David Sparks, director of basketball analytics for the Boston Celtics, graduated from Duke with a Ph.D. in political science in 2012, and earned his undergraduate degree in economics and political science from Vanderbilt University in 2006. Sparks is scheduled to give a talk sponsored by the political science department detailing the progression of statistical modeling in basketball analytics Friday afternoon at 1 p.m. in Gross Hall 111. Ahead of Sparks' talk, The Chronicle's Brian Pollack sat down with Sparks to discuss his career path, time at Duke and thoughts on analytics in sports:
The Chronicle: You earned a Ph.D. in Political Science from Duke—how did a degree like that lead to a career in basketball analytics?
David Sparks: I found that I needed a lot more math and stats to be a graduate student in political science. So I spent a lot of my coursework the first two years getting that. In the course of that, I learned statistical programming, first in Stata and then in R. So I took, with some of the other political science Ph.D. students, some courses in the stat department, and there was one in particular where we were trying to learn statistics and learn R at the same time. One of the ways I dealt with that was I found some baseball and then basketball data sets because they were very familiar to me, and you could sort of ask lots of direct and interesting questions to those and get answers very quickly.
In my spare time, I would kind of do the things I was trying to do in class—I would use Stata, and then later, R—to do the same sorts of things. I thought it was interesting, and I started making graphs. I really liked the graphing part of it. I thought it was sufficiently interesting that I wanted to share it. Online there’s a forum for basketball research called APBR Metrics [Association for Professional Basketball Researchers], and so I posted a few things there, and then I started my own little blog where I put things. I did that for a couple of months and Mike Zarren, who’s now the assistant general manager of the Celtics and my boss, he posted an internship opening to the message board and I applied and he had seen some of my work, so I got the internship. So while I was doing my Ph.D., I was sort of doing very part-time work for the Celtics from Durham.
What I found was doing things in political science and I’m getting better at statistics and getting better at statistical programming meant that I was a better basketball analyst. The difference in the subject matter wasn’t a hindrance. It actually ended up making it easier to get better at both, I think, and a lot of what I do now, I sort of describe as “social science for basketball”, where someone has an idea about what wins basketball games or what a good player is, and we can test that. It’s similar to ‘what creates democracy?’, ‘what creates a good economy?’, those kinds of questions, it’s just different data.
TC: How does the analytical work you do with the Celtics get filtered down to the management and coaching staffs?
DS: We do both personnel evaluation and strategic kind of things. So we work with both the general manager and the coaching staff. I don’t typically interact directly with the players. In general, things go through the coaching staff, which I think is the appropriate way to handle that stuff. There are very different kinds of problems—sort of macro-scale, ‘Who’s a good player? What makes a player good?’ and the more micro-scale, ‘How should we be playing in this certain game situation?’ So there’s a lot of interesting variety in those different kinds of domains.
TC: What has been the biggest development in basketball analytics you've seen in your time working there?
DS: To me, the big change is sort of obvious. Starting two seasons ago, the NBA installed SportVU cameras in all 30 arenas for all 30 teams and that data has been very interesting. It has let us answer, or start to answer, a lot of questions we couldn’t really get at with points, assists, rebounds, steals, things like that. So that has created a lot more potential for doing interesting things and building interest in different kinds of models.... The other difference is essentially every team has someone doing a similar sort of role [as me].
When I started, it was probably half a dozen teams doing it and I believe now I can at least remember reading about someone working for every team. I’m sure it varies from team to team the extent to which those groups are integrated. Maybe it’s a consultant for a team, in my case I’m full-time. I think that it’s not a secret anymore. I feel like it’s something most teams recognize it as something they should be working on and paying attention to.
TC: What exactly do SportVU cameras track and how is that data used?
DS: What SportVU does is there are six cameras in every arena and those cameras track the x and y coordinates of every player on the court, and also x, y and z for the ball. So that obviously is a lot of numbers—I calculated it out, that’s 11 values 25 times a second…it’s like 2 billion values. That’s a lot of data. It needs to be processed significantly before you can get useful things out of it. But because the resolution is so fine, it means you can do a lot of things. Essentially, almost any question that you can ask, if you can figure out a way to wrangle x and y into things you can count, you can start to answer those questions.
TC: The Duke basketball team is one of a handful of collegiate programs that uses SportVU cameras as well. Is this something that could open the door to analytics being more prevalent in the college game?
DS: Absolutely. People will email me and ask, and like I told you, I suggest starting a blog and getting your work out there. The other thing I suggest is I think college teams will be increasingly interested.... I will say that the NBA, that the data available for NBA teams is really good. I don’t know how easy it is to get data for college teams. Part of what a college stats analyst might be doing is data gathering themselves. That’s an obstacle only insofar as you have to do some work before you can do your stats, but I think it’s just a matter of time. I don’t really forsee a lot of obstacles and I think coaches want to win and taking a statistical approach to basketball can help teams win. So I think it’s just a matter of time
TC: How would you compare the state of analytics in basketball to similar movements in other sports?
DS: My impression is that the statistical approach to the game is there, and its well-entrenched, and it’s there to stay. I think basketball is right on the edge of that, and then I think football and hockey are headed in that direction. I think football and hockey are really tough actually, there’s so many players. Hockey in some ways, is just like basketball, but there are fewer things to count. But I would just say that baseball, this movement if you can call it a movement, it happened earlier in baseball, and baseball is a little bit more discrete. I think in some ways, it’s easier to get to very useful things in baseball than it has been in basketball.
One thing I like about basketball more than any other sport is the five guys on the court have ostensibly different positions, but each of them can do all the things the others can. In soccer, there’s only one goalie who can use his hands. In hockey, there’s one goalie in front of the goal. In baseball, there’s a pitcher. Different positions mean very different things. In basketball, a point guard can shoot a three, a center can create a turnover, and so can a point guard. I like that, but that's sort of hand in hand with the fluidity of the game. Good offense makes it easier for your defense, and good defense, if you can generate a turnover, makes it easier for your offense. So there's a lot of complicated interactions there.
Get The Chronicle straight to your inbox
Signup for our weekly newsletter. Cancel at any time.