@Alexfrog, I’m also interested as to whether you applied any cuts to your data set?
i.e. discount any games with abnormal amounts of influence (to remove illegal decks, or beginners using the core decks); turn 1 runner wins (since these are entirely luck based and not a true reflection of skill or ID performance).
You also say you discounted Fisk and The Collective, but what exactly do you mean by that? Do you mean that you just disregarded them for the purposes of analysis or did you actually discount those matches? The Collective will have contributed to a significant number of Corp losses. Similarly it will have biased your determination of “strong” and “weak” players - since some mid-range but good Corp players will have lost more than half, and mid-range bad runners won more than half due to a broken ID.
I think if you went back and applied a post hoc ELO (or rating system of choice) you could have a bigger dataset, because you wouldn’t have to restrict yourself to people who have played 20+ games. If someone only plays 5 games vs. high ELO players and wins them all that’s still a reasonable indication of skill.
You could then look at ID win-rates in 5% bands of skill. The data set would definitely then be smaller, but it would be a start. We’ll never have a large enough dataset in a stable meta to make any firm conclusions about anything.
Possibly speak to David Sutcliffe (Magicdave on BGG). He’s very good at stats and probably already has some scripts written from his last analysis.