As many of you know, I did a thing where I averaged out 6 of the major NBA stats to form a "consensus player rating", which was/is at a tab on the left. I decided to redo it, but to try and weight each metric in the blend pool, as opposed to simply averaging them together.
The relative weight of each metric was determined by averaging its correlation to past wins with its correlation to future wins, thereby rewarding accuracy. Correlations for future wins was pulled from Neil Paine at APBR, and I did the past wins correlation for 2012-2013 myself.
I dropped EZPM from the pool, leaving us with Wins Produced, Win Shares, PER, ASPM, and XRAPM. I standardized each metric, summed and regressed them on 2012-2013 wins, weighted them, and added them together.
Surprisingly, there wasn't a huge difference in weights. PER had the lowest, at around 17% of the blend, and XRAPM had the highest, at about 22% of the Blend.
I'm also regressing each metric and its components (O-XRAPM, D-ASPM, etc.) on "basic" stats like TS%, ORB%, and OFF EFF, to see what makes them "tick". That will be released later in the week, but I just wanted to let everyone know what's up.
The new "Blend Score" Ratings for the Top 40 players will be posted on a tab at the left, in place of the Consensus Ratings of the past. The table includes the standardized Blend Score, as well as each players WP48, WS/48, ASPM, PER, and XRAPM, also standardized, so you can compare across metrics.
The relative weight of each metric was determined by averaging its correlation to past wins with its correlation to future wins, thereby rewarding accuracy. Correlations for future wins was pulled from Neil Paine at APBR, and I did the past wins correlation for 2012-2013 myself.
I dropped EZPM from the pool, leaving us with Wins Produced, Win Shares, PER, ASPM, and XRAPM. I standardized each metric, summed and regressed them on 2012-2013 wins, weighted them, and added them together.
Surprisingly, there wasn't a huge difference in weights. PER had the lowest, at around 17% of the blend, and XRAPM had the highest, at about 22% of the Blend.
I'm also regressing each metric and its components (O-XRAPM, D-ASPM, etc.) on "basic" stats like TS%, ORB%, and OFF EFF, to see what makes them "tick". That will be released later in the week, but I just wanted to let everyone know what's up.
The new "Blend Score" Ratings for the Top 40 players will be posted on a tab at the left, in place of the Consensus Ratings of the past. The table includes the standardized Blend Score, as well as each players WP48, WS/48, ASPM, PER, and XRAPM, also standardized, so you can compare across metrics.