Stats and results

What wins games

For 5,390 graded games (Oct 18, 2022 to Nov 8, 2026), how well each team's numbers going into the game lined up with the final margin. Every comparison is home team minus road team, with the sign set so that "better" always points the same way.

Which numbers line up with winning

Correlation with the final margin · how often the better side won

Pregame gap in… Correlation with final margin Better side won

Elo ratingmodel input 0.42 64%

Season scoring marginmodel input 0.41 63%

Net rating, last 10model input 0.37 63%

Offensive rating, last 10model input 0.29 60%

Defensive rating, last 10model input 0.27 59%

Rotation points missingmodel input 0.22 59%

Opponents' shooting allowednot an input 0.21 57%

Shooting (eFG%)not an input 0.20 58%

Turnover ratenot an input 0.19 56%

Opponents' offensive reboundingnot an input 0.18 56%

Offensive reboundingnot an input 0.08 51%

Opponents' free throws per shotnot an input 0.08 52%

Back-to-backmodel input 0.08 57%

Free throws per shotnot an input 0.08 52%

Turnovers forcednot an input 0.07 53%

Rest daysmodel input 0.06 55%

model input four-factor stat the model doesn't see directly (10-game average). Home teams won 56% of these games regardless.

What stands out

Elo rating lines up best: the better-rated team won 64% of the time. Measures built from results (Elo, season margin, net rating) beat any single box-score stat.
Among the four factors, opponents' shooting allowed leads. Shooting and shooting defense matter most; offensive rebounding and free throws barely separate winners from losers on their own.
Injuries matter: the side missing fewer rotation points won 59%.
Back-to-backs: the rested side won 57% when only one team was on the second night, yet the model gives that input almost no weight (0.06 pts). Part of that edge is home court, since road teams are more often on back-to-backs, but it's worth a closer look.

How to read this

Correlation runs from 0 (no relationship) to 1 (the gap fully predicts the margin). Single-game NBA results are noisy, so 0.4 is strong. These stats overlap heavily: shooting feeds offensive rating, which feeds net rating and, through results, Elo. A stat can matter for winning and still add nothing to a model that already has the summary. This is correlation over past games, not proof of cause.

What the model leans on

Average push on the projected margin, 118 explained forecasts

Elo rating3.6

Defensive rating, last 101.4

Season scoring margin1.0

Rotation points missing0.7

Pace, last 100.7

Games played (how much to trust this season)0.7

Offensive rating, last 100.6

Rest days0.1

Back-to-back0.1

How many points each input moved a forecast's projected margin, on average, across the forecasts that can be explained exactly (how). Compare with the ranking above: the model leans hardest on the same results-based measures that line up best with winning. The sample is this season's forecasts only, so treat small values as rough.

What builds a rating

Method

offense defense

Shooting (eFG%)+1.51 rating pts per 1 pt ±4.1

Turnovers (TOV%)−1.42 rating pts per 1 pt ±2.0

Offensive rebounding (OREB%)+0.45 rating pts per 1 pt ±1.7

Free throws (FTM per FGA)0.29 rating pts per 0.01 ±0.9

Offensive and defensive ratings are the model's view of recent form. The four factors explain 95% of a game's rating. Bars show the typical swing each factor gives a team's 10-game rating, in points per 100 possessions. Shooting moves ratings about twice as much as anything else. Every game page traces its teams' ratings back to these factors under "Built from the box scores".

Home court, measured

Does it shrink against better teams?

Home teams win more often, and the model knows it: Elo adds 70 rating points, about 2.5 points of margin, to every home team before the score models take over.

55.6%

home teams won

5,390 games

+2.1 pts

average home margin

actual results

+2.2 pts

average forecast margin

what the model expected

63%

home team favored

share of forecasts