2017 Sloan Sports Conference: Day 1

The 2017 MIT Sloan Sports Analytics Conference is being held in Boston Friday and Saturday. Several Nylon Calculus contributors are there covering the event and we’ll be sharing our thoughts and reactions to panels and research papers. Check back throughout the day for updates.

Even the most analytically savvy GMs are susceptible to bias

“People are bad at making decisions.”

That’s how Daryl Morey describes why he, Billy Beane, Sam Hinkie and Farhan Zaidi are participating in “Moneymind: Overcoming Cognitive Bias.” Moderated by Cade Massey, this panel focuses on the deleterious heuristics that plague every human being — even front-office decision-makers who surround themselves with rigorous analytics, systematic processes and other safeguards.

Indeed, in some ways, one can argue that the most rational and conscientious GM might end up being uniquely susceptible to decision-making pitfalls. How? As Zaidi notes, you can be so aware about cognitive bias — and protecting yourself against it — that it perverts into paralysis. “You almost have to worry about overcorrecting,” he says.

Moreover, you might establish a system to minimize human error, but end up exacerbating it or creating an entirely new problem. One example is a secret ballot. To overcome groupthink, GMs might ask their staff to file their votes blindly so that they arrive at conclusions on their own. But how much independence are you truly getting, Zaidi asks? After all, the staff are still using the same suite of tools based on the same set of data. I’d add that the staff have presumably been hired in similar fashion, which introduces selection bias into the mix. The result is a false sense of security against blind spots that have now been transformed or enlarged.

The panelists also discuss more common issues. Notably, they bring up situations when a contract negotiation goes well or a trade is deemed successful. Teams subsequently try so hard to replicate these deals that they either miss opportunities or make mistakes in a different context. “Every decision is independent,” says Morey, but this nugget of wisdom is easier said than followed.

In general, based on their understanding of behavioral economics and related fields, analytics folks tend to emphasize humility in decision-making. The panelists are no different. They realize that, in their very efforts to sidestep cognitive bias, they end up becoming potentially exposed to it in different ways. This realization itself is a key insight.

— Positive Residual, @presidual

Silver on Silver

FiveThirtyEight Editor-in-Chief Nate Silver and NBA Commissioner Adam Silver went one on one in what was certainly the best-named panel of the conference. At least for this year.

Nate Silver asked most of the questions, with the focus on both the current strengths and successes of the NBA, as well as the challenges the league is currently facing. With the Golden State Warriors and Cleveland Cavaliers presumably heading for a third consecutive NBA Finals matchup (Kevin Durant’s health notwithstanding) the topic of parity came up and how responsible the league is for ensuring that as a baseline dynamic.

Adam Silver characterized the NBA’s mission as, “doing the best job we can to create equality of opportunity,” with the implied counterpoint being equality of outcome. That is to say, the potential for parity is more important than parity for parity’s sake. The league’s intention is to create the opening for mobility. The end result is less important.

Adam Silver then turned the question back to Nate Silver, asking about his perception of parity in the league and how essential or non-essential it feels for journalists and fans.

“If you’re reporting on the sport and covering it as a journalist, to have recurring storylines is more compelling,” was Nate Silver’s answer and he want on to talk about the relative satisfaction an NBA season provides with the best team usually winning, as opposed to baseball or the NCAA Tournament which he described as sometimes feeling like a lottery.

The NBA gets credit for being one of the most progressive professional sports leagues when it comes to the use of analytics. For the second time today (see Evan Wasch’s presentation earlier this afternoon), it was refreshing to hear how much careful, rational thought and thorough analysis goes into the league’s decision-making about how to structure itself.

— Ian Levy, @HickoryHigh

Identifying offensive possessions from the ground up

Each NBA offensive set, like the Hammer and the Weave, consists of sequenced player movements at specific locations on the court. The trained eyes of video coordinators and coaches can easily identify them. But, given the massive volume of game film and data now readily available, even experts can often struggle with the task.

This type of problem tends to invite analytical attention, as machine learning can help ease the burden of categorizing play after play after play. Previous work has centered around such labeling. Based on standard definitions of particular offensive sets, algorithmic models have been developed to look for particular player interactions and appropriately classify them. Yet this method is still dependent on “known structures” when it’s entirely possible that undiscovered patterns exist.

