Read on Twitter
Welcome to Improv Explosion! To set the scene, I am:
Being interviewed by <news org>: BUZZFEED!
About <subject>: STUDY DESIGN!
On <noun>: SUPERSPREADER EVENT!
For a <event>: MOTORCYCLE RALLY!
In <place>: STURGIS, SD!
Headlined by <band>: SMASH MOUTH! https://www.buzzfeednews.com/article/peteraldhous/coronavirus-superspreader-sturgis-motorcycle-rally
Being interviewed by <news org>: BUZZFEED!
About <subject>: STUDY DESIGN!
On <noun>: SUPERSPREADER EVENT!
For a <event>: MOTORCYCLE RALLY!
In <place>: STURGIS, SD!
Headlined by <band>: SMASH MOUTH! https://www.buzzfeednews.com/article/peteraldhous/coronavirus-superspreader-sturgis-motorcycle-rally
This paper is far, far from perfect, but I am more surprised than anyone to discover that ... I don't hate it? The overall argument is weirdly compelling (emphasis on the weird)?
Will share some more detailed thoughts on this tomorrow, need sleep now.
Will share some more detailed thoughts on this tomorrow, need sleep now.
To kick it off, there are a few well thought-out threads on this that are well reasons, and which I largely agree with
https://twitter.com/ashishkjha/status/1303536487259148291
https://twitter.com/RexDouglass/status/1303379252742479872
I find myself in the strange position of being more positive about this paper than others. How to square that?
https://twitter.com/ashishkjha/status/1303536487259148291
https://twitter.com/RexDouglass/status/1303379252742479872
I find myself in the strange position of being more positive about this paper than others. How to square that?
In fact, I'll add to it: I think the synthetic control methods they use are sketchy at best. The stats overall are pretty meh, need huge improvement.
Broadly, the paper underplays uncertainty, and is probably broadly an overestimate. Not great.
So why am I weirdly positive?
Broadly, the paper underplays uncertainty, and is probably broadly an overestimate. Not great.
So why am I weirdly positive?
Set aside the stats and their models (yes seriously, no I have not been kidnapped). Ignore the synthetic control stuff in particular.
Look at the data, and think about this first as an ITS (within Sturgis), then as a CITS (Sturgis vs SD)
Checks all the sniff test boxed for me
Look at the data, and think about this first as an ITS (within Sturgis), then as a CITS (Sturgis vs SD)
Checks all the sniff test boxed for me
Zoom out a little. Realistically, it almost doesn't matter what method they used; those data, warts and all, speak the causal inference story.
I don't think there is a real debate about "did something happen here?" What we're debating now is how big, how certain, etc.
I don't think there is a real debate about "did something happen here?" What we're debating now is how big, how certain, etc.
Maybe I've just been spending so much time in the muck these past few months that my guard is down, but I am surprised to say that I find the overall story, if not the specifics, compelling.
Except the economic $ value part; that's junk.
Except the economic $ value part; that's junk.
So lets pick those nits! It's crucially important that we do so. Tear it apart at the seams!
But also, let's step back and have some perspective. This is honestly pretty ok. The material on which those seams are made is sturdier stuff than most.
But also, let's step back and have some perspective. This is honestly pretty ok. The material on which those seams are made is sturdier stuff than most.
As @TheAuburner pointed out, I underplayed the economic impact bit.
Its actual role in the paper is minimal. It's just a tacked on multiplier from another paper's estimate.
But it's fair to say this was main conclusion, so it being weak is a major issue.
Its actual role in the paper is minimal. It's just a tacked on multiplier from another paper's estimate.
But it's fair to say this was main conclusion, so it being weak is a major issue.
Re-reading this; one of the things that's striking to me is just how differently this is being discussed in the media vs. how and what the claims were made in the paper itself.
Not saying the original paper is blameless here, but they are VERY different.
Not saying the original paper is blameless here, but they are VERY different.
The media-selected and interpreted claims are (very reasonably, I think) likely contributing a lot to why people are reacting so negatively to this paper.
Always a problem here; to what degree do we treat the paper as an independent entity vs how it might be translated?
Always a problem here; to what degree do we treat the paper as an independent entity vs how it might be translated?
The paper makes a LOT of claims. The large majority of those claims are pretty reasonable.
It also makes some very big claims. Those big claims are not so reasonable. Those big, not so reasonable claims are the ones you see plastered pretty much everywhere.
It also makes some very big claims. Those big claims are not so reasonable. Those big, not so reasonable claims are the ones you see plastered pretty much everywhere.
@drjenndowd weighs in, with a different take than the previous threads on this.
https://slate.com/technology/2020/09/sturgis-rally-covid19-explosion-paper.html
Worth noting: that 266,796 figure that's going around is described in the discussion section, but not in the abstract or the results section.
https://slate.com/technology/2020/09/sturgis-rally-covid19-explosion-paper.html
Worth noting: that 266,796 figure that's going around is described in the discussion section, but not in the abstract or the results section.
