← All Articles

November 19, 2026

The Real Scheduling Penalty Is Having No One to Beat

Texas wasn't punished for playing Ohio State. Four seasons of AP and committee rankings say the credit goes to beating good teams, and not everyone gets the chance.

Share

Every few seasons a team with a big-name non-conference game misses the playoff, and the same argument starts up. Scheduling up doesn't pay. Lose to a great team in September and the polls hold it against you all year. Play a cupcake instead, win by 40, and nobody remembers. The smart move, the argument goes, is to schedule soft.

Texas in 2025 is the cleanest test case there is. The Longhorns opened the season at No. 1, went to Ohio State (NovaModel's No. 2 team that year) and lost 14-7. They finished 9-3 and the committee ranked them 13th, three spots outside the last at-large bid. If scheduling up was ever going to cost a team a playoff spot, it was here.

So we tested it. We took every AP Top 25 and Playoff Committee ranking from the 2022 through 2025 regular seasons and every non-conference game in those four years, and asked: once you account for how good a team actually was, does a tougher non-conference schedule leave it ranked lower? The short answer is no. Scheduling up doesn't cost you. The teams that really lose out are the ones that never get the chance to beat anybody good.

How we measured a tough schedule

A team's non-conference schedule strength is the average final NovaModel rating (our v4 ratings) of its non-conference opponents, with any FCS opponent counted at that season's lowest FBS rating. A game counts as non-conference when the two teams were in different conferences that season, when either one was an FBS independent, or when the opponent wasn't FBS. That's 2,145 non-conference team-games across 534 FBS team-seasons, 485 of them against FCS teams.

"Equally strong" means the same final NovaModel rating: how good a team turned out to be once its whole regular season was in. Because AP points climb steeply near the top (No. 1 gets about 1,550 points, No. 25 about 100), every comparison also holds teams inside the same strength tier (1-5, 6-10, 11-15, 16-25, 26-40), and holds power-conference teams apart from everyone else. To keep the numbers readable, we report the gap between a soft schedule and a tough one (the 10th and 90th percentiles, 38.7 and 49.7 rating points) in poll spots. One AP spot is about 60.6 points.

Finding 1: Scheduling up isn't punished

Method: OLS regression, with an equivalence bound

For every FBS team-season we ran an ordinary least squares (OLS) regression of its final ranking on its final rating, its strength tier, a power-conference indicator and its non-conference schedule strength. The AP version uses its points in the last regular-season poll (0 if unranked). The committee version uses 26 minus its rank in the final committee ranking (0 if unranked), so the units are spots.

  • Final AP poll: going from a soft schedule to a tough one moves a team +0.03 spots. 95% CI -0.62 to +0.72.
  • Final committee ranking: -0.12 spots. 95% CI -0.73 to +0.54.

"Not significant" isn't the same as "no effect," so we set a stricter bar for calling something zero: the whole 95% interval has to sit inside one poll spot either way, the smallest step a ranking can show. Both clear it. We set that one-spot bound after the first run came back empty, and we're saying so here because it matters.

We also asked whether voters give credit for a tough schedule at the same record. Here the AP poll leans that way, +0.55 spots (95% CI -0.07 to +1.22), but the interval reaches past one spot and crosses zero. We're calling that one unresolved.

Finding 2: It's who you beat, not who you play

Method: OLS regressions on the poll after each game, and on the final rankings with quality wins added

Week to week, AP voters do pay attention to who a team played. For every non-conference game played by a ranked team, we regressed its AP points in the next poll on its points before the game, the margin, the venue and the opponent's final rating. Losses (63) and wins (329) were fitted separately.

  • After a loss: for every 10 rating points of the team that beat it, a ranked team keeps 100 more AP points, about two spots. 95% CI +16 to +185, Freedman-Lane permutation p = 0.003, adjusted q = 0.006.
  • After a win: for every 10 rating points of the team it beat, a ranked team gains 48 more AP points, about 0.8 spots. 95% CI +34 to +64, p < 0.001, q = 0.003.

So a loss to a great team costs less than a loss to a bad one, and a win over a great team is worth more than a win over a bad one. By season's end that mostly comes down to one thing. We refit the final-ranking regressions with three things added: regular-season losses, wins over teams in NovaModel's final top 25 (conference games included), and a blue-blood indicator for the fifteen programs from our blue-blood study.

  • Each win over a top-25 team: +1.6 committee spots (95% CI +0.99 to +2.08) and +97 AP points (95% CI +60 to +127). Both q = 0.002.
  • The blue-blood name: -0.4 committee spots (95% CI -1.83 to +1.08) and -31 AP points (95% CI -122 to +49). Nothing clear once wins are counted.
  • The schedule itself: with wins counted, a tough slate is worth +0.35 committee spots (95% CI -0.19 to +0.90), still inside one spot.

That explains Texas better than the schedule does. Three-loss Texas sat ahead of two-loss Vanderbilt (No. 14) and Utah (No. 15) because it had beaten three NovaModel top-25 teams: Oklahoma, Vanderbilt and Texas A&M, all in conference. Utah had beaten none. Vanderbilt had beaten two and lost to Texas.

Texas 2025: would a cupcake have helped?

Method: a replayed season and two committee models

Here is what each of Texas's losses did to it in the AP poll:

WeekLossAP rank beforeAP rank afterAP points
1at Ohio State (NovaModel No. 2), by 717-259
6at Florida (NovaModel No. 48), by 89unranked-1,071
12at Georgia (NovaModel No. 7), by 251017-397

Losing at Ohio State cost Texas six spots. Losing at Florida knocked it out of the poll entirely. That's the week-to-week finding playing out in a single season.

