Exploring Lineups and Fit
Projecting the upcoming NBA season for any team, it can be tempting to focus on the expected starting lineup, how good those players are how that unit might fit together. Both “analytics people” and long time observers of the game[1. As if those are totally separate categories.] realize how lineups perform on the court is often something either more or less than the sum of the parts. In fact, analytics have already helped solidify some significant insights into on court fit issues such as the value of spacing, having rim protection on the floor, the necessity of sufficient shot creation, or the rivalrous nature of rebounds.
So while obsessing over the exact fit of pieces in a starting five might seem natural, one probably shouldn’t bother. It turns out that is probably not the most practical way to think about teams or project their success. Because, while fit matters quite a bit, the starting unit is likely to play together much less than one might expect.
The NBA regular season has 3,936 minutes of scheduled playing time for each team before any overtime minutes are added. In contrast, over the last five NBA seasons, 2011 to 2015, only 14 five man lineups have played over 800 minutes together during any season[1. About 20% of the regulation minutes in a season, per Basketball Reference]. A further 22 groups played between 600 and 800 minutes together in a season[2. Approximately 15% to 20% of the team’s minutes.] over that same period
As the chart below shows, there are relatively few lie ups that have been put on the floor for significant amounts of time and the bulk of minutes in the NBA are played by line ups that shared the court as a unit for only a modest amount of time.
Overall, five man lineups that played together for more than 200 minutes only played about 17% of the total minutes scheduled in the NBA regular season[2. Based on my calculations via Basketball Reference data, not adding overtime minutes to the denominator, meaning I am erring to overestimate the percentage of time higher continuity spent on the floor]. Frankly, this is something of a pain in the backside for analysts. Lineups that only share a few minutes over the season are subject to too much noise to reliably say much about how the players meshed or did not.
In one sense the general competence of NBA coaches robs analysts of the most robust experiments, there aren’t any five man point guard lineups or four on five defenses[3. Yet], but the change of just one player can alter the context for the entire offense or defense.
This can be seen simply looking at the top two five man lineups for the Dallas Mavericks or Memphis Grizzlies last season. Each had team had two relatively high minute lineups with four of the same starters with one spot in the lineup changed. For the Grizzlies the common players were Marc Gasol, Mike Conley, Zach Randolph and Courtney Lee, with either Jeff Green or Tony Allen at the three. With Allen on the floor, Memphis’s always suspect spacing was reduced even further with Allen on the court even compared to Green in a down shooting year. Of course, this represented a tradeoff with significantly improved defense. For the Mavs, the lineups were the starters before and after the replacement of Jameer Nelson with Rajon Rondo at point guard via trade. Much like the Griz, this switch produced similar spacing issues, only without the same level of defensive improvement.
In addition to the performance of complete lineups, analysts have developed other methods of searching for synergies between players, factors that go beyond simply adding player’s abilities together. For example, team-level analysis can reveal connections between certain tendencies and winning. Differences in the production of two and three man combos can illuminate specific skill sets and position combos. Of course one has to be careful to account, even informally, for the missing variables of the other two or three players on the court.
Similarly, applying adjusted plus/minus techniques to facets of the game beyond the scoreboard such as team rebounding or teammate effective field goal percentages can also add to our understanding of how certain skills provide or subtract value when deployed in concert. Five man lineups, with a large enough sample of longer term groupings, can also help draw some inferences comparing players types and skills that compliment each other, but it is probably not going to help you project the success of any particular team.