COVID-19 May 18, 2020

I know it's tempting to look at individual daily numbers and make inferences about what is happening.  This is especially true if the daily number comports with your beliefs.  As a scientist, it's important to recognize and the short circuit the insidious nature of this confirmation bias.  As an example, the large number of new infections (1801) reported in Texas the day before received a lot of press, and many were quick to jump on that as evidence that there was a cause and effect relationship between that number and Texas's loose quarantine policies.  The number of new cases in Texas yesterday was 785.  That is one of the lowest numbers reported by Texas in the last two weeks.  Will that make the news?  Will those who jumped all over the 1801 number mention it?  The answer is probably no and no. 

It is generally unwise to focus on an individual daily number.  All the numbers must be taken together and understood in the context of the situation.  That situation includes an abysmal testing and reporting system that lacks nationwide standardization.  Daily numbers go up. Daily numbers go down.  I've taken to using a 7-day average because that seems to at least smooth out the weekly cycle that is inherent in the data.  The daily variations tell me that looking at the data with a fidelity shorter than about a week is probably bad data analysis.   If the numbers in Texas are going to respond to their policies, it will show up in the longer-term average.  It's not going to happen overnight, which means it's not going to show up suddenly as a big increase in just one day.  At least it's not likely to turn out that way.

And now, here are the usual plots: