One modeler’s thoughts on the UIUC thing. Because while there are some valid criticisms to be had here, I think there’s also some unfair aspects running around #epitwitter .

So here we go - a long thread.

First, yes, it's fun to drag physicists for modeling epi problems. And there is a tendency for some fields to show up and assume your problem is an easy subset of their problems. And to want to mock them

I'll admit to having done so in the past too. @NoahHaber will confirm ;)

But there is, on occasion, just as much armchair modeling from non-modeler epis as there is armchair epi from non-epi modelers.

Y'all make some silly assertions too.

At the risk of subtweeting, some of those silly assertions are published.

The *principles* behind modeling epidemics are approachable from a number of fields. And I know skilled people doing it with backgrounds in ecology, and epi, and physics, and economics and computer science, and...well, the list goes on.

The irony here is that modeling is very much *not* necessarily welcomed in a lot of epi circles. You can see this in the discomfort some modelers who work entirely or primarily on disease systems calling themselves "epidemiologists" or submitting to epi journals.

I had a faculty member ask if what I was doing was really epi when I was in graduate school. And I had a single course (a good one) on modeling that was definitely framed as "What to do when you encounter someone else doing this". The rest I picked up *outside* my epi coursework

There are conferences I go to where I feel like I have to defend what I do less, and feel more at home from a methods standpoint.

A lot of these aren't epi conferences.

Some of the colleagues I most admire? Have never been and likely will never be at SER. Or only went once.

I was once invited to a vector ecology conference to give a keynote about policy-driven modeling, where one of the opening slides was me admitting I don't work on vectors and that "Nurses are basically just fancy mosquitos".

That got a chuckle - and not pushback.

Modeling is, for a lot of epi, something of a fringe topic. It involves calculus, and isn't really doing what a lot of the rest of epi does (except when it sometimes is).

Modeling is, in my mind, a case of convergent evolution with most of what we consider "epi methods"

Enter Stage Left: The evolutionary biologists telling me to stay in my lane.

What that means is, if you need modeling expertise on short notice, you often *have* to reach to someone who doesn't call themselves an epidemiologist.

And that that person may very well be better qualified to *do* epidemic modeling than an equally random draw of the people who do call themselves epidemiologists.

On the "intellectually thrilling" comment - I think that's a fair one. Part of academia is pursuing what engages you intellectually - it's basically the whole reason to do something that is, economically, stupid.

There are things *I* don't find intellectually thrilling, but I would work on in a crisis. But I'd probably miss epidemiology if I had spend several months of my life working on a problem that happens to use the same methods as it.

Like if I got pulled into modeling the spread of disinformation over social media. I've got the methods background to do it, it's an existential threat to the Republic, is worth working on, and I'd probably not find it particularly thrilling or want to do it for years.

The final thing is that I think non-modelers underestimate just how hard it is to create a model, under scrutiny and constraints from the outside, about an utterly unprecedented event.

No, they didn't model students being students and going to parties.

In our models, neither did we. Why?

Because which of you has good estimates for how much mixing a college party is compared to just living in an apartment? A good estimate for the amount of students who will comply with policies that can't be enforced off campus?

We know frats are a problem.

How much of a problem?

I need your answer by Monday. Yes, I know it's a weekend. The decision is going to be made on Tuesday, with or without you.

If we incorporate stuff like that, we are (rightly) accused of conjuring parameters out of thin air and putting our thumb on the scale to get a model back that confirms our priors.

That's a good way to get your model dismissed out of hand.

I am working on a model right now coping with exactly this problem.

My answer is going to be wrong.

Because none of this was studied much before hand. There's a huge amount of extrapolation, guessing, and "Lets try 25% and 50% and see if it matters..."

And honestly, assuming random mixing of the entire student population *is* a pretty good at least "bad" case scenario.

Yes, the UIUC model missed. And UIUC is probably a good example of "This just isn't actually possible to do safely" because they did make a genuine attempt.

But the problem here? I very much doubt was the physicists as much as it was trying to work in an utterly unprecedented space, where there is no data, very large financial, institutional and political constraints, with very, very tight deadlines.