What irked some RAPM fans however, was the new defensive component of XRAPM, which rewards players simply for being tall and playing in the post. Players like Al Jefferson who were considered mediocre defenders, were suddenly vaulted to the top(ish) of the XRAPM charts, not necessarily for their play, but because of their size/position.
With that in mind, I decided to take a stab at position adjustments for XRAPM (2013 season only). My first pass used the position designations from the NBAGeek, which for the record includes 9 different categories. The results weren't that good. There was only a 30% correlation between the position adjusted XRAPM values, and the original ones. This was primarily due to the Defensive component of XRAPM, which after position adjustment, only had a 14% correlation to the original.
Remember, the goal isn't to warp XRAPM to the point where its unrecognizable, its to try and correct for some of the position bias.
After encountering similar problems with traditional 5-man positions, I decided to use only 3 position groupings. Guards (Irving/Harden), Wings (Durant/Batum), and Bigs (Boozer/Dwight). This proved to be a wise move, as the correlation between positioned XRAPM and the original was now 94%.
A simple glance at the "positioned" XRAPM table reassured me. Now guys like Dwyane Wade weren't buried beneath guys like Amir Johnson and Omer Asik.
I then tested the explanatory power of XRAPM and Positioned (P) XRAPM for the 2013 season.
Original XRAPM
Position Adjusted XRAPM
I didn't test the prediction capabilities and I don't plan to. At worst, the position adjusted XRAPM is way off, and at best, it correlates identically to the original XRAPM's prediction power. I suspect its somewhere in between, although much closer to the latter.
I'll be posting position adjusted XRAPM scores for the top players in a tab at the left. I plan on going farther with this subject, but I'm content with this first-pass, for now.
*Original XRAPM's correlation to wins for 2013 (91%) IS higher then original RAPM's 86%.
*Based purely off explanatory and predictive correlations, D-XRAPM (original) is probably the best defensive metric in the world, surpassing D-ASPM (my previous favorite).