Ohio State gets more out of its players than any team in college football. Alabama, which recruits better, doesn't.

Ohio State recruits the 2nd or 3rd best roster in America almost every year. That part is not interesting; everyone knows it, and a team stacked with five-stars is supposed to win.

The part sitting on top of the recruiting is the part worth measuring. Take every FBS team from 2015 through 2025, 1,434 team-seasons. For each one, measure two things: how good the team actually was (points per play, adjusted for who it played) and how good its roster was on paper (its recruiting rankings). Across all 1,434 seasons more talent means more winning, and that relationship gives you an expectation: a roster this good normally produces about this much. How far a team beats its own expectation is the part recruiting does not explain. Ohio State beats its expectation by +9.05 points per game, among the largest of any team in college football.

There is an objection here that kills most posts of this kind, and I have not seen anyone else deal with it, so I will break my own number first.

That number comes from fitting a straight line through the relationship between talent and winning, and the real relationship curves. When you fit a straight line to a curve, the very best rosters get handed a head start before they play a down. Sort all 1,434 seasons into ten equal groups by recruiting, where group 1 is the worst-recruiting tenth of college football and group 10 the best, then average how far each group finished above or below its own expectation. If the method were clean, every one of those averages would sit near zero, because expectation is supposed to account for talent already.

The bottom nine groups land where they should. The top tenth clears its expectation by more than two points a game simply for sitting at the top of the range, and the 46 most talented seasons in the sample (Ohio State, Alabama, Georgia, Clemson, LSU, Florida State) clear it by nearly five before anyone plays a game. Ohio State sits in that top group every single season, so a real chunk of its +9.05 is the shape of the method rather than anything about football.

The fix is to stop comparing Ohio State against the whole country and compare it only against teams like it. For each Ohio State season, take every other team whose recruiting that year landed close to Ohio State's, about 87 teams a season, and ask how Ohio State did against just those. And leave Ohio State out of its own comparison group: it and Alabama together are 22 of the 46 elite-roster seasons in the sample, so otherwise they would be half of their own grading curve.

Compared only against teams whose rosters were as good as its own, Ohio State is +6.95 points a game better (t = 5.20).

Smaller than the headline number. It is third of 130 programs on this measure rather than first, behind Air Force and App State, and I originally wrote that it was the largest in the sport. Every number from here on is this like-for-like version.

For scale, here is where that leaves the programs recruiting in the same neighborhood. "Average class rank" is where a program's recruiting classes finished nationally, on average, across the eleven years; "average finish" is where its teams actually ended up in the ratings.

Alabama recruits better than Ohio State over this window, a 1.3 average against 2.8. It converts that into an average finish of 10th, against Ohio State's 4.3.

The same talent-matched test run on Alabama returns a premium of +1.77, t = 1.38, which is not statistically distinguishable from zero. That surprised me more than anything else in the data. Against other teams holding equally elite rosters, Alabama has been roughly average at conversion over these eleven years. Its enormous raw residual is mostly the artifact I just described.

Ohio State's survives the same test. +6.95 against +1.77, over the same window. That reverses how the two programs usually get ranked: Alabama's edge over the field has been its recruiting, and Ohio State's has been what it does after the recruiting.

Talent-matched, leave-one-out, 2015-2025. Zero means 'exactly as good as the roster says'.

Ohio Statet = 5.20+6.95

Michigan+3.66

Clemson+2.76

Georgia+1.86

Alabamat = 1.38 — n.s.+1.77

LSU-0.02

USC-2.94

Texas-3.27

Alabama recruits better than Ohio State over this window and converts it into +1.77, which is not statistically distinguishable from zero (t = 1.38). Ohio State's +6.95 carries t = 5.20.

GRIDPEX.COM

Two more objections, both worth answering.

"This only exists in your rating." The most important objection anyone can make about a model's output, and it is testable. SP+ is independent, widely cited, and built differently. Refit the identical construction — per-season regression of rating on recruited talent — using SP+ as the outcome:

Two ratings built by different people from different inputs land within 0.2 points of each other on an eleven-year residual. The effect survives outside my model.

"The Big Ten is soft." The rating is opponent-adjusted, and if the adjustment were too weak every Big Ten team would look inflated. The conference's average residual across 162 team-seasons is +0.13, t = 0.27 — dead neutral. There is no conference bonus hiding in this.

