I’ve been tracking #COVID19 data in Chile for a while now, and some interesting patterns arose that I thought would be useful to put together in sth more coherent.
Did small-area lockdowns have heterogeneous effects on the daily progression of new cases in Chile? A thread 1/n
Did small-area lockdowns have heterogeneous effects on the daily progression of new cases in Chile? A thread 1/n
What I was most interested in was seeing whether there were differences in the effectiveness of these policies by income level, and potentially analyzing why these differences happened. Spoiler: I do find differences 2/n
During the first two months of the pandemic in Chile (March and April), the government implemented lockdowns at the municipality level, which are small areas even within a city (I’ll focus in Santiago). Here’s a map: 3/n 
Through the end of March and mid-April, the Government implemented lockdowns in different areas in Stgo, where some were high-income and some were lower-income municipalities. I’ll particularly focus on the results of those that entered quarantine at the end of March 4/n
One advantage we have here for identification is that the government was kind of random when applying them. High number of cases (total and daily) were a factor, but did not fully determine quarantine status, so I use Augmented Synthetic Control Method (ASCM) 5/n
With ASCM I build a counterfactual based on other untreated municipalities that follow a similar pattern in number of cases before the quarantine was announced (I use announcement as t=0 to avoid anticipation effects). 6/n
Here are some results of ASCM! For high-income areas number of daily cases starts decreasing after the 12th day* of the quarantine (Q). However, for lower-income areas that entered early Q, we see a different story 7/n
*I use a 12-day mark to assess the effectiveness of Q. because they are not immediately effective. 7b/n
(To avoid potential spillovers of the Q, I also use buffer zones for areas that people might have moved to during Q, and find the same results) 8/n
Why could these differences be attributed to? Here is where I start looking at two factors: 1) compliance with Q and 2) testing information. Lower-income areas might have higher costs of staying at home, and less availability for testing 9/n
Let’s see about compliance. I look at a mobility index by municipality, and again use Augmented Synthetic Control Method to assess the differences between areas that were subject to Q and those that were not, by income level 10/n
Before school closed (for the whole country), differences seem noisy but centered at 0. After schools closed, high-income areas responded more heavily reducing their mobility, and even more in Q. Otoh, by day 12 of Q, lower-income areas showed little reduction in mobility. 11/n
For testing, num. get more difficult because @ministeriosalud doesn’t release these num. at the municipality level. So what I did was: 1) measure correlation between positivity % and private testing %, and 2) estimate time from 1st symptoms to test results by income level 12/n
There’s a correlation of -0.61 b/w positivity % and % of tests done by private centers (opposite is true for public centers). This indicates that private centers are testing more people that end up being (-) (most people w/ priv. health insurance are in high-income areas) 13/n
In terms of delay b/w 1st symptoms and test results, I have to compare epi reports and changes in num. between epi weeks (?). I leverage the fact that they come out every 3 days or so, and find that patients from lower-income areas take longer to get a diagnosis. 14/n
This is an observational study, and it does rely on some assumptions that I describe in more detail in the paper, so that needs to be taken into account. I run some robustness checks and argue how I handle potential threats, but still this is not an RCT. 15/n
More than anything, data is something that is lacking. I’m still waiting on some better data for mobility and testing analysis, and some changes in reports for num. of cases make things difficult. 16/n
But in the end, what I wanted to highlight is that average treatment effects can mask important heterogeneity that is useful for policymakers, especially when designing new measures for fighting something like a pandemic 17/n
This is not to say that lockdowns are bad and should not be adopted in lower-income areas, but probably that other measures need to complement these quarantine policies, especially for those more vulnerable. 18/n
If you want to check out the paper it’s here: magdalenabennett.com/files/sub/mben… , and if you want to check out the daily progression of #COVID19 in Chile, you can go here: maibennett.shinyapps.io/corona_app/ 19/19
Hilo en ESPAÑOL: 20/19
