Your experiment has a flaw in that it's not taking into account other factors, such as:
1) Consistently playing in a party
2) Game Mode Normally played
3) Mid-match quitting
A high K/D in any event doesn't necessarily correlate into a high W/L, nor does a high W/L correlate to a high K/D. There's just too many underlying factors that your experiment/analysis doesn't factor in to be accurate.
Sure I know the flaws, but the only way to get rid of them would be to do quite a bit of data work... which would just take too much time. sometimes... in analysis you have to make assumptions... and simplifications to account for the real world problems of data analysis. This analysis if I get a large enough sample... would probably tend to the average beta anyways.... mitigating the problems you listed.. eq 1. is y = xB + error Test for correlation of x and error, and correct with AR(1) if a problem exists. Use a t-test for significance etc... Hypothesis is that there won't be a statistical relevance.