APBRmetrics

I looked at the top 50 lineups that played 230+ minutes last season. I compared the sum of the players' usage to the lineup's offensive efficiency and found an extremely weak +.05 correlation. I had thought a sum of average usage rates would have at least a moderate correlation with offensive efficiency. Ben F.'s XOHoops simulation was built around such an assumption or data analysis finding. But maybe I didn't get the finding I was expecting because this is a extremely small sub-set of all lineups.

I found the average usage distribution by position within this group for those that were at least +5 per 100 possessions on raw +/- and those were were below that threshold. The top performing group's usage distribution across positions (in rounded terms) was

PG 21%, SG 22%, SF 21%, PF 22%, C 18%.

The top 2 positions were SG and PF (pretty classical, I think).


The lower performing group's usage distribution across positions (in rounded terms) was

PG 23%, SG 19%, SF 22%, PF 21%, C 19%.

The PG leadership on usage strikes me as a modern trend (but I haven't studied it exhaustively yet). SFs in second place on usage for this group might be modern too.

The difference between the two averages was biggest for PG and SG. But pretty small differences. Are they significant?

I looked at the correlation for these 50 lineups between raw +/- performance and variance from the average usage pattern of the top 28 lineups that were +5 or better and found a low (but not as microscopic) +.26 correlation. But I guess that was comparing a larger group including a subgroup to the average of a subgroup so you'd think therewould be at least some correlation. Not much though.

The lineups with the closest fit to this top performing lineup average using absolute difference in position by position usage were for the Grizzlies and Celtics. Philly, Memphis, Phoenix and Atlanta had lineups that fit pretty close as well and most of these 6 teams actually had more than one leader in conformity to the usage of the top performing lineup average. The 4 lineups most different from this average were Chicago's 2nd most used lineup and 3 of OKC's including both the beginning and end of season starting lineups. Doing things differently, with a couple of very high usage players instead of the fairly flat distribution of the top performing lineup average.

While that team specific information seems somewhat interesting to me (to least be aware of and think about), the correlation between offensive efficiency for the 50 lineups and their absolute difference from the usage pattern of the top 28 most used lineups in the league was -.02. Again no pattern.

I am not be completely over interest in usage distribution but this data will probably moderate my concern about it. It might prompt additional research, when I think of a new angle.

Of the top 50 minute lineups with sum of player average usage of 110+, the average +/- was +8.4. With sum of player average usage of 105-110, the average +/- was +7.4. Not much marginal net gain for the first group's higher sum of player average usage.

With sum of player average usage of 100-105, the average +/- was +3.8. With sum of player average usage under 100, the average +/- was +3.5. Again not much difference between these segments.

Of the top 50 minute lineups with sum of player average usage of 110+, the average offensive efficiency was 111.5 and the average defensive efficiency was 103.1. For the 105-110 group, it was O 112.3 D 104.9.
The higher usage group actually had a lower average offensive efficiency, in line with what was found with the single point example of Dallas. Oddly they did better on defense than the second level of usage group.


Of the top 50 minute lineups with sum of player average usage of 100-105, the average offensive efficiency was 109.8 and the average defensive efficiency was 106.1. For the under 100 group, it was O 108.5 D 105.
The lower usage group of these two offset its offensive loss with defensive gain.

These trends might not be multi-season stable. But one could check. They are at least food for thought about what usage level to strive for and what to expect from one kind of lineup or another. If you are not getting near the average return (for top lineups or simple overall average), maybe you need to adjust or move on to working / testing another lineup option, if any are promising based on data or coaching insight / hunch. The function appears to be dynamic rather than strictly linear.

If you are limited in number of high average usage players (almost everyone is) you might not want to burn them all up in the minimum of lineups. It might make sense to try to spread them out at least a little bit. Depending on what the unique players and unique lineups actually do.

Somewhat related, put here to avoid opening a new thread:

It appears that the title winners have had 2 starters over 25% usage in the playoffs in 9 of the last 22 seasons. The 2007 Spurs had 3. But about 60% of champs had just one big usage starter (and presumably a more even usage distribution beyond him, though I haven't computed that yet for champs.)

Shaq-Kobe in 2001 was the only recent occasion were the champ had 2 over 30% usage in the playoffs though they were also close in 2002 and the 2007 Spurs had 2 over 29%. Still that is only once in 22 years and teams were close to doing this in less than 15% of recent seasons. Miami and OKC are trying to buck that history.