Projected CFP Rankings: Week 10

The first College Football Playoff rankings of the season will be unveiled tonight, and while they will of course change as more results come in, this is one of the most informative sets of rankings we’ll see all year. In future weeks, the committee will be reacting to individual weeks of data points. This week, they’re reacting to two thirds of most teams’ seasons.

The core piece of our college football model is how it estimates the committee ranks teams. We have a detailed explanation of how the model approaches this estimate, but the two high-level things to know are these:

• It’s 36-for-36 in ultimately predicting the playoff teams.
• It adjusts each week, reacting to the committee’s rankings.

To react to the rankings, the model needs a starting point, so that it knows what surprises it. Here is that starting point. This is where our model thinks the rankings will lie tonight, based on committee precedent. Some notes below:

The top five do seem likely to be these top five in some order. They’re the five undefeated power conference teams, and three of the five have beaten someone else expected to be in the top 15. Three of the five are rather proven. It’s the other two who are the most interesting.

Michigan comes out on top in our model’s estimate because of how thoroughly they’ve been beating their competition. They haven’t played a single team our model expects to be in the top 25 tonight, but our model does have Rutgers 26^th, and Michigan’s closest game remains that 24-point victory. They’ve beaten every other opponent by four or more possessions.

Georgia comes out fifth for two reasons: The first is that they haven’t played anybody all that good. Our model ranks teams 1–131 (JMU and Jacksonville State are excluded), and at 44^th, Florida is Georgia’s best opponent so far. The second is that they haven’t beaten teams all that decisively. Their most impressive wins came against two of their three best opponents, but they only beat Auburn by a touchdown and South Carolina by ten points, and in their early-season buy games, the results were decisive but not emphatic. Adjusted Point Differential (APD), our model’s metric which stands in for some of the fringier committee criteria, like advanced ratings and the eye test, only has the Dawgs 14^th in the Power Five. They haven’t played that well so far, going by margin and opponent.

But, Georgia could certainly exceed our model’s expectations. Georgia could be ranked first in the country, in fact (though I’d personally be surprised if it was anyone other than Michigan or Ohio State). One cause? Inertia, or credibility. Georgia has done so much in recent years that the committee members likely believe in them more than they’d believe in a Kentucky team with an identical body of work so far. The other? We’ve seen some signs our model systematically undervalues SEC teams in the rankings to begin the year. We haven’t yet found a good way to fix this, but it’s on our radar.

Wherever Georgia lands, our model will react, and that reaction will come from learning just how high this committee is on this Georgia team. This is why we do the process.

Among the one-loss power conference teams, Missouri and Louisville come in behind Notre Dame and LSU. For Missouri, this isn’t so much because of their head-to-head loss to LSU (though it’s convenient for us that our model has LSU on top of Mizzou, because our model never considers direct head-to-head but the committee sometimes does if teams are directly next to one another) as it is because Missouri has won less impressively than any other one-loss Power Five team. They have some good wins, but they’re held back by their struggles against Middle Tennessee and Memphis.

For Louisville, the issue is the loss to Pitt. To find another team with a loss as bad as that loss (our model considers opponent, location, and margin while scoring wins and losses), you have to go all the way to Oklahoma State at 29^th. The Pokes lost by 26 to South Alabama. Louisville’s loss to Pitt was almost as bad, especially now that Pitt just lost by 51 points in South Bend.

Again, maybe we’ll see Missouri and Louisville get some love. Louisville’s win over Notre Dame is the seventh-most impressive in the country so far, going by how our model scores them. It would be a slight deviation from precedent if we did, though.

Two Group of Five teams make our top 25, and I will admit that I only expect one to make the committee’s, if that. Our model is not all that well-equipped to estimate the committee’s evaluation of the worst FBS leagues, and no FBS league is worse than Conference USA. Our model has Liberty 16^th, but I would guess Liberty is unranked tonight. It’s generally an inconsequential issue for the model once we’ve gotten these first rankings, so we have yet to figure out a good solution, but this edition of Conference USA is also an historically bad FBS conference, and it’s hard to build off precedent where no precedent exists. The model should be better, but this is why it isn’t.

As for Air Force? Our model is calibrated to where the committee ranks Group of Five teams at the end of the year. Sometimes, they take a while to begrudgingly move a team up their list. Air Force has no impressive wins, but Air Force is undefeated, and while they rarely blow a team away, they’ve also played only one team within one score. They are an unusual team, and if we were putting error bars on these estimates (we need to create that feature in future years), theirs would be one of the widest.

Replacing Liberty, or replacing Air Force, or replacing one of the two-loss Power Five teams? We mentioned that Rutgers is 26^th, but for whatever it’s worth, the rest of the top 30 goes: Iowa, Duke, Oklahoma State, Tulane. Behind them, it’s UNC. If anyone behind UNC makes the top 25, it will mean a three-loss Power Five team not named Duke is ranked, and I personally don’t expect we’ll see that. The committee generally reflects a fairly median opinion, and tossing Texas A&M or Arizona or Wisconsin into the top 25 doesn’t seem in line with any median.

What’s the most consequential piece of tonight? Beyond Georgia, I’d say to keep an eye on how our model’s 6^th through 10^th-ranked teams line up. That’s Texas, Oregon, Oklahoma, Alabama, and Penn State. All those teams are very much alive in the playoff picture, but we’re still at a point in the season where being named the best one-loss team is a valuable thing. Beyond that? Washington is a team to watch. They have a great win over Oregon, and the road win over Arizona is looking good with age, but they’re coming off a bad two weeks. Not a lot of people would say Washington is actually a better team than Oregon, but the committee has to walk a tricky line between what it says it’s doing (ranking the best teams) and what people want it to do (rank teams based on a weighted average of accomplishment and ability).