By Doug Liebe on 2019-03-22

Introducing Codstats Rating

Plain and simple: we have been using KD as a catch-all statistic for too long. Having a KD of 3 only means that you killed 3x as many people as killed you. This could mean you went 3-1 or 30-10. A high KD doesn’t always mean that your team wins, either. Players frequently cite KD as a reason for getting cut from teams or a reason why they are the best player, but the truth is: if you are not getting your team WINS you are not helping the team.

Our Codstats Rating is the probability a player gave his team of winning, based on his performance, divided by 50% (league average win%, obviously). This is done to make the average Rating = 1.0. This system is done much the same vein as the csgo website's Rating 2.0.

What is Codstats Rating? Rating takes all your stats from a game and tells you how likely an average team would be to win given your performance. I will go into more detail on the formula for Rating, but let's pretend for now that the formula, f, is simply:

_f(all stats) = Rating_

Importantly, the stats that go into Rating do change by game mode. Another shortfall of KD is that kills and deaths alone are not as important in Hardpoint as they are in Search & Destroy, where getting 5 kills wins the round. Rating solves this problem by allowing us to change the Rating function by mode and by player type.

Subs and AR players play very different roles in BO4, and our win probability models indicate a significant difference in the statistics that best describe their contributions to a team win. Rating allows us to break Sub/AR play apart, while still comparing all players on the same scale, win percentage. That is the most important takeaway from this whole section: every game, every type of player gets a Rating thats on the same scale 0-2.

Block Diagram of Ratings Algorithm Inputs

What goes into determining Rating?

Rating is broken down by mode and by Sub/AR role. Logit models are built to predict win probability for each of the 6 combinations (HP x sub, HP x AR, CTL x sub, … ). The predictions from the models are used to generate Ratings. Combinations of the following rate statistics are used in the 6 models to predict win probability:

  • SPM - Score per minute
  • DPM - Damage per minute
  • KPM - Kills per minute
  • DeathPM - Deaths per minute
  • APM - Assists per minute
  • KD - Kill Death Ratio

For now, the exact combinations of the 6 models will be kept proprietary, but we will consider updating in the future.

Is this better than KD?

Yes. Teams are trying to win games and the team with the higher KD is not crowned the champion. By modeling past performance using all of Black Ops 4 stats to train our models, Rating performs significantly better than KD.

Is this the best thing?

No. With any stat, attempting to summarise performance with one number will have shortcomings. This stat will not encompass the shot-calling prowess a player may possess or the leadership a team needs to succeed. Conversely, a team with great Ratings may lack patience or finesse. However, we firmly believe the Codstats Rating system is a huge step forward for the entire Cod Community in terms of valuing players.

All single-mode Ratings are available in our leaderboards for CWL Fort Worth and will be available going forward. *Overall Ratings are the average Rating over all games played, which is to show the average win probability of playing with a player, not the sum of all his scores. Summing stats would show the win% of the “average round” which could be biased by lopsided results.

An Example

Here are two players compared by their performances at CWL Fort Worth:

Comparison of Unknown Players at CWL Fort Worth

By KD, the right-hand player had a much better event. I think we all know why I picked this matchup, so let’s dig deeper and look at the underlying stats by percentile in each mode. (*Note: in Deaths per minute - higher percentile is good still, the scale is inverted to keep things sensible)

Comparison of Per Minute Stats of Unknown Players at CWL Fort Worth

So players were much more even in categories that matter. The left-hand player had more score per minute in SND and HP and fewer Deaths in Control. The left-hand player might be described as playing more at the extremes, high score and more deaths, where the right-hand is lower score and fewer deaths (except in control). Keep in mind that more than just these two variables go into the Codstats Rating, this is just a cross-section.

So who are these two players?

Comparison of Ratings between Nagafen and Sukry at CWL Fort Worth

Obviously, in general, aKD correlates more strongly with Rating, but I choose an example where it did not to show where KD can be misleading. Nagafen plays his role very well and gets very high win percentages for his team in the process.

Here is a graph of every player at CWL Fort Worth in overall aKD vs. Rating. The blue line is the average relationship between the two.

Graph of aKD versus Ratings for all players at CWL Worth

Above the line = aKD doesn’t give them enough credit; Below = Kills alone not helping the team enough; Near the line = aKD and Rating similar descriptors of talent

Codstats Ratings added in all player categories in our leaderboards