re: Wash Your Hands Thread

quote:

---

Problem we have is that lack of testing / quality statistics makes it very challenging to have accurate models/estimates

---

Even with relatively clean data, doing any kind of regression model to map a logistics curve to that data and extrapolate it to find, say, the eventual number of cases or even just when you expect the apex on a curve showing number of new cases with respect to time; is going to have huge confidence intervals due to the exponential nature of it. It's just extremely sensitive to initial conditions, so small errors in data lead to large errors down the pipe.

3Blue1Brown has recently released some videos on the matter I find interesting, where he goes into a high level breakdown on some simulations he set up on the matter. Specifically, he attempts to show how variables such as adherence to social distancing and the time scales that this adherence takes place, the virulence of COVID-19, and the general dynamics of human travel impact a series of simulations that are run. As expected, the simulation based models share the same sensitivities to initial conditions (as well as revealing an immensely nonlinear and quite sensitive relationship between the simulation parameters and the results) as the analytical models and reveal large confidence intervals.

Just an immensely difficult phenomena to model accurately based on early data even with an extreme amount of simplifying assumptions.