New box score metric: Estimated Impact

[shutupandjam]

Just wanted to share my box score metric, Estimated Impact. It is based on a regression of elements from the box score against long term RAPM. It is better at predicting future outcomes and, in my opinion, more reasonable than other box score metrics (even other SPM metrics). And obviously, it's much more stable and less biased than regression based plus minus metrics (like APM and RAPM). Plus I went all the way back to '52 with it, so it gives a nice historical perspective.

Read more about it here.

And see the results here.

[blabla]

What exact elements from the boxscore? Are you just doing straight up linear regression or are you using interaction terms?

[shutupandjam]

What exact elements from the boxscore? Are you just doing straight up linear regression or are you using interaction terms?

I'm using a number of interaction terms, and I'm not going to go into exact specifics at this point, but I will tell you that pts*ast (adjusted for pace, etc. obv) was shown to be very significant (which is in line with what some other folks have found).