Then we asked the playoff question directly. We replayed the 2025 season through our v4 ratings twice: once as played (the replay reproduces our stored ratings exactly), and once with the Ohio State game replaced by a win over an FCS team. Swapping in the cupcake barely moved Texas's rating, from 61.64 to 61.39, still No. 17 in NovaModel. We then moved every team's committee score by the change in its predicted score from the committee regression, fitted two ways: with the strength tiers, and with a smooth curve in the rating (a cubic) in their place, because crossing a tier line can move a prediction a long way.

  • With the strength tiers: Texas finishes No. 13 (95% range No. 12 to No. 13).
  • With the smooth curve: No. 13 (95% range No. 13).

The last at-large team in the 2025 field was No. 10 Miami, with Notre Dame and BYU between Miami and Texas. In 2,000 bootstrap refits of each model, the cupcake never moved Texas inside that cut. Whatever kept Texas out of the playoff, it wasn't playing Ohio State.

The real victims: teams with no one to beat

Method: a description of who played whom, not a test

If the credit comes from beating good teams, the question that matters is who gets to play them. Among team-seasons that finished in NovaModel's top 25, here is how many regular-season games each group played against other top-25 teams, on average:

Contenders fromTeam-seasonsGames vs top 25Non-conferenceWins
Group of 571.00.70.3
ACC122.71.01.5
Big 12132.80.51.4
Notre Dame43.03.01.3
Big Ten233.40.41.7
Pac-1293.40.31.7
SEC324.70.52.3

An SEC contender gets almost five shots a year at a top-25 team. A Group of 5 contender gets one. SEC contenders turned those shots into about two more top-25 wins a season, and at about 1.6 committee spots a win, that's roughly three spots. Most of it comes from conference schedules, though, not from refused games: the Group of 5 contenders actually played slightly more non-conference top-25 games than the SEC, Big Ten or Big 12 ones did. Some groups here are a handful of seasons, so treat the table as a description rather than a test.

The Big 12 sits in the middle of that table, and it's the conference this summer's scheduling reporting is about. In 2024 and 2025, six Big 12 teams finished in NovaModel's top 25: Colorado, Arizona State and Kansas State in 2024, and Texas Tech, Utah and BYU in 2025. None of the six played a non-conference top-25 opponent.

Our data can't say why, because no data we have shows who asked whom. The reporting is clearer. Texas Tech coach Joey McGuire said this summer, "We're talking to everybody and anybody. So far, everybody's told us no," naming talks with USC, Texas A&M and NC State (OutKick, Aug. 11, 2026). The Deseret News reports that Texas Tech's 2027 game with NC State was canceled and that the Red Raiders have no non-conference game against a power-conference opponent scheduled until 2030, while Arizona State lost future series with LSU, Florida and Texas (Deseret News, Aug. 29, 2026). Alabama dropped a future series with West Virginia and called off one with Oklahoma State by mutual agreement, cancellations the reporting ties to the SEC's move to a nine-game conference schedule (Bay Area News Group via KSL, Aug. 6, 2026). "Changes in the SEC and ACC the last two years have caused some disruption," Big 12 football chief Scott Draper said there.

Put the two halves together and the real penalty is clear. Nobody gets punished for scheduling a great team. Teams get held back by not having great teams to beat, and at least one Big 12 contender trying to add them says it keeps hearing no.

What we can't say

This is four seasons of correlational data, not an experiment. Rankings record what voters did, not why. The Texas swap assumes nothing else about its season or the committee's other choices would have changed. Quality wins are counted against NovaModel's final top 25, not the committee's. The one-spot bound for "no effect" was set after the first run. The table of opportunities is a description of 2022-2025, and the reasons behind any team's schedule come from public reporting, not from our data.

Every number, interval and method note is on the full study page, which updates itself if the study is re-run.

How to reproduce it

  • Data: AP Top 25 and Playoff Committee Rankings ballots and game results from CollegeFootballData (CFBD), 2022-2025 regular seasons (pulled 2026-10-13). Team strength is NovaModel's v4-state-space rating at the end of the regular season. Conference membership is each season's own, from CFBD.
  • Effects: every number is a coefficient from an ordinary least squares (OLS) regression with an intercept: AP points or committee spots per 10 rating points of schedule or opponent strength, or per top-25 win.
  • 95% confidence intervals: percentile cluster bootstrap. Teams are drawn with replacement 2,000 times, keeping every row of each drawn team, and the regression is refit on each draw. The interval runs from the 2.5th to the 97.5th percentile.
  • p-values: Freedman-Lane permutation test. The regression is fitted without the tested term, its residuals are shuffled across rows 2,000 times and added back to its fit, and the full regression is refit each time. p = (shuffles whose coefficient is at least as far from zero as the real one, plus 1) / (shuffles plus 1), two-sided.
  • Multiple comparisons: Benjamini-Hochberg false-discovery-rate adjustment (q-values), once across the five main questions (final AP poll, same record, committee, after a loss, after a win) and once across the six quality-wins tests. A result counts as real only when its 95% interval excludes zero and its q is under 0.05.
  • No effect: a schedule counts as neither a penalty nor a reward only when the whole soft-to-tough 95% interval sits inside one poll spot either way.
  • The Texas swap: the 2025 season replayed week by week through v4 from 2022 (as the stored ratings are built), as played and with the Ohio State game made a win over an FCS team. Committee scores shift by each team's change in predicted score under the committee regression with losses added, fitted with strength tiers and again with a cubic in the rating. The cut is the lowest rank with every team at or above it in the playoff field.
  • Checks: the three final-ranking questions refit on NovaModel's top 40 only, and on FBS opponents only with the number of FCS games as a control. Both agree with the main results.
  • Random seed: 42, for both the bootstrap and the shuffles.