APBRmetrics

For 5-man units playing over 25 minutes on the season, BV.com offers Unadjusted (Overall Rtg) and (1 Year) Adjusted +/-
Maybe we can call these UPM and APM, unless there's a more standard convention.

For some reason, these unit (lineup) UPM are almost always higher than the APM. Look at the most average team last season, the Suns:
http://basketballvalue.com/teamunits.ph ... 1&team=PHX

The 4 most-used lineup units are way positive in UPM (Overall Rtg, right-most column), and notably lower in APM.
The rest of the lineups are about half each, positive and negative in UPM; and somewhat lower to much lower in APM.

Every team you check will follow this trend.
I can understand that a given unit may have a nice point differential if they have played mostly weak opponent units. But that can't be true for a whole team, certainly for a whole league.

Summing all 810 units in the league with 25+ minutes, I get a point differential of +4110 in 124,523 possessions, for a net UPM of +3.3
I suppose these relatively 'stable' lineups may be beating up on a lot of more 'experimental' lineups, but that seems pretty large.
Minutes-weighted average APM for these lineups is -0.2, and this seems low, no?

The net difference is that unit APM's are on average 3.5 (pts/100 poss) lower than UPM.
How can this be right, across the whole NBA?

Interesting data, creating impetus for even more analysis.

Looks like about half of all minutes go to lineups used 25+ minutes for the season and about half go to lineups used less than that leaguewide. I wonder what percent of lineup selections used less than 25 minutes were really called for by the matchup or situation. That these dink lineups got about 50% of all minutes strikes me as too high, not optimal based on the average performance of the few teams I've checked and that survey of all lineups by minutes used that somebody did awhile ago that showed that average performance fell pretty constantly on average as minutes of use declined.

I wonder what percent of all league minutes are lineup with over 25 minutes of use in the season vs. the same level of lineup use (over or under 25 minutes), how much is over 25 vs under 25 and how much is under 25 minutes vs, the same.

I wonder what the average performance results would be if anyone who can generate Adjusted +/- lineups estimates simplified the lineup data to these groups and these 3 scenarios and found the average Adjusted +/- for each group in each scenario. That might be interesting to see and consider with respect to whether one should try to run a general lineup strategy that emphasizes using lineups used over 25 minutes against lineups with the same scale of use or emphasizes trying to get against the dink lineups. In a specific game your substition pattern should be based on not only these results for your team but the results for the opponent. You might go on offense seeking opportunities for your lineup used over 25 minutes vs lower against many teams but maybe not against some others depending on how your profiles compare and what their strategy is overall and at specific parts of the game. It is not all one contest, it is a contest of probably 10-20 segments.

Of course one could take this study down to individual lineups as well to help decide when they specifically are better or worse off on average and then look at specifics and think about why and what to do to improve the performance where that is most needed.

With manual compilation one could see the unadjusted +/- scenario data for these overall group scenarios and for specific lineups without needing the assistance of an Adjusted +/- modeler.

Maybe this further analysis would help answer the questions Mike G raises. Maybe it would raise additional questions. I'd think something useful would come from it, on the initial questions or something else.

The over / under 25 minute standard for group separation may not be the best or only useful standard. With sufficient budget for analysis one could look at it several ways. I'd probably want to look at lineup splits using 100, 200 and 500 minute cutoffs as well.

The 810 lineup units playing 25+ minutes sum to just about 55% of all NBA minutes last season; avg 27 units per team.
I assume all adjustment is due to a unit's opponents, since all teammates are part of the unit being adjusted for. Could this not be right?

The performance of lineups used over 25 minutes for all teams is lower than I would have expected. I am not sure about everything that goes into that result.

It might help if the producers of Adjusted +/- (traditional APM, RAPM and other) would indicate if they check the minutes weighted sum of their ratings after the values are produced to ensure that the set of ratings are properly balanced to a net of zero for all minutes. That is the organizing principle but is it checked?

I did look at the 29 teams with tables at BV sorting by .500 or better record vs below and sorting into lineups with 100+ minutes and 50-99 minutes.

For teams .500 or better, the 100+ minute lineups averaged +8.1 on unadjusted +/- and +5.6 on Adjusted +/-. The average offensive efficiency was 111.6 and the average offensive efficiency was 103.4. The 50-99 minutes lineups averaged +6.8 on unadjusted +/- and +3.2 on Adjusted +/-.
The average offensive efficiency was 111.1 and the average offensive efficiency was 104.4.

For teams under .500, the 100+ minutes lineups averaged -1.3 on unadjusted +/- and -4.3 on Adjusted +/-. The average offensive efficiency was 107.0 and the average offensive efficiency was 108.3. The 50-99 minutes lineups averaged -3.3 on unadjusted +/- and -6.7 on Adjusted +/-. The average offensive efficiency was 106.9 and the average offensive efficiency was 110.2.


The performance of these lineups for the teams under .500 are a drag on the average for all teams / all lineups over 25 minutes. Their lineups from 25-49 minutes are probably even more of a drag. I am not sure if lineups of that size for teams .500 or better are positive or if they dip into the negative by that point.

For teams .500 and above there were an average of 6 100+ minute lineups and 8 50-99 minute lineups. For teams .500 and above there were an average of 5 100+ minute lineups and 7 50-99 minute lineups.

Winning teams probably should maximize the time given to lineups over 100 minutes and to a lesser degree those between 50 and 99 minutes until they stop working adequately or start causing bigger declines in other lineups than the gain they produce. Teams below .500 may be wise to do the same to minimize their rate of loss but they probably should carefully review what lineups they've used and which ones they haven't and perhaps pick a different mix as well as different rates of use. Of course winning teams can also refine their exact mix of lineups and rates of use too in addition to following a greater emphasis on existing lineups with bigger minutes.

OK, good.
I'd already created a column for team SRS, so I'll break these units (lineups: a baseball term?) the same way, by 25-49 minutes, 50-99, and 100+, and this only moves Atl units into the 'losers' bracket.
First the winners: diff is just (Adj - UnAdj) -- that is, the avg adjustment made.
These are minutes weighted, not just averages of each unit. Poss = (OPoss+DPoss)/2

And the aforementioned losers:It seems these less successful teams have had their worst luck with the 50-100 minute units.
And, perhaps not by choice, they've gone less with their top units.

For the upper table, of above-avg teams, there's a clear drop in the magnitude of the adjustment when more minutes are played.

Once again, unadjusted +/- averages are very high (relative to SRS), adjusted +/- are universally much lower -- and especially for below-avg teams.

Total SRS for all these units is 0.49, UnAdj 3.30, Adj -0.1
Shouldn't they all be same as the SRS (+.49) ?

Are you set up to find the winning / losing team data for lineups used under 25 minutes? That is the only piece of the data puzzle remaining to be seen and considered. It could be found from team totals - the 3 minute level groups or compiled on its own.