Playing with numbers

Covid-19 cases and deaths in Michigan. Data supplied by frequent commenter Coyoteken

The error implicit in the presentation of this data is that the cases that died rarely died the day they were diagnosed.

There is a lag between exposure and testing positive and an even longer lag between exposure and death.

While it is not possible to estimate the time lag between exposure and death, it may be possible to estimate the time lag between diagnosis and death.

One way to perform this estimate is to calculate the correlation between the number of positive tests and the number of deaths. If you do that starting with March 18 (the date of the first death in Michigan) then you get a correlation of 0.983

If you slide the death data up five days so you are comparing the number of deaths with the number of positive cases recorded five days earlier you get a correlation of 0.996 which is much better correlation than 0.983

Sliding the data six days yields a correlation of 0.997 while sliding it seven days yields a correlation of 0.995

That suggests that the typical time-lag between a positive test and death is approximately six days.

Now the scary part
Lining up the number of deaths with the number of cases diagnosed six days earlier suggests an 10% fatality rate.

Data started March 19 when testing became more generally available. Wild jumps in computed death rate suggests small numbers issues.

Yes, I know that the Number of Cases is garbage data. The actual number is undoubtedly much, much higher than reported because many people are not going to go to the doctor or hospital until the symptoms are severe.

Nevertheless, it does paint a picture of the freight train that is hitting hospitals across the country, starting with the ones in major, metropolitan areas.