On Saturday, I attended the New England Symposium on Statistics in Sports (NESSIS). Compared to the Sloan Sports Analytics Conference, I found NESSIS much smaller, focused, and academically rigorous. The presentations revealed fascinating dimensions to analytics in a variety of sports, and the small size afforded me the opportunity to speak with many of the presenters. I even talked to some statistical legends I've been following for years. I will certainly be returning; unfortunately NESSIS only occurs every other year.

Ryan Brill, a PhD student at University of Pennsylvania, stole the show with his talk "Analytics, have some humility: a statistical view of fourth down decision making" (Abraham Wyner also credited on the paper). He revealed some troubling flaws in widely-circulating fourth-down calculators based on Expected Points Added (EPA) and Win Probability Added (WPA). Specifically:

• Good teams run more plays in the red zone than bad teams, so EPA calculations based on statistically averaging without regard for team quality overweight the chances that good things will happen for the offense. Accordingly, fourth-down calculators based on EPA also show this flaw.
• While there are hundreds of thousands of plays that EPA / WPA models pull from, WPA formulas autocorrelate to one outcome (win or lose), drastically reducing the data set (Brill estimates to about 4000 data points over 15 years). Accordingly the error bars are much higher than fourth-down / win probability calculators want to admit, an 8% Win Probability Interval on average. Fourth down bots should be a lot less confident.

There were a lot of other great talks, and I filled a notebook with terms to look up and try to understand better: Extreme Gradient Boosted Regression Trees, Markov Chain Monte Carlo, Bayes Additive Regression Trees, Random Forests for Conditional Density Estimation, etc. By the time 2025 rolls around, I'll understand more of this stuff. Probably. Either way, I found both the presentations and the conversations inspiring, and I'm looking forward to learning more.