130-team S&P+ rankings: The gap between Bama and Clemson is now tiny

Below are the new S&P+ rankings with just one game remaining in the 2018 college football season

Good predictive ratings are not résumé ratings, and they don’t give you bonus points for wins and losses. They simply compare expected output to actual output and adjust accordingly. That’s how a given team can win but plummet or lose and move up.

The S&P+ rankings have performed pretty well this year, hitting 53 percent against the spread and 52 percent on the over/under point totals for the year.

(You can find full unit rankings, plus a yearly archive, at Football Outsiders. The offense and defense pages are updated by Monday at the latest.)

Five days before the national title game between Clemson and Alabama kicks off, the Crimson Tide are around a 6-point Vegas favorite. General sentiment I’m seeing on Twitter is that the Tigers will be lucky to stay that close, even though gamblers moved the early line a little bit in Clemson’s direction. It appears Alabama has beaten a lot of fans into assuming another blowout like we saw in the 2017 semifinals between these two schools (a 24-6 Bama rout).

The numbers agree with the people actually laying the money, however. ESPN’s FPI separates the two teams by merely 0.3 points and gives Alabama only a 52 percent chance of winning; S&P+ is nearly as bullish on Clemson’s chances, putting the Tigers only 1.0 adjusted points per game behind the defending champs, which also projects to about a 52 percent win probability.

Better yet, if you’re a Clemson fan: if we look only at Clemson’s last nine games — i.e., the games in which quarterback Trevor Lawrence was both the starter and healthy — the Tigers’ S&P+ rating rises to plus-30.2 adjusted points per game, 0.6 points ahead of the Tide. Granted, if we look only at the games in which Alabama’s Tua Tagovailoa was most healthy, we see a similar bump.

But while our heads always assume Bama’s going to win any game easily, the spreadsheets suggest otherwise.

Bowl season was incredibly all-or-nothing, with 16 games decided by a touchdown or less and 14 decided by 24 or more. There are typically a few extremes during this time of year, but that’s impressive.

And it offered a bunch of teams opportunities to move up by a decent amount. Twelve teams improved their adjusted points per game average by at least 1.6 PPG, including four teams now ranked in the teens.

On the flip side, 12 teams regressed by at least 1.6 adjusted points per game, including two top-10 teams in Michigan and UCF.

Both teams were far enough ahead of the rest of the field that their rankings remained pretty static — UCF remained eighth because the No. 9 team (Washington) also fell a bit, and Michigan fell only from fifth to sixth because the previous No. 6 team (Notre Dame) also regressed by a hair. Still, know that each of these team’s statuses fell a bit.

The stats back up what the eyes saw: the AAC had a horrid bowl season — the league went just 2-5, earning wins over UL Lafayette (Tulane’s) and Virginia Tech (Cincinnati’s) and losing five other games by an average of 23 points (15 without Houston’s miserable performance). Most of the other leagues were pretty up-and-down.

By the way, I will be revealing a new layer of S&P+ adjustments later in January. It appears I’ve found a way to make projections more accurate by figuring out stronger conference-level adjustments. That’s good for me and bad for Fresno State and other members of mid-major conferences who will see their rankings slip a bit when that adjustment is made. So stay tuned for that.

One last note: the 2019 season preview series — which begins in February and goes in order from worst conference to first — will begin with the MAC for what I believe is the first time. Usually the Sun Belt finishes last in average S&P+, and when it doesn’t, it’s Conference USA instead. But the Rust Belt league takes the ignominious distinction this time, and with ease.

You’re on the clock, Bowling Green.

When I get the chance (so, maybe in the offseason), I will update previous years of S&P+ rankings to reflect these formula changes, too.

Basically, I’ve added a step to the the rankings process: after the rankings are determined, I go back and project previous games based on those ratings, and I adjust the ratings based on how much the ratings fit (or don’t fit) those results.

The adjustment isn’t enormous, and it diminishes dramatically as the season unfolds.

Testing this process for past seasons improved performance against the spread a little and, more importantly, decreased absolute error (the difference between projections and reality) quite a bit. I wouldn’t have made the move if it didn’t appear to improve performance.