There is broad misunderstanding of how to interpret and best use models. Infectious disease models a la @neil_ferguson are best interpreted as providing a picture of what happens if we don’t respond. Their purpose, in a sense, is to be wrong; 1/n

i.e. trigger responses so worst scenarios are avoided. These rely on understanding of transmission dynamics. The @ihme model, in contrast, was meant to inform hospital planning on a scale that, according to Chris Murray @FiveThirtyEight, does not influence disease dynamics. 2/n

So their goal has been to be precise, and they rely on data that implicitly account for policy responses. An unintended consequence has been that the models were misused (at federal level) and rosy scenarios may have caused behavioral responses that increase transmission. 3/n

resulting in predictions that were off. Economist @tylercowen @MargRev seemed to embrace the objective of being precise (as econ models aim to be), and raised the question of whether the behavioral response to the models could be incorporated into the models themselves 4/n

Theoretically, the answer is yes. But if the model is aimed to inform public policy, then incorporating behavioral feedback will probably not generate the desired outcome. Consider a system of two equations: 5/n

1) an infectious disease system characterized by Rc = Rc(c) where c is contact rates. For clarity here I am referring Ro to be Rc given that it changes with c, but it is otherwise the normal definition; and

2) human behavior system, c=c(Rc). 6/n

(If you are an economist, c is an optimum value derived from a utility fxn where contact is good and disease is bad). Importantly, c is a fxn not of the real prevalence of disease (which individuals don’t know, but rather the model predictions. 7/n

One can solve for an equilibrium contact rate here that accounts for feedback. Graphically, it could look something like this. 8/n

The purple line shows the equilibrium value of Rc as a function of contact rates (its linear because c is just a scalar in Rc). the green curve is the individual (optimal) c as a fxn of Rc. Its curved because at very high values of R0 its impossible to avoid infection, 9/n

C*R* is the equilibrium. If we used THAT contact rate in the disease models, it would converge on reality because the model accounts for response.But that's a bad outcome. It implicitly encourages high contact rates so that the model is right. 10/n

But there is a negative externality to contact because individual contact benefits the individual more than it benefits society (it harms society). The blue curve represents the socially optimal c that account for externalities (11/n).

This can be derived explicitly-its the marginal effect on R0. In other words, we can theoretically incorporate behavioral response into the model so that they are more precise. But that would undermine the real value of the model, which is to get to the right contact rates.12/12