I've a data set that contains 10 years worth of player & team data by game. I'm in the works of developing models to predict players stats (i.e. PTS, AST, TRB, BLK, STL, & TOV per game) and I'd like to gather the thoughts of the APBR Community:
-what should I control for? i.e. control for min by converting to PTS per 36 MIN
-what should I include as predictors? i.e a team with a high average PTS scored against might predict player X will score more PTS when playing against that team
if you're kind enough to provide some sound advice & I'd be happy to email you a zip file of the raw data sets for you to play with.
if you really want to work at building a strong predictive model & are willing to share it I'd be happy to create you your own custom data sets.
Without A PBR what would you do on the weekend?
Predictive Models...
Re: Predictive Models...
With the exception of brand new rookies, I think just about all systems use as their main predictor some measure of what the player did before (in previous seasons or previous games this season), possibly regressed to overall league averages (RAPM does a form of this). So then it's a question of what factors might cause us to change those predictions. You can pretty much just name all the usual things which we take into account either formally or intuitively when trying to predict who's going to win a given game: home/away, back-to-back games, quality of opponent, game pace of opponent, etc. Where it gets more interesting is when we start looking at individual players: matchups (is someone going up against a particularly strong or weak defender), injuries, and on a season-to-season basis, age and/or experience. Most of these things might also be affected by player type, either the player's discrete position or a statistical description.