I couldn’t stay up late enough to watch the NBA Finals, they started after 9 PM ET and finished close to midnight. But when I looked at the Game 6 box score the next morning, I was struck by how the basic basketball box score is actually an incredibly insightful piece of data presentation. And I wondered if Pickaxe could be trained to write summaries of the game without ever watching a clip.
Currently, there are automated services that write game summary articles but they use event based analysis like quarterly summaries, etc and they often fail to capture the “why” and “so what”. And there’s an entire school of APBR metrics, but it’s hard to read
Exponent = 1.5 log(R+RA/G)
etc and translate that into a sensible understanding of what happened in a game.
I wanted to see if we could take just the raw end-of-game box score and summarize the game accurately for the casual fan. So we took the data and ran it through our insight engine process and here’s what it had to say:
What’s interesting is that it called out defensive statistics more than offensive. Traditionally, sports game summaries start with the high scorer. But to the engine, those defensive metrics were more unexpected given the playoff and season averages. For example, Steph Curry scoring 34 points has happened a lot of times, but the Warriors hadn’t averaged anywhere near that many turnovers this year and Andrew Wiggins had never had this many blocks and steals in a single game.
One thing the engine didn’t pick as one of the top three insights was that Jason Tatum scored only 13 points compared to his season average of 26 points per game. While that was significant, it was probably weighted lower than the other three bullets because he had an even worse game recently (in Game 1, he scored 12 points) so not only was this only 50% below his average, it was also not that big a deviation from the mean if you look at his previous 5 Finals games.
So, all in all, this was a pretty accurate summary. Now, none of these insights are things that good reporters missed and automated analysis isn’t meant to replace in-person coverage. But the ease with which Pickaxe’s Insights Engine was able to write such a sensible summary of the game is a testament to the usefulness of the basic basketball box score is and why it is such a great example of good data analytics!
Further Reading
If you enjoy nerding out about basketball box scores, I recommend BasketballReference’s archive and history of the box score evolution and this Grantland story of how that came to be.
Disclaimers
- This was a rough – and highly unscientific – extraction of stats from ESPN.com and Basketball-Reference.com. It only included numbers from the Celtics & Warriors and only their season averages (not all 82 games), playoffs average (not individual games), and the previous 5 games of the Finals. This means the insights could lack context depending on the kurtosis of individual players’ scoring patterns.
- I only looked at the basic set of box score player and team stats. So this doesn’t include any fancy stats like Player Efficiency Rating,
- We don’t currently have the NBA as a client and this was done for non-commercial purposes, and because we’re curious about this kind of stuff.
- Also, as a Nets fan, I need something to distract from all the depressing Brooklyn news this summer. Or maybe we’ll turn this on next season so I don’t have to watch any of the games.
