Marksman strength over time

tl;dr Is Jinx actually a late game hyper-carry? If not, who is?

Champion power at different times during the game is one of the most interesting and difficult-to-quantify ideas in League of Legends. This week I have taken some baby steps toward the holy grail of plotting each champion's power-level (whatever that is) in the early, mid, and late game.

The data for this post includes 736,383 NA ranked solo-queue matches played in season 5 (that is, on patches 5.1-5.6) by players of all skill levels.

I made these plots with the following procedure. I first filtered by marksmen. These are players that spent the majority of the early game in the bot lane, bought carry items, got much more cs than their support in the early game, etc... The heuristics are not perfect (obviously), but this data does not include any spurious top-lane Ashe nonsense. Then, for each such champion, I put each match in a bucket based on its length. I ignored games shorter than 20 minutes (~1% of games end in less than 20 minutes in my dataset), and I also ignored games longer than 60 minutes. Then, I split the time from 20-60 minutes into 2-minute buckets. So, 20-22, 22-24, etc...

Then, for each champion, and each bucket, I calculated that champion's "Advantage". This is basically a simple conditional probability of that champion's team winning given that game length. If some champion had their team win every game in that bucket, the advantage would be 1, if they lost every game in that bucket the advantage would be -1. As you will see; this, modulo some difficulties with sample-sizes, gives a nice visual that is hopefully related to champion strength over time.

There are many assumptions and conflating issues in doing an analysis like this. League of Legends is a team game. Having some specific champion on your team far from guarantees success or failure at any certain point of any given match. Individual game-time decision making and mechanical skill always trumps champion select. But over a large enough sample set, some of those differences wash out, and we can begin to glimpse some of the underlying structure of champion balance.

Take a deep breath, and enjoy the plots.

Here are some plots of ranged ad carry champions that enjoy more success in the early game:

Corki and Ezreal fall off quite hard in the late game, which agrees with common conceptions of those two champions, I think. They can have a big effect on a game with certain item power-spikes (e.g., Trinity Force), but no one expects them to take over and win 1v9 at 60 minutes.

The shape of Jinx's plot is one of the most surprising results of this analysis. She has a good win-rate over all (most of the bars are above the x-axis), but she wins shorter games more. The consensus in the community is that she is a late-game hyper-carry, but this data is not confirming that.

Sivir also maintains a decent advantage at different points in the game. There is a clear downward trend for her as games go on, though.

Kalista and Varus have similar plots, both definitely trending down over time. The very last bar on Kalista's plot is likely spurious. Despite having over 80,000 matches played by her in the data, only 93 of those matches ended between 58 and 60 minutes (~.1%). She does have a positive win rate between 56-60 minutes in this set of data, but I would want many more super-long matches of hers before I drew any conclusions.

Here are four champions that peak in the mid-game:

Ashe has the clearest mid-game strength out of the champions in this dataset.

Graves and Lucian have interestingly similar plots. This data actually kind of supports the hypothesis that Graves and Lucian are actually Graves and worse-Graves. That first bar on Grave's chart is pretty severe. In this case, it may actually carry some weight, as the sample-size on that bar exceeds 20,000 matches. Note that the y-axis scales are different for each plot here. Graves may force the most surrenders, but he never really carries that large of an advantage.

I did not know where to put Draven in this post. The plot makes it look like he has both early and late spikes with a slight lull in the middle, but he is not played frequently enough in the season 5 meta to get really good data on him. As ever, read what you will into that plot.

Here are two more champions with less clear moments of strength, but that appear to lean toward the late game:

Again, make sure to check out the y-axis scales. Miss Fortune has much larger swings than Quinn, even though the plots appear similar. The wild swings on the right hand side of Miss Fortune's plot are unfortunately most likely just an artifact of too-little data. Though I have close to 20,000 matches played with her, those last two bars are only based on ~150 games. Though, for both her and Quinn there does seem to be some upward trend in the charts.

Here are some plots that indicate late-game strength:

Caitlyn and Kog'Maw have win rates that are relatively flat until the super-late game. This agrees with my personal experience on these champions. One needs a significant number of items (and more than other bot lane carries) in order to take over a game with these two.

Twitch, Tristana, and Vayne all have clear upward trends in their plots. Sadly, for Twitch, I have essentially no games longer than 54 minutes. The last three bars on his plot are made from vanishingly little data. It is sad to see Tristana's win rate so low after her season 5 changes. I wonder if they will buff her for the summer split. Regardless, it still looks like a good idea to take games late with her.

And this Vayne plot is a good one to end on. This chart matches my expectations about Vayne quite well. The longer the game goes, the more she can do for your team.

The sun sets on another post. Thanks again for reading, and making it all the way to the bottom. I literally could not do this without you guys.

I think there are some interesting insights to be had, here. Where is the late-game strength of Jinx we hear so much about? Was the Tristana rework an epic nerf in disguise? Rito pls. I am sure I missed a lot here, so get at me with your ideas and questions. Would a similar analysis of other roles be interesting? I think so, but are there any obvious improvements that could be made?

Until the next post, follow this three step guide:

1. Go to the homepage and read everything in sight.
2. Subscribe to the rss feed so you can get the next post as soon as possible.
3. Play more ranked solo-queue so I have lots of data to chew on.

And then hold on tight, because I will be back soon with more League of Legends math insights here at LeagueMath.com.

No instalock Teemo adc or feed players were harmed in the making of this article.

Peace.