2021-2022 MAVS NEWS: 4th in West | WCF loss [ARCHIVED]

[mavsluvr]

I think line-up data, 5-man, or even 3-man or pairing data is useful. Seems like the closer you get to 5-man, the more noise you remove. But even with 5-man, there is noise, because you don't know, just by the numbers, who that lineup has been playing against. 

But individual +/- is impacted by so many things, and the number doesn't give you the context. 

Regarding Luka, he hasn't been LUKA, but he's still been good compared to an average NBA player. You can point to things that Luka has done that could negatively impact a lineup (not getting back on defense because he'd rather chat with the ref), AND you can point to lineup issues that would impact Luka's +/- number. He has spent lots of minutes in the dreaded 2-big starting lineups that have tanked. And Luka has spent a lot of time in 4th quarter lineups that have tanked. I'd say Luka not being LUKA has contributed a little to the tanking of the 1st and 4th quarter lineups, but the lineups themselves have a been a huge problem because of non-Luka factors.

Partnow discusses the use of +/- to evaluate lineups (as opposed to an individual player) at some length. This is an area where the stat (as adjusted according to the team's algorithm) has proved useful. However, it still much be approached with knowledge of its limitations. As you point out, you still need contextual information to evaluate it. 

And an even bigger issue with it is sample size. These stats tend to start providing a meaningful basis for analysis, he says, at around 500 possessions (250 minutes or so), and very few lineups reach anything close to that much playing time together. (In the season he references, about .16% of all 5-man lineups played). When you go down to smaller combos, more of them reach the minutes threshold, but there is more external context to be accounted for. 

Coaches and organizations can appear to have widely different views on +/-, but that may be more superficial than anything. He cites a couple of examples. One is the Mavericks, who are openly believers (while acknowledging the limitations). Carlisle relied on this type of analysis in choosing to start JJ Barea in the back half of the Miami series; they had lineup data indicating that the team performed particularly well when JJ lineups countered Miami lineups containing Mike Bibby. Pop, on the other hand, is dismissive of the nightly +/- stat, saying he normally doesn't even look at it, and if he does, he gives it little weight. That doesn't mean that the Spurs don't make excellent use of this type of analytics -- they do, and Pop is just correctly noting that the single-game individual plus-minus stats reported in the box score tell you the aggregate scoring margin in that game when a particular player was on the floor, but are 100% useless as far as predicting what the scoring margin will be when that player takes the floor again. 

He thinks that these stats, as adjusted, can in some situations do a pretty fair job of categorizing lineup/players into good, medium, and bad tiers. However, they have next to no value in predicting outcomes with specificity. For example, if a lineup has a +5 in 250 minutes played, that would tend to indicate that a deeper look might suggest that it is generally a good lineup. However, it provides no reason to think that the next time the lineup takes the floor, it will have a +5, or even that its average play in the future will be +5.