Clairvoyance Blog

Article by Bilby & Jeopardy, Design by Nancymon

PICKING YOUR CHAMP IS A BIG DEAL. A champ can motivate you to play, it can carry you to challenjour, it can say something about you (for better or worse) even before you load into a game. More often than not, the other players in the game don’t even call you by the name you’ve given yourself, but by the name of the champ you’ve picked—hopefully preceded by adjectives like “amazing,” “godlike,” or “super fed.”

For us (hi, we’re Jeopardy & Bilby), a champion can reflect how you connect with the game: Jeopardy came from a fighting game background (WHEN’S MARVEL), so he instantly clicked with Riven's combo-heavy playstyle. Bilby, meanwhile, started playing Fizz because he’s a Moby-Dick nerd, and Tundra Fizz had a whale ult.

There are plenty of perfectly valid ways to pick a champ, but what do those choices say about you? And, once you've found one that clicks, which other champs are you also likely to play, and what ties those champs together? Do they split into cliques, like kids in a cafeteria? If so, who would sit together? Thanks to data that we gather during every game, we've found some answers.

---

Here's what we analyzed

JEOPARDY'S DATA CORNER: We sampled from the population of all players who played at least 25 games during the first half of March this year. The key stat we looked at was "number of games played on each champion" for each individual player.

In the data we gathered, two champions are correlated if players tend to play those two champions together—if many players play Riven and Azir, then Riven and Azir would have a higher correlation. If many players play Riven but proactively avoid playing Malphite for whatever reason (perhaps because you perceive him as too easy and binary and rewarding for being passive in lane, for example), then Riven and Malphite would have a much lower correlation.

Node

Edge

To help us better understand these champion correlations, we created a network visualization of the champions, which you can explore below. We call this handy tool the CHAMP FRIENDSHIP COMPATIBILITY DEVICE (CFCD). The CFCD network graph consists of nodes and edges: a node represents a single champion, and an edge is a line that links two champions together. The edges of this graph are undirected—a line from Annie to Olaf is treated the same as the line from Olaf to Annie. For the network graph, we filtered out all correlations that were below a certain threshold, so only stronger relationships would appear in the visualization.

Cluster

Correlation

---

CHAMP FRIENDSHIP COMPATIBILITY DEVICE

For your research and entertainment, here is the tool we used. You can explore the champ groupings (sorted by color), or look at individual champs and their strongest correlations to other champs (shown at the bottom, in order). Maybe you'll find a new champ to try out!

---

Here's what we found

We analyzed the modularity of the network to sort champs into communities based on playrate correlations. Modularity measures the strength of a community within a larger network—high modularity means that communities are well-defined, such that nodes within a community correlate stronger with each other than with nodes from other communities. We also factored in edge weights: stronger correlations were weighted heavier toward defining tighter communities, and weaker correlations were weighted less.

Okay class, are you still awake? If you wanna stretch and walk around for a little bit, that’s cool. Here we go: Segmenting a larger population into groups like this is useful when you’re trying to identify similarities and differences between data points—in this case, we can look at the clusters to find relationships between champions based on play rate. Long story short: if you play a champ in one cluster, you’re more likely to play other champs in that cluster than ones that aren’t. Here’s where they fell:

---

A few funny things that jumped out in the details

While poking around the CFCD, we found some other interesting groups that we wanted to share.

---

What did we learn?

Unsurprisingly, it does seem that a champion’s role has a high impact on which group they fall into, which makes sense as players get better in specific positions and start to specialize. But it’s also clear that a champ’s kit isn’t the only thing that matters. There are clusters that span roles like the high-skill-ceiling club, or ones that seem to be based on a champ's (un)popularity. Looking at this data gives a great window into what different types of players are out there, but the reasons that a specific champion resonates with a player seem to be as diverse as players themselves.

What did you find when playing with the CHAMP FRIENDSHIP COMPATIBILITY DEVICE? Let us know in the comments below.