It's too damn hot. Thank goodness for AC.
RECORD COMPARISON, 2007-2015
Record vs Top 25
Win Lk Comparison
|Conf||Record||C-Rec||Tot Rat||rk||Adj Rat||rk||Res Rat||rk|
|# Tms||Prj Rec||Prj C-Rec||Adj Off||rk||Adj Def||rk||Prj SOS||rk|
|Perf W||1 Loss||Bowl||Ret Off||rk||Ret Def||rk||Rec 3yr||rk|
|Gm||Wk||Date||H/V||Opponent||Opp Rk||FCS||C-Gm||Prj Lk|
|Opp Rk||Team||B10 Avg||Opp %ile|
One of the inevitable truths that you have to face when getting in the hobby of doing up projections is that you will be wrong. The hope is that you're not wrong all that much, but every year some team like justNorthwestern will come along and bust out a 10 win season on the basis of potato magic. It happens, life goes on.
Still, it's worth taking an evaluation of how pre-season guesses compare to final season results. While I'm sure your favorite team is unique and special, I've always favored taking a look at general trends, so you'll just have to deal. We're using 1990-2015 for the years in question here. First, how my projections fail at various percentiles ranks (so the better pre-season teams are on the right):
(RMSE is root-mean-square-error, in case that needed clarifying.) The first thing you'll probably notice is a definite trend in the average error between pre-season and final ratings. Highly ranked teams in the pre-season tend to under-perform, shlt teams in the pre-season tend to over-perform. Ideally, this trend would average zero across the board, but given how my pre-season ratings are made I can understand why this exists.
Pre-season ratings are mostly based on recent history. Teams highly ranked in the pre-season probably performed rather well in recent history. Unfortunately, some of those teams will fall victim to inevitable regression, hence the trend you see here (flip it for the crap teams). I used to have a regression coefficient added into my pre-season guesses, but for various reasons I decided to drop it. C'est la vie.
Probably more interesting is if teams can consistently over/under perform their pre-season ratings here. Again, we'll look at over-all trends, but feel free to hold on to your anecdotal accounts as proof:
I am now thinking I should've provided a count N in the figure somewhere. Whelp, too late now. It's probably somewhere in the ~2500 range.