US LG3 Fit and Prediction
Talking with SB, a faculty member of my department, he mentioned C19 data fit he has done at the county level using the logistic function. He said it was a damn good fit and is matching new daily numbers. I have shown before that C19 deaths in countries during the first wave were much better modeled by a lognormal distribution than a logistic curve. But after the first wave things get complicated and diverge from anything a single lognormal function can fit. For example, the US now has two distinct peaks and we are upslope of a third. But can the US data be fit with three lognormal curves, one each associated with the observed peaks and implied future one? Thanks to the countless hours of work done earlier this year on my python codes, I could answer this question in an afternoon. Indeed, the US data is well modeled by the sum of 3 lognormal functions. The actual fit is done on the total death data, but the fit can be cast to overlay with the daily death data as well. The result not only fits the data like a glove but allows extrapolation beyond the data, I run it out to the end of 2021. The implication is sobering, US daily death count rising to 6000/day by Mar 2021, total US death count passing 1M in late Apr 2021 and when it is all over there could be 1.2M dead in the US. Of course these predictions are without a vaccine, so they must be viewed as the worse case scenario.
