Thread by @joel_c_miller

Dear everyone (especially people reading
@tylercowen's blog) and discovering the fact that epidemics can be modeled as spreading on networks.

We know about this fact. We also know it affects some of the predictions.

Before you dive into it, why don't you work through the course I taught in January?

scholar.harvard.edu/joelmiller/tea…

Or you can read through the notes of another course I've taught with Tom Hladish for over a decade:

tjhladish.github.io/sismid/

Or if you're ambitious, you can buy our book: springer.com/gp/book/978331…

Focus on chapters 5 and 6.

If you're hoping to astound us with your network-based simulations, you can probably save yourself a lot of effort if you use the software package I developed.

epidemicsonnetworks.readthedocs.io/en/latest/

Otherwise, so far all of the results I've seen people talking about are contained somewhere in my publication list:

scholar.google.com/citations?user…

You should be aware that most of the agent based models have some form of implicit network structure within them. There are a lot of these out there, which have been developed and refined for more than 10 years, and tested against many epidemics.

The fact that heterogeneity can have dramatic impacts on R_0 is known, and that removing the highest degree individuals (even if by infection) reduces R_eff is also known.

I believe the result dates back to 1980, but probably was known before then.

sciencedirect.com/science/articl…

There are some problems with trying to apply these assumptions to the current epidemic.

One thing is that the heterogeneities in numbers of contacts are actually only one source of heterogeneity. There are other issues that are equally important.

If we act as if all heterogeneity is explained by contact variations, we'll get a lot of stuff wrong.

Particularly for respiratory diseases, I think there is much less variation in contact rates than there is for sexually transmitted diseases.

When you look at a seemingly simplistic model and say "but they are ignoring heterogeneity", you should realize that often that's okay. I'm not including heterogeneity in *most* of the modeling I'm doing right now, even though heterogeneity is what my career is based on.

The reason is that often the goal of our model is to explain some effect and see whether a particular policy is going to be important.

For that we usually want to use a simple model, and the intuition gained from decades of research to guide us for whether we need heterogeneity.

So please - before you start announcing that epidemiologists haven't thought about this or that aspect of network structure:

Read some of the review articles or at least the table of contents of some of the books.

It's possible that it's only new to armchair epidemiologists.

[and I'm still waiting for the armchair epidemiologists to discover clustering and degree-degree correlations]

Also some credit to give out - I said "our book", but didn't mention the coauthors: @istvanzkiss and Peter Simon.

The course I've taught with Tom Hladish will be taught again this year, with @bansallab joining. biostat.washington.edu/suminst/sismid