Michigan's Big-Game Gap

A gap that survives adjustment

Success rate, all 13 FBS games · marquee = AP Top 25 opponent at kickoff (Oklahoma #18, Ohio State #1, Texas #12)

Facing better defenses accounts for a bit under half of the raw disparity. The other half is a real, opponent-adjusted shortfall in marquee games. Every game's "expected" value uses each opponent's full-season defensive rating, fit by this repo's independent opponent-adjustment formula — not a leakage-safe forecast, a retrospective grade. See Method at the bottom.

Overall offense, adjusted to opponent defense

Success-rate residual = actual minus opponent-adjusted expectation, chronological by week.

The quarterback, game by game

Two views of Bryce Underwood's passing, both run through the identical opponent-adjustment formula, restricted to dropbacks: success rate (stayed on schedule) and net yards per dropback (overall throughput, sacks included).

Unlike a team-level gap that's partly explained by opponent quality, the pass-specific success gap is larger, not smaller — and yards/dropback stays clearly negative in marquee games too. For Michigan, the passing problem in big games wasn't hidden by the run game. It was the whole story.

Good calls, bad calls: fourth downs

Every fourth down Michigan faced in 2025, graded against a from-scratch expected-points model trained on 1.5M+ plays across 11 seasons (2014–2025), and three historical rate curves fit the same way: go-for-it conversion rate by distance, field-goal make rate by kick distance, and punt outcome by field position. Not a win-probability guess — what has actually worked, at scale, in these exact spots.

Rush vs. pass, confirmed a different way

The same expected-points model used to grade fourth downs also grades every offensive snap. This is a fully independent build from the opponent-adjustment formula above — and it lands on the same story.

Passing was worth less than half of what rushing was worth per play across the season (+0.091 vs. +0.249 EPA/play) — and against the three marquee defenses, passing EPA went negative (−0.278), while rushing merely flattened (+0.012, essentially replacement level). Across 819 graded offensive snaps, this independently-built model confirms the same shape the opponent-adjusted passing numbers above show: Michigan's run-first identity traveled to big games. The pass game didn't.

Three games, one problem, three mechanisms

Michigan only had three marquee games in 2025, and all three were pass-game underperformances — but not in the same way.

Ohio State (L, 9–27) is total collapse: pass success, team success, and yards/dropback were all clearly below what a #1-ranked defense should have allowed. Nothing offsets anything here.

Oklahoma (L, 13–24) is a pure consistency failure: pass success and team success both cratered, but yards/dropback landed almost exactly at expectation. A boom-or-bust day — long stretches of failed dropbacks broken up by enough big gains to keep the yardage total normal.

Texas (L, 27–41, CFP) is the one the box score hides: team-level success was barely dented — the run game and other phases picked up real slack — but the passing-specific numbers were still clearly poor, the worst throughput mark of the three. A team-only view of this game would miss the quarterback's day almost entirely.

How this was built