The 2-for-1 is real, coaches actually do it, and it pays

With thirty-some seconds left in a quarter, a team with the ball faces a genuine strategic fork: shoot quickly and buy an extra possession before the buzzer, or hold for the last shot and deny the opponent any reply. Coaches disagree about this in public. A full season of play-by-play (all 1,230 games of 2023-24) lets us ask two separable questions: do teams actually play the 2-for-1, and does it pay? Two more NBA seasons and three WNBA seasons then test whether the answer holds.

Both questions are read off the final minute of a quarter. The diagram below marks the clock points the study uses on one real 2-for-1, and shows how the team with the ball is sorted by what it chose.

A spike at 31 seconds is a decision, not basketball

The first question has a sharp answer, and it comes from a discontinuity rather than an average. Of the 7,221 team-periods with a shot in the final minute of quarters one through three, 78% took their first shot with 32 or more seconds left. But that share alone is weak evidence, because later in the window there is simply more time in which to shoot. The strong evidence is where the shots bunch: teams launch at 31–33 seconds at 201 team-periods per second, versus 147 in the 36–40 second band. That is 1.37× the rate, in the band with fewer natural shooting opportunities. Nothing in the flow of a basketball game produces a spike exactly there except a decision rule.

The extra possession is real, and so is the edge

So teams do it. Should they? Counting a team's own points, 2-for-1 windows produce +0.78 more points per team-period than holds (95% bootstrap CI [+0.68, +0.88]). The mechanism shows up exactly where it should, as roughly +0.94 extra field-goal attempts per window. The extra possession is real, not an accounting trick.

But the case for holding the ball was never about your own points. It is that holding denies the opponent a reply. So the honest outcome is net points, and the honest comparison controls for the fact that the team with the ball early is a different kind of team-period. On net points, in the comparable-sample analysis, the edge shrinks to +0.24 per opportunity (95% CI [+0.03, +0.44]): the opponent's reply claws back about half of the gain, but not all of it. Pooled over three NBA seasons that net edge is +0.22 [+0.10, +0.34], clearly above zero. The decision favors shooting early: you give up nothing on shot quality, and you keep a real, if modest, edge after the opponent answers.

Everybody plays it. Sacramento just plays it hardest.

Two natural follow-ups deserve their own figures. First, who actually hunts the 2-for-1? The behavioral dial is team-controlled, and the league spread is real but modest: Sacramento leads at 85% of its final-minute windows, Chicago trails at 73%, and every team sits above 70%. This is a league-wide habit with team accents, not a strategy some teams run and others refuse.

The clock decides, not the quarter

Second, does the quarter matter? Almost not at all. The 2-for-1 rate is 78.4% in the first quarter, 78.0% in the second, and 76.8% in the third. The payoff ordering also holds in every quarter: taking the 2-for-1 wins each one on own points and on net. That flatness is itself informative. The decision is driven by the clock, not the scoreboard or the stage of the game, at least through three quarters (the fourth is excluded by design, because endgame strategy changes everything).

The habit has stopped spreading, and the holdouts came back

One season could not say whether the 2-for-1 was still taking over. Three can. Looking only at the team that has the ball at the 40-second mark, the share that goes quick sits at 84.5%, 82.4% and 83.0% across 2023-24 to 2025-26: flat, not rising. Holding for the last shot, nearly extinct at 24 team-periods in 2023-24, ran 50 and 48 in the next two seasons. The bunching at the 32-second threshold also softened, from 1.37× to 1.19× the rate in the 36–40 second band. The play is standard, but the league has not finished converging on it.

The WNBA plays it less, and less on schedule

The WNBA plays 10-minute quarters with the same 24-second shot clock, so the same 60-second window and 32-second threshold apply. WNBA teams with the ball at the mark go quick 76% to 78% of the time, below the NBA's rate, and they do not time the first shot to the threshold at all: the 31–33 second band never reaches even the rate of the 36–40 second band. The hope for the replication was a larger group of teams that hold, which would make the comparison better powered. Holds are about twice as common as in the NBA, but that is still only 16 to 19 a season.

Does it pay there? The possession-level comparison says yes: quick minus hold pools to +0.71 net points over 2024-2026 (95% CI [+0.25, +1.18]). The shot-timing comparison restricted to first shots at 20–45 seconds does not: −0.13 [−0.36, +0.11]. The possession design is the more direct of the two, because it compares only teams that had the ball at the mark, so it carries the conclusion, but the disagreement is real and worth stating.

These numbers were wrong twice. Each time, a check caught it.

The first version of this analysis silently sorted play-by-play event numbers as text, putting "100" before "2". The family's rule that every analysis is implemented twice, in R and Python, caught that. But a second implementation cannot catch a mistake both implementations share, and there was one. Both ordered events by action number, and the league's feed inserts late events with a high action number but an early clock. Those strays carried the end-of-quarter score back to the start of the final minute, which zeroed or shrank the window points of 15% of 2023-24 team-periods and understated every estimate. An independent feed caught it: the NBA's own liveData play-by-play disagreed with the old window points, and agrees with the corrected ones on 100.0% of team-periods. The diagram below shows the mistake on one real period.

A second pass the same day, prompted by the Elam Ending study's own check against that feed, fixed how scores are read: Instant Replay rows carried stale scores, and free throws at one clock reading could arrive out of order. It changed 28 more team-periods across all six league-seasons. Every season in this study is now checked against that second feed, and the README's correction note lists each published number next to its corrected value.

Method, code, and verification

The full technical write-up (data provenance, model specification, validation gates, and stated limitations) is the README. The analysis is implemented independently in R and Python and reconciled to numeric tolerance; run_checks.sh replays every gate.