Thursday, December 4, 2014

T-Rank Note

In case you care, I've rejiggered the T-Rank algorithm a little. The changes:

1) Originally preseason ratings had an influence through 10 games. I've upped that to 13 games (or so - because, as I explain below, some games don't count as a full game now, preseason ratings will probably still have an influence till the 15th game or so for most teams).

2) I've rigged it so blowouts and mismatch are counted less than games between well-matched teams. I noticed that the Pythag of the upper teams was getting out of hand. For example, T-Rank was projecting Kentucky as likely to go 30-1 and 18-0 in conference. Those just didn't seem right to me, and de-emphasizing mismatches make the predictions look better to me.

3) This hasn't kicked in yet, but eventually more recent games will count more than old games. (Games within the last 40 days all count fully; games between 40 and 80 days ago drop down 1% per day; games 80 days and older are counted as .6 of a game.

I'm done tinkering with it for the year.

No comments:

Post a Comment