Read on Twitter
I guess I have to do a response to the response now, because that's how arguing on the internet works. I'll keep it brief, given that I made most of my points in the original thread. https://twitter.com/lymanstoneky/status/1253219512591106051
Let me be blunt in response: this argument is just plain dumb. Just because the infections variable is measured with noise, doesn't mean an analysis employing it is useless. Furthermore, as I discussed before, *deaths are also measured with noise*. https://twitter.com/lymanstoneky/status/1253219743042920449?s=20
For the noise on infections to really be a concern, you would have to argue that it biases the results somehow. I don't see any argument here for how that could occur. They exist, but not in this thread.
Do county and time fixed effects address all potential sources of bias? Obviously not. But they're sure as hell more trustworthy than most analyses that don't include them, esp naive estimates employing finicky arguments about peak death timing worldwide https://twitter.com/lymanstoneky/status/1253220657107628034?s=20
County-to-county spillovers are indeed tricky to model, but they're also likely second order to community transmission. Just because our analysis abstracts away from them does not mean that it is not credible. It means that it's a first pass approximation https://twitter.com/lymanstoneky/status/1253220657107628034?s=20
This argument is especially odious given that I see nothing in @lymanstoneky's original analysis that deals with spillovers at all. So if this was such a threat to credibility, where was that concern when you were asserting confidently that lockdowns don't work?
If you think about it, the logical implication of this argument is that national level analyses are more credible as a rule than subnational analyses. That's just bonkers. As a rule we want more granular data, b/c it includes more relevant variation https://twitter.com/lymanstoneky/status/1253221544945311744?s=20
So you weren't confident enough in your estimates to show us the actual specification or the numbers, but you were sufficiently confident in them to base a broad policy argument on them? Ok. https://twitter.com/lymanstoneky/status/1253221194540744706?s=20
I don't get this argument at all. People who get tested either die or they don't. So infection data directly measures a share of actual morbidity. Is that measurement more correlated with total morbidity than deaths? Almost surely yes, mechanically. https://twitter.com/lymanstoneky/status/1253221959116009472?s=20
This is the most telling tweet of all. In science, if you come across a claim you believe is overblown, the right response is not to "fight fire with fire", but to correct the record to the truth. https://twitter.com/lymanstoneky/status/1253222243192066048?s=20
Given this rhetoric, it's hard to feel like the original article is anything but thinly-veiled punditry.
