COVID-19 July 06, 2020

Way back when, when all this was starting, I cautioned about applying national statistics too broadly.  At the time, New York was surging and the bulk of national statistics were coming from the state.  Those numbers almost swamped out data and trends from other places.  Many states also had growing infections and deaths, but New York was a standout.  If you applied the average national numbers you would get the wrong impression of what was actually happening nationally and you'd get the wrong impression of what was actually going on in New York. 

There's a name for the phenomena that I've described above: Simpson's Paradox.  Wikipedia has a nice summary: https://en.wikipedia.org/wiki/Simpson%27s_paradox .    The bottom line is, be very careful how you interpret statistics.  The statistical number crunching can be accurate, but the interpretation can be deceiving.  You really do need to understand the nature of the problem and some amount of detail about the origin of the numbers.  This goes for most everything: COVID trends, police and race, bias in proposal selection, and bias in college admissions. 

I've previously given some possible explanations for the rapid rise in infections and the lack of such a rise in deaths.  Simpon's paradox is one more possible explanation to add to that list.  Deaths may be decreasing or steady on average across the country, but that is not necessarily the case in Texas, Florida and Califonia the bulk of new cases are coming from.  To my eye, there is at least some indication that the death rate in the first two states has been inching up over the last couple of weeks.  It's not dramatic, and I haven't formally run the numbers, but the trend does appear to be there.  You can look at various other websites that track the data.  Texas deaths look to be going up since mid-June. Same in Florida.  California, however, has been pretty flat. 

And now, I bring you the official numbers from yesterday.  It was a typical down day with the oscillation effect amplified somewhat by the holiday and the weekend.  Let's see the numbers later this week.