We Updated Our College Football Model, And Now Michigan Looks Slightly Less Screwed

We’ve been getting a lot of angry notes from Michigan fans. And even though I’m from East Lansing, I think they have a point. Sort of.

Here’s the rub: Michigan fans claim their Wolverines have a shot at the college football playoff, even though they rank fifth (the top four teams make the playoff) and have finished their regular season (Wisconsin and Penn State are playing for the Big Ten championship instead).

Even the most rabid Michigan backers don’t expect their team to displace any of the current top four if everyone wins out. Undefeated No. 1 Alabama is one of the strongest college teams in history and will become the SEC Champion if it beats Florida on Saturday. No. 2 Ohio State — which, like Michigan, is idle this weekend — just beat Michigan last Saturday. And No. 3 Clemson and No. 4 Washington are potential one-loss conference champions, while two-loss Michigan is neither of those things.

Alabama would have a decent shot at the playoff even with a loss, but if either Clemson or Washington falls, another slot could open up. Apart from Michigan, the most plausible contenders to fill it are Wisconsin and Penn State — whichever one wins the Big Ten championship — and Colorado, if it beats Washington for the Pac-12 title.

Wisconsin, Penn State and Colorado would each be 11-2 conference champions, as compared to 10-2 non-champion Michigan. The playoff selection committee explicitly accounts for conference championships as part of its selection criteria. So Michigan has a hard argument to make, it would seem.

Except for one thing: Michigan played Wisconsin, Penn State and Colorado. It beat all three of them. And although the committee says it considers conference championships, it also says it takes head-to-head results into account. How would the committee weigh everything? Nobody’s quite sure.

Our college football playoff model, however, had been putting a fairly heavy thumb on the scale against Michigan. That’s because we’d programmed it to account for conference championships, but not for head-to-head results. Why not? There wasn’t any particularly good reason; we’d intended to build in a head-to-head adjustment earlier this fall and then got distracted by that whole presidential election thing. Also, because head-to-head results didn’t happen to matter very much in the first two years of the committee’s rankings — there was no case analogous to the one Michigan faces this year — we didn’t have much data on how much the committee really cares about them.

Still, we think making some effort to account for head-to-head results is better than nothing, even if we’re basically just making an educated guess about the magnitude of the effect. So we’ve built an adjustment into our model. As before, the program runs a series of simulations in which it plays out the remaining games and estimates how the committee will rank the teams. Then there’s a new step: It checks to see if teams that are ranked in close proximity played one another. If in one simulation it initially had Colorado ranked No. 4 and Michigan No. 5, for instance, it might flip them because of the head-to-head result. Or it might not: The magnitude of the head-to-head adjustment is randomized a bit from simulation to simulation but generally set to a fairly conservative value. (We’ll recalibrate everything next year; how the committee untangles Michigan and the other teams will tell us a lot about how much it really cares about head-to-head play.)

As a result of this change, Michigan’s chance of making the playoff increases to 6 percent. That’s still not very good — the model thinks it’s unlikely that the committee will put an idle team into the playoff, especially when it has two losses and didn’t win its conference championship. (The loss to Ohio State is especially complicating, because the committee would have to take two teams from the Big Ten but not the conference champion?) The committee could also evade the head-to-head question by taking two-loss Oklahoma or Oklahoma State, whichever one wins the Big 12 this weekend, instead of a second Big Ten team. But 6 percent is an improvement for Michigan from the 1 percent chance our model gave it before the adjustment.

Apart from Michigan, this change also affects Ohio State’s calculus a bit. Aren’t the Buckeyes shoo-ins? Would the committee really demote a team all the way from No. 2 to No. 5?

Probably not, but consider what happened in 2014. The committee — to our model’s surprise — dropped TCU all the way from No. 3 to No. 6 in its final rankings despite TCU having won. We learned from that experience that the committee isn’t necessarily all that consistent from week to week. So Ohio State, which won’t be the Big Ten champion, might be just a little bit nervous if the committee decides it values conference championships highly.

But it matters which team wins the Big Ten instead of Ohio State. If it’s Wisconsin, the Buckeyes have less to worry about because they beat the Badgers head-to-head. (Wisconsin might make the playoff as a second Big Ten team, but probably not without Ohio State making it as well.) Ohio State lost to Penn State in the regular season, however. So if Penn State wins the Big Ten, it will be able to cite both a head-to-head victory and a conference championship in its case to get in ahead of Ohio State. Our model expects that Ohio State would probably still make it under such circumstances — quite possibly alongside Penn State — but it isn’t quite as safe. (In the new version of our model, Ohio State has a 97 percent chance of making the playoff if Wisconsin wins the Big Ten and a 91 percent chance if Penn State wins instead.)

So almost no matter what happens, we’ll be left with a bit of a mess. One solution? Expand the playoff to six or eight teams, so the close calls stemming from janky conference-championship scenarios are resolved on the field and not in a conference room.