Can you Moneyball League of Legends? Applying sports analytics to the LCS

by Daniel Rosen Apr 18 2017
Thumbnail image courtesy of Riot Games

Jeffery Liang has been a League of Legends fan since Season 2, and has been a basketball fan for even longer. Following two pro scenes made him wonder if he could try to put the two together, leading him to apply tradtional sports analytics to League of Legends in order to solve some important questions.

Which champions are the most powerful across several pro seasons? Which are the most useless? Who is the most likely to be broken? Liang's paper, presented at the MIT Sloan Sports Analytics conference in March, set out to answer these questions. theScore esports got a chance to chat with him about whether or not it's possible to apply Moneyball strategies to League of Legends, which champions can never be viable without a rework and why the jungle is hard to balance.

How did you get interested in applying these kinds of analytics to League of Legends?

My interest comes originally from traditional sports analytics. I've been a long-time basketball fan and a follower of the analytics movement in basketball ever since Moneyball came out as a book when I was a freshman in college. Then, this past year, I got interested in the idea of applying the concept of sports analytics to esports, which on the timeline of development is very much in the early phases of adopting sports analytics (at least in my view). So I really just started with the idea of surveying the esports landscape for potentially interesting angles at which to tackle bringing sports analytics into esports. The honest answer was that a lot of my decision was based on data availability. If you think about a lot of interesting stuff you could do if you had like, really robust data about in-game timing and decisions and that stuff, it was beyond my ability to gather that data.

So what I knew I could analyze was champion pick ban outcomes, which are out there, published by the community. In that, I felt that there was a really rich data source because it encompasses all the hundreds of hours and time and resources invested by all these professional teams into champion research, strategy, development, to me it represented the market efficiency, I guess, of professional play. Additionally it's just kind of like a new dimension that traditional sports don't have. They have their players but they don't have the extra layer of having to think about which champions should their players play.

Do you think that pro teams could use these findings?

For me there's no reason why teams can't and shouldn't really be exploring and investing more heavily into this space. I think the success of traditional teams that have used sports analytics to win championships, there's countless examples there. For me there's no real obstacle or systematic reason why those benefits and those theories can't be applied to esports as well. If I think about drawing any sort of lines, I notice there's less dedicated staff towards more statistically based or heavily analytical analyses. A lot of the teams have a lot of their in-house analysts, but I feel like most of them are relying on very basic statistical measures or metrics or stats as opposed to exploring the potential benefits of more rigorous approaches.

What do you think your findings could be used for?

I feel like my research does two or three main things. The first is if you think about what analysts and the community thinks of now in terms of measuring champion viability, for the most part most people think of pick ban rate. But then you have to pick an arbitrary point in time to start counting from, usually the start of a split or the start of a season. That makes it difficult to compare pick-ban rate across early versus middle versus late parts of the season because you're talking about different expanses of time. And so I think the first primary benefit of this research was to quantitatively identify what is the appropriate window of recent games to be thinking about which champions are in the meta. Just by solving for an optimal smoothing factor, or a discount factor, it tells you how quickly or not quickly you should discount prior matches, and what's the window of recent matches you should be looking at that are most predictive of what champions are going to show up in the next game. I think that's very useful information for analysts and coaches that are thinking about their pick ban strategies to know what the data says is the appropriate window of matches to be looking historically at. At what point do old data points become obsolete, and at what point are we looking at too narrow a horizon of recent matches that we're not getting the full picture.

A lot of the data suggests that newer champions are harder to balance, why do you think that is?

I feel like part of it is driven by the dynamics at play behind why Riot releases new champions. They release new champions to keep the game fresh and to have exciting new content, and potentially content that's interesting for the playerbase to buy. Personally, I don't feel like they release new champions thinking about what would enhance the competitive pro scene — they're thinking about the game published for the masses. I feel like these new mechanics over time are harder to balance. The meme right now is that the mobility of new champions totally outclass the mobility of the original, let's say 50, and that's just one example. It's a combination of creating interesting new characters, and also character that feel good to pick up as opposed to have players pour 6,300 IP into a new champion and it feels boring and weak.

