Trying to fill an off-season without an end-date, I took a stab this summer at applying statistical analysis to the performance of NBA coaches. I’ve presented two different methods over the past few months, both relying on the same premise and focusing on the same facet of coaching. The only available set of statistics which can be comfortably placed wholly in the domain of a head coach is the distribution of minutes. Assessing how well this one aspect of the job is managed gives us one measure of the effectiveness of a coach.
The first method I used was to run a correlation between the minute-weighted Net Rating of each five-man unit and the number of minutes their coach had them on the floor. The more effective a unit is, the more often they would be playing together, theoretically. The stars of that first method were coaches like Phil Jackson and Doc Rivers. I received a fair bit of feedback on that piece along the lines of, “Duh, they have the best players. It’s easy to decide who to play when you have a loaded roster.”
The second method I used was an attempt to hone in on coaches who might not excel at consistently running dominant lineups on the floor, but who succeed in keeping their negative combinations off it. Here I calculated the percentage of a team’s lineups that outscored their opponents, and then the percentage of the team’s minutes that went to those lineups. The difference between the two should be an indicator of how well a coach is managing their rotations.
Both of these methods are obviously limited in their scope. There is more to being an effective coach than managing rotations. There are factors beyond the coach’s control which affect these numbers – injuries, suspensions and trades. Still I believe they give us a slightly-less-than-hazy picture of who knows their personnel and uses them effectively.
The logical next step in this analysis was to take these two methods and apply them to numbers from the playoffs. I’ve combined the outputs from both methods into the single table below. Positive% is the percentage of a team’s five-man units which played at least five minutes together and also outscored the opposition. Min% is the percentage of the available minutes which were played by those positive units. Diff. is the difference between those two percentages. A positive result here indicates a coach who squeezed more out of what they had, a negative result indicates the opposite. Correl is the correlation between how effective each unit was and how often they played. The table below shows the results from the last five seasons.
|2010||CHI||Vinny Del Negro||0.500||0.652||0.152||0.038|
|2009||CHI||Vinny Del Negro||0.538||0.374||-0.164||-0.193|
|2007||HOU||Jeff Van Gundy||0.455||0.749||0.294||-0.743|
|2011||ORL||Stan Van Gundy||0.357||0.240||-0.117||-0.108|
|2010||ORL||Stan Van Gundy||0.615||0.809||0.194||0.708|
|2009||ORL||Stan Van Gundy||0.583||0.384||-0.199||-0.345|
|2008||ORL||Stan Van Gundy||0.667||0.806||0.139||0.261|
|2007||ORL||Stan Van Gundy||0.286||0.134||-0.152||-0.846|
This next table shows the cumulative results for each coach over the past five seasons. For some coaches this is multiple seasons, for some it is just one.
|Vinny Del Negro||0.524||0.474||-0.050||-0.036|
|Jeff Van Gundy||0.455||0.749||0.294||-0.743|
|Stan Van Gundy||0.547||0.533||-0.014||0.089|
When looking at the playoff numbers we’re dealing with a much smaller sample size. This creates some problems, while solving some others. Smaller sample size means it’s easier to identify patterns. But unfortunately some of these patterns stand out for a freakishness born of small samples.
Doug Collins was the only coach with a single season correlation ranked in the top 10 who didn’t win a playoff series in that season. His 76ers were overmatched by the Miami Heat, but Collins did his best to keep positive rotations on the floor. The biggest factor was that his starting, and most used lineup, posted a Net Rating of +37.76. The next six most used lineups in that series were outscored handily by the Heat. Collins’ numbers look good here because the starting lineup of Jrue Holiday – Jodie Meeks – Andre Iguodala – Elton Brand – Spencer Hawes was so much better than any other combination he could put forth. However, it seems conceivable that the series could have been closer had Collins relied on those starters even more. That wildly successful unit played just 56.38 of a possible 240.0 minutes together in the series.
In Indiana we found the opposite – a coach riding his starters when he probably should have eased off the reins a little. Frank Vogel played his starting lineup of Darren Collison – Paul George – Danny Granger – Tyler Hansbrough – Roy Hibbert 80.72 of 240 possible minutes in the Pacers’ playoff series against the Bulls. That unit was absolutely destroyed, cobbling together a Net Rating of -16.17. The Pacers’ next two most used units posted Net Ratings of +50.91 and +15.48 respectively, but they combined to play just 26.95 minutes. There are no sure things in sports, but it seems like the Pacers might have been able to steal another game or two if Vogel had been willing to go away from the patterns he established in the regular season and follow what was working at the moment.
In many ways the regular season is an experiment. Coaches have a chance to dabble, assemble different rotations and try to identify the ones which work best in different situations. In looking at numeric stories big and small, it’s amazing how often coaches seem unwilling or unable to make drastic changes when the matchups overwhelm the rotations they’ve established. When the matchups align a coach can look like a genius. When they don’t, they can look decidedly Dunleavyish.
One of the constants across teams, seasons and techniques was the volatility of the numbers. This was no different when we look at the playoff numbers. By these numbers, Mike Brown had the single best coaching performance over the past five seasons, in 2009 with the Cavs. He also had the 20th worst mark in 2007. We also find coaches like Stan Van Gundy and Phil Jackson bouncing between the top and the bottom.
There are lots of ways to interpret this radical shifts in performance from season to season. Injuries, trades, and the simple fact that players change, all have a huge part to play here. Maximizing rotations is a skill that is dynamic to the nth degree. The available tools and the scenarios within which this skill is performed are constantly changing. Very few coaches appear to do it well all the time. It seems the mark of a good coach is someone who can do it well some of the time.
While these techniques I’ve been using are probably not new (The Mavericks’ pursuit of the same idea is reportedly what led them to Rick Carlisle), to my knowledge, this is the first time these numbers have been on public display. The phantom and statistical sorcerer, “Crow”, who haunts the APBRMetrics forum and the comments section of Daily Thunder has done some digging, using these methods as a launching pad to look at some smaller patterns. Between the two of us, the surface has barely been scratched. If anyone else is willing or interested in using these numbers to initiate an exploration, micro or macro, of how a coach or coaches manages rotations please borrow, borrow, borrow. Just let me know what you find.