Creating a Solo-Laner Index: Higher-Order Statistics & League of Legends Esports
In the world of League of Legends esports, there’s no hotter debate than that of “who is the better mid-laner”. In an effort to silence the multitude of biased, opinion-driven drivel I heard on a weekly basis during my tenure as a photographer for Team Liquid and the LCS, I dove into data provided by Tim Sevenhuysen over at EsportsOne and (This link open in new tab) (This link open in new tab) Oracle’s Elixir to find out exactly what qualifies a successful solo-laner (yes, both mid and top), and how could we quantify it in an effective and easily digestible way.
What is a solo-laner?
In League of Legends (LoL) there are three major lanes, top and bottom which take an “L” shape at the top and bottom of the map respectively, and middle which cuts through the middle of the map diagonally, effectively shortening the lane and thus the distance between you and your opponents base, or “Nexus”. There’s also a quadrant system called the Jungle, which I’m saving for future exploration, but there will be mentions of Junglers in this article.
Though the top and bottom lanes are equal in length, and mirror images of one another, the top is a solo lane while the bottom is a lane that is occupied by a primary damage dealer, the ADC, and a Support player who plays somewhat of a healing or peeling role. We will only be looking into Top- and Mid-laners during this exploration as the variables are much more explainable through individual stats.
What does a good solo-lane performance look like?
Certain stats are tracked for all players in LoL, such as their gold income over time, the number of kills, assists, and deaths, and even whether or not they delivered final blows on neutral objectives such as the Rift Herald and Baron Nashor. These metrics are helpful, of course, but there’s a more abstract layer of play on a professional level that is not currently (to my knowledge) tracked or even quantified. This includes a player’s ability to pressure a lane enough to roam (often called Lane Priority), how often they’re helping on key neutral objectives, and whether or not the player is mechanically better or worse than his or her opponent.
At the end of the Early Game or Laning Phase, laners often switch lanes, or form 1–3–1 or 1–4 compositions, or group as 5. This phase ends right around 12–15 minutes. This is around the time you start to see the fruits of your early game labors in the form of snowballing, whether that’s through exponential increases in kills, objectives taken, or both, resulting in growing gold leads. This can skew endgame statistics, so Riot has blessed us statisticians and data scientists with some key metrics tracked at 10 and 15 minutes. These are the versions of these metrics we will be utilizing for this Solo Laner analysis. This will give us a clearer image of how they perform on their own, rather than as a part of the team.
Through regression analysis, I was able to both prove what most already knew as well as discover a few new ways to more accurately classify players in some of these key untracked areas. One of the key metrics that people look to when comparing laners is their Creep Score or CS. Each wave over the first 10 minutes of the game averages out to 127.5 gold; with 18 waves (114 total CS), that’s 2295 gold at 10 minutes that can be attributed to CS. Gold is a key in snowballing after the laning phase, so by measuring the proficiency a player has in terms of CS-ing, we can also provide an accurate look at how much gold we can expect them to have from that one lane-specific metric. The equation used to measure proficiency is a simple percentage calculation, taking the average CS of a player at 10 minutes and dividing it by the total possible CS by 10 minutes. When calculating this, it’s understood that a player may roam to a new lane and pick up CS not dedicated specifically to their lane, however, this does even out in the CS missed in their lane thanks to the constant spawn and movement rates of CS. After acquiring their average proficiency rating, we can multiply it by our gold-per-wave average of 127.5 to estimate their amount earned through CS-ing.
That brings us to our next key metric, gold earned outside of CS. Since we already have the amount earned through CS at 10, we can subtract it from our Gold at 10 stat to get this number. We average this amount from all of their games, and through custom (i.e. proprietary) methods of weighting derived from weeks of analysis, we are able to accurately provide our second statistic, what I’ve named the Gold Difference Rating, not to be confused with the player’s Gold Difference at 10/15 minute stat that is tracked and given via the Riot API. Instead, this is an example of a statistic we can use to assume a player’s aggression, team play, and more when compared with other statistics like Kills Per Minute, Baron Kill Rate, etc. This is a major step in a positive direction when it comes to accounting for the “human factors” in professional play!
What if I told you that PowerOfEvil is actually the best Mid-laner in the LCS right now? You’d probably call my bluff, and you might be right from some (or many) angles of analysis, but in terms of raw laning ability, the numbers don’t lie; he’s the most proficient CS-ing and gold-earning Mid in the LCS! What’s less shocking is the amount of LEC and LCK representation, specifically Mid-laners, when it comes to the top 10 Solo-laners as rated by Lane Score. These types of higher-order statistics are what we should be working towards as esports data scientists to better represent and quantify (to a degree) these abstract concepts analysts have spoken about for 10 years now.
As far as the Top 10 Top- and Mid-laners:
- Nuguri (LCK) — 76.8
- Doran (LCK) — 76.5
- Orome (LEC) — 73.0
- Ruin (LCS) — 73.0
- Kiin (LCK) — 72.5
- Alphari (LEC) — 71.2
- Impact (LCS) — 69.1
- Rascal (LCK) — 68.5
- Bwipo (LEC) — 65.3
- Hauntzer (LCS) — 64.0
- Tempt (LCK) — 81.4
- Nemesis (LEC) — 81.4
- Febiven (LEC) — 80.0
- Jenax (LEC) — 79.2
- ShowMaker (LCK) — 79.2
- Chovy (LCK) — 79.0
- Kuro (LCK) — 79.0
- PowerOfEvil (LCS) — 78.3
- Bdd (LCK) — 78.0
- Fly (LCK) — 77.5
There’s a notable gap in scores between Top and Mid, which may be a result of the position of the lane itself. As mentioned in the introduction, the middle lane slices the map in half, therefor the Jungler has to cross the lane multiple times while pathing, opening the lane up to a higher gank probability. In the current meta, as of writing this (patch 10.3), Junglers begin ganking as soon as Level 2 or 3. A large contributing factor to the success of all Solo-laners was their ability to earn gold outside of CS-ing, but there is a notable difference in the CS-ing proficiency of the Top-laners, which may help support my Jungle gank theory when paired with Kills/Assists data. When a gank is successful, it opens the lane up to uncontested wave clearing; something Top-laners may not have as often, putting them at a more average CS proficiency rating.
There’s no doubt in my mind that there are holes unaccounted for in the scoring system for Solo-laners, even when constrained by our assumptions, but I’m confident this is a step on the path towards my overarching goal of quantifying the more abstract concepts analysts have been using to judge players for the last decade of pro play. With more data scientists entering the space, I can only hope this is a small drop that creates a ripple effect, so please share this among your esports-loving friends and discuss ways you think I’m wrong and how you’d evolve this concept!
The next exploration is a secret, for now, but rest assured there’s more coming! I’ll have to set up some type of email subscription system to notify you all. I don’t have any social media plugs, but feel free to contact me via Discord (KMH#6227) if you ever want to chat about LoL esports, statistics, or data science!
Visit my website for more written data science and esports content, https://kevinmhaube.com/