What is it that breaks a champion? Is it in the kit or more about their numbers?

I absolutely feel like there are certain champion kits and abilities that are extremely hard to balance around. For example, Taliyah's kit is really feast or famine. She's super squishy and she's mostly about control and speed and a little displacement, so her ability to move around the map and the amount of damage that's packed into two spells really, I feel like makes her hard to balance. I think she's top three or four in terms of kurtosis, which is how extremely that champion's viability score wavers up or down. Other champions are like Soraka, which you can imagine is hard to balance around because of mass healing. Yasuo is number five and even something like Wind Wall is a good example of a new mechanic that people probably didn't think was possible in like Season 3, the technology wasn't there. But that is even a really extreme ability in my mind. That is only one of the three or four things in Yasuo's kit that makes him extremely hard to balance as well.

I wouldn't say it's a function necessarily of Riot trying to create overpowered champions with each release. It coincidentally happens with interesting new mechanics they want to release and trying to keep a general interest and freshness to the game.

Who are the least varied champions? Are they also usually the worst, because they have the least variability?

The champions with the least kurtosis are really the ones that don't get picked ever so there's no variability in their viability score. So we think about the champions with the lowest viability score: Master Yi, Teemo, Fiddlesticks, Rammus. Interestingly some of these have become a bit more meta, so for example Talon is down here as well but with his rework he's seeing a lot of play.

So for those champions, is there any way to fix them beyond just totally reworking their kit? Or are they lost causes?

Barring a major rework, I would say yeah. For example, Mordekaiser is like, eight from the bottom but he would be literally at the bottom, never having seen a pick or a ban up until Worlds 2015, when they reworked him. Then they nerfed him again and i only until this season with one team., FlyQuest playing him. Before that rework, and given his kit in general, it boils down a lot to kit, champions like that would be, in my view, would be pretty hopeless to see play in competitive. Kit is something that's very interesting in my mind. You can look at base stats and damage and defenses as much as you want, but the ability set and what it's able to do is probably what dictates the potential for a champion. Eventually, every champion's numbers will be balanced within reason, so it's really the ability kit that gives the potential for the champion. so a champion like Mordekaiser who has no CC and really slow base movement, and is really just damage and tank, has very little potential.

What about the jungle makes it so hard to balance?

I don't feel like the jungle champions themselves are necessarily overpowered, it's just that jungle is the role with the least diversity by far. If you think of the five roles, jungle is the least diverse followed by ADC, but jungle is almost 50 percent less diverse than AD carry, which is already the second least diverse role. There's very specific characteristics that champions need in order to be successful in the jungle. Jungling is more an optimization type role. It's why people like Meteos back in Season 3 or 4 were the most farm-heavy and efficient jungler which led him to be one level above his opponents. It's what made him a dominant jungler that season, because the jungle is mostly not a game against players, you're playing against the game's meta and it's an optimization problem. Because of that, only a few select junglers are S-tier, that can optimize the jungle as it exists at any given time. They're high champion viability scores aren't necessarily because they're overpowered, it's just that there's not other stuff you can pick.

Do you think Riot could have use of your findings? Or is this all stuff they should probably already know given their own analytics?

I don't actually know quite honestly. Riot has their own commentators and analysts and Jatt does a lot of his own analyses and publishes it for the community and does a great job of helping shepherd along the analytical community. So there's probably people at Riot that would find this kind of stuff interesting, this research and potentially anything like it. I'm also in the camp of wanting more of this kind of analysis to exist in esports in general. There are potentially pockets of people at Riot that might have an interest in this and could explore other topics or even take it a bit further.

Can you Moneyball League of Legends? Could you use analysis to put together a theoretically perfect pro team?

I would say yes, because I think that's the opportunity that analytics kind of closes. Until advanced analytics is several years down the line in terms of adoption and development in esports, I think that opportunity exists and is ripe for the taking for some team to go for it. A lot of the analytics that would need to be used to successfully moneyball League of Legends probably doesn't exist today. A lot of the analytics you can do probably hasn't been invented or approached.

Daniel Rosen is a news editor for theScore esports. You can follow him on Twitter.