This is very much in line with the regular season rankings, with notable exceptions being Cloud9 and Unicorns of Love. UOL tend to spiral into a gold deficit early game, and make a moderate comeback. C9, on the other hand, often struggled in converting their leads into major objectives after getting a decent lead early in the game.
With how drastically different every team is, we can learn more by separating each game into two stages: pre-15 minutes and post-15 minutes:
Click for full size
This plot provides a clearer picture of how each team’s games tend to play out. Teams in the top right quadrant should generally be considered strong teams, while the teams in the bottom left tend to struggle.
The Unicorns of Love have a negative Snowball before 15 minutes, and a positive one afterwards; on the other hand, EnVyUs generates a large early Snowball, but slowly loses it past that point. Three NA LCS teams–TSM, CLG, and Dignitas–stand out in the top left quadrant. All three are semifinalists that tended to have weak early game control. TSM made up for it with their superior late-game Snowball, while CLG’s and DIG’s shortcoming can be attributed to large negative Snowball in the early game of their losses: CLG and DIG rank 17th and 20th respectively in first 15 minutes’ difference between Snowball in losing games and Snowball in winning games.
Advantages and Shortcomings
Snowball is an indirect measure of teams’ shotcalling. Unlike the Major Leads & Deficits metrics introduced recently by the EU LCS Stats team, Snowball accurately rewards the full process of a comeback as long as a team is shrinking their deficit, not only rewarding them once they finally pull the gold past a certain threshold. TSM’s road to comeback from a 10k deficit against Immortals in NA LCS Finals Game 4 should be reflected in any statistics measuring gold leads. Snowball is also a symmetric score: as much as a team should be rewarded for a good play, the enemy team should be punished. The scoring can measure the impact of each time period, showing whether the status quo of the game was preserved, or whether there was a major change in the game.
Unfortunately, Snowball has plenty of weaknesses as well. Due to limited data availability, the team’s total gold can only be measured each minute. As Snowball is a derivative of % gold share over time, more frequent sampling (such as every second) would produce a more accurate Snowball metric.
Snowball also considers every minute of the game of equal importance. In reality, the first minute is pretty much irrelevant, outside of an occasional First Blood, and there may be other ways that different game stages mean more or less overall, based on things like the state of the meta game. Further, team compositions can affect Snowball quite a bit: with scaling compositions, for example, it may be acceptable to slowly lose in the early game, though you will be expected to make large turnarounds in the later stages. Weighing each minute depending on the team composition may be too tall of a task, given the limited number of pro games played each year.
Dan is currently an analyst for Team Vitality in the EU LCS. His background is in Applied Mathematics and Economics, as well as software development. Follow him on Twitter @Dannerlame.