Week 2 · Sun, Sep 20, 1:00 PM
Model picks
NE
Model spread
NE -5.8
Market spread
4.5
the market forecasts these games better than this model
How the teams compare
heaviest model weight first · 0–100, oriented so higher is always better
Expected Points Added per play· weight 11.8
Offense
EPA allowed
Passing efficiency· weight 8.0
Offense
EPA allowed
Drive-sustaining rate· weight 6.6
Offense
EPA allowed
Position-weighted availability score· weight 6.6
No data yet for this metric. Nothing has been published for these teams at this point in the week. It will fill in once it exists.
Sacks taken per dropback· weight 5.7
Offense
defensive rate
Rushing efficiency
Offense
EPA allowed
Completion % Over Expected
Offense
EPA allowed
Giveaways per play
Offense
defensive rate
20+ yard completions per attempt
Offense
EPA allowed
This chart is not the model. It shows the inputs that read well as a side-by-side. 7 of the model’s heaviest-weighted inputs have no head-to-head form and are not drawn here: point differential, overall matchup, team win rate, pass matchup among them. The full list of all 55 is on the methodology tab.
Around the game
- SchemeNE
NE faces a 4-3 front
PIT's roster profiles as a 4-3 base defense.
Base front is inferred from roster composition (agrees with snap data at Spearman 0.50). Most teams play nickel by default, so this is a weak descriptor: measured defensive efficiency carries far more weight in the model.
- Recent formNE
NE +6.6 pts/game trend
Over the last 10 games (exponentially weighted), NE has been outscoring opponents by 6.6 points per game.
- Recent formPIT
PIT -3.9 pts/game trend
Over the last 10 games (exponentially weighted), PIT has been being outscored by opponents by 3.9 points per game.
- VenueNE
NE at home
Home teams have won 54.7% of games in the training window; the model prices this in rather than applying a fixed bonus.
- RestNE
NE holds a rest advantage
4 more days of recovery than the opponent.
Every figure here comes from Calibrated Logistic Regression (core + market features), validated walk-forward by season (train on all prior seasons, test on the next). For how the model is built, what the metrics mean and how it has actually performed, read the methodology.