I& #39;m late to the party on this, but the IHME model is apparently just trend extrapolation, and relatively crude 70-year-old laurel-resting full prof trend extrapolation at that https://statmodeling.stat.columbia.edu/2020/03/29/the-second-derivative-of-the-time-trend-on-the-log-scale/">https://statmodeling.stat.columbia.edu/2020/03/2...
I predict that the current IHME model will be clearly, catastrophically wrong by mid-May, and similar to the constantly changing Levitt predictions referenced by Gelman, the IHME model will be revised multiple times to match the stubbornly not-declining death rate
I will also explain what I think will be found to be wrong with the IHME model, as it& #39;s not *necessarily* a horrific idea to predict the epidemic by fitting sigmoid curves to the data.
Fitting sigmoids only works if the *reason* for the curve is logistic (self-limiting) growth.
Fitting sigmoids only works if the *reason* for the curve is logistic (self-limiting) growth.
If the disease& #39;s future rate of increase is primarily determined by its current prevalence (relative to everyone it *could* possibly infect) then growth should follow a nice symmetric logistic curve.
This will not be the case if growth is changing due to social dynamics.
This will not be the case if growth is changing due to social dynamics.
Despite the recent barrage of mostly flawed antibody studies, I believe the prevalence of COVID is still relatively low - simply because not enough people have died from it, given the 1-2% IFR we can reasonably expect from countries that have contained their epidemics.
The rate of growth in deaths has, thankfully, stagnated, but this is mostly because of lockdown measures, not because the disease is so prevalent that herd immunity has begun to limit its expansion (as sigmoid fitting assumes).
Increasing rates of immunity can combine with social distancing measures to drive R0 < 1 and lead to an extinguished epidemic well before 100% of people are infected. But when your prevalence rates are in the single digits nationwide, the growth dampening power of immunity is low