It probably shouldn’t be a surprise that S&P+’s absolute error (the absolute value of how much you miss each game by, basically) in a given week is tied pretty closely to its success against the spread — the closer you get to getting the actual score right, the better you’ll do against Vegas’ estimates, right? Makes sense.
A large portion of the nerd crowd will tell you that absolute error is a better indicator for how you’re doing from year to year; it’s more under your control ... or at least, it’s less out of your control.
Still, there’s an obvious correlation. Using my latest method for S&P+ projections, here are the averages since the start of 2014:
- Absolute Error (for a given week) between 11 and 12 points: 55.4% against the spread
- Between 12 and 13 points: 53.5%
- Between 13 and 14 points: 51.6%
- Between 14 and 15 points: 50.5%
- 15+ points: 46.0%
(Note: From an absolute error perspective, the most successful systems in a given year tend to settle in the low- to mid-12s.)
The error for S&P+ tends to float between about 13.5 and 15 points for the opening weeks of a given season, as S&P+’s preseason projections are exposed as accurate about some teams and drastically inaccurate about others. In theory, it goes down from there, hopefully eventually settling in the 12-13 range.
Example: Last season the absolute error was about 13.5 for the first nine weeks of the season, then 12.4 over the next four. (It then popped up to about 14.0 for the postseason for one reason or another.)
It was in the low 14s for each of the first three weeks of 2016 — not a great number and certainly a sign that the preseason projections weren’t as effective as I’d like. It’s been drastically off in games involving Ohio State, Louisville, Memphis, Army, Houston, Bowling Green, and Marshall in particular (average error for each is over 26 points per game, which is pretty incredible), while it’s had teams like BYU, Pitt, Utah, South Carolina, UCLA, and UConn pretty much nailed (average error under 6 for each).
In terms of absolute error, S&P+ broke through into the low 13s last week. A nice start.
Oh yeah, and S&P+ went 20-32 against the spread. A cool 38%. Good times. Good, good times.
Ball State and Appalachian State covered by 0.5 points. Temple, Utah, and Boise State covered by one. S&P+ went 0-5 in those games. Stanford scored on a fumble return as the buzzer sounded to win by nine and cover. S&P+ had UCLA.
In the games in which one team covered easily (10 or more points), S&P+ went 12-10. In the games that finished closer to the Vegas line, S&P+ went 8-22.
It was a really dumb week, and it made me a little bit insecure. But win probabilities are faring as they are supposed to, and the error really was in a decent place. So I’ll take a deep breath and proceed.
Thus far it appears that S&P+ has been a little lucky in Week 2, a little unlucky in Week 3, and a lot unlucky in Week 4. And if the absolute error is any indication (and it should be!), performance should regulate pretty soon. Here’s to hoping because a 20-32 week knocks you into a pretty deep hole for the season.
Here’s the updated Google doc with all picks. And here’s a completely useless embed, just for fun: