The Toronto Raptors have been on quite a run this past month, vaulting themselves onto the top half of the Eastern Conference standings and establishing themselves as a team to watch across the NBA. Sports analysts have been quick to assign various reasons for the turnaround, but one of them comes down the team’s smart use of data and analytics to make better decisions about personnel and gameplay.
Always interested in how organizations across Ontario are using data to make better decisions, we sat down with Alex Rucker, the senior analytics consultant for the Toronto Raptors, to learn a little about the data modelling system they’ve created and the impact of data and analytics on not just the NBA, but on decision-making in all kinds of organizations.
Alex, thanks for speaking with us. Can you quickly introduce us to who you are and what you do?
Thanks for the chat. I’m Alex Rucker, a founding partner of KBAR Consulting LLC and, at present, the senior analytics consultant for the Toronto Raptors. I head a small team that provides basketball analytics exclusively for the Raptors. My long-time business partner, Keith Boyarsky, and I have built a modelling and predictive analytics system that can directly inform coaching decisions such as shot selection on offence and positioning and rotation on defence.
How did you get involved with basketball analytics, and with the Raptors?
I got started in the business of basketball in the mid-90s working first for the University of Western basketball team, and then with an internship at Simon Fraser University. Jay Triano (the previous coach of the Raptors and the current coach of the Canadian men’s national team) was the head coach of SFU, and we stayed in touch. When Triano became the head coach of the Raptors, he invited me to demo my analysis to the coaching staff.
Keith, who has a background in engineering and computer science from UC Berkeley, and I worked closely for three seasons building our models with the Raptors. A couple of years ago, we added Eric Khoury, a University of Toronto graduate, to the analytics team.
Can you tell us a little bit about how the basketball analytics system you’ve developed works?
Data-driven analysis of players and strategies has been hugely influential in baseball, to the point where the term “moneyball” is familiar to casual fans and Nate Silver is a household name. Basketball is a more complex sport, with a wider range of possibilities within each sequence. Moreover, the traditional box score metrics don’t provide a granular enough level of analysis to be truly informative for coaches. Some progress has been made by applying logic from other sports like hockey, so that plus/minus (+/-) statistics is now available for individuals and units of players, but the next step is to look deeper into each possession and assess strengths, weaknesses and possibilities within that system.
Starting two seasons ago, the Raptors (along with several other teams) installed SportVU cameras in their arenas. These provide a bird’s-eye view of the entire game and an associated data feed that provides player, ball and referee locations (more on that below). The NBA has begun publishing some of the data in aggregate on their new stats portal, and analysts are figuring out how to slice the data and develop actionable recommendations for players and coaches. This season, SportVU cameras have been installed in all NBA arenas, and the analytics team will now have even more data to work with going forward.
The SportVU cameras provide an XML feed with the location of all the players, the ball and the referees. The granularity is 1/35th of a second for a single frame of video. A game generates over a million rows of data: 14 data points x 35 frames per second x 2880 seconds per game (48 minutes). The analytics team parses this data and builds models of optimal shots, defensive positioning and substitutions. The models can also provide input into player personnel decisions, such as playing time and combining players into units, as well as insight into potential trades and contracts.
The data loading and warehousing is done on a standard SQL server setup. The analysis, modelling and visualization is done with custom Java tools developed largely by my business partner over the past several years. They have also developed a visualization front-end to generate reports for the coaching staff that highlight opposing players’ tendencies at a very specific level (for example, how often the player goes right vs. left off the dribble).
Is this significantly different from what was being done before? Or from what other teams or other sports are doing?
Most basketball analytics has been done with the available summary data. However, box score summaries do not include much valuable information on defence, and there is a limited amount of data to work with. Play-by-play data adds another layer of depth to potential analysis, but still does not provide a complete picture. While you may know how the sequence ended (a shot was made or missed, or the ball was turned over), the data still lacks context about how the play developed, where the players were positioned and what options were available at each juncture.
Camera data allows the analysts to model and predict at a much more specific level. For example, the analytics team has developed a model that shows the optimal defensive positioning on every position, which can be plotted against the team’s actual performance. This video demonstrates a simple version for a single play: https://www.youtube.com/watch?v=5Sq_Z6Um3UM
How is all of this making the Raptors a better team? What does the future look like?
The notion that objective, data-driven analysis should inform decision-making and will result in better decisions is absolutely correct, and that can and should be applied in any industry. Leveraging data in professional basketball for competitive advantage will result in a superior organization (more wins over the long run, relative to the other 29 teams) and, based on the current landscape of how the game is played, a more exciting brand of basketball. Those two key factors—improved performance and a more engaging, entertaining style of play—will result in a superior fan experience and, and as a result, accelerated economic growth for our organization.
Basic analytic concepts have become relatively mainstream in the NBA. A number of front offices have an analytics staff, and successful offences have been built around the core quantitative findings: the value of three-point shooting, the importance of getting to the foul line and the sub-optimal efficiency of mid-range jump shots. While data is driving decision-making, it’s not possible for every shot to be a dunk or a three-pointer; teams still need to figure out how to optimize within the existing constraints.
Serious statistical modelling is just beginning, because micro-level data is just coming online now. SportVU includes a summary and analysis tool, but the Raptors team found the raw data requires additional smoothing and cleaning to make it more useful. Much of the work is still around data quality and getting things into a useful form, since the optical recognition and location coding from the software can be somewhat inexact. The next step is applying machine-learning algorithms to figure out when a certain ball movement trend is a shot versus a pass versus a dribble. After that, you need to build the models and evaluate what should be happening, what is happening, and where are the clearest locations for improvement. It’s a complex set of problems and there’s a lot more work to be done.
Any chance we can take a sneak peek at the reports or visualizations you produce?
The analytics service is proprietary to the team, so I can’t share them right now. If you want to learn more about the methods, we’ve gone into depth about them in a two-part series on ESPN’s Grantland.