Many junior high students like to watch sports, but few use statistics to accurately predict the NBA's most valuable player. Zachary Knowlton, who is graduating this month with a master's degree in statistics, is the exception.
Knowlton has always been interested in math, sports and sports statistics. He became hooked on sports statistics in 2004, when he and his dad created a formula that correctly pegged Kevin Garnett, a forward playing for the Minnesota Timberwolves at the time, as the league's most valuable player that year.
“That was my science fair project, and I correctly predicted the MVP that year,” Knowlton said. “Ever since then I’ve loved sports and stats.”
That love led him to BYU’s statistics program, from which he’ll graduate with a master’s degree this month. A key part of Knowlton's graduate work has been focusing on predicting player performance for the BYU Football team.
Knowlton based his graduate work on a 2002 paper by professional statisticians Chris White and Scott Berry. The paper ranked players based on expected points per play according to field position, downs and yards to go. But while this paper's NFL rankings were based only on raw points, Knowlton and his colleagues added coaches’ rankings to enhance the model.
“We connected player grades... in an innovative way that we’re pretty sure no one else does right now,” Knowlton said. “We think it’s pretty cool, and hopefully it helps the team on a week-by-week basis, if not at least a season basis.”
Once they had developed the model, Knowlton used his enterprising nature to get in contact with the BYU Football staff in an unexpected way. He posted a tweet to head coach at the time, Bronco Mendenhall, offering help from the statistics department.
“The director of football operations at the time saw that,” Knowlton said. “He tweeted me back, got my email and we set up a meeting.”
Knowlton’s team met with Mendenhall and they hammered out the details of the project. After doing some basic summary statistics for the team in 2014, Knowlton started collaborating with football staff on a weekly basis during the 2015 football season.
“Every week they sent us these play-by-play grades,” said Knowlton. “The coaches went through and graded players based on how they did each play – kind of a plus-minus grade. They either did their job or didn’t do their job.”
Knowlton and his fellow researchers combined the coaches’ grades with expected points per play, determined by a statistical model they built.
“The expected points [include] every play, based on down, distance and field position,” Knowlton said. “So we can rate those players by connecting the coaches’ grades with expected points.”
With this information, the coaching staff had a data-based way of rating their players.
“A coach will have a coach’s eye,” Knowlton said. “They’ve been playing football forever; they’ll know who played well or not that game. But now they can actually attach an added points value to that player and say, ‘Oh, they actually did add points to this game,’ whether or not they scored a touchdown.”
This analysis should guide weekly coaching decisions. Knowlton said it could also define position importance and therefore affect recruiting strategies.
Knowlton wrote a paper on the project with statistics professor Gilbert Fellingham. The paper won first place when Knowlton presented his research at the 2015 Joint Statistical Meetings in last August.
“It was cool to be able to go up there and know that we’re actually doing something that people see as valid and as being effective,” said Knowlton.
While he doesn’t plan to pursue sports statistics as a career, Knowlton will use the skills he’s learned through his football project to be an effective statistician—maybe even the MVP—wherever he works.
