When I saw Cornell's epi model, I had many of the same reactions as @WStevenHolbrook. (I was on the committee that wrote the reopening report, but a different sub-committee from the one that wrestled with health concerns and did the epi modeling.)
1. A lot of faith being put on results from a student "survey." Survey was always intended to be quick-and-dirty snapshot of opinions, not one that meets social science standards. E.g., no attention to response rates, nonrandom missingness, question wording/ordering, etc.
2. Survey was in late spring. Life was different then. US didn't have 130,000 deaths. The unemployment rate hadn't hit 15%. Covid-19 cases in NE concentrated in NYC & NJ. Few appreciated the extent of physical damage to survivors. We hadn't seen rise in cases among 20-35 yos.
3. Survey was of students, not parents or guardians. Many students may want to come back to a college town to be with friends even if classes are on-line, but whoever is footing the bills may have quite different ideas on the subject. (So speaks the parent of a college student.)
4. The either-or nature of the modeling. Assumption seems to be that if campus is closed, Cornell cannot implement testing of local students. But why not? Can't use lever of access to physical resources if campus closed, but can use lever of access to digital resources.
5. Related argument is that we can't require testing for off-campus students in online case because we don't know where they are living. Aside from digital or GPS surveillance (which to Cornell's credit doesn't seem to be on table), must rely on students to self-report location.
6. I like to think that most students are honest and will report location accurately for sake of public health, or at least to avoid penalties if caught. Even if this is naive, it's no more naive than believing students will wear masks and distance for sake of public health.
7. Another other core assumption is that students have an average of 8.3 close contacts, defined by distance and duration of contact. This value is based on estimates of average # of contacts for adults w/ no or minimal physical distancing. Is this reasonable?
8. The inevitable problem of lag. The national covid-19 situation looks much worse now than it did even 2 weeks ago, especially for young people. And reopening just became more complicated w/ mandatory 14-day quarantine of students coming from ever-changing list of states.
9. Good news is that the modeling team put enormous effort into checking robustness of two sets of estimates (residential instruction vs. online only) to the assumptions baked into the models. The "residential safer than online" conclusion is robust *for Cornell.*
10. But, model as presented - especially in press, where devilish details lost - makes assumptions tilted toward the "open up" conclusion.
It will be too tempting to think that the same conclusions hold outside this lovably odd university and centrally isolated town. /fin
It will be too tempting to think that the same conclusions hold outside this lovably odd university and centrally isolated town. /fin
