The Curious Case of Rick Carlisle

Posted by Ian Levy on October 20, 2011 under Commentary | Be the First to Comment

Screen shot 2011-10-19 at 4.17.00 PM

Ian Levy is the author of Hickory High, a contributor to Indy Cornrows, and a part of The Two Man Game family. You can follow Ian on Twitter at @HickoryHigh.

For several Maverick veterans, last year’s championship run was the final component in cementing a legacy. It verified Dirk Nowitzki as a legitimate Hall of Fame candidate and a top-five NBA player (I’m looking at you #NBARank). It was the capstone on Jason Kidd’s claim as one of the best point guards in NBA history. It helped shift the focus from Tyson Chandler’s injury history to his status as an every-night warrior and his strong defensive foundation. It validated the years of specialized and all-around production provided by Peja Stojakovic and Shawn Marion respectively. And, although it hasn’t been mentioned nearly as often, I believe it also established Rick Carlisle as one of the game’s elite coaches.

After his ninth season as a head coach, Carlisle has won a total of 443 games with a 0.600 win percentage. He’s led his team to the playoffs in eight of his nine seasons and has posted a playoff win percentage of 0.535. He’s achieved a great deal, but has also had the benefit of coaching some great players. In evaluating the worth of a coach, how do you examine team success and separate the contributions of coach from player?

During one his NBA Finals recaps, John Hollinger mentioned that one of the reasons the Mavericks pursued Carlisle was that their statistical studies showed he had a tendency to give the most minutes to the most effective lineups. This idea struck a chord with me. The biggest stumbling block in using statistics to analyze the performance of coaches is simply finding numbers which can be directly attributed to the coach.  The idea Hollinger mentions seems to me to be the one domain where just such a set of statistics exist. Coaches decide which players are on the floor and in what combination at any given moment. The effectiveness of those choices provides us with one quantitative measure of a coach’s effectiveness.

Over the past few weeks, I’ve used two different techniques at Hickory-High to try and examine this issue, both of which relied heavily on the statistics BasketballValue provides for five-man units. Their data only covers the last four seasons, which means that my analysis is restricted to that time frame. In my first piece, I ran a series of correlations between the effectiveness of each five-man unit, using their Net Rating weighted by minutes played, and the number of minutes they were played by their coach. This technique was an attempt to directly represent the statistical studies Hollinger mentioned, with the idea being that the higher the correlation is, the better job a coach has done of judging the effectiveness of each of his units and allocating minutes accordingly.

With this first technique, Carlisle came out looking very good, but not necessarily great. His individual season correlations ranged from 0.488 in 2009 to 0.645 this past season, none of which ranked in the top 25 of the seasons I looked at. Carlisle did stand out somewhat for his consistency. Many coaches showed volatility in their numbers from season to season, possibly a result of injuries, roster changes or even an irrational infatuation with a particular player or configuration. Carlisle’s cumulative correlation over the last three seasons was 0.585. Of coaches who worked multiple seasons over the last four years, that number ranked 8th, trailing only Doc Rivers, Phil Jackson, Stan Van Gundy, Flip Saunders, Alvin Gentry, Byron Scott and Gregg Popovich.

In my second piece, I started with the same set of data but used a slightly different technique. This time I calculated the percentage of the team’s lineups that finished the season with a positive Net Rating, and then compared that percentage to the number of minutes those positive lineups were allotted as a whole. For the purposes of the study the difference between those percentages — positive or negative — is what we’re considering the coach’s ability to manipulate their lineups. This second technique may be a more blunt measure than the first, but I think it helps solve some of the inflation from stacked rosters that occur with the first technique.

With this second technique, Carlisle again stands out: not for his superior performance, but for his consistency. None of his individual seasons made the top 25. Over the past three seasons, 50.4% of the Mavericks lineups which played at least five minutes together have had a positive Net Rating. Those positive lineups have played 65.5% of the possible minutes, an increase of 15.1% over a random distribution of minutes. That number ranks fifth among coaches who worked multiple seasons over the last four, trailing only Jackson, Van Gundy, Rivers and Gentry.

The two techniques I’ve put together are not exact and not comprehensive. However, I think they do give us some quantitative information, general though it may be, about a coach’s ability to optimize their lineups. As with any question of basketball analysis, if we are left with questions after an initial look, the task is to find more data and narrow the focus. The numbers I’ve put together show that Carlisle has been among the league’s best at managing his rotations. For the skeptics out there, I was able to find some more specifics on how those rotations shake out.

I started by combining all of the lineups Carlisle has used during his three-year tenure in Dallas, trying to find some natural groupings for them based on the number of minutes played. I settled on these categories and general (subjective) classifications:

  • Over 145 MP – This group represents the primary starting lineups for each season and the go-to bench rotations.
  • Between 82 and 145 MP – This group represents the deeper bench rotations, and situational lineups.
  • Between 40 and 82 MP – This group represents deep, deep bench rotations, extreme situational lineups, and some accommodations for injuries.
  • Less than 40 MP – This group represents garbage time lineups, minutes scraped together for rookies, and strange experiments.

For each of those categories I counted the total number of lineups Carlisle has used in Dallas. I also calculated the cumulative Net Rating for all the lineups which fit into each category.

CategoryNumber of LineupsNet Rating
>145 MP12+10.4
145-82 MP12+10.1
82-40 MP38+1.7
<40 MP329+1.4

The way his numbers worked out are unique on several fronts. The first is how many lineups were found in each of the first two categories. With 12 lineups in each category over 3 seasons, we find the Mavericks under Carlisle using an average of 8 lineups a season which play more than 82 minutes. There is absolutely no pretense of riding a formidable starting lineup into the ground. The Mavericks’ most played lineup last season, Barea-Terry-Marion-Nowitzki-Haywood, logged just 350.45 minutes. That mark was eclipsed by 25 lineups from 21 different teams. Instead of relying on one dominant lineup, Carlisle is able to spread the minutes around in many combinations. But the most important factor is how effective those combinations are. With cumulative Net Ratings over +10.0 for both lineup categories above 82 minutes played, Carlisle is able to maintain a first tier level of performance among several lineups.

As the advanced statistics movement trudges forward, there will continue to be a vocal segment searching for a one-size, fits all, comprehensive measure, a number which says definitively that one player is better than another. To be honest, I find that search to be counter-productive. A visual observation can usually tell us who the best players are. For me, the benefit of advanced statistics is the increased ability to delve into details. I don’t want one number to explain everything. I want all the numbers. When it comes to looking evaluating coaches with statistics, I see similarities everywhere. A single numeric representation of a coach’s ability is certainly out of reach at this point, and to be honest, I’m not sure it’s needed. The fun is in the thin slices, in digging into the specifics and unique situations to try and find some answers. These techniques I’ve put together aren’t meant to be a step toward that single unifying theory of evaluation, they are meant to be another thin slice. They are meant to answer a few questions, and hopefully raise just as many.