Learn
The Thin Schedule Fix: how to rate college football teams when nobody plays everybody
Here is how to rate college football teams when almost nobody plays each other: the Thin Schedule Fix, three fixes you apply in order. Connect teams through common opponents, start each season from a prior instead of zero, and check every rating against the market price before you trust it. This checklist walks through each fix with the steps to copy and ends with the break-even table you need for the price check.
This is the checklist from the video. It follows the same three fixes, in the same order, and then goes deeper.
The three fixes:
- Connections. Rate teams through common opponents, so a win counts for more when the team you beat links to the rest of the field.
- The prior. A season doesn't start from zero. Start from last year's rating adjusted for returning production.
- The price check. Compare your number with the market, and know the win rate you need just to break even.
Why the schedule is thin
The FBS has 138 teams in 2026, and most play 12 regular-season games. A team meets 12 of the other 137, under 9 percent of the field, and most of those come from its own conference. So a 12-0 record tells you a team beat the dozen teams it happened to play, not where it stands against the other 125.
This is a known problem in the research. Park and Newman pointed out that because each team plays only a small fraction of eligible opponents, rankings built from win-loss records alone may have no unique solution, may contradict themselves, or both.
Fix 1, the connections
What it means. You rate a team by the chain of games that links it to everyone else. If Team A beats Team B, and Team B beat Team C, that result says something about A against C even though they never met.
Steps to copy:
- Build a game table: date, home team, away team, final margin, and whether the game was at a neutral site.
- Fit a margin rating, the simplest version being: predicted margin = home rating - away rating + home edge. Solve it with least squares over every game so far (a spreadsheet solver or a few lines of code will do).
- Count each team's links: how many of its opponents also played teams outside its conference. A team with few links has a less certain rating, so treat its number with a wider range.
- Consider capping blowout margins (for example at 28 points) so a 63-3 win over a weak team doesn't swamp the rating.
Watch out for: conference islands. Osting, Brune and Osher analyzed a 2011-12 schedule and argued that better-connected schedules give more informative rankings, with more out-of-conference games linking the separate communities. In practice that means the non-conference weeks carry a lot of the information about how conferences compare, so check how much your ratings move after them.
Fix 2, the prior
What it means. With only 12 games, the current season can't carry a rating alone, especially in September. So you start from what you knew before the season and let results move you.
ESPN's Bill Connelly describes SP+ preseason projections as built from three pieces: last year's rating plus returning-production adjustments, which he says make up about two-thirds of the formula; recent recruiting rankings, with the newest class weighted most; and a sliver of program history from roughly the last four seasons.
Steps to copy:
- Start each team at last season's final rating.
- Pull it toward average for lost production. Quarterback changes matter most: Connelly notes that replacing a starting quarterback with a transfer leaves returning yardage at roughly half.
- Nudge it for recruiting strength.
- During the season, blend the prior with the new results, giving the prior less weight each week. A simple schedule is to let the prior count like several games in week 1 and fade toward zero by late season.
Watch out for: a prior that never fades. If week 10 ratings still look like the preseason ones, the model isn't learning from the games.
Fix 3, the price check
What it means. A rating only matters for betting when it disagrees with the price, and the price is a strong opponent. A PLOS ONE paper on soccer notes that betting odds are frequently found to beat mathematical models as forecasts. That is soccer, not football, but it is why a new model should be assumed worse than the market until it proves otherwise.
Steps to copy:
- Turn your predicted margin into a fair spread and a win chance.
- Compare it with the market line and write down the gap for every game, even the ones you don't bet.
- Bet only where the gap is large enough to clear the break-even rate for the price.
- Track your number against the closing line, not just wins and losses.
The break-even table
At -110 you need about 52.38 percent wins just to break even. The same math for other common prices, so you can see the bar each bet has to clear:
| Price | Break-even win rate |
|---|---|
| +110 | 47.62% |
| +105 | 48.78% |
| +100 | 50.00% |
| -105 | 51.22% |
| -110 | 52.38% |
| -115 | 53.49% |
| -120 | 54.55% |
| -125 | 55.56% |
| -130 | 56.52% |
For a negative price, the break-even rate is the price divided by the price plus 100 (110 / 210 for -110). For a positive price it is 100 divided by the price plus 100. The fair odds calculator does both.
The one-page checklist
- [ ] Game table built with margins, home and neutral sites
- [ ] Margin ratings fit on all games, blowouts capped
- [ ] Link count per team, wider ranges for poorly linked teams
- [ ] Preseason prior from last year, returning production and recruiting
- [ ] Prior fades week by week
- [ ] Fair spread and win chance for every game
- [ ] Gap to the market logged for every game
- [ ] Bets only where the gap clears the break-even rate
- [ ] Results tracked against the closing line
How to tell it worked
Three signs. Your ratings stop jumping wildly after the non-conference weeks. Your preseason and late-season numbers differ in sensible ways. And over a full season, your fair spreads move toward the closing line more often than away from it. Even then, a season of college football is a small sample, so read early results as information, not proof.
The college football betting model page covers the data problem in more depth, and the sports betting model scorecard explains how a model should be graded.
Quant+ is analytics for informational purposes. 21+. Nothing is guaranteed. If you or someone you know has a gambling problem, call 1-800-GAMBLER.
Questions people ask
How many games does an FBS team play against other FBS teams?
Most FBS teams play 12 regular-season games, and some of those can be against FCS teams, so a team sees only a small slice of the 137 other FBS programs.
What is returning production?
It is the share of last season's production a team brings back, such as passing yards, receiving yards and tackles. Preseason ratings like SP+ use it to adjust last year's rating for roster turnover.
Should I stop using a preseason prior once the season starts?
Not all at once. The usual approach is to blend the prior with current results and give it less weight each week, so early ratings lean on the prior and late ratings lean on the games.
What win rate do I need at -110?
About 52.38 percent, because you risk 110 to win 100. Any rating that can't beat that rate after the price is only a rating, not a bet.
Sources
- NCAA Division I Football Bowl Subdivision, Wikipedia
- A network-based ranking system for American college football, arXiv (Park, Newman)
- Optimal Data Collection For Informative Rankings Expose Well-Connected Graphs, arXiv (Osting, Brune, Osher)
- Spring update of 2025 college football SP+ rankings, ESPN via ABC30
- The Betting Odds Rating System, PLOS ONE
- Moneyline odds, Wikipedia
