This is an update as of April 7th, 2020.
We are analyzing the existing available data on daily deaths caused by the SARS-CoV2 virus and use that in conjunction with certain simple models to predict the evolution of the disease in certain geographical areas. You can read the full introduction here.
The generalized logistic fit is optimized today. The model has 4 parameters and it is skewed. Because of the skewness, the parameter a is not the timing of the median anymore, but rather, the median is:
The parameter a represents the timing, it is not the median anymore.
The parameter c still represents the size of the outbreak.
Parameter α represents the skewness. In general, for the SIR model the outbreak is right-skewed, meaning that α > 1, but there are some instances where α < 1. In those cases, either the outbreak is before the peak, or the fit is not that great yet and the points don’t align well with the hypothesis.
Because the numbers span many orders of magnitude, the more recent points have an outsized influence on the fit, forcing some parameters towards un-natural values. We will change the weights of the fit tomorrow, to give a slightly higher importance to the points early in the outbreak, in order to capture better the shape of the disease progression.
On another note, we notice that more than half of the considered places are already passed the median point (median < 42) and that is very good news as it shows that the worst is behind. One caveat is that the regions that were selected, were chosen based on having an early outbreak, so we might have many regions that were not chosen yet and are still not passed the median.
California is about to pass Washington in terms of number of deaths.
In Washington the more recent uptick in deaths from yesterday was mainly in Snohomish county.
Error Function Fit
Generalized Logistic Fit
Your feedback is important to us. Please send it to firstname.lastname@example.org