I’m a little off my game to start the new NFL season, as evidenced by the fact that in one fantasy league I drafted two guys who aren’t even on teams (to be fair, it’s a super-deep league where each of the 10 teams has 30 guys) and in another I found out that my kicker was hurt when I was out at dinner and the Packers-Saints game came on (to be fair, it was my kicker). So with apologies to missing the game last night, here are some predictions for how the first weekend should play out.
I’m going to follow the same method I used last year, which involves predicting week 1 winners from the team’s record/point differential in the previous year. This is very crude, but you can read that post for some discussion on why predicting from one year to the next is virtually impossible anyway. For some comparison, here are Bill Simmons’ picks for week 1. They’re against the spread, but since the winner also covers the spread most of the time, you can view my winner choices as spread choices as well if you’d like. As a side note, I’ll be making my predictions throughout the season like I did last year, and I’ll post Bill’s as well if he makes them public again. I might also add the win probabilities from Advanced NFL Stats, although they only cover a certain range of the season.
I went a little further this year and fit five models to the data: a ‘wins’ model that predicted the week one winner from the home and away teams’ wins last year; a ‘differential’ model that did the same but using point differential; a ‘home’ model that used only the home team’s info (since last year that was what was significant); an ‘away’ model that used only the away team’s info; and a ‘full’ model that used both home and away wins and differential. I compared the models using a likelihood ratio test as well as leave-one-out cross validation to see if the home-only result was still true with an extra season in the data, and all the models performed virtually identically with a slight preference for using both teams’ number of wins.
For simplicity then, and because it presumably won’t change the accuracy much, I’m going to use last year’s win totals to predict week 1 victory. I should note that the away team record was again not-significant, but I included the term anyway. Here goes:
There are a few games just begging to turn out differently. For example, with Peyton out, Houston should probably be a bigger favorite over the Colts than basically 50/50. You’ll also notice there’s still a strong preference for the home team, with only two road teams being favorites (the Patriots and Raiders) and even then not by much. For that reason I might not be so bullish on the Rams over the Eagles, for example. But all in all, football is a tough game to predict. The model already got the Packers pick correct, so we’ll see how the rest go in the next few days. Football!!!