I was just auditing the predictions from this season and it looks like there's a problem with the home-court adjustment that's causing home teams' predicted scores to be about 6 points higher and road teams' predicted scores about 6 points lower than they actually wind up being. I did some tinkering with the model at the start of the season in an effort to improve accuracy and I must've stepped on something in the code when I was mucking about in there. I'll try to figure out where the problem is tonight, but I won't be posting any daily predictions until I get everything sorted out.
Update (11:08 PM): The reason everything was so symmetrically wrong was that I was comparing home team predictions with the actual results of the visiting teams and vice-versa. The lesson here is simple: I shouldn't try and do a quick audit of the data on my lunch break.
No comments:
Post a Comment