APBRmetrics

[dquinn1575]

I would like some insight into the sample size impact here:

I was thinking that a study that showed 5 man units expected points per possession versus actual points per possession would identify those units where a player did the "little things" that didn't show up in the box score. If I then identified the players who consistently turn up positive, I could see the impact of those players.

My concern as I looked at the 5 man unit numbers was sample size. Very few units play together even as much as 100 minutes. How valid is the sample and analyses for this data? It would appear that a lot of the data generated by a 5 man unit would be chance, and not reflective of the true talent of the group.

If I look at the most used 5 man unit by team, the most used unit (MUU) by team was only the best (minimum 100 minutes) for 9 of the NBA teams- and 1 was Miami, which only had 1 such unit (its second best unit, with 94 minutes was better).

This is saying that either most of the NBA coaches do not use their best line-up, or they intuitively know that the unit comes out on top due to sample size.

Now, I think my original study might still be valid. My concern is the impact of different with and without you studies using these small sample sizes.

[Mountain]

Coaches play "hands" / lineups and they want to win them but ultimately they want to win on net at the end. Winning your most used lineup would be nice but it is about winning enough hands, for enough chips, however- early, steady or in big runs or at the end. Some coaches play straight from the start, use a lot of their best talent / plays / tricks early and try to win big and as much as possible with most used / starting lineup. A "big 5" approach.

Other coaches (and GMs) play more of a bench game. They seem to emphasize chipping away in the middle a little more with later match-ups, slipping in productive plays and coaching tweaks subtly on the edges, as much as they can get away with without drawing a more effective response from the opponent. Not using as much of their best right away.

Of course the end game is often decisive, the final hands, often the biggest "pots" with the opponents having a lot more information about each other... but sometimes still guessing or making the other guy guess and guess wrong. Different ways to play the game (against different opponents- players and coaches).


Looking at performance of specific lineups has value and should continue to be done more and more deeply by serious basketball circles. How good the internal study is surely varies. From the outside many teams' lineup utilization look less than highly optimized. Small sample size is a real issue but as I've noted before coaches (at least those of the best teams with more slack to work with) could get more decent sized samples for lineups they rely on (especially in playoffs) with more tightly disciplined lineup management during the available 4000 or so minutes of regular season play aimed at gathering more reliable performance data about key lineups. As shaky as small sample lineup data is, coaching gut feelings about lineups seem shaky / error prone too. The best bet is the most / best lineup testing & analysis you can accomplish weighed along with coaching feel for the lineups and the opportunities and needs of specific moments in the course of many hand / possession game.


It is important to remember the context though, the effort to win the general war, not necessarily each and every battle.


Comparing expected points per possession of the players put on the court versus actual points per possession of the lineup as you suggest is one side by side to look at.
I am starting to look at the adjusted +/- of individual players and the adjusted performance of lineups. Seem like good approaches for understanding what works and how well at the 5 man lineup level and what changes to make to the lineup rotation "dance".