Signature stats
Gas Tank and Schedule Edge: measuring fatigue in shifts and schedules
Everyone agrees tired players play worse, but proving it is hard: coaches shorten shifts when things go badly, so long shifts and bad outcomes are tangled together. The icing rule gives us a clean test, because it forces a tired unit to stay on the ice.
Updated · Method version sequences v1
How it's calculated
We know every player's shift start from the NHL shift charts, so for every event we know how long each skater on the ice had been out (shift age).
- Icing trap: find every 5v5 defensive-zone faceoff that follows the defending team's own icing (no change allowed) and compare it with every other 5v5 defensive-zone draw. Measure faceoff win rate and xG against in the next 20 s.
- Shift-age curve: split those draws by the defenders' average shift age at the faceoff.
- Tired penalties: for each shift-age bucket, compare the share of minor penalties taken with the share of ice time spent at that age (rate vs exposure).
- Schedule Edge: for every game, compute rest days, back-to-back status and travel distance for both teams, and compare win rates and shot share by situation, season by season.
rate(age) = (penalties taken at that shift age / all penalties) ÷ (ice time at that shift age / all ice time)
Penalties taken vs ice time, by shift age
Research sample: 200 games from 2025-26 (2025020500–699)
Relative to exposure, penalty rates rise from 0.48 times in the first 15 seconds of a shift to 3.52 times past 90 seconds.
After your own icing: xG against in the next 20 s
Research sample: 200 games from 2025-26 (2025020500–699)
xG allowed after an icing rises with defender shift age: 0.016 when fresh, 0.020 at 10 to 50 seconds, 0.024 past 50 seconds.
Back-to-back team vs rested opponent: win %
Research sample: 200 games from 2025-26 (2025020500–699)
Teams on the second night of a back-to-back against a rested opponent won between 37.9% and 43.7% of games in each season.
Worked example
What an icing costs
Real research numbersLeague numbers from 200 games of 2025-26: after its own icing, the defending team allows 0.022 xG in the 20 s after the draw, against 0.017 after other defensive-zone draws.
- Extra xG against per icing: 0.022 − 0.017 = 0.005 (95% confidence interval of the difference: 0.002 to 0.008).
- Relative increase: 0.005 ÷ 0.017 ≈ +29%.
- Over 10 icings: about 0.05 extra expected goals against, plus the lost draws (43.9% won instead of 52.7%).
- If the defenders had already been out 50 s or more, the figure is 0.024 rather than 0.016 for a fresh pair.
An icing is a small but real cost, and most of it is fatigue: the effect grows with how long the defenders have been on the ice.
Shift-age tables
| Defender shift age | xG against in 20 s |
|---|---|
| Under 10 s | 0.016 |
| 10–50 s | 0.020 |
| 50 s+ | 0.024 |
| Shift age | Penalty rate vs exposure |
|---|---|
| 0–15 s | 0.48× |
| 15–30 s | 0.98× |
| 30–45 s | 1.1× |
| 45–60 s | 1.43× |
| 60–90 s | 1.92× |
| 90 s+ | 3.52× |
Schedule Edge by season
| Season | B2B team vs rested: win% | Rested team vs B2B: win% | B2B after 1,000+ km travel: win% |
|---|---|---|---|
| 2021-22 | 38.1 | 61.9 | 27.5 (n 51) |
| 2023-24 | 43.0 | 57.0 | 41.4 |
| 2024-25 | 37.9 | 62.1 | 39.3 |
| 2025-26 | 43.7 | 56.3 | 40.5 |
Part of the back-to-back effect is the backup goalie, who starts many second nights. We are separating backup-goalie starts in the model so a fantasy user can tell whether the drop is about the skaters or the goalie.
Reliability
These are league-level effects with confidence intervals, not player ratings. The schedule effect has held in every season we checked.
| Measure | Split-half r | Sample | Verdict |
|---|---|---|---|
| Icing trap xG against difference | — | 200 games; 95% CI 0.002–0.008 | Stable |
| B2B vs rested win%38–44% every season. | — | 4 seasons | Stable |
| Quality of competition (QoC)Barely varies (0.485–0.514); shown with a note. | — | Season | Noise |
| Quality of teammates (QoT)Meaningful context. | — | Season | Moderate |
Split-half r: we split each sample in two (for example odd and even games) and correlate the two halves across players or teams. Near 0 means the stat is mostly noise; near 1 means it is a stable trait.
Limitations
- The icing experiment tells us fatigue matters at the moment of an icing. It does not give a per-player stamina rating, and we don't publish one.
- Shift age uses the NHL shift charts, which round to whole seconds and occasionally contain errors; we drop rows from other games and check totals against the boxscore.
- Schedule effects mix fatigue, travel, backup goalies and home ice. Win rates by situation are descriptive; the model separates the pieces it can.
- The Schedule Edge table covers the four seasons in the research sample; 2022-23 and earlier seasons will be added with the full backfill.
FAQ
Do NHL teams lose more on the second night of a back-to-back?
Yes. Against a rested opponent, teams on the second night of a back-to-back won 38.1% (2021-22), 43.0% (2023-24), 37.9% (2024-25) and 43.7% (2025-26) of games.
How do you know it is fatigue and not just bad teams icing the puck?
We compare the same situation (a 5v5 defensive-zone draw) with and without a forced no-change. The effect also grows with how long the defenders have been on the ice: 0.016 xG against when fresh, 0.024 after 50 seconds or more.
Do tired players take more penalties?
Yes. Relative to the ice time spent at each shift age, penalties are 1.9 times as common 60–90 seconds into a shift and 3.5 times as common past 90 seconds.
What is quality of competition, and why do you downplay it?
QoC is the ice-time-weighted quality of the opponents a player faced. In our data it barely varies between players (0.485–0.514 xGF%), so it rarely explains much. Quality of teammates varies far more.