Thanks for bringing up this discussion. To speed things up, here is the relationship between EAPMS and ELO, taken from our database. Please note that only ranked 1v1 RM matches were taken into account and that there might be a bias because only matches someone generated an analysis for were used.
Very interesting, thank you for sharing. So, like most activities, performance is multi-factorial, correlated with EAPM, but far from completely determined by it. For EAPMs at the lower end of the spectrum, there appears to be an ELO ceiling that it is very hard to exceed with that EAPM. E.g. with an EAPM around 20, it seems that pretty much nobody manages to exceed 1700 ELO.
One very important thing that I want to highlight, though, is that it’s easy to fall into the trap of thinking “the fact that there are players at 2200 with 30 EAPM shows an EAPM of 30 is no obstacle to reaching that level”. This is not a valid inference. The players who are at 2200 with around 30 EAPM are performing at the same standard as people who have over 60 EAPM by virtue of being exceptional in other ways that are compensating for their lack of EAPM. These other factors are almost certainly affected by both talent and training, such that not everyone will be able to achieve the same level of performance with 30 EAPM, as they may not have the same talent for the other factors.
It would be interesting to analyse other factors in performance via automated analysis, but obviously many of the other factors that differentiate players are massively harder to automate an assessment of. There are some that would be fairly simple, such as dark age TC idle time, and time spent housed. In isolation, these things would be much like EAPM - correlated with ELO, but with a spread at each ELO level. But if they were combined into a weighted score, I’d expect the relationship to become a little tighter, the more things are accounted for, i.e. some people may waste their high EAPM by doing the wrong things and still having TC idle time or getting housed, which would lead to them having above average EAPM for their ELO, and a weighted score would capture the different factors cancelling out.
Thanks for sharing your thoughts. I also would like to add, that the chart displays certain players within certain matches. In my humble opinion, that has some important implications, e.g
The chart says nothing about winning or losing a specific match, so a high ELO player might underperform in a certain match and also gets punished for that
The dataset contains also very short/aborted games (< 1 minute), which does not represent the overall EAPM performance during a full match
Therefore I don’t think there is much value in focusing at the outliers. Also I believe there is a substantial dependence between decision making and EAPM in the way that a player might be capable of a higher EAPM, but incapable of making more decisions.
We are analyzing the saved game files and count every action associated with a user. Actions in the sense of a saved game are whatever has impact on the game.