Friday, April 10, 2020

About Models

The President and his Corona Virus Task Force have used models to predict how bad the pandemic may be in the U.S., how many will sicken and how many will die. Governors have used them as well. Many have criticized the models and modelers, ergo we need to understand the process.

In some ways modeling future events is just a way to systematize your guess. Building the model, you make explicit your estimates of future behavior and states that a person making a seat-of-the-pants guess does intuitively.

Are there reasons to believe modeling provides better answers than SWAGs? Yes. Should you assume the modeler was better able to estimate "future behavior and states" than yourself? Not necessarily, although thinking things through in an orderly way should help

The "garbage in, garbage out" principle is very much operative in modeling. A model is no better than the completeness of the list of factors influencing outcomes, the estimates of the relative strength and directionality of such factors, and the degree to, and fashion in which, they interact. 

In modeling something like the corona virus pandemic in the U.S., most minor variables are omitted. Plus, predicting human behavior is, needless to say, a very, very inexact thing to attempt. And different regions will cross-contaminate differently, depending on urbanization, use of public transport, etc.

Will people take social distancing, mask wearing, and hand washing seriously? Will different subgroups of the population do so differently? And if you believe they will, is it something you can admit without destroying your reputation or career? 

A factor that gets built into disease contagion models is the need to be gloomier than necessary to get people to take contagion-reduction steps seriously. As a modeler you know you'll be in worse trouble if you predict a thunder shower and experience a hurricane, than vice versa. 

Did you ever notice that predicted rainstorms tend to arrive a bit later than the TV weather guesser says? That is no accident, if rain comes 2-3 hours late no complaints, if it come 1 hour early, you'll hear about the ruined picnic, the rained out ballgame. Predicting disease behavior is similar in its need to be pessimistic.

I won't be surprised if we end up with fewer deaths, fewer hospitalizations, etc. than predicted. I won't be displeased but also not surprised, as I understand why it needs to be that way.