APBRmetrics

I've written plenty about lineup management design in theoretical and applied terms but usually for a season or going forward for an unspecified time subject only to a general understanding that it will be adjusted over time.

Here I am going to sketch a general structure for lineup management thru a season.

Before first 10 games, analytics staff works with coaching staff on a plan to meet joint team factor goals primarily using 16 lineups (the average game probably uses about that many). Coaching staff produces a final list of 16 lineups they intend to emphasize. The target imo would be use those 16 lineups at least 50% of total game time. An explanation would be expected if the actual time allocation did not meet this minimum. The rest of time (here and below), Coach can freelance as he sees fit.

Time allocation goals and reviews would be made over the segmented game blocks and would not to be expected to rigidly identical in every game.

Rinse and repeat with freedom to change for each of next 3 ten game segments but trying to increase the minute allocation to the top 16 to at least 60%.

After 40 games, the goal would be to chose the top 12 lineups to get at least 65% of total time over the next 20 games. Brand new lineups could be added but the expectation would be that at least 2/3rds would probably come from the bigger minute, better performing / looking lineups of the first half of the season.

For the final 22 games, the goal would be to have no more than 10 lineups receiving at least 70% of time. Some variance could allowed but the whole point is more concentration than the status quo.

Within the 10-16 priority lineups, there should be a top priority group of 4-6 getting much higher concentration.

The final design might be a 10 minute lineup, a 6, two 4s and 4-6 transition lineups averaging about 2 to 2.5 minutes each. But it could be different, if believed that a different design is better.

Playoff design would likely be about the same but with a recognition for matchup strategy and judgment based quick adjustments.



Some study of actual lineup management behavior across segments of a season is probably appropriate to try to uncover the pacing and logic of change within the vast conventional lineup chaos, but this is my first cut, blue sky conceptual approach free from exactly knowing or using those norms.

For the season to date the Celtics have used 237 lineups with just 1 lineup over 4 minutes per game and 1 more over 2 minutes. The top 4 have gotten 34% of time and the top 16 got 57%.

The Nuggets have used 280 lineups with just 2 lineups over 4 minutes / gm and 1 more over 2 minutes. The top 4 have gotten 37% of time and the top 16 got 51%.

The Thunder have used 461 lineups with none over 4 minutes / gm and 3 over 2 minutes. The top 4 have gotten just 15% of time and the top 16 got just 28%.

There are some similarities in these actuals compared to my design and some differences.

I might check other teams later.

The Grizzlies have used 254 lineups with just barely 1 lineup over 4 minutes / gm and 1 more over 2 minutes. The top 4 have gotten 22% of time and the top 16 got 43%.

The Cavs have also used 254 lineups with 2 lineups just short of 4 minutes / gm and no others over 2 minutes. The top 4 have gotten 21% of time and the top 16 got 42%.


Higher or lower lineup concentration in the playoffs than the competition could be a significant factor in degree of success.


The Thunder have around double the lineups used as these playoff teams and at least 50% less concentration on the top lineups.

For the 2022 playoffs, the Celtics used 80 lineups with 3 lineups over 4 minutes per game and 1 more over 2 minutes. The top 4 got 42% of time, up from 34% in regular season. The top 10 got 62% compared to the 57% the top 16 got in regular season.

The most used lineup (with G Williams) was a dog though at -8pts / 100p. And the 3rd most used lineup was almost as negative.

Celtics in playoffs were pretty close to my conceptual design. But sticking with 2 bad performing lineups probably did them in. Doesn't take much to fall short at some point.

2 of 4 most used Celtics lineups were very bad against Warriors. Not that near good enough.

3 of 4 most used Warriors lineups were bad to very bad against Celtics. But...the biggest one was very very good. They got an edge on a much less effective overall rotation.

Even modest differences can have huge consequences but Warriors pulled away with 3 straight victories at the end.

Neither did great at the top. Dink lineups certainly played a role. For how they performed and for being selected instead of bigger minute lineups that were working (or believed to should work).

It should be possible with machine learning to predict what lineups will be played and when for teams. In general and maybe in particular game matchups. How accurate I dunno.

In addition to predicted actuals could be "should", though the shoulds should cascade with every change. But something could probably be learned from the exercise, even if the model loses definitive tracking.



Updates on lineup management by Celtics, Nuggets and Thunder by end of 2022-23 season appear in other threads, so nothing added on here.

Talk about shortcomings in lineup design by team. Whose will be least bad / best? May get clues in first month or two. Better sense early next year. Final answers or guesses in spring / early summer.