Lineup Optimization or: How I Learned To Stop Worrying And Love Sam Mitchell
Measuring the effect and importance of coaching is a notoriously thorny issue. A reliable and reasonable statistical method to evaluate coaches is the Loch Ness monster of the APBR community. There are many who don’t believe such a thing exists, but the legend persists. The crux of the challenge is sifting out what part of performance is purely attributable to the players, and what part a coach has had their hand in. Statistics are generated by players and using them to evaluate a coach inevitably means the inclusion of a multitude of other variables.
There is however, one place where a clear distinction between player and coach can be made. In this one instance, the responsibility for a statistic, which is attributed to a player, falls primarily in the domain of the coach. I’m talking about minutes played, and specifically, how minutes are distributed to the various players on a team. It’s one of a coach’s most basic duties – deciding who will play, when, for how long, and in what combination. Other than in cases of injury or suspension, this distribution is usually the sole discretion of the coach, and therefore provides a set of statistics which can be attributed directly to them. The question then becomes – Can you quantify how adept a coach is at distributing those minutes?
I have a few different ideas to try and answer that question. To start with, I’m going to re-introduce a technique I used briefly while writing for The Two Man Game during the Finals. In one of his Finals’ recaps, John Hollinger mentioned that the Mavericks had pursued Rick Carlisle, in part, because their statistical analysis showed he had a clear tendency to give the most minutes to the most effective lineups. To check this idea, I borrowed the playoff five-man unit data from Basketball Value and ran a correlation between the number of minutes each lineup played for the Heat and the Mavericks, and the Net Rating (ORtg. – DRtg.) of that lineup. To account for absurd Net Ratings from small samples, I weighted the Net Rating for each lineup, multiplying it by the number of minutes played. Rick Carlisle and the Mavs had a 0.692 correlation between the effectiveness of the unit and the number of minutes that unit played. The correlation for Erik Spoelstra and the Heat was -0.177. Those numbers certainly dovetailed with the more qualitative and subjective evaluations of the way each coach handled their lineups.
For this piece, I’ve taken that simple correlation method and applied it to all the regular season five-man unit data from Basketball Value, covering the last four seasons. There were, unfortunately, a few exceptions to that data. Since Basketball Value’s data covers the entire season, I wasn’t able to include any year in which a team changed coaches mid-season. Aaron Barzilai who runs the site is also a consultant for the Memphis Grizzlies which means their data isn’t listed on the site and therefore, is not included.
The table below shows the correlation between minutes played and weighted Net Rating for each unit, of each team (other than the Grizzlies) who had the same coach for the entire season, for each of the last four seasons. I’ve also included the Win % of each team.
[table id=22 /]
Doc Rivers is one coach who looks incredibly effective by these numbers. Three of the top six individual seasons in this sample belong to him, including the highest correlation I found, his 0.953 mark in 2008. Rivers has had the benefit of coaching some incredibly talented rosters, but has also clearly demonstrated the ability to maximize the available talent.
Sam Mitchell also stood out quite a bit. His 2008 Toronto Raptors won just half their games, but Mitchell’s minute distribution showed a 0.763 correlation between the effectiveness of the unit and the number of minutes they played. This was the 17th highest mark in our sample. There are several different ways to interpret these numbers. In this case, I see a coach with a limited roster who did a terrific job doling out minutes and managing his rotations. In a cruel twist for Raptors’ fans, the coach who replaced Mitchell, Jay Triano, had two of the ten worst numbers in our sample.
I also wanted to try and extend these numbers, looking for patterns across seasons. This next table shows the correlation for each coach covering the combination of all the seasons they worked since 2007-2008. For some coaches this is just a single season.
[table id=23 /]
Because these numbers reflect a single season for some coaches, and multiple seasons for others, it’s difficult to draw strict comparisons. However, some of the names which rose to the top (or fell to the bottom) of this analysis were no surprise. Of the coaches who worked multiple seasons over this stretch, the top five, by these numbers, were Doc Rivers, Phil Jackson, Stan Van Gundy, Flip Saunders and Alvin Gentry. The bottom five were Kurt Rambis, Jay Triano, Mike Dunleavy, Jon Kuester and Lawrence Frank. I was also surprised to see Jerry Sloan, George Karl and a few other notables hanging in the middle of the pack.
While these numbers have a lot to say, there are some obvious issues with both the method and results. Weighting a unit’s Net Rating by minutes played may not be an accurate reflection of how that unit would have played given more time on the floor. The options a coach has in determining units can also be limited greatly throughout the season because of injuries or roster moves. In short, the numbers may not be perfect, and they may not entirely reflect a coach’s ability to use what is at his disposal.
Although it will take some time to put together, I have another method I’ll be trying in the next few weeks, looking at this issue from a different angle. Hickory-High has been on a lengthy hiatus, but the vacation is over!