Oracle's Elixir

Win Bias and Damage Per Minute

In response to a Reddit post yesterday which listed the top Damage per Minute numbers from the NA/EU playoffs, one Redditor made this comment:

“This data is very skewed because CLG only won and never lost. It would be a lot less so if you just compared DPM from wins or just losses.”

I thought this was worth responding to in a bit more detail, to help people understand the concept of “win bias” and why, in my opinion, DPM isn’t one of the stats it heavily affects. Here’s an expanded version of what I wrote.


In League of Legends, there are some statistics that are a consequence of winning, rather than a cause of winning. We can call these stats “win-biased”.

For example, Towers per Game (TPG) can be misleading because the winning team usually gets some free Towers as a consequence of winning. They may be ahead 8-6 in Towers, then win the game-deciding team fight and go pick up two or three free Tower kills en route to destroying the Nexus. Those extra Towers didn’t really help them win the game; they were just a consequence of winning. The really useful number we want to know is the 8 Towers they had before they won (but that’s difficult to pin down). So ultimately, part of the TPG stat is just measuring whether or not your team won.

A win-biased stat is not a useless stat: I don’t think we need to throw a stat out just because it’s win-biased, which is why I still include TPG in my team stats tables. But we do need to be very careful how we use those stats in our analysis and team/player comparisons, because the numbers may not quite say what we think they’re saying.

Now to damage (to champions) per minute (DPM): in my view, DPM is not a heavily win-biased stat. Doing more damage is a big part of why you won: you don’t win and then do more damage.

I want to be careful how I present this, though. When I say DPM isn’t win-biased, I’m not claiming that there aren’t meaningful differences between the DPMs of winning players and losing players. There is a demonstrable DPM “win bias” from a purely numerical standpoint.

Note: Based on all games currently in my database.

But this gap between winning and losing DPMs does not mean that DPM is a win-biased stat when it comes to interpretation. As I said above, damage output is something that leads to victory, not something that is incidental after you’ve already won (like killing Nexus Towers).

The one angle on DPM that does hint at possible win bias is when a team wins the early game hard, getting themselves fed, which then increases their damage output. As a player, if your team has a large lead in a game, your superior damage in team fights is then a consequence of that lead. But by getting fed, your team earned that damage increase, and your higher damage output is now a key part of helping you win the game. It can be argued that some games are a foregone conclusion halfway through, and the rest of the winning team’s damage output is now just win-biased damage, but in my opinion, that doesn’t negate the value of that damage: it’s still earned damage.

To be clear on another point, I’m also not defending DPM as a perfect or clean statistic. There are many ways DPM is not an ideal stat, but win bias isn’t one of my DPM concerns.

For the record, some things that do concern me about the validity of the DPM statistic are the following:

Some of these things can be addressed by looking at damage share whenever you look at DPM. Other questions take a bit more digging and reviewing to address. But just because there are questions and context to consider it doesn’t mean the statistic is not useful or meaningful.

TL;DR: Win bias is real, but it doesn’t render stats useless: we just need to be more careful in our interpretation. In my opinion, DPM is not a win-biased stat.

I’ll continue to explore the subject of win bias and the other kinds of context behind various statistics, and I welcome your thoughts and input into how we can best use these numbers to inform our analysis.