Slate Stats › NFL › Positions

Running Backs: What Actually Predicts a Fantasy Season

A running back's fantasy season is decided by how many times he is handed the ball, and almost nothing else that is commonly cited survives a reliability test. Carries repeat at 0.92 between halves of a season. Yards per carry repeats at 0.14.

The handcuff is the biggest edge in the dataset

When a lead back does not play, his backup gains +5.29 carries and about 20 rushing yards against his own trailing baseline, net of control (t=15.6). Nothing else measured in this project comes close — it is three to five times the size of the best matchup signal, and unlike a matchup it is known with certainty once the inactives are out.

His touchdown chance moves too, and that is unusual

Touchdown rate is the least reliable thing a skill player does (0.56 between halves of a season), so a touchdown projection is normally close to noise. The exception is this one: in a lead-back-out spot the backup's chance of scoring rises 24.6 points (t=6.8), because he inherits the goal-line work along with the carries.

Favorites run; underdogs do not throw much more

Bucketing team-games by the pre-game spread, rush attempts move cleanly across the range (29.0 for a big favorite down to 24.6 for a big underdog) while pass attempts move about two plays and not even monotonically. The usable claim is favorites run more, not that underdogs throw more.

The line matters less than the front they face

A 2x2 of shrunk offensive-line and opposing-run-defense ratings moves rushing yards by 7.0 a game between the extreme cells — and entirely through efficiency, with carries moving 0.6. Holding the line good, the opposing front is worth +4.6 yards; holding the front weak, the line is worth +1.3. The defensive half carries about three and a half times the weight, which is the opposite of how the rule is usually stated.

Running back leaders: carries per game

2016-2026. only those with at least 4 games, applied because the question set no minimum and a rate over a handful is not a ranking.

Every figure above was measured against a control and carries its sample size and t-statistic. Verified 2026-09-29. A lift measured without a control is not merely larger here — it is often significant and wrong, which is why each one states what it was compared against.