The statistics being reported on Covid-19 keep pouring in—far too much information by my reckoning. Per the nation’s top infectious disease expert, Dr. Anthony Fauci, I focus on positivity rates as a predictor of the ups and down of the coronavirus. However, the calculations for even this one statistic cause a great deal of controversy, especially in times like now with rising cases of Covid-19.
For example, as reported by The Las Vegas Review-Journal last week, positivity rates for the Nevada now vary by an astounding five-fold range depending on the source of the statistics. It doesn’t help that the State went from 7-day to 14-day moving averages, thus dampening down an upsurge.
“We’re trying to get that trend to be as smooth as possible, so that an end user can look at it and really follow that line and understand what’s happening.”
State of Nevada Chief Biostatistician Kyra Morgan, Nevada changed how it measures COVID’s impact. Here’s why., The Las Vegas Review-Journal, 10/22/20
My preference is 7 days over 14 days, but, in any case, I would always like to see the raw data graphed along with the smoothed curves. The Georgia Rural Health Innovation Center provided an enlightening primer on moving averages this summer just as State Covid-19 cases spiked. Notice how the 7-day averaging takes out most of the noise in the data. The 14-day approach goes a bit too far in my opinion—blunting the spike at the end.
I advise that you pay attention to the nuances behind Covid-19 statistics, in particular the moving averages and how they get shifted from time to time.
PS My favorite method for smoothing is exponentially weighted moving averages. See it explained at this NIST Engineering Statistics Handbook post. It is quite easy to generate with a simple spreadsheet. With a smoothing constant of 0.2 (my preference) you get an averaging similar to a moving average of 5 periods, but it is far more responsive to more current results.