Last season’s first-ever College Football Playoff might have miscalibrated everyone’s sense of what it takes to make it to the final four. Six power conference champions or co-championsdubious decision by the Big 12.">1 — Alabama, Baylor, Florida State, Ohio State, Oregon and TCU — were undefeated or had one loss against reasonably good schedules. There’s plenty of room to critique how the committee went about leaving Baylor and TCU out, but there was no one inherently correct way to slot six fairly equally matched teams into four playoff positions.
But a field as crowded and qualified as last year’s was atypical. Most of the time there are a couple of teams that are weak links: a three-loss conference champion here, a one-loss team that played an incredibly weak schedule there. A mess like last year’s isn’t impossible, obviously. But usually the knot will untangle itself through conference championships, rivalry games and upsets that knock teams out late in the season.
So don’t despair, Michigan State fans. (Of which I’m one.) Yeah, you probably lost on a bad call last weekend. But you’re still highly likely to make the playoff if your team wins all of its remaining games, in which case you’ll have defeated Ohio State and (probably) Iowa in the Big Ten championship. Pretty much every one-loss team from a power conference is more likely than not to make the playoff if it wins out.
And an undefeated power conference team like Oklahoma State shouldn’t fret, even if it is currently outside the committee’s top four. Some of the teams ranked in front of it are almost certain to lose — and even if they don’t, there’s a good chance Oklahoma State will leapfrog some one-loss teams if it keeps winning.
Our subjective perceptions of the playoff picture aren’t the only thing that may be miscalibrated, though. The same could be said about the FiveThirtyEight College Football Playoff model. At least, that’s the conclusion we came to when we were conducting research for this article. Although the model seems to give basically reasonable answers, a couple of things left us scratching our heads when we examined it more deeply.
For instance, it posited a conspicuously large gap between Iowa’s chance of winning the Big Ten championship (27 percent) and making the playoff (8 percent). Iowa is undefeated, and while it’s possible they could win the Big Ten with one loss or more, the internal calculations in the model also implied that they’d have only about a 55 percent chance of making the playoff even if they ran their record to 13-0. One can see why a computer might come to that conclusion — Iowa has played a pretty bad schedule, and its margin of victory hasn’t been impressive — but human beings are going to vote an undefeated Big Ten champion into the playoff unless almost everything else2 is working against them.
Readers had some questions for us too. Why was USC, which already has three losses, given any realistic chance by the model (granted, it was just 4 percent) of making the playoff? And why was one-loss Alabama’s chance of making the playoff so much higher than its SEC championship chances? There are some good reasons for that one,spectacularly upset by Arkansas.">3 but even accounting for those, the gap seemed to be too wide and the model seemed to be too optimistic about Alabama still making the playoff if it endured a second loss.
The theme here is that human beings pay a lot of attention to wins and losses — more than our computer seemed to be doing. An undefeated power conference team is going to get in except under rare circumstances. Two-loss power conference teams have historically finished in the AP top four more often than you might think, but it’s still a hard road. And a three-loss team making the playoff? Almost impossible unless there’s total carnage everywhere else.
Since the whole point of our model is to mimic human intuition, reader feedback made us think it had some blind spots. So we re-examined the historical data4 and concluded that our model should be placing more weight on plain-vanilla wins and losses. Or at least, it should be doing so for power conference teams (and for Notre Dame); minor conference teams historically haven’t been treated that kindly by either poll voters or the committee. Even if the committee currently ranks a one-loss team ahead of an undefeated team, or a two-loss team ahead of a one-loss team, it may re-examine the case in future weeks, and the team with fewer losses will often get the benefit of the doubt. (For a more technical explanation of how this is implemented in the model, see the footnotes.change we introduced earlier to the model is that it reverts the projected committee standings each week toward a team’s Elo rating. Elo ratings are a fairly simple method that often correspond pretty well with human judgment in ranking sports teams. But there’s an even simpler method: a rating based solely on a team’s win-loss record and whether it plays in a power conference. The new version of the model reverts a team’s ranking based on a combination of its Elo rating and its win-loss record, instead of its Elo rating alone.">5)
This relatively simple change has little impact for most teams, but it does affect a couple of the cases that had bothered us (and some of our loyal readers). Iowa’s chances of making the playoff are now 18 percent instead of 8 percent. Alabama’s are 42 percent instead of 54 percent. USC’s are 1 percent6 instead of 4 percent. Here’s how everyone’s odds were affected by the change:
And here’s a new summary table showing the playoff picture heading into tonight, when the playoff committee will release its new rankings at 7 p.m. Undefeated Baylor has moved slightly ahead of one-loss Notre Dame in our forecast, but otherwise the top six are unchanged.
|Team||CFP||Elo||FPI||Conf. Title||Playoff||Nat. Title|
|Ohio State 9-0||3||2||5||44%||57%||15%|
|Notre Dame 8-1||5||6||9||—||30%||7%|
|Oklahoma St. 9-0||14||4||13||37%||23%||5%|
|Michigan St. 8-1||7||8||22||12%||10%||1%|
|North Carolina 8-1||—||18||20||30%||5%|
|Mississippi St. 7-2||20||16||15||3%||3%|
|Florida State 7-2||16||24||17||0%|
|Texas A&M 6-3||19||49||23|
|College Football Playoff (CFP) rankings as of Nov. 3.|
But back to football substance. I mentioned before how lots of teams, even if they don’t technically control their own destiny,7 are favored to make the playoff if they win the rest of their games. Now that the model is (hopefully) doing a better job of mimicking the emphasis that human voters place on wins and losses, we can be more precise about that. Specifically, the model estimates that 14 teams have a 50 percent or greater likelihood of making the playoff conditional on winning out.
The model figures that Michigan State, for instance, has an 86 percent chance of making the playoff if it wins out. Even the lowliest one-loss major conference team, North Carolina, which wasn’t ranked by the committee last week, is about even-money to make the playoff if it wins out. And undefeated Iowa (91 percent) and Oklahoma State (97 percent) are all but assured of making the playoff if they finish the year without a loss, even if the committee doesn’t have them in the top four tonight.
So the tough part for teams like Michigan State isn’t sweating the committee’s decision if it wins the Big Ten; it’s getting to that point in the first place. With Ohio State and Iowa still in the way; the Spartans have only an 11 percent chance of running the table.