Grinnell assistant professor forecasting models of COVID-19 spread at colleges

August 06, 2020 — As a researcher in the field of network analysis, Nicole Eikmeier, a Grinnell College assistant professor of computer science, mostly conducts research that is theoretical in nature.

Nicole Eikmeier
    Nicole Eikmeier, assistant
    professor of computer science

“I get interested in a problem because problems themselves are interesting,” she says. “Then it’s an effort to find an application and market the research. Obviously studying the COVID-19 virus is completely different. This research is 100 percent relevant. It’s definitely exciting. It also is somewhat intimidating.”

In May, the National Science Foundation awarded Eikmeier; Matthew Junge, assistant professor of mathematics at Bard College; and Felicia Keesing, the David and Rosalie Rose Distinguished Professor of Science, Mathematics, and Computing at Bard College, a Rapid Response Research grant for their project “Quarantined Networks and the Spread of COVID-19.” The professors received $60,000 to develop network models that better capture the geographic and social complexity of the pandemic.

True to its name, the Rapid Response Grant has a much quicker turnaround time than traditional grants. In fact, most of the modeling has already occurred. A paper could be published about the findings in the next couple of weeks in PLOS, an interdisciplinary, open access journal that also has a fast turnaround time.

“Rapid Response Grants are supposed to cover extremely topical, relevant issues where having the research funding immediately would be super beneficial and it wouldn’t make sense to wait,” Eikmeier says. “It’s perfect for this type of research. When we found out about the grant, I was delighted of course both because it’s awesome to have the opportunity to work on this topic, and it’s awesome to get feedback from the NSF that what we were thinking of working on would be useful, relevant, and exciting to them.”

As professors at two liberal arts colleges, Eikmeier, Junge, and Keesing chose to model the forecasts in their backyard, so to speak, by studying how the virus could spread at small liberal arts colleges.

The researchers used agent-based modeling to forecast infection rates. They first built a network structure of places students and faculty members would likely step foot in, such as classrooms, residence halls, libraries, dining halls, and work out facilities. Then the agents were created, which in this case are students and faculty. They assigned the agents a schedule, so they have an idea of how they moved around the network.

This work set up the question every college is grappling with right now. If someone is infected, how does it impact the other people that are in the same spaces at the same time? And what level of intervention is needed to prevent an outbreak?

“I think the results that we have thus far are probably not too surprising,” Eikmeier says.

Graphic: On the y axis, Policy and on the x axis, compliance. In the lower left corner with low compliance and low policy, it is bright red to show a high number of cases. The colors get lighter as you work up and out. The upper right corner is almost all
A graphic created by the researchers shows how the total number of students infected throughout the course of a semester depends on both the policies enacted by administration and the compliance of students.

If there’s no intervention at all – college resumes as normal – the models show just about everyone on campus gets sick. If minimal intervention is done – such as a little bit of testing and some mask wearing – then the cases do go down. However, 15 to 20 percent of students still get sick over the semester. “It’s unlikely any college would be OK with that figure,” Eikmeier says.

Grinnell’s approach to having a low density or very low density campus with frequent use of PPE and social distancing further reduces the chance of an outbreak.   

“I think what we’re seeing with our model is essentially backing up what scientists have been saying all along,” Eikmeier says. “The steps that we need to be taking, and probably taking them all at the same time, is going to be the thing that actually stops this spread.”

The caveat to the modeling is it’s difficult to forecast human behavior, i.e. get the agents to follow the schedule assigned to them in the network.

“You can make all the restrictions you want in the classroom, library, and dining hall, but you can’t control what students are going do on their own time,” Eikmeier says. “We’ve tried to look at a few different scenarios about what level the students will take the recommendation to heart. We assume that if students get together to hangout or socialize, they probably are not going to be wearing their masks or sitting far away from each other.”

If the researchers model closing the campus fitness center, for example, they have to assume what students would be doing instead of working out.

“Do we do we assume that they just stay in their rooms? Because if that’s the case, then that sort of intervention works pretty well,” Eikmeier says. “But if we assume that instead they go out and visit with friends, then that becomes riskier than letting them go to the gym in the first place.”

The research has not yet been shared with Grinnell College’s administration, but the general finding of “reducing the density, reduces the spread of infection” jives well with decisions made to have a low or very low destiny campus this fall.  

Eikmeier and her collaborators want to caution against treating the models as strict recommendations. She says they have talked a lot about the reaction people could have to the research, especially since the topic seems to be endlessly deliberated in politics and social media.

“Would people take it as a recommendation, and would we feel comfortable if that’s how they read it,” she says about those internal discussions. “There’s definitely an extra layer of responsibility and pressure. Ultimately, what we’re doing is modeling. So we can predict what might happen based on our model, but whenever you model something, you inherently have to make decisions that might end up not quite reflecting reality. So nothing is for certain.”

Eikmeier earned a Ph.D. in mathematics from Purdue in 2019, focusing in the fields of network analysis and random graph models. At Grinnell, she taught Functional Problem Solving and Analysis of Algorithms during her first year on the computer science faculty.

“The students have been wonderful to work with,” she says. “The academic support assistance, the Center for Teaching, Learning, and Assessment (CTLA), and Susan Ferrari and Laura Nelson Lof in the grants office have all been amazing. There are so many people eager to help me succeed. I would not have gotten this grant if it weren’t for Susan and Laura’s help. It’s a great opportunity despite the horrible mess of a situation.”

—by Jeremy Shapiro

For your information:

For the latest updates on fall semester plans and a range of pandemic-related information, view the College’s Meeting the Challenge webpages.

To read more alumni news, check out our news archive.