Building A Winning Model
n 1971, the seeds of a data revolution in sport were planted when an enterprising group of baseball enthusiasts with access to early desktop computers (and the occasional mainframe) founded the Society for American Baseball Research (SABR).
Soon, baseball analysts like Bill James were mining hidden statistical gems and surfacing new insights about the game. Sabermetrics, as the approach was dubbed, became indispensable after 2002, the year General Manager Billy Beane used sabermetrics to build a low-budget Oakland Athletics squad that could compete with big-money franchises. That team, whose story was told most famously in the Michael Lewis book and subsequent Hollywood hit Moneyball, won 103 games that season, including an astonishing 20 in a row, all with a payroll smaller than the cost of the feature film that recounted its exploits.
Baseball has always abounded in statistics. Newspapers were featuring rudimentary box scores by the 1870s; data on runs, hits, batting averages, and earned runs has piled up for more than a century and a half. But it was the advent of the modern, accessible computer that gave baseball boffins the means to exploit the information fully.
Brooke Coneeny ’23 is a player-development hitting analyst with the Reds. She double-majored in mathematics and computer science, and captained the varsity softball squad as a senior, leading the Garnet to their first-ever Centennial Conference softball title and into the finals of the NCAA Regional.
Cole Hebble ’21, who majored in economics and history, recently worked as an operational analyst with the Phillies. He played varsity baseball at Swarthmore, then used his extended eligibility from the COVID shutdowns to continue playing as a graduate student at the University of Massachusetts and Duke University, where he earned master’s degrees in sport management and business administration, respectively.
Isaac Kleisle-Murphy ’20, a catcher in his youth, made the switch to lacrosse and played goalkeeper for Swarthmore’s varsity team. He majored in mathematics and was hired by the Phillies for their analytical crew as a junior. He went on to earn a master’s degree in statistics at Stanford University and is now a lead quantitative analyst.
“When I was young, I always went to bed listening to baseball,” he says. “And in high school, I read about it with an analytical bent and loved the statistical richness of the game.” For the Phillies, he builds statistical models and forecasts to predict baseball outcomes, review player performance, and analyze game strategies.
If the term “baseball outcomes” seems a bit vague, it’s because it can mean almost anything you can examine in the game, from individual pitches to season-level forecasts.
“It runs the gamut. It could be whether a ball is going to be fielded, or the chances a pitch will induce a swing and a miss, literally any baseball outcome you can imagine,” says Kleisle-Murphy.
“You’re really trying to identify what metrics and data points are most predictive, and which are noisier,” he explains. “I fall back on a lot of what I learned at Swarthmore, putting things into a predictive modeling pipeline and building a good generative model for the problem we’re trying to solve.”
While Kleisle-Murphy works frequently on in-game strategy, preparing and working with his colleagues, by game time he is mostly an observer.
“During the game, I want to lock in and think about whatever problems I’m working on,” he says. “We’re not just on our computers. We’re watching the games to learn about baseball. There’s no crystal ball, and there will always be blind spots and edge cases.”
“A huge part of our role is to take in all of the information from scouts, player development, and research, and synthesize that,” evaluating all kinds of players throughout the organization and the league, says Hebble. “We pull it all together and try to figure out, ‘Where does this player sit? How will he affect our roster?’”
Collaboration across the organization is something that sets the Phillies apart, says Kleisle-Murphy.
“The scouting department relies on their eyes, and we’re on our computers more, but we’re all rowing in the same direction,” he says. “We’re trying to flag things to supplement their evaluations, and they are identifying things that we need to pull into our models to see areas we’re missing.”
As a hitting analyst with the Reds, Coneeny works primarily with minor league players. While most of her communication is through coordinators and coaches, she does visit the teams periodically, and talks with coaches and players directly.
“When I’m in the dugout, I feel more like I’m part of the team than when I’m working in the office, which is fun,” she says.
“Some players will ask me questions, and coaches and coordinators may invite me to smaller meetings to give data feedback.” Having her own experience of batting at a high level helps Coneeny connect with players, and translate what the data is saying effectively.
“You don’t want to overwhelm them. They’re already feeling a lot of pressure, and you don’t want them to get too much in their head,” she says. “You might bring some simple visuals to a player and say, you know, you’re already doing well, here’s one thing we think you can improve.”
Coneeny has noticed a difference in the way younger and slightly older players work with data-driven assistance.
“A player who is 26 and did not grow up with all this technology might be a little more hesitant about it. Since I work with the minor league guys, the majority are younger, which has made things easier for me than if I’d started this job five years ago,” she says. “Baseball technology has grown so much. We now have 19-year-old players who were drafted out of high school, and have been using this technology since they were in sixth grade.”