Best of #econtwitter - Week of October 17, 2021 [2/2]
Welcome readers old and new to this week’s edition of Best of Econtwitter. Thanks to those sharing suggestions, over email or on Twitter @just_economics.
This is part two of two, this week. Part one is here.
Paper summary threads

New @AnnenbergInst working paper posted! edworkingpapers.com/ai21-476
We use a fuzzy RD design with national panel school-level data to ask:
What effect do police in schools (#SROs) have on school environments and student outcomes?
🧵1/n


New working paper: @HendrikDoepper, @NateHMiller, and @JoelStiebale, and I examine whether product-level markups are increasing over time. We estimate IO-style models with flexible consumer preferences, allowing us to look at potential drivers of change in both supply and demand.

“Labor Supply Responses to Learning the Tax and Benefit Schedule” has been in my care for six years. It has found a new home in AER’s Nov issue. The idea was born in 2014: What % of labor supply mistakes can be explained by people lacking information about their incentives? 🧵1/8

The complexity of the language in FOMC statements at the end of Bernanke's tenure required a reading grade level of 20 to 21, equal to a doctoral degree level of education
It has averaged between grades 16-17 thus far under Powell, equivalent to a BA or MA federalreserve.gov/econres/notes/…


Within a precinct, individuals who move to blocks where people vote more themselves vote more, by 0.5% for every 10% block-turnout difference. Findings based on 12 million voters / 1 million movers in Mexico.



Using more than 1 million images scraped from news websites, I apply machine learning tools to detect the faces of politicians and measure the emotions expressed on their faces.
I find:

My paper w/Maryam Kouchaki, “Virtuous Victims”, is out today in @ScienceAdvances!
We show that people often see victims of wrongdoing as morally virtuous—not because of anything that they have done, but simply because others have mistreated them.
science.org/doi/10.1126/sc…
🧵👇


We study the impact of offering repeated HIV testing (every 4 months for 2+ years) to women + partners in rural Malawi. Testing increases dramatically and people perceive their partners as safer. Three key findings:

The in-game performance of professional soccer players who had a recent COVID-19 infection, drops by 6%.
Performance is still 5% lower 6 months later with notable negative spillovers on team performance.
Cool paper by @schmal_w & @jjreade:
dice.hhu.de/fileadmin/reda…


^related: big cross-country data effort on mental health during pandemic

Neat paper by @cqcampos and Caitlin Kearns: LA implemented "Zones of Choice" — areas of school market choice with more options.
Improved competition led to better outcomes, which parents value. Not just selection.
dropbox.com/s/kf8j3g1yeau2…




More: yield curve inversion, video games and well-being, fiscal stimulus something, Airbnb review reciprocity, exchange rate classification, foreign UST demand, opioids bad for municipal finances, dowry law
Public goods

I'm super excited to announce the release of glum (github.com/Quantco/glum/), a Python-first generalized linear modeling library. We focused a lot on correctness, performance and a powerful range of features. glum gets a 3-10x (or more) speed up over glmnet. It's also fun to use!


**ADOPT A PAPER**
AaP matches papers of junior economists to experts in their field. It aims to expand & diversify access to high quality feedback.
Eligibility: APs in research intensive universities in the US, Canada & Europe.
Apply: adoptapaper.org/apply



#EconTwitter hive mind: once upon a time someone posted a link to a website that was an index of a ton of applied micro studies searchable by title and indexed by method (so you could filter by diff-diff, RCT, RD, IV, whatever). Does anyone know where that is?
^link
Interesting discussions

Here's a basic queuing theory fact that may have behavioral/psychological implications.
If customers take on avg 10 minutes to serve and arrive randomly at a rate of 5.8 per hour, then with one bank teller working, expected wait is 5 hours. With two tellers, 3 minutes.
1/

this is super interesting! shows that the STATA default "reg y x, robust" may be problematic and makes a case for using "reg y x, vce(hc3)" instead.

Data Colada @DataColada