Do touchdowns regress?

The question

"His touchdowns will regress" is the most common sell-high argument in fantasy football, and it rests on a claim about the number: that a touchdown rate above what a player's yardage would predict is luck, and luck gets paid back. This study measured both halves of that claim on six NFL seasons, expecting the conventional answer, and got a different one.

What was tested

Two things, on DataBaller's box scores from 2020 through 2025. First, reliability: how well a player's touchdowns per touch over the first stretch of a season predict his rate over the rest of it, for rushing touchdowns per carry, receiving touchdowns per target and passing touchdowns per attempt. Second, persistence: whether a player whose touchdown rate over his last 100 carries, 40 targets or 200 attempts sat above or below his own two-year level stayed there over the rest of the season, at ten mid-season dates from 2021 through 2025. The claim written before the run was that it would not.

What turned up

Touchdown rate never stabilizes inside a season. Receiving touchdowns per target correlate with themselves at 0.14 to 0.16, flat from 20 targets to 80. Rushing touchdowns per carry sit at 0.33 to 0.34 across 50 to 150 carries, passing touchdowns per attempt at 0.24 to 0.38 across 50 to 300 attempts. None come near the 0.7 that counts as a stable statistic. Receiving touchdown rate is the batting average on balls in play of football.

And a player's touchdown rate persists anyway. Displacement from his own baseline carried into the rest of the season at a rank correlation of +0.433 for rushing, +0.269 for receiving and +0.284 for passing, each far beyond chance. The claim that it would not persist was falsified on every side.

Yards per touch in the same data tells the opposite story. Its reliability is just as low, yards per carry at 0.30 to 0.33 and yards per target at 0.31 to 0.39, but its displacement gives itself back: players far above or below their own level closed the gap over the rest of the season while players in line with themselves drifted away from it.

What it means when you ask

A touchdown rate above a player's own history is noisy and partly real. A goal-line role, a red-zone target share and a quarterback's eye for the end zone are things that exist, and they show up in the persistence. So DataBaller's Sustainability Score for football reads yards per touch, where the give-back is measured, and reports a touchdown surge as context beside it rather than discounting it. The Scoring Shift forecast for football leaves touchdowns out entirely, as target and as input, because a rate that is mostly noise and partly role is the wrong thing to build a forecast on either way. Ask whether a player's touchdown pace will hold and the answer says what his yardage supports and reports the touchdown rate against his own history, without promising it back.

What it does not mean

It does not make one hot stretch a forecast of next season's touchdowns; the reliability says the number is mostly noise even when its sign persists. It says nothing about a team's red-zone touchdown rate, which is a different quantity and does regress toward the league average in public studies. And the persistence measurement shares a baseline term on both sides, so a role that changed for good mid-season, a back who became the goal-line back, inflates the correlation. That confound is on the record and does not rescue the original claim.

The record

Measured 31 August 2026 with derive_football_thresholds.py and the first round of backtest_football_sustainability.py, five seasons of persistence cells. The football Sustainability Score shipped in the form the result demanded, and the 2026 season is its first test on games no model has seen. The full portfolio and its testing rules are on the decision metrics page.