League of Legends stats: some findings

I’ve been noodling with some League of Legends data as an excuse to learn more data analysis tech. I’ve blogged about the software, here’s the actual results I found.

League of Legends is a huge game, some 100M a month playing 1M+ games a day. And the data for those games is readily available. I took a look at third party summaries of game statistics from champion.gg and na.op.gg and tried to divine some meaning. The dataset mostly is for ranked games by Platinum+ skilled players in North America for patch 6.17. But I also looked at players of different skill and trends across game patches.

The most interesting conclusion I found was that some champions win much more in the hands of skilled players. Nidalee, for instance, has a 45% win rate in Silver but a 54% win rate in Diamond. That’s not exactly a surprise, she’s known to have a high skill cap, but it’s good to see hard data. And there are surprises; I would not have guessed Pantheon also was highly improved or that Dr. Mundo suffered at higher skill levels.

Despite looking, mostly I didn’t find any unusual trends or big shifts in the game over patches. To me that’s a testament to the quality of the game, in particular Riot’s game balance team.

Below find some analysis and graphs. I’ve put my notebook output online for viewing here:

Basic stats for patch 6.17

The first thing I did was look at the basic champion.gg stats for patch 6.17. Each row in my dataset is one summary line for all players of a champion in a role. For instance Nami / Support. Note there’s a bit of bias here; champion.gg omits rare cases like Nami / Jungle.

There’s relatively little diversity in ADC and Support champions. Simple result: there are only 17 regularly played ADC champions and 25 Supports. There’s ~50 champions played Top/Middle/Jungle. This fact skews a lot of my later statistics, particularly with the ADC pool where a single champion (Lucian) dominates with a 32% play percentage. The top mid, by contrast, is Ahri with only 13%.

download (1).png

Players generally don’t pick champions with higher win rates. This surprised me a bit; I thought there’d be a clear bias towards the OP champs. There isn’t.

download.png

Players do tend to win more on champions they have more experience with. No big surprise here, but more games = more wins. The stat “experience” here is average games played on that champion. (It’s in the champion.gg dataset, but not displayed on the website.) The correlation coefficient is +0.24 with a p value of 0.0005, which seems likely to be significant to me. The causation is less clear. For example Galio-Top players have an average of 107 games on the champ and a 55% win rate. Do people win on Galio because they play him a lot, or do they play him a lot because they win? (Or in the specific case of Galio; do relatively few people play him but the ones who do are really good?)

download.png

Stats for players of different skills from na.op.gg

I was curious how these statistics varied by tier, so I went to na.op.gg and downloaded stats for players at all skill levels. I mostly focussed on Silver through Diamond, since those are the most “normal” player populations. This data is for the month of Aug 15 – Sep 15 2016 and spans patch 6.17 and 6.18.

Some champions have much higher win rates at higher tiers. This result is the most interesting I’ve found. I feel like each player at the extremes has a story. I’m not surprised that Nidalee’s win rate goes up 9.37% from Silver to Diamond; she’s hard to play well and so presumably does better in the hands of a skilled player. Aurelion Sol goes up 5.8%, maybe because he depends on skilled teammates? But why does Pantheon go up 7.5%? He seems so simple! I’m guessing Aatrox and Garen do worse in higher tiers because players know how to exploit their weaknesses. And Amumu’s 4.1% drop may be as simple as Diamond players being less vulnerable to Amumu cheese. But there’s more to say here and I’d like to develop this data further. Here’s the 10 most improved and 10 who suffer the most.

Screen Shot 2016-09-21 at 4.32.54 PM.png

Some champs are significantly more popular at higher tiers. The companion result to win rates; what about pick rates by tier? Janna is the most more frequently picked, she goes from 0.8% to 2.9%. (As a frequent Janna player my theory is she is much more rewarding to play when your team is more skilled.) Leona drops from 1.6% to 0.6%. The ADC champion pool attenuates at higher tiers too; Lucian dominates and Jinx drops off, for instance.

Screen Shot 2016-09-21 at 4.37.11 PM.png

Silver-to-Diamond win rate is correlated with Silver-to-Diamond pick rate. In the previous section I reported that for Platinum players, win rate is not correlated with pick rate. However the champs that have the most improvement rate from Silver-to-Diamond also tend to have higher pick rates from Silver-to-Diamond. This suggests that higher tier players are picking champions that have more impact at their skill level. Well maybe; I have a suspicion this effect is pretty small.

download (2).png

 

Trends across patches

Finally I was curious how stable the game statistics were over patches. I collected historical data from champion.gg’s GitHub going back to patch 5.10, over a year ago. These are all ranked Platinum games.

Stats are pretty stable between patches. In general there’s not a lot of change in the game since 5.10. For instance, players averaged a 2.55 KDA in 5.10 and a 2.48 KDA now in 6.17. There is a general trend up in some of the cumulative stats like “goldEarned” or “totalDamageTaken”. That could easily be a symptom of games just taking longer over time; unfortunately I don’t have stats on game length to normalize with. Two highlights though. Experience has a big drop at the beginning of Season 6. I take this to be new accounts and/or people changing up how they play. Also totalHeal had a big bump up around patch 5.22. That could be the same general cumulative stat effect, but I think it may also indicate the addition of more healing abilities and the rise of healing supports like Soraka and Nami.

download (2).png

Champion diversity is pretty stable between patches. Up above I showed the distribution of playPercent across 5 roles for patch 6.17; here’s an animation of that distribution across patches. It’s pretty stable. So are the distributions for all the metrics I’m most interested in; goldEarned, KDA, winPercent, in general the distribution across champs looks the same between patches. I find that a bit surprising because I think of LoL game balance being pretty dynamic with a constantly changing meta. But it’s not really reflected in the stats.