Where are my college football rankings?

If you’ve been waiting for me to start talking about my college football rankings, there is a reason I’ve been quiet. The calculations are running, and the numbers are available. I just want more football before putting much confidence in them.

Three weeks is enough to give us some interesting results. It isn’t enough to reliably compare schools across the country using this season’s results alone.

Three games, three very different schedules

My power ratings account for how a team performs and the strength of its opponents, with adjustments to keep extreme scoring margins from overwhelming the calculation. As explained in my ratings methodology, teams begin at the same baseline, and the system repeatedly works through the results to develop the ratings.

The problem in September is that we are still learning how good the opponents are, too.

A team can start 3–0 by beating three struggling opponents. Another can go 2–1 against much stronger competition. Even two schools with identical records and similar scoring margins may have faced very different tests. Strength of schedule helps account for that, but after three weeks, strength of schedule is itself an uncertain estimate.

There are also too few cross-matchups to tie everything together. Most schools haven’t played one another, and there may be little overlap among their opponents. We need more common opponents and more chains of results connecting different schedules before those comparisons become dependable.

Suppose one school dominates its first three opponents. Were those opponents weak, or did a great team make them look weak? When those opponents play more games against other schools, we gain evidence that helps answer that question. A team’s rating can change even during a bye week because we learn more about the teams it has already faced.

Repeating the calculations cannot create that missing evidence. A model can settle on a numerical answer while the evidence behind that answer is still thin.

Look at the current FBS top 10

Here is a snapshot of my 2026 FBS power ratings from the Week 3 report generated September 20. These are the ratings at the time of this post; the linked report will change as the season progresses.

Texas leads this group in both power rating and estimated strength of schedule. Indiana has outscored its opponents 145–16, but its schedule rating is the lowest in the top 10. That is why simply sorting teams by record or scoring margin doesn’t tell the whole story.

Ohio State is second despite a loss. These are power ratings, so an undefeated record doesn’t automatically put a school ahead of every team with a defeat. Meanwhile, Ole Miss and Michigan are separated by just 0.028 rating points. Printing three decimal places doesn’t mean we have enough evidence to confidently distinguish fourth from fifth.

And then there is Tulsa at No. 10. That doesn’t prove the model is broken, and it doesn’t establish Tulsa as one of the country’s ten best teams. It tells us where its first three results place it in a comparison that still has a lot of uncertainty. More games will help determine whether that position holds up.

The current Elo top 10

For comparison, here are my 2026 FBS Elo ratings from the same Week 3 report generated September 20.

Texas leads both lists, but Michigan is second in Elo compared with fifth in power, while Ohio State is tenth in Elo compared with second in power. Miami enters the Elo top 10, and Indiana drops out. The methods weigh performance differently, so they can produce different orders from the same limited set of games. Neither list removes the uncertainty that comes with having only three weeks of results.

Why carrying over Elo ratings used to help

One way to address the early-season problem is to bring information forward from the previous season. An Elo system can do that.

Elo updates a team’s rating based on its result relative to the expected result against its opponent. Beating a highly rated team generally earns more credit than beating a team you were already expected to beat. If a school carries some of its previous year’s rating into September, the system has a starting estimate before the new schedule provides much evidence. Some implementations move those ratings back toward average during the offseason.

That approach made more sense when rosters were more stable. Think of the old four-year roster cycle: roughly one class, or about 25% of the roster, would leave each year, while the others returned with another year of experience. That is a useful rule of thumb, not an exact historical average; redshirts, early NFL departures, and other attrition always complicated it.

With a substantial returning core, last year’s team performance gave us a reasonable starting point for this year’s team. Carrying over an Elo rating helped fill the gap while the new season’s results accumulated.

The school name stays. The players may not.

The transfer portal and the financial opportunities associated with NIL have made that continuity harder to assume. Schools still lose players to graduation and the NFL, but they can also lose experienced players with eligibility remaining and replace them with transfers from other programs.

The scale of player movement has grown. In May 2023, Axios reported, citing ESPN’s data, that football portal entries across the NCAA had more than doubled from the inaugural 2018–19 cycle. Portal entries aren’t the same as completed transfers or a team’s total roster turnover, but they illustrate why the old continuity assumption deserves more scrutiny.

For programs undergoing substantial portal turnover, annual roster replacement can be much greater than that old one-quarter rule of thumb. There isn’t one percentage that describes every school. And the identities of the departing players matter as much as the count: replacing a starting quarterback and most of an offensive line can change a team far more than losing the same number of reserves.

An unchanged school name can therefore conceal a very different football team. Carrying over its old Elo rating may reward the new roster for what departed players accomplished, or penalize incoming talent for the struggles of players who are no longer there.

That doesn’t make Elo useless. Coaching, returning players, and program resources still provide information, and a model can reduce its carryover or adjust for roster changes. But simply carrying a school’s rating from one season to the next is less convincing when so much of the team has changed. My view is that those old ratings deserve less weight now.

I’m waiting for more evidence

Starting everyone at the same baseline avoids leaning on last season’s reputation, but it comes with a cost: patience. Three games per school leave too much uncertainty about both team quality and opponent quality.

As the season progresses, more matchups will connect schedules, opponents will reveal more about themselves, and any single unusual result will account for a smaller share of a team’s record. There isn’t a magic week when every ranking becomes correct, but the comparisons should become better supported.

I usually like to wait until after the first weekend in October to release my rankings. This year, I’ll be releasing them on September 27. That gives the results a little more time to develop, even if the early rankings will still need some patience.

I’m impatient too.