Read on Twitter
New paper: "The Causal Effect of Social Distancing on the Spread of SARS-CoV-2"
1
George Wood and I have a new article where we use week to week variation in weather conditions within counties as a natural experiment that led to changes in social distancing in early March.
George Wood and I have a new article where we use week to week variation in weather conditions within counties as a natural experiment that led to changes in social distancing in early March.
1.1
George will join NYU Data Science next fall - his work is described here: https://gwood.me/
George will join NYU Data Science next fall - his work is described here: https://gwood.me/
2
Findings: Every 1 % increase in non-essential visits in early March led to 7-8 % increase in new cases the following week. Results were stronger in high-density counties, close to zero in lower-density counties
Findings: Every 1 % increase in non-essential visits in early March led to 7-8 % increase in new cases the following week. Results were stronger in high-density counties, close to zero in lower-density counties
3
Here's the causal diagram. Changes in weather are used to instrument for changes in distancing within counties.
Here's the causal diagram. Changes in weather are used to instrument for changes in distancing within counties.
3.1
We find that within county changes in precipitation and temperature have a strong relationship with distance traveled outside the home and non-essential visits
We find that within county changes in precipitation and temperature have a strong relationship with distance traveled outside the home and non-essential visits
3.2
Note that results are the same using just precipitation as IV or precipitation and temperature
Note that results are the same using just precipitation as IV or precipitation and temperature
3.3
We controlled for relative humidity in all models to make sure there was no "backdoor" path from weather conditions to new cases.
We controlled for relative humidity in all models to make sure there was no "backdoor" path from weather conditions to new cases.
4
In the second stage we estimate the effect of distancing in a given week on new cases in the next week, controlling for existing cases.
In the second stage we estimate the effect of distancing in a given week on new cases in the next week, controlling for existing cases.
4.1
All estimates use county fixed effects or first differences, meaning we are examining changes in distancing occurring within the same places - no unobserved differences across counties are affecting results
All estimates use county fixed effects or first differences, meaning we are examining changes in distancing occurring within the same places - no unobserved differences across counties are affecting results
5
We find that every 1 percent increase in distance traveled or non-essential visits in early March led to a 7-8 percent increase in new cases the following week
We find that every 1 percent increase in distance traveled or non-essential visits in early March led to a 7-8 percent increase in new cases the following week
5.1
Equally important: results were driven entirely by high-density counties. Effects were close to zero in lower-density counties (note that first stage worked just as well in high and low-density counties)
Equally important: results were driven entirely by high-density counties. Effects were close to zero in lower-density counties (note that first stage worked just as well in high and low-density counties)
5.2
Results are extremely robust. They are the same using just precipitation as the IV, same w both measures of distancing, same excluding the lagged measure of cases.
Results are extremely robust. They are the same using just precipitation as the IV, same w both measures of distancing, same excluding the lagged measure of cases.
5.3
Placebo test also shows no relationship if we use lagged measure of new cases as the outcome.
Placebo test also shows no relationship if we use lagged measure of new cases as the outcome.
6
Yesterday we heard about another paper using the same idea, but w slightly different approach looking across places rather than within https://twitter.com/AjaycencyMatrix/status/1262395929862209537
Yesterday we heard about another paper using the same idea, but w slightly different approach looking across places rather than within https://twitter.com/AjaycencyMatrix/status/1262395929862209537
6.1
It looks like a really good paper, ends up with same basic findings but the differences are nuanced enough that each paper stands on its own
It looks like a really good paper, ends up with same basic findings but the differences are nuanced enough that each paper stands on its own
6.2
If we're thinking in terms of academic status systems and publication races, I'll be honest and acknowledge that it sucks to see another paper so similar.
If we're thinking in terms of academic status systems and publication races, I'll be honest and acknowledge that it sucks to see another paper so similar.
6.3
But if we're thinking about the advancement of science, this is a great outcome. Two groups of researchers w similar design, analyzing the data in different ways, and ending w findings that complement each other
But if we're thinking about the advancement of science, this is a great outcome. Two groups of researchers w similar design, analyzing the data in different ways, and ending w findings that complement each other
6.4
Note: I realize I have the security and privilege of tenure, so it's easier for me to take a different perspective - I still think this alternative perspective is the right way to think about this.
Note: I realize I have the security and privilege of tenure, so it's easier for me to take a different perspective - I still think this alternative perspective is the right way to think about this.
6.5
all that said: Nice work by @AjaycencyMatrix and co-authors!
all that said: Nice work by @AjaycencyMatrix and co-authors!
7
Our paper has been under review for the past week. We originally decided not to publish preprint b/c some journals that we were considering do not allow it
Our paper has been under review for the past week. We originally decided not to publish preprint b/c some journals that we were considering do not allow it
7.1
Now that the other working paper is out there, and given that the outlet where ours is being reviewed does allow pre-prints, we're posting ours as well here: https://www.patricksharkey.net/distancing
Now that the other working paper is out there, and given that the outlet where ours is being reviewed does allow pre-prints, we're posting ours as well here: https://www.patricksharkey.net/distancing
7.2
It will also be posted on socarxiv once their pre-moderation concludes. I think the permanent link will be here (it's dead now, fyi): https://osf.io/preprints/socarxiv/hzj7a/
It will also be posted on socarxiv once their pre-moderation concludes. I think the permanent link will be here (it's dead now, fyi): https://osf.io/preprints/socarxiv/hzj7a/
8
Comments welcome
Comments welcome
Created with @ThinkNested
