“Well,” the student remarks, “there are two potential risks of including that many variables in the model. The first is that you’ll probably get some false positives simply by running so many t-tests. The second is that you’ll risk overfitting to the sample—do a great job here, but a poor job on future datasets.”
We are three months into the first semester of the Master of Science in Decision Analysis program at Minerva, and one of my students is starting to persuade me that something is going terribly right. The students appear to be “getting it.” We are examining the efficacy of economic sanctions and the students have just read a seminal work on the subject, alongside one of its sharpest critiques. Their homework was to determine whether quantitative models of the impact of sanctions can capture the full picture, or whether we also need some kind of qualitative measure—a narrative to provide context and tell a story that numbers alone cannot.
A second student joins the conversation: “My concern is not with the number of variables, but their type.”
“These are all economic measures,” she says. “But as we know from Prospect Theory, people are generally risk-seeking in loss scenarios. So when faced with sanctions, they might take non-economic risks, such as political risks. Without some kind of narrative, we could miss that element of the sanctions’ impact.”
Now I know that they are getting it. The first student invoked a statistical concept from another course in the program (Formal Methods for Analysis and Decision Making), and this second student has invoked Prospect Theory, a core concept from a third course (Decision Making in Complex Social Systems). I did not ask either of them to make these connections; they did so spontaneously. And this is just what we want because when the time comes for them to work on their master’s theses, there will be hundreds—even thousands—of fruitful combinations of ideas, and we want the best ones to suggest themselves automatically.
To help all our students prepare for such an open-ended challenge, we do something unusual. Rather than going deeper and deeper into content, we change topics every two or three weeks. For example, in my Research Methods class—the source of the quotes above—we were in our second week of studying economic sanctions. Before that, we did a cross-cultural comparison of professional compensation strategies. Before that, we tried to understand the psychology and demography of climate change denial. And before that, we explored the use of multi-agent computer simulations to explain patterns in crime statistics. It would be natural to ask, “What is the focus of this class? Is it about economics? Business? Psychology? Criminology?” The answer: all of the above and more.
By designing the courses this way, we make it apparent to students that we are trying to give them broadly-applicable tools, rather than make them into experts in a specific domain. I do not yet know which problems they will choose to work on in their theses, but in a way, it is better that way, and it is better that they are not preparing to tackle just one. After all, life is fluid, careers evolve, interests change… and the world changes, too. We want our graduates to be powerful, yet agile, prepared to adapt to whatever they encounter.
Speaking of agility: it is a good thing all of the professors in the program meet every week to discuss our experiences and that we are all immersed in the most collegial teaching environment any of us has experienced. Without that support, I am not entirely sure I would have been able to respond appropriately to my student’s invocations of concepts from the other courses. Even if I had been perplexed, though, everyone in class knows that the purpose of the program is not for the professors to shine in the spotlight, showing off our intellects. Rather, the purpose is to put the students themselves in the spotlight, giving them a chance to conduct precisely the sort of intellectual experiments illustrated above.
Since we do not lecture at Minerva, there is plenty of class time for each student to do this repeatedly. If the experiments do not go well at first, that is okay. Everyone receives plenty of feedback and has the opportunity to try again the next day—or in the next class, or on the next assignment—with minimal risk to their grades. I suppose this intensive setup is a slight shock at first—there is no back row to hide in—but we do not structure classes this way to be cruel: our pedagogy is informed by an enormous body of scientific research demonstrating that this kind of fully active learning is by far the best route to cognitive growth.
A few weeks later, the semester drew to a close. As I write this, the students are resting during their winter break, getting ready for Spring. In keeping with the shifting-topics method, my class’s Spring semester will examine subjects ranging from political freedom to biodiversity, from legal testimony to computer-assisted medical diagnosis. But this time, there is a twist: the students will run the class. They will decide how to approach each topic and which facets they want to examine. They will select the readings, and they will facilitate the class discussion. Again, this is great educational technique: learning through teaching is extremely effective. Plus, it is excellent practice for the broader freedom they will have in their thesis work… and I cannot wait to see what else they teach me.