Humans Don't Understand Models Because Of Course They Don't

The grim news out of the White House today is that Dr. Fauci and therefore President Trump expect between 100,000 and 240,000 deaths in an unspecified time frame due to COVID-19 (aka., the Wuhan Flu, the Kung Flu, or the Chinese Virus).

This number is based on a model. A model makes assumptions about (assigns values to) relevant variables X₁, X₂, ... , X_n, and plugs those variables into some equations to make a predictions about outcomes, in this case the estimated deaths which will be caused by the Kung Flu.

Simplifying, those equations are the model. (For grisly details, read here.) Using different values for variables X_i, you get different models runs. For each model run, it is those assumptions about values that really count. One model run is particularly important. It is the worst case based on worst case assumptions.

Humans who don't use models to make predictions don't understand any of this because of course they don't. So the default assumption is that if the model says 100,000 (or maybe 240,000) people are gonna die in the worst case, then we must prepare ourselves for the grim reality that 100,000 (or maybe 240,000) people are gonna die.

The worst case model run has now been given the status of an omniscient God. 

We can find some insight into all this in the NPR story Models Of Epidemic Predict Huge U.S. Death Toll; Scientists Hope For Better Outcome (March 31, 2020, yesterday, emphasis added).

The models – in particular, a model developed by Chris Murray, at the University of Washington – assume that the future of the epidemic will play out much like the recent past, Birx said. "If you ask Chris Murray, he would say he's using the information coming out of New York and New Jersey and applying that to potentially other states having the same outcomes."

Got it? Murray assumed in one model run that everywhere in the country becomes just like New York City and New Jersey. That was presumably the worst case model run. That's the 100,000 to 240,000 deaths case.

Birx is hoping that New York and New Jersey turn out to be unusual. Some states, like Washington and California, have managed to avoid the spikes in coronavirus infections that overwhelmed New York City. If other states manage to duplicate that experience, it would change the model's assumptions about the epidemic's trajectory.

So we already know that the worst case is not happening in places like California or Washington state, which was initially hard hit. Or anywhere else outside of New Orleans as far as we know. So, mitigation efforts do in fact work, especially in areas which are not as densely populated as New York City and New Jersey.

Why is the worst case being put forward as the outcome we can expect? What the fuck?

Now, it is perfectly OK and necessary—it is standard procedure—to model the worst case outcome. What is not OK is to publicly air only that worst case outcome. Doing so means that this worst case estimate quickly becomes set in stone, regardless of how realistic that outcome seems to be. And so general perceptions of the threat level rise, and so does the general fear and hysteria.

This has the effect of terrifying the shit out of already terrified people. Portraying the worst case as the most realistic possibility also makes fearful people far less willing to question the wisdom of the extremely destructive policies which have been put into place. This framing also makes Those In Charge look good when what was unrealistic to begin with never comes to pass.

And so the Wuhan Flu hysteria marches on.