Ohio State beats equally-talented rosters by five to seven points a game

Ohio State recruits the 2nd or 3rd best roster in the country almost every year. Nobody needs a model to know that, and a team full of five-stars is supposed to win. What I wanted was the part on top of the recruiting.

It's +6.95 points a game over eleven seasons, 95% interval +4.33 to +9.57. That number's talent-matched: every Ohio State season scored only against teams within a hair of the same recruiting level, about 87 of them a season, with Ohio State held out of its own peer group. 1,434 team-seasons, 2015 through 2025.

Two things before anyone runs with that. Ohio State is 3rd of 130 on this measure, not 1st: Air Force sits at +11.22 and App State at +6.98. Air Force is up there because this rating under-credits triple-option offenses, which is a separate finding of ours, but third is still third.

I also picked Ohio State because it was near the top, which inflates the estimate. Shrink it toward the league average the standard way and you get +5.42. Call it five to seven points and you're safe either way.

Alabama recruited better over the same window, a 1.3 average class rank against 2.8, and comes out at +1.77 with an interval of −0.74 to +4.28. That's not distinguishable from average, though the interval is wide enough to contain a three-point Saban premium. The gap between the two teams is the part that holds: +5.18, standard error 1.85, t = 2.80, interval +1.55 to +8.81.

One definition, because Alabama won three national titles in this window. Every rank here is average season rank on my own opponent-adjusted per-play rating: Alabama 10th, Ohio State 4.3. That's not an AP finish and it's not a trophy count.

The draft says the same thing from a different angle. Fit NFL picks on recruited talent across 817 team-seasons and you get picks = 0.98 + 2.10 x talent in SD units. Ohio State's rosters project to 5.47 a year and produce 8.50. That's an excess of 3.03 picks a season, SE 0.97, over ten drafts, interval +1.13 to +4.93.

It is not the quarterbacks

This was my first guess and it's wrong.

The quarterback play has been absurd. Percentile against every FBS quarterback with 200+ plays:

Eight straight seasons at the 91st percentile or better, against a median top-10 quarterback at the 83.5th. The reputation is deserved.

It still doesn't track the residual. (Per-season numbers below are the raw talent-versus-results gap, which averages +9.05 — a different construction from the +6.95.)

The two seasons the edge nearly vanished are 2018 at +1.0 and 2020 at +3.4, and those quarterbacks graded 91st and 94th. Barrett's 2016 is the weakest passing season in the sample at the 34th percentile, and its raw gap is +7.1. That's two points under the eleven-year average, not the collapse you'd expect if the quarterback were driving this.

What the defense actually does

The reputation has this backwards, and the honest table here is smaller than you'd expect.

Twenty measures, each scored against the talent-matched peer group. Twenty tests means a Bonferroni bar of |t| = 3.02. Two clear it:

Four more are positive and don't clear: points per drive allowed (+0.76, t 2.73), line yards allowed (+0.67, t 2.82), success allowed on passing downs (+0.54, t 2.45), power success allowed (+0.53, t 2.33). Ignore them. Fitting a straight line on talent flatters the best rosters, which is the whole reason I'm scoring against peers instead.

Then the two a fan would name first. Explosiveness allowed: +0.05, t 0.16. Highlight runs allowed: −0.18, t −0.33. Third-down distance forced: −0.18, t −0.78. Flat.

So the picture of five-star athletes making splash plays isn't what shows up. What shows up is field position. Opponents start further back and reach scoring range less often, and once they get there Ohio State gives up about what anybody would.

Seven coordinators

Luke Fickell in 2015 with Chris Ash as co-coordinator, Greg Schiano 2016-18, Jeff Hafley 2019, Kerry Coombs 2020-21 with Matt Barnes taking play-calling mid-2021, Jim Knowles 2022-24, Matt Patricia 2025. The premium spans all of them.

Then there's 2021. Oregon came into the Shoe in week two, ran for 269 yards and won 35-28, and it wasn't close in the way the score suggests. The front crashed, the edge and the second level didn't hold, and the scheme never adjusted. Coombs said it himself that afternoon: "My response is that I'm responsible. That's my job. We have to play better." Day's Monday presser got headlined "there's enough blame to go around." By the next week Coombs was in the press box, Barnes had play-calling, and the base had moved from single-high to two-high. Third-down defense and three-and-outs forced are the lowest of the eleven years. If you asked a Buckeye fan to name the worst defense of the decade they'd say 2021 and they'd be right.

And the raw gap that season is +11.7, one of the better ones in the sample. Stroud graded 98th percentile.

So the season the defense was worst is a season the premium was high. I don't have a good answer for that.

Then Knowles left for three years at $9 million after a national title, and Ohio State replaced him with Matt Patricia, who had never coordinated a college defense in his life and whose last head-coaching job ended 6-26. Eight defensive starters left too. I thought it was a disaster hire. I said so at the time to anyone who'd listen.

2025 came out as the best defensive rating in the eleven-year sample. I've been chewing on that for a month and the honest position is that I have no idea how he did it.

Which is where the coordinator argument falls apart. Season to season this residual moves with a standard deviation of 6.40 points. The gaps you'd want to build a scheme story on, +9.07 against +13.33, sit inside that band. They can't separate a coordinator from a coin flip.

The Penn State control group goes the same way. Knowles took his scheme there, their 2024 defense had graded +13.7, and in 2025 it came in at +1.2. I read that as a 12.5-point collapse and used it as my cleanest evidence. But seasons grading +13 or better fall by 7.27 points the next year on average no matter who's coaching, standard deviation 5.74. Penn State's fall is about five points past expectation, inside one SD. Franklin was fired six games in on top of that. It's regression to the mean wearing a coordinator's face.

Seven men have called this defense in eleven years and the premium is there across all of them. That's consistent with a program trait.

The exception is Larry Johnson, who has coached this defensive line since 2014 and is therefore present for every season I have. In this data Ohio State and Larry Johnson are the same column, and no amount of Ohio State seasons will separate them. If he were the whole story I'd expect it in line yards and the run front, and those are the two that don't clear the bar, so I don't think he is. But that's an argument, not a test, and I can't run the test.

There's one hole in this and it's the one that bothers me. Four straight losses to Michigan sit on top of a piece claiming Ohio State converts talent better than anyone, and the question that actually matters is whether the premium holds against opponents that good. My game-level data starts in 2023. I can't split the eleven-year number by opponent quality, and three seasons won't do it. That's the number I want most and don't have.

If this is a program trait, the 2026 defense holds its field-position edge in the +0.9 to +1.1 band with Patricia in year two and another round of turnover, and the raw gap stays north of +5.

One season moving six points means very little, so next year won't settle it. Three will.

Come back in January and tell me I'm wrong.

Quotes: Coombs after the Oregon loss, September 2021 (The Ozone, The Lantern). Day's "enough blame to go around" presser (Eleven Warriors). Knowles on giving up penetration, September 2023 (Eleven Warriors). Happy to post the per-season tables, the peer-band sensitivity or the block-bootstrap intervals if anyone wants to pick at the construction.