(Implicit) Event Study for Utah

An event study is a technique in finance, where you take data before and after an event (and sometimes during), and check whether the data behaved differently around the event.

Johns Hopkins now has a graphical device for doing something like an event study with public policy choices related to COVID-19 for individual states.

******************************************

WARNING: Proceed with care. Doing event studies with trending data is not straightforward. You have been warned. There's a reason I am not writing up something too sophisticated about this (and it's not because I think I understand why this looks the way it does).

******************************************

So here's a time series showing new cases:

Obviously cases are trending up. Part of that is our state isn't one of the ones that's doing OK right now. But another one is that as we test more and more, we find more cases (which implies we should see some arcing up to the right, but it should be fairly smooth ... and it's not).

Anyway, what's cool about this is that the interactive graphic (and you should go to the site to play with it) shows public policy events: towards lockdown are in orange-ish, towards opening are in bluish-green, and neutral ones are in gray.

QUICK UPDATE: I chose April 28, a neutral one, for zero reason ... just showing an example.

The problem is, with trending data, especially when we don't understand the underlying mathematical process (and we're sure it can't trend forever), that You. Can. Not. interpret the orange moves as keeping cases under control, and the green ones as taking the lid off the pot.

You. Can. Not. But everyone does anyway. And they're full of BS. Do note that I'm not denying that this may be the true story, just that I know better than to open my big mouth with this sort of data.

And here's a chart showing new deaths:

Same issues here, although it helps that they're not trending, and don't have an obvious kink like above. Even so, the warning still applies.

Hat tip to MJ.

UPDATE: As always, if growing data isn't logged, beware. In this case, it's a double problem. Really, we ought to start with the series we know is growing — total cases or deaths — then log that, and take differences to get something like what they're trying to get at here. But, by skipping two steps, my guess is that these cool charts are just junk in this form.