Where can I find out more details about Coronavirus modelling?
July 4, 2020 6:12 PM   Subscribe

I live in Canada and I work in data analytics and predictive modelling. I read articles about the Coronavirus that refer to estimates from a model used by the Canadian government. Is there anywhere that I can find the "nitty-gritty" details of the model and the assumptions being used? I'm interested in learning more about how the modelling works.
posted by NoneOfTheAbove to Science & Nature (4 answers total) 3 users marked this as a favorite
Best answer: FIveThirtyEight has a good roundup of most of the common models, with brief descriptions of each, and links out to the home pages for each specific model. Those homepages typically contain much more detail, and in many cases there are published (or pre-pub) research papers digging deeper into the model. Most projections also have source available (e.g., here's the source to MIT's model), so if you can read code that might be the best place to look for the nitty gritty.
posted by jacobian at 7:00 PM on July 4, 2020

Best answer: Oh, and related: What Happens Next? by Nicky Case and Marcel Salathé is a lovely intro-level exploration of epidemic modeling. It introduces the Susceptible/Expose/Infectious/Recovered (SEIR) model, which as far as I can tell is at the core of most epidemiological modeling.

I should add, in case it's not clear: I'm a software developer, not an epidemiologist, so don't trust me without doing some research.
posted by jacobian at 7:04 PM on July 4, 2020

A basic intro to pandemic modeling can be found in a Numberfile video here. I wrote a version of this model (with a slight change) in C#. The actual calculation is about 10 lines long. If you (or any other MeFite) is interested, I'll be happy to share the code and some of my observations about it.

More to the point the IMHE projection pages here have been elaborated since the beginning and now show the projected effects of mask usage, easing mandates, etc. This gives a little insight to the "sub-models" they've built to elaborate the generic models to give better results based on local conditions.
posted by SemiSalt at 6:26 AM on July 5, 2020

IMHE has all their data available for download which makes it possible to analyse the question of how good the forecasts have been. I have kept a chart for my home state (CT) which illustrates one way it can done.

This shows the first two weeks (approximately) for each of a sequence of forecasts. The first (orange) was one shown was on April 1, and was too low. The second (grey) was too high, but called the timing of the maximum pretty well. I think that was remarkable since we were still in exponential growth.

You may remember that the published confidence limits were comically wide at that time for projections only a week out.

Note: IMHE revised their projections more often than shown. I only show a few to keep the chart easy to read.
posted by SemiSalt at 8:19 AM on July 5, 2020

« Older how to offload a US car in Canada   |   Cool Canadian sheets Newer »
This thread is closed to new comments.