Consequently, Andrew C. Miller and Luke Bornn strive to take a different approach. In “Possession Sketches: Mapping NBA Strategies,” they use “unsupervised structure discovery” to break plays down into small components and build a “vocabulary” of player movement from the ground up. They then treat the problem as one of “topic modeling.” Just as text mining can uncover hidden semantic structures in a document (for example, a scientific-journal article on cancer might have “tumor” and “cell” while another on neuroscience might feature “synaptic” and “hippocampal”), the Miller and Bornn algorithm can spot specific cuts and screens and associate them with Hammer action — even if the team executes the play at different speeds with their own nuances. It makes for a more flexible tool that’s aware of game context.

The project is definitely fascinating and promising, especially from the standpoint of drawing parallels between offensive possessions and text documents. The latter field has many existing tools that can be brought to bear to the NBA. If basketball analytics can unearth new ways for defenses to stymie offensive flow or, more ambitiously, outline entirely new types of plays, then Miller and Bornn will have truly advanced the movement.

— Positive Residual, @presidual

Tracking shot quality down to a shooter’s pose

The analytics community has developed increasingly granular ways to measure “shot quality.” Beyond the standard box score, which merely classifies field goal attempts as either 2- or 3-pointers, researchers have parsed play-by-play data to determine how game factors like assisted opportunities, shot clock and preceding action affect the likelihood of a bucket. Some have created shot charts to visualize high-value zones. Others, including our own Krishna Narsu, have developed models that incorporate dribbles, defender distance and other player-tracking data. SportVU even collects the “z coordinate” of the ball, which Mike Beuoy of Inpredictable has used to analyze shot trajectories for successful and struggling free throw shooters alike.

Enter “Body Shots.” In this research paper, Panna Felsen and Patrick Lucey exploit recent advances in computer vision and dynamic time warping to identify 17 attributes that describe player movement during a 3-point attempt. They then link these attributes to shot outcomes and identify the specific body motions that correlate with shooting success.

Felsen’s and Lucey’s findings seem to match what coaches might observe with the “eye test.” On open 3-point attempts, for example, made shots come in higher proportions when the player has no movement prior to receiving a pass, refrains from pump fakes, uses a narrow set foot stance and splits his legs during the fall. Interestingly, a case study on Steph Curry shows that “he moves more than other players in every phase of his shot.”

While the general results may simply confirm conventional basketball wisdom (at least in the limited scope of the study), there is certainly value in developing a new branch of the shot-quality literature with a fairly different technique. It’s also great to see objective methods applied to what might otherwise be a predominantly subjective project. Felsen and Lucey help us see how the “bottom-up” approach that many coaches and scouts take can be enhanced by analytics.

— Positive Residual, @presidual

NBA macrobiology

The NBA’s Special Vice President of Basketball Strategy and Analytics shared an interesting peek behind the curtain, in his talk about how the NBA thinks about itself when considering changes to the structure of the league. Specifically, he used the analogy of the life cycle of an NBA season, and viewing the relationships between those events, and the teams and the league, as an ecosystem. The draft lottery leads to the draft which leads to free agency, then summer league, schedule creation, training camp, roster finalization, the regular season, the trade deadline, the playoffs, and on to the Finals.

Wasch talked about the three goals the the league considered around any changes — 1) integrity, competitive balance, and fairness, 2) player healthy and wellness, 3) competitive quality and fan engagement. Adjustments that the league makes need to serve those goal but also need to be considered in the context of how they affect all the other aspects of the eco-system.

Adjusting the schedule and spreading out games theoretically improves competitive quality and player wellness. It also may change how teams build rosters — maybe willing to sacrifice some depth knowing their stars will get more rest baked into the schedule. That market adjustment changes the dynamics of the trade deadline and of the summer free agency period.

Wasch also talked about some specific changes the league has made over the past few years, and shared a proposal for draft lottery reform that had been brought to the NBA Board of Governors, before being ultimately rejected. The idea would have given the four worst teams in the league equal odds at the No. 1 pick, eliminating that race for the absolute bottom among the least competitive teams in the league. The other component was to add three additional lottery picks, so that the top six picks would be determined by lottery.

This would have improved competitive balance and increased fan excitement around the lottery. However Wasch shared that the league’s owners were nervous about making multiple adjustments to the team-building piece of the league’s ecosystem. With the rapidly changing salary cap structure, adjusting the odds around lottery picks felt like too much change, too quickly.

