Momentum in the NBL is mostly a story

Shooters do not get hot in any way the data can find, runs fade at the same rate with or without a timeout, and a ten-point half-time lead still slips away about one time in eight.

01 · Leads

A lead is not safe until it is large

A model of how scores evolve calls 74% of games it has never seen correctly at half-time and 91% with two minutes left. What it shows about leads is plainer than that. Up 10 to 14 at half-time, a team won 87.2% of the time. Up 5 to 9 it won 71.1%. Only at 20 or more does a half-time lead become close to certain (98.9%).

Comebacks are part of the NBL. In 21.6% of games (344 of 1,591) the team that lost had led by ten or more at some point; 85 teams have blown a 15-point lead and 8 a lead of 20 or more. The record is 26, by the Adelaide 36ers against the Perth Wildcats on 2021-03-14.

Nor does any club close games measurably better than another. The league converts 91.8% of late positions where it is an 80% favourite. Clubs range from 89.8% (Taipans) to 94.7% (Phoenix), but the largest gap from the league rate is 1.1 standard errors, inside what chance produces across 10 clubs. Closing ability looks like a myth at this sample size.

02 · The hot hand

After a make, players shoot differently, not better

After one make, a player is 4.5 points more likely to take a three and 4.0 points less likely to shoot at the rim. Those are harder shots, which is why the raw make rate dips after a streak. Hold the difficulty of the shot fixed and the change in accuracy is −0.1 points (range −0.6 to +0.4), and no larger after two or three makes. Confidence changes what players try, not what goes in.

The one place a small streak effect shows is the free-throw line, where nothing about the shot changes: a player is 2.6 points more likely to make the second shot after making the first (range +1.0 to +4.1), consistent with rhythm rather than heat. Individual shooters look no streakier than chance would make them: the spread of results across 220 players matches random variation almost exactly.

03 · Runs and timeouts

Runs end on their own, with or without a timeout

A run is an ordinary night: 67.7% of games include a 10–0 run, 95.3% include an 8–0 run, and the biggest unanswered run in an average game is 11.5 points. It does not decide the game by itself: in 29.8% of games the eventual loser had a 10–0 run of its own.

If momentum were real, a team that has just finished a big run would keep outscoring its opponent. It does not. After a run of ten or more, the scoring team is outscored by 0.61 points over the next two minutes, the same as after any other basket (0.51). With no timeout, the team being run on outscores the running team 50.1% of the time; the run continues only 37.3%.

Teams that call a timeout mid-run outscore the running team by 0.59 points over the next two minutes; teams that play on, by 0.53. After adjusting for run size, score, quarter, time, venue and team strength, a timeout is worth +0.06 points. Across 9 ways of setting up the comparison every estimate sits between −0.04 and +0.16. Coaches call one in 16.3% of 6–7 point runs and 20.2% of 10–11 point runs where they could.

What this does not show

• Timeouts do other things. They rest players, set up a play and manage the end of a game. None of that is measured here; the finding is only that a timeout is not what stops a run.
• A run is a definition. Here it is a team outscoring its opponent by six or more over two minutes. Timeouts are attributed to a team only for 2015-16 to 2024-25, because the league's own feed leaves the team blank from 2025-26.
• The hot hand test has limits. A planted +3 point effect was recovered as +2.1, so the test errs slightly cautious, but a real effect of more than about a point would very likely have shown up. Smaller effects cannot be ruled out.
• Closing ability needs more games. With 10 clubs and about 92 to 241 late positions each, small real differences between clubs would not show.