Ice hockey analytics takes faceoffs into consideration, but current approaches don’t go much further than the idea that winning more faceoffs than losing is good for a team.
“How much do faceoffs help teams to win or lose games?” asks Tad Berkery, a fourth-year computer science major. “Existing analytics haven’t really mapped faceoffs to scoring outcomes or quantified the value of winning a faceoff.”
That’s where his project comes in. Collaborating with a team of students from Hopkins Engineering’s sports analytics research group, Berkery has harnessed the power of AI and machine learning to determine the true value of the faceoff.
The dataset Berkery and the research group used considers more than 5.2 million plays from three hockey analytics sources: Evolving Hockey, Money Puck, and analyst Corey Sznajder.
Berkery and his collaborators—engineering student Max Stevens, Justin Nam, Engr ’23, and Syracuse University graduate Chase Seibold, under the guidance of Whiting School professors Anton Dahbura and Donniell Fishkind—calculated that one faceoff is worth about .015 goals. While that may sound minuscule, with an average of 60 faceoffs per game, the statistic becomes significant.
If a team wins six additional faceoffs per game, it could add 15 goals over the course of an 82-game season, Berkery says. Teams could use the findings for roster construction, he says.
“The project is showcasing how faceoffs are an underpriced market inefficiency that can be exploited: faceoff performance can win games, and you can get people who are good at faceoffs for near league minimum salaries,” he says. “It’s an answer to the question of not only who’s a good player, but also who’s a good player that I can get on a below-market deal.”
The NHL has taken notice: He presented to one team this summer.
Berkery has always had an interest in sports analytics. In high school, he worked with the University of Maryland softball team and sat in on graduate data science classes at George Mason University. He also created elaborate projections in fantasy football leagues—a skill that won him several championships.
Through the engineering analytics group, Berkery worked on a project with the Baltimore Ravens that analyzed how kickoffs can help win games, and another with the Baltimore Orioles on predicting hitters’ performances.
“Sports can be a game of bounces and a game of inches,” Berkery says. “There’s something really special about combining analytics with athletic talent and luck—to elevate the game and potentially turn a would-be loss into a well-deserved win.”