Pythagorean Expectation in College Football: Turning Points Into Expected Wins
Points scored and allowed predict a team's record better than the record itself. How Pythagorean wins work in college football, with the 2.37 exponent.
The advanced numbers, built from scratch. What each metric measures, how it's actually computed, and where it misleads.
Points scored and allowed predict a team's record better than the record itself. How Pythagorean wins work in college football, with the 2.37 exponent.
A team that goes 7-1 in one-score games gets called 'clutch' — but from first half to second half of a season, close-game records correlate at just r=0.11.
A preseason ranking is a prior; each game is evidence. Bayesian updating says results outweigh the poll after about 5 games under moderate trust.
A point spread converts cleanly to win probability under the normal model. A 3-point favorite wins about 57% of the time; a touchdown favorite, about 67%.
If your team beat the team that beat the playoff team, are you better? Model it: 'A beat B beat C' implies 'A beats C' just 51% of the time between even teams.
Every rating system has one dial — the K-factor — deciding how much one game moves a rating. Crank it up and ratings chase flukes; turn it down and they lag.
Slowing a college basketball game doesn't make the underdog better — it makes the favorite's edge matter less. The random-walk math behind 'shorten the game.'
A 7.2 yards-per-carry start is a mirage with an error bar. Uncertainty shrinks with the square root of carries, so a 20-carry sample is ±2.6 yards.
"Two-score game" isn't a category — it's a probability that depends on the clock. Modeled as a random walk, an 8-point lead is 81% safe early, 99% safe late.
A kicker's "range" isn't a yard line — it's where the expected points of a try fall below punting. Why a 52-yarder (≈1.0 EP) is a genuine coin-flip decision.
Convert two-point tries just under half the time and expected points says always go for two. The break-even is 47%; the average college team sits on it.
A live win-probability graph that lurches 40 points on one snap isn't broken — that's leverage. Build the logistic model and see where the swings come from.
Yards per attempt, adjusted net yards per attempt, completion over expected, and EPA per dropback — what each adds beyond NCAA passer rating.
Yards per carry can't tell you whether the line or the back earned the run. How line yards, opportunity rate, highlight yards, and stuff rate split the credit.
Points per game is contaminated by pace. Why points per drive, available yards, and drive-finishing rank offenses and defenses fairly — and how to read them.
Stuff rate counts the runs a defense stops at or behind the line. What it measures, how it relates to havoc, and how front-seven play shows up in the numbers.
Red-zone defense is its own skill: holding offenses to field goals, the field-shrinks dynamic, and why TD% allowed and points-per-trip beat scoring percentage.
Go, punt, or kick on fourth down — by field position and distance. The win-probability framing, and why analytics pushes toward going for it.
Sack rate allowed versus created, why pressure and sacks aren't the same thing, and how much credit belongs to the line, the quarterback, and the scheme.
Red-zone scoring percentage flatters teams that settle for field goals. Why touchdown rate and points per trip tell the truth about red-zone offense.
Third-down conversion rate is a headline efficiency stat — but third-and-1 and third-and-12 are different games. How distance-to-go ties it to success rate.
The exact NCAA passer rating formula, worked term by term, how it differs from the NFL's capped rating, and the real limitations of a single quarterback number.
Third-and-1 and third-and-12 aren't the same game. How splitting plays into standard downs and passing downs sharpens every efficiency number.
Turnover margin decides games but barely predicts them. Why fumble recoveries are close to a coin flip, how turnover luck regresses, and what's actually a skill.
Havoc rate counts the plays a defense blows up — tackles for loss, forced fumbles, and passes defensed — as a share of snaps. How it's built and what it reveals.
Not all yards are equal. How expected points turns a spot on the field into a scoring value, and why field position quietly decides close games.
The Fremeau Efficiency Index grades whole possessions, not single plays. How drive-based, opponent-adjusted efficiency complements SP+ and explosiveness.
Success rate measures how often you stay on schedule; explosiveness measures the big plays. How yards per play and marginal explosiveness complete the picture.
A worked strength-of-schedule calculation from a real schedule, and why unadjusted stats reward the soft slate.
Possessions, points per 100, and opponent adjustment — the tempo-free framework that makes basketball stats comparable.
The NCAA's NET, quadrants, and the selection committee: what the metric rewards, what it ignores, and how to read a team sheet.
Compute success rate and expected points added by hand from a real drive, and see why down-and-distance beats raw yards.
What SP+ actually measures, why opponent adjustment matters more than raw stats, and how a tempo-free efficiency rating is built.