Metacognition: NBA narrative mapping, data-driven approaches


Each NBA season is a sprawling story with thousands of characters and infinite arcs, large and small. How can we organize this massive narrative?

The NBA season itself is an eight-month slog of daily basketball. Tack on another four months of draft prognostication, free agent shenanigans and optimistic offseason ramblings. That stretches the story of an NBA season over 12 months, binding together the micro-narratives of nearly 600 players (counting draft prospects and G Leaguers), 50 teams and thousands of support staff, scouts, coaches, trainers and executives, not to mention the tendrils that stretch beyond this season contextualizing the present with both the past and future.

Keeping track of it all is impossible and even just keeping track of the essential arcs is a monumental challenge in executive functioning.

Over the past few months, I’ve been intrigued by the idea of trying to put down on paper (actual or digital) the mental models and narrative structures we, collectively, use to try and follow and understand an unfolding NBA season. In the next few installments of Metacognition, I’m going to share some of the ways I’ve been trying to illustrate my thinking.

As is my tendency, numbers were a comfortable place to start, putting an easily understandable scaffolding around a sports story. The map below shows the week-by-week progression of Google Trends search data for the Phoenix Suns, going back to the end of last regular season.

I’m not sure what would explain the small spike of interest in late May, but otherwise, this seems like a pretty accurate portrait of the last nine months for the Suns’ organization, as defined by what generated national interest. There is a spike in search traffic leading into the NBA Draft, where the Suns held the No. 6 pick. They ultimately traded that pick to the Timberwolves for Dario Saric and the No. 11 pick, which they used on shooter Cam Johnson. Interest surges again through the beginning of free agency but drops off precipitously after the signing of Ricky Rubio, which largely signaled the Suns were finished making big moves.

General interest rose leading into the season but spiked again with their opening night win, the PED-related suspension of Deandre Ayton and their win over the Clippers. As they fell back towards the middle of the standings, interest dropped and we can see later changes at least somewhat pinned to big wins or losses.

One of the issues with this mapping structure is it’s not actually descriptive of the nature of the Suns’ season, other than the events I cherry-picked to highlight. Interest could be surging for positive (beating the championship favorites just a few days into the season) or negative (second-year cornerstone suspended for testing positive for PED masking agents) reasons.

The alternative structure below takes a slightly different approach, adding more data points to the events side and putting the outside observers in as a static marker. The graphs show a five-game rolling average for the net rating of the Oklahoma City Thunder, for last season and this season. The blue line represents 538’s preseason projection for the team’s net rating, as a stand-in for the general public expectations and perception of the team.

In 2019-20, the Thunder have already gone into 20 games carrying a five-game rolling average of net rating that was above their preseason expectation. They finished all of last season with just 41 such games and didn’t hit 20 until their 56th game of the year. In plain English, last year’s Thunder team was objectively better but this season’s team has exceeded preseason expectations by a wider margin. So, on some level, this feels like found money whereas last season was a disappointment.

Sequencing is another important aspect of a team’s story. A team that wins 41 consecutive games and then loses 41 games in a row would have had a very different experience than another team with a .500 record but who never won or lost two games in a row. The visualization below looks at the sequence of wins and losses for the Lakers and the Bucks, the two teams who have been, by far, the most dominant in the first half of the season.

It’s an extremely simple visualization but offers all sorts of interesting insights. The Bucks have yet to lose two games in a row, while the Lakers had that four-game losing streak leading through Christmas. And while the Bucks have that enormous 18-game win streak in the middle of the season, they otherwise haven’t been able to rip off many long runs. Their second-longest win streak is just five games, whereas the Lakers’ losses have been more evenly distributed, giving them win streaks of 10, nine and seven games (twice).

In each of these cases, the statistical information is meant to be narrative and descriptive rather than evaluative or predictive. A win is a win. A loss is a loss. But it is the sequencing, rhythm, size and causes of those victories and defeats that write the story.

I’m just beginning to play around with different formats for mapping team narratives so stay tuned for some additional experiments, along with some that are completely separate from statistics and results at all. In the meantime, if you have any tools or structures you use for thinking about these kinds of extended narratives, please share!

Metacognition is an irregular column series, thinking about how we think about basketball. Check out the entire project at A Unified Theory of Basketball.