Statistics PSA: We're about to enter a depressing phase where in various communities we get a one-day "ooh, the rate went down, hooray" followed by ":-( the rate went up again". There's day-to-day random variation (called Poisson noise) that can give you false optimism.
The size of this variation is roughly equal to the square root of the number of cases. So if you have 900 cases one day and 870 the next day, that's a "1 sigma fluctuation" and not significant.
That's even without taking into account other causes of variation that could cause a spurious drop, like an occasional snafu causing delays in reporting results. So you probably also want to average (smooth) over several days at least
Sorry to be the bearer of bad mathematical news, but I want you to breathe a sigh of relief at the *right* time and not have your hopes dashed. As @KennethRSloan points out to me, such dips *are* evidence that the growth rate has slowed below exponential, at least.
But if you look at the data for Italy, for example, the top of the curve is very flat and lasts a long time before there is a real drop.
This is one of those things like political polls, where it always annoys me that the media say "X is 1 point ahead of Y, he is winning!' when the margin of error on the polls is 3 points or something and the correct conclusion is "they are about even and we just don't know".
I'll finish this thread with my regular mantra: science isn't about knowing numbers precisely, it's about knowing how big the uncertainty is.
You can follow @planet4589.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: