College running back ratings

Every qualified FBS running back’s season, rated on EPA per rush, share of his team's carries and receiving yards per team pass play. It describes the season, not a projection or a betting number.

What it measures. Every pass target and designed run in FBS games since 2014, taken from play-by-play. A rating is scaled like the college QB rating: 100 is an average qualified player and 15 points is one standard deviation. The two receiving ratings and the running back rating use the same three ideas.

Running back = 30% EPA per rush + 40% share of the team’s rushes + 30% receiving yards per team dropback. EPA per rush is CFBD’s predicted points added on each designed run, adjusted for the defense faced and pulled toward average.

Yards per team dropback. A player’s receiving yards divided by the number of pass plays (attempts plus sacks) his team ran in the games where he was credited with a target or a carry. It stands in for yards per route run. A team that throws fewer than 25 times in a game is credited with 25, so an option offense does not turn a handful of catches into a huge number. Each game’s yards are then scaled by the pass defense faced: league-average yards per dropback divided by what that defense usually allows (from a model fit to every FBS game of the season, limited to 0.6 to 1.6), so yards against a stingy defense count for more.

Small samples. Each number is pulled toward the position average in proportion to how little volume sits behind it (empirical-Bayes shrinkage, strength set from the data). A player with two big games reads as above average, not as a star. Plays after a game is decided (a lead of more than 38 points in the second quarter, 28 in the third, 22 in the fourth) count a quarter as much in the EPA-per-rush number; yards and shares count every play.

What was tested and left out. Efficiency per target (EPA per target, yards per target, catch rate) does not hold up in college data. From one season to the next it correlates at 0.07 to 0.14 for receivers, against about 0.5 for yards per team dropback, so it is not in any rating. Two other adjustments were tested and made things worse: one for the quarterback who threw to the player, and, for backs, one for the team’s blocking and scheme.

What the data does not have. There is no routes-run or snap data for college football, so a receiver is measured against all of his team’s pass plays, not the ones he ran a route on. Targets are often not named on incompletions, so target share counts only games where the feed names the target on at least 80% of incompletions. Players whose name cannot be matched to an ID, or who have no usable roster position, are not rated.

Low team pass volume. Navy, Army and Air Force run option offenses and throw far less than anyone else, so their receivers can rate high on few catches. The math is not changed for them. A player is flagged when his team throws no more per game than the lowest 5% of FBS teams that season. That is mostly the option teams, plus any other team that rarely passes. Read a flagged rating as resting on fewer plays than it looks.

A season in progress. It is scored against the 2014–2025 averages and spreads, never against its own few weeks, so a 2026 rating means what a 2019 one does. The qualifying minimum is the full-season minimum times games played over 12, so it rises every week. Ratings update weekly.

How well it holds up. Weights were fit on 2014–2018 and checked on 2019–2025. For players with a real workload two seasons running (15 catches, 12 for tight ends, 50 carries for backs), the rating correlates with itself at 0.43 for receivers, 0.52 for tight ends and 0.45 for running backs, against 0.33, 0.48 and 0.38 for raw yards. It describes a season; it is not a projection or a betting number.