Feel like there's a bit too much talk in the media about how much spread of COVID is growing or shrinking (R) and not enough about *levels*. Both are important. Sustaining a plateau (R=~1) can be a disaster if the level of cases is high. Or a pretty decent outcome if it's low.
— Nate Silver (@NateSilver538) May 23, 2020
Nate Silver has a circular definition of "news."
What causes an indicator to move? News. How do we know it was News? Because an indicator moved.
This issue had seemed to be confined to his political analysis.
But now he's doing the same thing on the Coronavirus. There was one example a few weeks ago where he stepped right up to the edge of the Regression to the Mean fallacy, to explain a group of states outcomes that were "stuck in the middle," with an associated epidemiological concept of "partial herd immunity." In other words, states where not much "seemed" to be happening -- no News.
Now here is again talking about a "plateau" which can arise if R (the reproduction rate) is approximately equal to 1.
Here's the problem. If R=0.99, infections decline. And R=1.01, infections grow. And these are exponential processes. There is no plateau.
So why does he do this? Because if R is approximately equal to 1, according to the "media," there's no clear "news" to explain the variation in levels, so he cobbles together a story about levels and changes -- which is the road back to the Regression to the Mean fallacy. Hair of the average dog that bit him, many times.
UPDATE 25 MAY: He keeps doing it. This thread is textbook example of Regression to the Mean fallacy. He's selecting states based on high deaths, and claiming that they will head to a "plateau" based on some theory about reactions to R. And if it doesn't happen -- he'll say that there was "News!"
No comments:
Post a Comment