"The recruiting rankings undersell them." They don't — if anything the opposite, and it's measurable. Fit NFL draft picks on recruited talent across all 817 team-seasons:

picks = 0.98 + 2.10 × talent (SD)

Ohio State's roster should produce 5.47 draft picks a year. It produces 8.50. That is three extra NFL players every season, on top of an expectation already set by the best recruiting in the country. Whatever is happening in Columbus is not a scouting-service illusion.

The explanation everyone gives

Quarterbacks. Haskins, Fields, Stroud, Howard, Sayin. That is a decade of the position played at a level almost nobody else has. The record, primary quarterback each year, as a percentile against every FBS quarterback with 200+ plays:

Eight straight seasons at the 91st percentile or better. The median top-10 team's quarterback sits at the 83.5th, so this is genuinely exceptional and the reputation is deserved.

It is also, as an explanation for the residual, mostly wrong.

What the defense is actually doing

The lazy version of this section says the defense drives it and stops there. That is a residual wearing a jersey. So, the decomposition.

For all seventeen defensive metrics, measure Ohio State against teams holding an equally good roster. Talent-adjusted, every season, sign-flipped so positive always means better:

Seventeen metrics were tested, so the bar is Bonferroni-corrected. The six rows below were computed by fitting each measure on talent across all 130 teams. Scored against talent-matched peers instead, which is the method this article argues for at the top, only two of them clear.

What Ohio State's defense does better than equally stocked rosters:

What it does not:

That is the answer, and it runs against the reputation. The picture of an Ohio State defense is five-star athletes making splash plays. On splash prevention they are indistinguishable from teams with the same roster. Explosive plays allowed: +0.09. Big runs allowed: +0.14. Neither is different from zero.

What they do instead is refuse to let a drive start well or continue. Opponents begin further back (+0.89), get stopped at the line (+0.75 line yards, +0.75 in power situations), go three-and-out more (+0.73) and convert fewer third downs (+0.61). Then the largest effect in the table: they reach scoring position far less often (+1.21). When an opponent does get there, Ohio State gives up roughly what anyone else would.

They do not stop you from gaining yards. They stop you from getting the ball into a place where yards matter.

Season by season the same thing shows up as a slope. The two years Ohio State's defense was merely league-average, 2018 and 2020, are the two years the edge over recruiting nearly vanished, and both had quarterbacks at the 91st and 94th percentile.

Split the sample in half and the top of that table holds in both halves: scoring opportunities allowed runs +1.04 early and +1.37 late, line yards +0.80 and +0.71, opponent field position +0.69 and +1.10. So this is no five-year fluke inside an eleven-year window. If anything the advantage is growing.

Four coordinators, one uncredited trade

The decomposition above is the eleven-year average. Split it by who was calling it and a real argument appears, one the fanbase has already been having.

Greg Schiano 2016–18, Jeff Hafley 2019, Kerry Coombs 2020–21 (demoted mid-2021, Matt Barnes took play-calling), Jim Knowles 2022–24, Matt Patricia 2025–.

Two things jump out, and both match what people were saying at the time.

2021 is the worst year in the sample, and the data says exactly what the complaints said. Third-down defense and three-and-outs forced are the lowest of the eleven years. Line yards stayed at +0.9, so the interior was fine. The problem was everywhere else, which is precisely the diagnosis after Oregon ran for 269 and won 35–28 in the Shoe: the front crashed, the edge and the second level did not hold, and the scheme did not adjust. Kerry Coombs said it himself that afternoon: "My response is that I'm responsible. That's my job. We have to play better," and, on the scheme, "I don't think there's any question that we have to do things differently going forward." Ryan Day's presser that week was headlined "there's enough blame to go around." Coombs went to the press box, Barnes took play-calling, and the base moved from single-high to two-high. The residual for that season: +0.50, the second-smallest in eleven years.

2023 is the hard break, a year after Knowles arrived. His first season, 2022, still looks like a Schiano defense: line yards +1.5, stuff rate +0.6, explosives allowed −1.4. The change lands in 2023, the year he dropped the Jack, blitzed less and played straighter up, at Ryan Day's request for something more conservative. Knowles described the trade himself that September: "If you live in that world against teams where you have a skill advantage, it can look really nice. But when you get into the matchup games, I've found that it can hurt you." Living in that world is what the 2016–22 numbers are. He gave it up on purpose.

Look at what that actually cost and bought:

The complaint that followed — fewer tackles for loss, fewer sacks, not getting off the field, and by September 2024 an Eleven Warriors thread titled "Is Jim Knowles Being Throttled?" — was correct as an observation. Stuff rate fell. Third-down distance forced collapsed to the worst mark in the sample. They stopped living in the backfield.

Everybody drew the wrong conclusion from it.

That defense gave up penetration and bought explosive-play prevention, and the exchange rate was excellent: explosives allowed swung 3.3 standard deviations, points per opportunity swung 2.8, and the season residual went from +9.07 to +13.33, then +13.18, then +12.45. The three most valuable defensive seasons in the eleven years are the three since the scheme got more conservative.

There is an irony in it. Asked about bend-but-don't-break when he was hired in 2022, Knowles rejected the whole idea: "I would never call a defense that I was associated with that. To me, defense is a right now proposition… I never think about bending but not breaking." The numbers say the 2023–25 defense is the closest thing to it Ohio State has fielded in a decade, and the best defense Ohio State has fielded in a decade.

So was it Knowles?

This is the question the section above begs, and 2025 answered it in the cleanest way anyone could have designed.

Jim Knowles left. Not fired — Ohio State was slow with an extension and Penn State offered three years at $9 million, making him the highest-paid defensive coordinator in college football. He had just called the nation's best defense on a national championship team.

Ohio State replaced him with Matt Patricia, a career NFL coach who had never coordinated a college defense, and did so on an explicit instruction: Ryan Day wanted something simpler, an NFL model, after the perceived complexity of the Knowles scheme. Eight defensive starters also left.

New coordinator. New scheme, with Patricia bringing front variation and a Penny front Columbus had never seen, based out of Cover 3 nickel. Eight new starters. If the coordinator is the cause, this is where the number falls over.

It did not fall over. 2025 is the best defensive rating in the eleven-year sample, better than either Knowles year, and the shape is identical: the line still gets conceded, explosive plays are still refused, and opponents almost never reach scoring position. A different coach with a different scheme and eight new starters reproduced the profile almost metric for metric.

And there is a control group, because Knowles took his scheme somewhere else.

Penn State in 2024, before he arrived: stuff rate +2.0, line yards +1.5, defensive rating −9.3, season residual +13.7, one of the best defensive seasons in the country. In 2025, with Knowles calling it: stuff rate −0.5, line yards −0.7, rating −1.8, residual +1.2. Every metric in that chart moved the wrong way. The reporting from State College said the players could not absorb the playbook in the time available.

That season is genuinely confounded, and the article does not lean on it: Penn State started 0–3 in the Big Ten, James Franklin was fired six games in, and a program in that condition is not a clean test of anything. But two things survive the confound. The metrics are talent-adjusted, so the roster turnover is already priced in. And Penn State did not retain him afterwards — he is at Tennessee now.

So the ledger reads: the coordinator left, and the defense he left behind got better. The coordinator arrived somewhere else, and that defense got much worse. One season each, and the direction is the same in both.

What was actually happening in Columbus

Put the two scheme changes next to each other and they are the same intervention, run twice.

2023. Knowles drops the Jack, blitzes less and plays straighter up, reportedly because Ryan Day wanted something more conservative. Penetration falls. The defense gets much better.

2025. Day replaces a coordinator coming off the nation's best defense with an NFL coach hired explicitly to run something simpler. The defense gets better again, to the best mark in eleven years, with eight new starters.

The two largest defensive improvements of the decade share a common factor, and the coordinator is not it. Both followed the same instruction from the same head coach: make it simpler. And the one season in the window where a coordinator got to run his own complex system at full stretch, 2022, is Knowles's worst year at Ohio State (+9.07, the smallest residual of his three).

That is also what the eleven-year number has been saying the whole time. The +6.95 premium spans four coordinators. Schiano, Hafley, Coombs, Knowles, now Patricia. Five men, four schemes, one number that does not move. A program-level trait does not care who is calling it.

The obvious objection

The natural reply is that all of this is just obvious. When you have the best athletes in the country you simplify, let them play fast and stop asking them to think. Complexity is for teams that need to manufacture an edge.

That is a testable claim. If it were right, the payoff to winning in the backfield should shrink as talent rises. Elite teams would be better served by structure and discipline, and havoc would be a poor man's substitute for talent.

Across all 1,422 team-seasons, split into fifths by recruited talent:

No trend at all. Winning at the line pays about the same whether you recruit 5th or 105th, and it out-earns bust-avoidance in every tier. Narrow to the genuinely elite, above +1.8 SD of talent, n = 90, and the gap widens: havoc r = 0.276, structure r = 0.033.

So college football does not support the rule that talent means keeping it simple. The sport says roughly the opposite: go get the ball.

An Ohio State fan should sit up at the next table. Over the full eleven years this is the most penetrating elite defense in the sample:

Alabama is less aggressive at the line than Ohio State. Georgia is the genuine structure-first blue blood, and Georgia's conversion premium is +1.86, which does not clear significance.

Which sharpens the whole thing rather than explaining it away. Ohio State built its eleven-year premium in the havoc years. Then in 2023 it gave the havoc away, stuff rate +0.6 to −0.8, third-down distance forced to the worst mark in the sample, and got better. It walked away from the thing the rest of football says pays, and it worked.

One limit, because it bounds the claim: this data has no blitz rate, no pressure rate and no coverage charting. "Havoc" here means winning at the line of scrimmage, which good players do in simple schemes too. The aggression payoff demonstrably does not fade with talent. Scheme complexity itself cannot be measured directly, and a reported storyline is not dressed up here as a measured one.

The 2026 test, with a number attached

If this is a program property, Ohio State's defense stays roughly where it is in 2026, with Patricia in year two and another round of turnover. That means a residual in the +1.0 to +2.0 band on scoring opportunities allowed and explosives allowed, and a team residual near +10.

If it was Patricia personally, it should slip when the roster churns again. If it was Knowles all along, it should already have slipped, and it did the opposite.

I would bet on the program, and the bet is falsifiable. Come back in January.

Three things a skeptic goes after

Every one of these is testable, so none of them stays an open question.

"Your peer group above +2.0σ is thin — a handful of elite rosters is doing all the work." Answerable directly: vary the matching band and see whether the number moves.

The premium runs between +6.27 and +8.68 across every band. The narrowest one, 42 peers a season and the thinnest comparison available, still returns +6.56 at t = 4.68. Band width is not driving this, and if anything the tight band is the conservative choice.

"Eleven consecutive seasons of one program aren't eleven independent observations, so your t is inflated." Correct in principle — good years tend to follow good years, which can make a result look more certain than it is. The test for it resamples the eleven seasons in unbroken runs rather than one at a time, so whatever carryover exists between neighboring years stays intact inside each run:

Even at four-season blocks, which assumes dependence running a third of the window, the 2.5th percentile sits at +3.92 and the interval never approaches zero. Read that last column as a count of draws rather than as a p-value. Zero hits in 20,000 draws puts the probability below about 1 in 6,700, which is the finest a 20,000-draw resample can resolve. Given a lower bound near +4, that is what you would expect. The serial-correlation objection is real, and the answer holds.

"You can't say why the defense is good." The section above does: drive denial and line play, not splash prevention. Six measures clear a bar corrected for having tested seventeen of them, and the two everyone would name — explosive plays and big runs allowed — are indistinguishable from zero.

What it adds up to

Ohio State converts elite talent into results better than any program in football: +6.95 points a game against teams holding an equally good roster, over eleven seasons, at t = 5.20. An independently built rating puts it at +9.24. Alabama, recruiting better over the same window, converts at +1.77, which is not distinguishable from average.

The quarterback is not the answer here, extraordinary as the quarterbacks have been. Eight straight seasons at the 91st percentile or better, and the residual barely tracks them.

The defense does it, through the part of the defense nobody talks about. On stopping explosive plays and big runs, the splash that five-stars get recruited for, Ohio State is indistinguishable from teams with the same roster. What it does is take away the field. Opponents start further back, get stopped at the line, and reach scoring position 1.21 standard deviations less often than anyone else with that roster.

And no coach does it either. Five men have called this defense in eleven years and the number has not moved. The best coordinator in the sport left for a record contract and the defense posted its best rating of the decade under an NFL retread with eight new starters; the coordinator's own scheme, run at full complexity somewhere else, produced the largest single-season defensive collapse in the sample. The two biggest improvements of the decade both followed the same instruction from the same head coach: make it simpler.

Ohio State does not have a great defensive coordinator. It has a machine that turns whoever is holding the headset into one.