Rest asymmetry: the differential moves outcomes, the shared load does not
Across 4793 NBA games the home win frequency rises from 0.4944 when the visitor is the better-rested side by two or more days (n=180) to 0.6409 when the home side is (n=259), a 0.1465 spread, while the two symmetric-congestion groups sho...
How to read this figure
Each cell reports n_games first, then the observed home win frequency and its game-cluster bootstrap interval. The contrast panel is a difference between two SEPARATE groups of games taken inside every bootstrap replicate, so its interval carries the joint resampling noise of both groups; it is not a paired within-game comparison, and a contrast is masked when either side is below the floor. In the market panel a gap whose interval contains zero means the recorded pregame reference forecast and the observed frequency agree within noise for that cell.
Published panels
Schedule population: 4,793 games over 4 seasons (2022-23, 2023-24, 2024-25, 2025-26). Priced population: 1,103 games, 2025-26 only. It is a subset of the schedule population and is the only population with a recorded reference forecast.
Coverage: 4,846 games in the source table, 53 dropped for a missing rest day, 0 dropped for a missing outcome, 4,793 measured, 1,103 of those carrying a recorded pregame reference forecast without a quote timestamp.
| Rest cell | Games (n) | Home win frequency | 95 percent interval | Masked |
|---|---|---|---|---|
| away +2 or more | 180 | 0.4944 | 0.4236 to 0.5668 | no |
| away +1 | 775 | 0.5161 | 0.4809 to 0.5519 | no |
| equal | 2,639 | 0.5517 | 0.5333 to 0.5709 | no |
| home +1 | 940 | 0.5755 | 0.5430 to 0.6062 | no |
| home +2 or more | 259 | 0.6409 | 0.5827 to 0.7016 | no |
Scroll horizontally for all columns
Schedule population. A masked row keeps its count and publishes no frequency.
| Rest cell | Games (n) | Difference vs equal | 95 percent interval | Excludes zero |
|---|---|---|---|---|
| away +2 or more | 180 | -0.0573 | -0.1307 to 0.0165 | no |
| away +1 | 775 | -0.0356 | -0.0739 to 0.0037 | no |
| home +1 | 940 | 0.0238 | -0.0142 to 0.0595 | no |
| home +2 or more | 259 | 0.0892 | 0.0257 to 0.1502 | yes |
Scroll horizontally for all columns
A difference between two separate groups of games, taken inside every bootstrap replicate; a row is masked when either side of the difference is below the floor.
| Congestion state | Games (n) | Home win frequency | 95 percent interval | Masked |
|---|---|---|---|---|
| both on a back-to-back | 245 | 0.5510 | 0.4857 to 0.6136 | no |
| mixed | 0 | masked | masked | no games in this cell |
| neither on a back-to-back | 2,394 | 0.5518 | 0.5325 to 0.5728 | no |
Scroll horizontally for all columns
restricted to games at an equal rest differential Similar observed shares here do not establish absence of fatigue.
| Season | Games (n) | Away rested more: freq (n) | Equal: freq (n) | Home rested more: freq (n) | Home-more minus away-more |
|---|---|---|---|---|---|
| 2022-23 | 1,230 | 0.5550 (209) | 0.5699 (665) | 0.6152 (356) | 0.0602 |
| 2023-24 | 1,230 | 0.4885 (260) | 0.5429 (687) | 0.5936 (283) | 0.1051 |
| 2024-25 | 1,230 | 0.4880 (250) | 0.5390 (692) | 0.6042 (288) | 0.1162 |
| 2025-26 | 1,103 | 0.5254 (236) | 0.5563 (595) | 0.5368 (272) | 0.0114 |
Scroll horizontally for all columns
Schedule population, split by season. The pooled gradient is an average over these four rows.
| Rest cell | Games (n) | Mean reference forecast | Home win frequency | Gap | 95 percent interval | Interval width |
|---|---|---|---|---|---|---|
| away +2 or more | 38 | 0.4795 | 0.4474 | -0.0321 | -0.1942 to 0.1186 | 0.3128 |
| away +1 | 198 | 0.5141 | 0.5404 | 0.0263 | -0.0340 to 0.0881 | 0.1221 |
| equal | 595 | 0.5476 | 0.5563 | 0.0087 | -0.0268 to 0.0444 | 0.0712 |
| home +1 | 221 | 0.5785 | 0.5294 | -0.0491 | -0.1094 to 0.0111 | 0.1205 |
| home +2 or more | 51 | 0.5824 | 0.5686 | -0.0137 | -0.1302 to 0.1008 | 0.2310 |
Scroll horizontally for all columns
Priced population only. The reference forecast is the devigged recorded moneyline pair; the source table carries no quote timestamp.
Verdict
The differential tracks the frequency; the two symmetric-congestion groups look alike, which is no visible difference rather than an absence of fatigue. Pooled over 4793 games the gradient is monotone across all five cells, and only the home +2 or more contrast against equal rest has a bootstrap interval that excludes zero. It is not stable season to season: the home-rested-minus-away-rested gradient clears 5 points in 3 of 4 seasons and is flat or reversed in 2025-26. The 1103 games that carry a recorded pregame reference forecast are all from 2025-26, and there the observed-minus-forecast gap stays inside its 95 percent interval of zero in every unmasked rest cell. So this corpus cannot separate 'the recorded forecast already carries the rest state' from 'this season had no gradient to carry'; the pricing question is recorded as a null, not as a finding.
What it means
Rest is a relative quantity here. A team that is better rested than its opponent wins more often, and the frequency climbs with the size of that advantage. At an equal differential the observed home-win share is nearly identical whether both sides are on a back-to-back or neither is; that similarity does not establish absence of fatigue, only that it is not visible in this outcome share. The pooled gradient is also a four-season average that one season does not reproduce.
Caveats and confounds
Descriptive conditional frequencies, not a forecast and not a causal estimate. Rest differential is confounded with travel, opponent strength and where a game sits in a road trip, none of which are controlled here. The market panel covers one season only.