A Note on Lineup-Based Analysis

Nov 8, 2015; Miami, FL, USA; Miami Heat forward Chris Bosh (1) talks with forward Justise Winslow (20) as guard Tyler Johnson (8) and guard Mario Chalmers (15) look on in the second half of a game against the Toronto Raptors at American Airlines Arena. The Heat won 96-76. Mandatory Credit: Robert Mayer-USA TODAY Sports
Nov 8, 2015; Miami, FL, USA; Miami Heat forward Chris Bosh (1) talks with forward Justise Winslow (20) as guard Tyler Johnson (8) and guard Mario Chalmers (15) look on in the second half of a game against the Toronto Raptors at American Airlines Arena. The Heat won 96-76. Mandatory Credit: Robert Mayer-USA TODAY Sports /
Robert Mayer-USA TODAY Sports
Robert Mayer-USA TODAY Sports /

There has never been more access to statistical information for the NBA fan. Whether derived from the box score or play-by-play, optically tracked or manually charted, the data is out there. Of course with great data comes great…something or other. The point is, it can be quite easy to get lost in cul-de-sac of misleading or trivial information that because it includes numbers, is given the imprimatur of importance and meaning. An example of this is the growing use of five-man unit data.

I’m as guilty of this as anyone, given the regularity with which I tout the magnificence of “Draymond Green at center” lineups. These lineups include the so-called “death lineup” of Green, Steph Curry, Klay Thompson, Andre Iguodala and Harrison Barnes and the replacements used[2. Often Shaun Livingston.] during Barnes’ lengthy absence with an ankle injury. In (quite coincidentally) 187 minutes through Sunday, Green-at-center has seen the Warriors outscore opponents by the equivalent of over 35 points per game. That works out to just over five minutes per game, indicating just how selective interim coach Luke Walton[2. Presumably upon advice and consent from ailing head man Steve Kerr.] has been in terms of deploying the unit. It’s incredibly unlikely the unit would be nearly as effective in a bigger sample, used in a broader range of situations and against more types of opponents, in larger part because of the high energy levels which are part and parcel of that incredible effectiveness.

Still, there must be some insight to be gleaned from A/B testing. For example, the Hawks appear to be three points or so per/100 possessions better off when Thabo Sefolosha plays alongside the Teague/Korver/Millsap/Horford core[4. According to the invaluable NBAWowy.com.] than with Kent Bazemore. With Baze, Atlanta is -2/100 in 327[2. Thru Saturday’s games.], while they are +0.9/100 with Sefolosha in 169 minutes. If those low minute totals stopped you in your tracks, good.

Cycling through play-by-play data[1. The following numbers could easily be off slightly, based on the occasional inconsistencies found in a season’s worth of play-by-play logs, but we’ll go with it for now.], 15,329 five-man units appeared during the 2014-15 regular season. Of those, nearly a fifth (2,923) logged less than a full minute on the floor together. Only 162 totalled 100 minutes or more. Thirty-six played at least 250 minutes. Only three, the preferred starting lineups for the Clippers, Hawks and Warriors played over 1,000 minutes together. In other words, the vast majority of five-man groupings didn’t play anything approaching a reasonable sample of minutes together.

We would rarely if ever make pronouncements on individual players over sample sizes that small. While a thousand minutes might get us part of the way to stabilizing things like rebound and block rates, it surely wouldn’t provide enough of an opportunity to tell with any certitude if a player is a good, bad or indifferent three point shooter — even the biggest gunner in the league would be hard pressed to launch even 200 3PA over that time period, when closer to 750 attempts are needed. And a single player is a relatively simple mix of skills and attributes, add the complexity of a lineup, with five separate sets of skills, not to mention the very synergies lineup-based analysis is meant to uncover, and it should be apparent that we don’t know enough about the vast majority of lineups to draw many conclusions at all.

Is the Sefolosha-for-Bazemore swap beneficial because of an actual difference between the players and/or how their individual skills fit with the four 2015 All-Stars? Or is the point differential discrepancy more the result of the variance of Korver shooting 30% for three in the Bazemore units and 41.7% with Sefolosha?

This isn’t to invalidate the entire effort, of course. Qualitative analysis of a given group can happen relatively quickly. A relatively experienced basketball mind can identify unbalanced groupings that lack the creativity or shooting to be offensively competitive, and while we might not know exactly how bad such a squad might be, there isn’t a great need to given it 2,000 test minutes to be sure. More specifically, lineups form the basis for the entire family of adjusted plus-minus metrics. While the results from this sort of analysis should not be taken as infallible rankings, they do form a very good first pass at who has performed well or poorly provided a big enough sample.  Jeremias Engelmann[1. One of the creators of ESPN’s Real Plus/Minus] has even gone the next step and created RAPM scores for entire lineups, which might be best understood as “strength of schedule adjusted net ratings.” Engelmann himself says these unit ratings are “just for fun,” providing little actual information on their own.

Better, to my mind at least, than looking at full five man units is focusing on categories of lineups. While Green-at-center doesn’t have enough minutes for us to be sure of the effect, it’s hard to argue against a style of play which has outscored opponents by nearly 0.75 of a point per minute over a small but not infinitesimal sample. When considering positional analysis, there are combinations which indicate a defined change in a team’s style. For example, the Heat with Chris Bosh at center are often A) playing against second units, B) playing a far more defensively active style, with Bosh typically joined by rookie Justise Winslow to great effect.

In fact, it is possible to look more deeply into just that sort of lineup. Again, using NBAwowy, we can see that in 375 minutes lineups with Bosh and Winslow, but without Dwyane Wade and Hassan Whiteside are +8.4/100, with a stellar 96.1 Defensive Rating. Swap Wade and Whiteside for Winslow and in 486 minutes, the Heat are +1.5/100, performing slightly better offensively, but far worse (105.1 rating) defensively. This tells us something, though given regular usage patterns with the Wade-Whiteside lineups facing opponent starters, the level of opposition should be taken into account as well. Such analysis would be hard to do if one were slavishly looking at full five man units. The most used Bosh-Winslow-but-not-Wade-Whiteside lineup has appeared for 48.9 minutes. The next most 31.6 and on down. But by identifying the important factors between differing styles of play, a meaningful sample can at least be approached.