The Back-to-Back Penalty: What Tired Legs Cost, 2023-24 and 2025-26

"They're on the second night of a back-to-back" is the most common excuse in basketball, and for once the excuse checks out. Reconstruct every team's 2023-24 schedule from the dates and the cost of a back-to-back is clean: a team playing its second game in two nights, against a rested opponent, won just 41.9% of the time and lost by an average of 2.8 points — while the rested team on the other side won 58.1%. Run the same count on 2025-26 and the penalty is still there but smaller: 47.5% against a rested opponent, a margin of −1.7. Two seasons, one direction, two sizes. The table below puts every definition we have ever published side by side.

Correction, October 29, 2026. This site used to publish two different second-night win rates for what reads like the same question: 41.9% on this page and 48.3% in a separate piece on rest and load management. Both were correct arithmetic, but they used different definitions and different seasons, and neither page said so. The 41.9% counts only second nights against a rested opponent in 2023-24; the 48.3% counts every second night, whatever the opponent's rest, in 2025-26. We have merged the two pieces into this one, recomputed every definition on both seasons from the same bundled game files, and put them in one table. We also removed a claim that the equal-rest buckets landing on exactly 50% proved the effect is real: they land there by construction, as explained below.

How we measured it

The game log doesn't label back-to-backs, but it doesn't need to — the dates do. A team is on the second night of a back-to-back if it played the day before; any team with at least one clear day since its last game is rested. We tagged every team in every game that way, then split each season two ways: by the team's own rest alone (second night versus rested, which is the cut the old rest piece used), and by the rest matchup — both teams rested, one team on a back-to-back against a rested opponent, the reverse, or both on a back-to-back. Then we counted wins and margins in each bucket. No model, no assumptions — only the schedule and the scoreboard. A season opener has no previous game, so it counts as rested in the matchup buckets and is left out of the own-rest comparison.

One question, two definitions, two seasons

Read it by row, not by number. The old 48.3% sits in the first row of the 2025-26 columns: every second night, including the ones where the opponent was tired too. The 41.9% sits in the third row of the 2023-24 columns: only the one-sided back-to-backs. Compare like with like and the two seasons agree on the direction and disagree on the size. Against a rested opponent the tired team won 41.9% in 2023-24 and 47.5% in 2025-26; across both seasons that is 272 wins in 609 games, 44.7%, with an average margin of −2.25.

How sure can we be about the size? Each one-sided bucket holds about 300 team-games, so its win rate carries a standard error of about 2.8 to 2.9 points and its margin about 0.9 to 1.0. The 2023-24 penalty sits well clear of zero; the 2025-26 one is about one standard error below 50% on wins and under two on margin. The gap between the two seasons is itself within noise. The honest summary is a real cost of roughly two points of margin, with the exact decimal moving from year to year.

The exhibit: rest is a differential, not a state

A note on what the chart can and cannot show. The two equal-rest bars, both rested and both on a back-to-back, are exactly 50.0% with a margin of exactly zero, but that is bookkeeping, not evidence: every game in those buckets puts its winner and its loser in the same bucket, so they must split 50/50. For the same reason the rested side's +2.8 is always the mirror of the tired side's −2.8. What the data does say is that the whole effect lives in the one-sided games. A tired team playing a fresh one gives up points; that is the only comparison in which rest can show up at all, and it does, in both seasons.

The same cost, seen across a whole season

Drop the opponent's rest from the question and the gap shrinks, because a share of those second nights came against a team that was tired too. Over all of 2025-26 the second-night side scored 114.8 points to the rested side's 115.7, shot 47.0% from the field against 47.2%, and its average margin fell from +0.27 to −1.17, a swing of 1.44 points. A team plays about 15 back-to-backs a season (441 across 30 teams), so this is a small cost paid often, not a cliff.

Why ~2 points is a big deal

A two-point swing sounds small until you place it next to the league's other famous edge: home-court advantage is worth only a couple of points itself. A one-sided back-to-back is, in effect, about large enough to cancel home court — a tired home team and a rested visitor are roughly even. It's also right in the range that flips close games: the NBA is decided by single digits often enough that two or three points of expected margin moves a meaningful number of outcomes from the win column to the loss column.

What's behind the number — and what isn't

The mechanism is exactly what you'd guess: legs. Tired teams defend with less close-out energy and lean on shorter rotations. That is also why load management exists. When a front office sits a star on the back end of a back-to-back, the defensible argument is not that he would be useless; the numbers say a tired team still wins close to half its games. It is that a healthy star in April is worth more than a point or two of margin in a January game, so the team trades a small, certain edge tonight for a larger, uncertain one in the spring. That is a different claim from "back-to-backs ruin players", which the data does not support. Three honest caveats keep the number in proportion:

• It's the rest differential, not a curse. The cost shows up when the opponent is fresher. Schedules are close to balanced on the back-to-backs a team plays, though not on the ones its opponents play; the rest ledger measures that imbalance and prices it at under a win a season.
• Schedule-makers aren't random. The league tries to avoid the harshest spots, and a team's back-to-backs aren't evenly distributed against strong and weak opponents. Some of the bucket differences carry a sliver of matchup-quality noise, not pure fatigue.
• Resting stars is folded in. Some second-night games feature a star sitting out, which is part of what a back-to-back costs but muddies "fatigue" versus "missing players".

Limits of this read

• Sample sizes shrink at the edges. The one-sided buckets are about 300 games each per season — enough to see the direction, not enough to trust the exact decimal. The "both on a back-to-back" bucket (114 and 140 games) is smallest.
• It ignores travel and distance. A back-to-back with a cross-country flight is surely worse than two home games on consecutive nights; "played yesterday" is a blunt proxy for true fatigue. Three games in four nights looks at the denser stretches.
• Two seasons, two sizes. The direction repeats; the magnitude moved from 41.9% to 47.5%. Expect it to keep wobbling with scheduling and rule changes.

The takeaway

The back-to-back excuse is legitimate, with a price tag that depends on how you ask. Against a fresher opponent, the tired team won 41.9% in 2023-24 and 47.5% in 2025-26, about two and a quarter points of margin across both seasons. Count every second night regardless of the opponent and the cost looks smaller, 44.1% and 48.3%, because some of those opponents were tired too. Whichever number you quote, quote its definition and its season with it. And when a good team drops a game it "should" have won, check whether it was the second night of a back-to-back against a team that slept in its own bed.

Reproduce it

Rebuild each team's schedule from the game dates, flag any game played one day after the team's previous game as a back-to-back, split every team-game by its own rest and by the rest matchup, and average wins and margin in each bucket. The 2023-24 chart is regenerated by chart_back_to_back.py against the bundled data_layer/nba_home_results.csv; the 2025-26 columns come from the bundled data_layer/nba_games_2025_26.csv, which matches the team game log behind the 2025-26 chart game for game. No network, nothing hand-entered.