Clemson weighted two rotation models for COVID classes and chose the fairer one

For fall 2020, Clemson University needed hybrid class schedules: students would attend partly online and partly in person, with the explicit goal of limiting COVID exposure inside classrooms. The question was how to rotate students through buildings so that in-person time survived without turning each classroom into a mixing chamber.

A Clemson team built and compared two rotational attendance models. The three-cohort model split students into rotating cohorts and cut weekly student interactions to about 57, with 1.6 in-person days per week. The once-a-week model brought every student to campus once weekly and held interactions near 83, with 1.9 in-person days. A do-nothing baseline averaged 84 interactions and 2.6 in-person days.

On the exposure metric alone, the three-cohort model looked better. But its numbers were bought with exclusions: about 1,600 of the 21,000 students who wanted in-person courses would have been forced fully online, and courses meeting twice a week would have to be attended twice inside a three-week rotation — a structural burden on both students and scheduling.

Weighing the interaction figures against those implementation costs, Clemson adopted the once-a-week model for fall 2020 and spring 2021: nearly the same in-person access for everyone, at a modest exposure cost, with a rotation simple enough to actually run.

The fairer model beat the numerically better one

The three-cohort model's headline win — 57 versus 83 weekly interactions — was achieved partly by excluding people: about 1,600 students who wanted in-person classes would have had none.

Twice-weekly courses under the three-cohort plan met twice per three-week rotation, a cadence that multiplies scheduling conflicts across the whole catalog.

The once-a-week model preserved universal in-person access at a small exposure difference, which mattered as much to the administration as the metric did.

Implementation burden was modeled as a first-class criterion, not an afterthought — the plan the university could administer on day one was the plan that got used.

Optimization output is an input to a decision, not the decision itself: the model with the best exposure numbers lost to the one students could actually live with.

Clemson ran the once-a-week attendance plan through fall 2020 and spring 2021. The comparison of the two rotation models — with the baseline figures, the exclusions each design required and the administration's reasoning — was published by the Clemson team in the INFORMS Journal on Applied Analytics' special issue on analytics remedies to COVID-19 (November 2022), making the tradeoff analysis itself part of the record rather than just the winning design.