For all those who gripe about the league’s inability to solve problems like intentional fouls and conference imbalance, this was a strong reminder that it’s a lot more complicated than it looks.

— Ian Levy, @HickoryHigh

The 3-point Revolution and drawing hard lines

Misconceptions around sports analytics are varied, multiple, and strongly ingrained. One of the biggest and most persistent is that the invariable end product of analytic work is hard rules about ideal strategies and systems. The 3-point shot is the most obvious example. The mathematical difference between 2-pointers and 3-pointers is rock solid and the increasing prevalence of 3-pointers in the NBA would seem to imply a direct relationship, blindly worshipping at the altar of 3>2.

At his talk, Truths and Myths of the 3-point Revolution in Basketball, Nylon Calculus’ Editor Emeritus, Seth Partnow worked to undo that perceived hard edge. Partnow addressed several popular myths around 3-point shooting: they lead to longer rebounds (true) and more transition opportunities (myth), that it’s killing the mid-range game (myth, it’s mostly reducing catch-and-shoot 3’s, stars still take their pull-up jumpers).

Partnow also talked about the structural nature of 3-pointers. They come from complex actions, for the most part teams aren’t just “taking” more 3-pointers. They’re building lineups, sets, and systems to take advantage of both the mathematical value and the byproduct — space — that can be used for other things. And then there’s defense — there are clearly good and bad ways to defend a 3-point shooter, but the best way is to keep them from shooting, something our defensive stats have a hard time capturing.

In the end, Partnow was banging some of our old favorite drums here at Nylon Calculus — don’t think in absolutes, make sure the words you use really capture your ideas, acknowledge context and the unknown.

— Ian Levy, @HickoryHigh

Fully Invested in Analytics

When a sports franchise — or any private company, for that matter — engages in “corporate social responsibility” work, it typically does so at arm’s length. It’s easy to see why. Social engagement isn’t necessarily a core competency for these organizations, so it’s often best that they provide support in appropriate ways, which is usually with some distance from the daily provision of community services.

Such context helps us understand what makes Step Your Game Up rather unique. In terms of program design, this Boston Celtics project is a fairly standard intervention, as it aims to identify at-risk middle school students, offer them contracts that link improvements in school attendance and academic grades with team-related incentives, then track objectively measurable outcomes. Its notable feature is the level of investment by the franchise.

From the start, one of the priorities has been to execute the program through internal collaboration between the Celtics’ community engagement and analytics departments. Rather than having an external evaluator, the team has marshalled the resources and expertise of its data unit, including director David Sparks, to monitor the impact of its social investment. Consequently, Step Your Game Up boasts, as assistant general manager Mike Zarren puts it, the “first statistically significant evidence” of a franchise’s direct impact.

I’m unsure how we might verify this statement, but either way, the Celtics have established a model for how analytics expertise can be leveraged for purposes beyond the court. At a time when “stick to sports” has become a common complaint, they have issued a constructive — and perhaps reproducible — rejoinder.

— Positive Residual, (@presidual)

Analytics as a teaching tool

Focusing on student interests is the most basic of engagement techniques teachers use with their students. If you’re working on some aspect of writing, give the student choice in the topic. If you’re trying to teach a mathematical or science concept, frame it around something the student is interested in, like trucks, or animals, or sports.

That was the topic of the research paper, From Sports to Science: Using Basketball to Broaden the Appeal of Math and Science Among Youth. One of the paper’s authors, John Drazan, shared the group’s work to try and answer this fundamental question:

“How can we create an authentic yet accessible introduction to STEM (Science, Technology, Engineering, and Math) using sports analytics?”

The most interesting piece of application here is that this research approach took things further than just designing math and science curriculum around sports questions and ideas. The research group built a open-source app and a structured program that actually had the students at their clinics shooting baskets and tracking data on their own shot locations and efficiency. In this case, sports is not just an abstraction, a contextual hook to do the same old sorts of science and math work. The students are actually doing the physical activity, playing  the sport that interests them, and then doing the work and analysis on their own work.

Drazan and his group were also clear that their work could be a way to reach out latino and african-american students who are significantly underrepresented in STEM careers. The group found that their program increased student engagement and built a positive perception of the science and math involved. The driving force, and one that the authors hope would cause the engagement and perceptual shifts to persist, is that it’s connected to the activity — STEM can help make you a better athlete.

— Ian Levy, @HickoryHigh

 

Load Comments