What am I missing? IMHE (University of Washington) Predicts 72,433 total deaths on August 4th 2020

[mike_c]

23. something to consider about model predictions

Up front let me say that I'm not an epidemiologist but I am a scientist who has spent many years modeling nature. Broadly speaking there are two general types of models. Conceptual models depict natural processes and are generally used to test our understanding about them. They are not meant to make accurate predictions about future events. The specific numbers they spit out are rarely accurate, if ever. No one really expects them to be accurate predictors. Rather, they articulate our assumptions and broad understanding of natural processes, such as how populations grow or how diseases propagate within our communities.

The second class of models is statistical. Statistical models are intended to make accurate predictions about future events, however they provide little or no understanding of the biological processes that underlay the data. A basic statistical model says "If the independent variable ranges from x1 to x2, then the dependent variable will likely (we think) range from y1 to y2," and we can know how confident we can be in those predictions. Statistical models are constructed after the fact, i.e. after we acquire sufficient data for analysis. It's also worth noting that statistical models become increasingly shaky when we use them to extrapolate beyond the observed data.

In the present case we still don't know much about this coronavirus' pathogenicity or its epidemiology, so we're trying to make informed decisions based largely on conceptual models of viral transmission and virulence. People need to realize that if such a model predicts X people will die today and Y people will die tomorrow, X+Y coffins is not likely to be the number needed, no matter what the model predicts.