Quadratic Coronavirus Epidemic Growth Model seems like the best fit

by Antimonic

it goes up to February 11th……

  • 05/02/2020 23435 cases 489 fatalities

  • 06/02/2020 26885 cases 561 fatalities

  • 07/02/2020 30576 cases 639 fatalities

  • 08/02/2020 34506 cases 721 fatalities

  • 09/02/2020 38675 cases 808 fatalities

  • 10/02/2020 43084 cases 900 fatalities

  • 11/02/2020 47733 cases 997 fatalities

First off, my original motivation was never about making “predictions”, as I explain further below. The fact that a quadratic model is enough to make accurate predictions is what I am putting into question. This should not work!

But alas, we wait another day, and get the new batch of data from WHO:

  • 24554 confirmed cases – that’s within 5% of my prediction
  • 491 death – that’s within 0.4% of my prediction

Bang! It worked again, but it shouldn’t have!

That seems pretty darn close for a quadratic fit of data that should be inherently exponential.

I would certainly not be advocating that one uses this to predict too far out into the future, because at some unpredictablepoint, the (political?) mechanism that is yielding the current quadratic rise will have to change.

Let me remind you that fitting consists of two steps: first, picking a function and [then] explaining your choice.

As a matter of fact, I started off by picking the only function (an exponential) that epidemics are supposed to follow. The explanation is that it was claimed by the WHO that 1 person infects around 2 more – but then I quickly realized that an exponential model does not suitably explain this data at all. This makes the data from this epidemic questionable!

So far a simple quadratic held up remarkably well for the last 2 weeks which defies all epidemic models published to date.

As for using this fit, or any other fit, to predict the death toll before the origin is just garbage. Fits have to be used within the bounds of the data set.


Please also note recent publications on coronavirus spread models:

Joseph T Wu, Kathy Leung, Gabriel M Leung. Nowcasting and forecasting the potential domestic and international spread of the 2019-nCoV outbreak originating in Wuhan, China: a modelling studyThe Lancet, Jan. 31, 2020 DOI: 10.1016/S0140-6736(20)30260-930260-9)

“The researchers estimate that in the early stages of the Wuhan outbreak (from December 1, 2019 to January 25, 2020) each person infected with 2019-nCoV could have infected up to 2-3 other individuals on average, and that the epidemic doubled in size every 6.4 days. During this period, up to 75,815 individuals could have been infected in Wuhan,”www.sciencedaily.com/releases/2020/01/200131114753.htm

That claim squarely contradicts the data being published by the WHO!