How Important Is the Quarterback?
A win-rate model that predicts a season to within 1.71 games.
An inference model that takes a team’s quarterback statistics for a season and predicts its regular-season win rate — a way to ask whether a QB is playing to standard, or whether the rest of the roster is over- or underperforming relative to them.
The data
I pulled 2006–2024 passing and rushing data from Pro Football Reference — roughly 2,800 player-seasons — and regrouped it by team and year, since the question is about a team’s season and not an individual’s line.
Two corrections mattered. Rate stats like QBR are weighted by games played, so a spectacular three-game stretch doesn’t stand in for a season. Counting stats are normalized per game, so the move to a 17-game schedule doesn’t quietly inflate recent years against older ones.
The model
Ridge regression with a cross-validated alpha and polynomial features. It lands at 1.71 games of mean absolute error, down from 2.25 for the baseline.
The residual is the interesting part: a team that wins more than its quarterback’s numbers predict is telling you something about the rest of the roster.
- 1.71 games of mean absolute error, against a 2.25-game baseline.
- ~2,800 player-seasons of 2006–2024 data, regrouped by team and year.
- Per-game normalization, so the 17-game era doesn’t skew older seasons.