This Week in Sports Analytics:
This is the second week of TWISA and in order to avoid too much overlap with StatStuffer.com I’m going to focus less on completing a comprehensive overview of sports analytics articles and more on highlighting the news or analysis in each sport that I found most compelling.
Dan Aucoin at Driveline Baseball took a look at methods to increase spin like increasing velocity, adjusting spin axis, finger strength, and friction. We’re firmly in the era of using data not just to evaluate but improve performance and that will be explored in the upcoming MVP Machine book by Travis Sawchik and Ben Lindbergh. This area of research is only going to grow in importance.
Jim Albert took a look at pitcher usage historically. His findings aren’t surprising, but I’d have to agree with some the main takeaways such as the diminished importance of the starting pitcher. Ben Lindbergh came to a similar conclusion back in October.
This article wasn’t written last week, but Ben Taylor’s deep dive on NBA advanced metrics is a great read. It gives us an idea of how much better new composite Box Plus-Minus metrics fare at predicting team performance.
There appears to be a big game coming up this Sunday and Aaron Schatz at Football Outsiders has a deep dive on the strengths, weaknesses, and strategy for both teams.
Statsbomb released a new metric called xClaimables that attempts to measure the expected number of claimable balls that will be claimed by a keeper. Diving into the data they were able to show that Hugo Lloris’s reputation for being error prone might be due to 3 mistakes on claimable ball attempts.
Alyssa Longmuir details her inspiring journey to bring analytics to Australian Women’s Ice Hockey League.
Aleksi Pietikäinen wrote about how DeepMind’s Starcraft 2 performance likely benefitted from allowing the model to execute commands at superhuman speed. I know next to nothing about Starcraft but the writeup has interesting takeaways for machine learning. Constraints need to be properly considered.
Zillow’s Kaggle Zestimate competition is finally over after launching in May 2017. Very interesting competition and hoping to learn more about the methods used in the final model.
Lukas Biewald gave some advice for managing machine learning projects.