Carnegie Mellon at center of hockey’s statistical revolution
It’s not every day that a graduate with a statistics degree becomes involved in sports. That, however, is not the case for visiting assistant professor Andrew C. Thomas, whose many interests include stochastic methods for modeling relational data and development of computational methods for hierarchical models. Hierarchical models organize the lowest-level units in a data set into successively higher-level units. As a hobby, he applies his research on stochastic modeling to sports, and most recently, hockey.
Thomas hosted the 2014 Pittsburgh Hockey Analytics Workshop on Nov. 8, where various sports writers and academia members involved with the subject presented their research findings on competitive analytics. Thomas started the event with a few opening remarks that emphasized the importance of interpreting data in terms of player potential and development, which can give coaches clues into how to improve players’ skill sets and game tactics in the long run.
For this workshop, Thomas was accompanied by two of his research assistants: statistics Ph.D candidate Sam Ventura (who co-hosted the event) and junior math and economics double major Benjamin Zhang. Each presented his own competitive analytics project. Ventura presented on zone transition time, which is the time it takes to move the puck from the offensive zone to the defensive zone, and vice versa.
Ventura’s presentation postulated that the time it takes for the defending team to remove the puck from their zone, as well as the time they spend keeping the puck in the offensive zone, can all be estimated from a real-time scoring system. This is a technique that makes calculations in the moment.
On the other hand, Zhang presented his findings on expected shot probability outcome, or ESPO, by blending chess methods with scoring rates. The ESPO is designed to track a team’s true merit by predicting game outcomes, shot ratios, and game ratios. A change in team skill causes the ESPO to change, useful in predicting real-time outcome at any point in the game.
The report can be expanded further to include factors such as home ice and score effects.
These findings by Zhang and Ventura have the potential to bring significant insight into how methods of comparing teams and players — as well as probabilistic analyses of live, in-progress hockey games — can be improved.
Thomas suggested possibilities for their findings to improve technologies such as SportVU, which the NBA currently uses. This technology collects data via cameras that are placed all around the court. These cameras can capture any player’s actions at any time, even more subtle ones such as muscle movements before taking a shot, down to the millisecond. This accuracy, Thomas said, allows commentators and sports analysts to predict when a shot will be made, as well as the quality of the shot, or when a move will either cause harm or improve the team’s progress during the games.
Thomas predicts that analyzing numbers derived from such instances can assess a team’s current potential and predict the team’s progress in future games.
“I’m encouraged that there’s a growing community of fans who are learning how we can ask these questions with the new kinds of data we have available,” Thomas stated during an interview.
Indeed, there are fans who keep the questions coming including Thomas himself, who expects his next big draw to be in basketball after his term at Carnegie Mellon ends and he takes up role as a visiting professor at the University of Florida.
It looks like the Florida Gators can look forward to a new team member who can greatly impact how they play.