Hi there! Really sorry for missing this til now, I have been offline the last few weeks.
Yes, I have done analysis on player correlations!
Here are some very rough numbers you can use for mean correlations for assets in the same match.
Yes, I have done analysis on player correlations!
Here are some very rough numbers you can use for mean correlations for assets in the same match.
Comments
Is this based on all players, or is it subsetted somehow?
I've dabbled with reducing the player set down a lot to a version of "all players I might realistically be interested in selecting", e.g. using something like TSB% as a proxy, and it can change outcomes quite a lot.
I remember discussing the balance of granularity (lots of mids have quite different points scoring profiles) with sample sizes (eg FWD-FWD pairs are quite rare), but I’m not sure there was a conclusion/next step.
EO sounds a good filter too I agree!
Made me "remember" there are a whole set of MIDs that play all the time that are still non-entities for FPL.
Could bucket players within positions by G/A output I suppose? Might hit the sample size issue though
There are also specific correlations that could be tested like set piece takers and aerial threats, or foul drawers and penalty takers maybe