Skip to main content

Ohio State Lost by 12 at USC. Our Ranking Moved It Up a Spot.

Week 9 PIP movement and what drives it

Ohio State lost 30-18 at USC and moved up a spot in PIP, because one category, Opponent Quality, erased four that got worse. We tested what fixing it would do.

· Ranktory Research · Polls: PIP editions, Week 8 and Week 9 · Data: game results through week 9

Our ranking has always included one ingredient that rewards a team for who it played, even when it loses. Ohio State's loss on Saturday, and the climb that followed, made it impossible to miss. Fixing it matters more than turning the win-loss dial. This is the first entry in a bigger question we're working through: what actually matters when you measure how good a college football team is?

The result that made us look twice

USC 30, Ohio State 18.

Our Performance Index Poll (PIP) the next morning: Ohio State #3 to #2. USC, the winner, #18 to #20.

Either a team can lose by 12 and get better, or something in the formula is off. We went looking.

What moved

PIP averages six category ranks. Here is what Ohio State's loss did to each one:

What Ohio State's loss at USC did to each PIP category
CategoryChange in rank
Results11.5 spots worse
Defense6.5 spots worse
Execution2.5 spots worse
Offense1.5 spots worse
Game Control2 spots better
Opponent Quality21 spots better

Four categories got worse, and then one category erased them all.

Opponent Quality asks how strong a team's opponents have been. Playing USC made Ohio State's schedule look tougher. The score of the game never enters into it. Think of a student who fails a hard exam and gets bonus points for taking a hard class.

It wasn't just Ohio State

The same thing shows up in the Top 25. Ohio State (#2) is ahead of Texas, which beat it in week 2, and ahead of USC. In all, 8 pairs of Top 25 teams are ranked opposite to how they played.

The obvious fix may not be the right one

Our first thought was to count wins more. We tested it. For the first eight weeks, we checked whether the higher-ranked team won its next game (448 games).

Picking the winner of the next game, Weeks 1 to 8 (448 games)
VersionPicks the winner
PIP as published64.7%
More wins, more opponent quality, fewer stats61.4%
Wins and opponent quality only, no stats57.1%
Statistics only65.4%
Opponent quality alone49.8%

Statistics belong in the ranking, and this is the clearest evidence yet. Take them out and accuracy falls from 64.7% to 57.1%, about 34 fewer correct picks, a gap bigger than the noise. Use them alone and you match or slightly beat the full formula. Whatever we change, the statistics stay in.

The statistics do most of the work. Opponent Quality on its own is a coin flip, which makes sense: knowing who someone played doesn't tell you how good they are.

What the numbers do and don't say

What the numbers say is that giving schedule strength more weight made predictions worse, while shrinking it or dropping it made them slightly better. Taking statistics out cost the most of any change we tried. That points toward crediting results against strong opponents, not just a hard schedule, and it keeps statistics in the formula.

What they don't say is that any new formula is better. 448 games is a small sample, and gaps under about ten wins are noise. Giving wins more weight barely changes how often a winner is ranked above the team it beat (77% now, 78% with triple weight), because records are coarse. Models that credit you for beating good teams lined the season's results up better (82%), and that is the angle we want to dig into. One caution: that 82% measures how well a ranking fits games already played, not how well it predicts the next ones, and the same models were no better at predicting (58% to 63%). Testing both on 2024 and 2025 is how we find out whether the fit carries over. And a ranking that predicts next week and a ranking that rewards the season you've had are different jobs. PIP v1 was built as a performance index, not a resume.

The limit is the sample, so we're making it bigger

Nine weeks of one season is the real limit here. Every caveat above comes back to it. A gap of a few wins out of 448 games can't tell a better formula from a lucky one.

So the next step is to back-test PIP over the 2025 and 2024 seasons, thousands of games instead of hundreds. That's where we can find out whether the Ohio State result is a one-off or a pattern, and whether fixing Opponent Quality actually predicts better on games the formula has never seen.

PIP v1 stays as published. Any change would be a new version, tested on those seasons first. Think of Ohio State and USC as Exhibit A. The case we're building is about which pieces of performance (statistics, wins, schedule) earn a place in a ranking, and we'll follow the evidence wherever it goes, including where it goes against us.

Join us next week. We'll bring the historical data to the site and run the test in public, with the numbers on the page. If you have a team you think PIP gets wrong, send it to us. We'll check it against the full history.