By MARTY LEVINE
When Lucas Mentch arrived as a statistics faculty member in 2015, he found the need for a new class on statistical learning and data science.
“In statistics in general,” Mentch says, “as data gets bigger and bigger, we have these more complicated functions and algorithms to analyze them.”
They present more difficult tasks for students and blur the line from statistics to data science to computer science. The whole field is undergoing a “stubborn transition,” he says, and so the new class was designed to “deal with a lot of modern problems that students are going to face when they go into industry.”
How could he make the transition and challenge the students — statistics majors and other majors alike?
“As a field, we do a really good job of teaching students specific techniques that work in specific situations,” he says. His new class would be designed to take the big-picture view and explore larger side questions: Which statistical models should I use for analysis? How certain can I be of answers? What do I do when two models are equally good but give contradictory answers?
The aim of the new class is to broaden students’ exploration of new techniques in the field, not just to teach them the tried and true.
“Statistics, in general, at least right now, is a very marketable skill,” he notes, “and it’s tempting for students to go in and say, ‘All right, I know how to make this model and that model and now I can go and get a job.’ I like to be able to take what might seem like advanced ideas and take them back to basic science — how we measure things and how we can be sure of what we are saying, taking a very complicated idea on the surface and breaking it down into simple components and being able to look at the motivation of each one of these simple components, seeing how these ideas build up.
“I want students to see why things work and when they break,” he adds. “I will often include homework problems that are a little outside the norm,” to push students to work until they run up against a puzzle they may not yet be able to answer. The goal is to get them to talk about these puzzles in class.
One potential stumbling block for Mentch is that statistical learning and data science is a lecture class of about 70 students. “The larger the class, the less likely it is that anyone is going to ask questions — or answer questions,” Mentch realized.
Last semester he implemented the Socratic method, calling on students at random: “It’s actually worked out surprisingly well. I don’t penalize students for not having the answer. It forces students to pay attention to what is being discussed and gives feedback to me on what is being understood and what is not being understood.
“It is rewarding to me to see them arrive at the right answer,” he says. “One of the things I have as a benchmark of whether the lecture is going well is if students ask about things I am about to talk about.”
Mentch says he’s been surprised to see that it’s rarely the students who have the strongest background in statistics and the highest ability in mathematics who do best in this class. Those who do well, he notes, are “students who come from a place of intuition. They see the need for this.”
“A lot of statistics students come in and they enjoy the class well enough,” he says. But he enjoys working most with student who face a new problem and want to solve it and discover the bigger picture in his class — often the non-statistics majors.
If there’s a secret to his early teaching success, he says, it’s understanding that “the most effective classes are the ones where you can stress why something is important and actually get students to believe it.”
Marty Levine is a staff writer for the University Times. Reach him at firstname.lastname@example.org or 412-758-4859.
Have a story idea or news to share? Share it with the University Times.