Here's a new GAO report on the federal statistical system's need to modernize and adapt to a rapidly changing data landscape--based on a forum GAO held in 2024 with experts and stakeholders.
#econsky
www.gao.gov/products/gao...

"This paper presents new methods for estimating long run treatment effects from short term experiments and observational data."

New paper from Athey, Chetty, Imbens, and Kang:

restud.com/the-surrogat...

#econsky
#REStud

Source: christinecai.github.io/PublicGoods/...

1/
What does the social fabric of an entire country look like?
We built a nation-scale social network of Denmark — 7.2 million people, 1.4 billion ties, 14 years of data.
Here’s what we found 👇
📄 doi.org/10.1038/s415...
#NetworkScience #Sociology

A Lightweight Deep Learning-based Model for Ranking Influential Nodes in Complex Networks
https://arxiv.org/abs/2507.19702
Identifying influential nodes in complex networks is a critical task with a wide range of applications across different domains. However, existing approaches often face t...📈🤖

Quasi-Experimental Shift-Share Research Designs
by Kirill Borusyak & Peter Hull & Xavier Jaravel (2018)

Economists love using linear regression to estimate treatment effects — it turns out that there are perils to this method, but also amazing perks

Come with me in this 🧵 if you want to learn about our now-published paper "Contamination Bias in Linear Regressions!"

1/ (Twitter rerun!)

In order to add some clarity on why the answer is just "use robust SE", here's a simulations that follow Abadie et al.'s paper "When Should You Adjust Standard Errors for Clustering?"

This is pure random assignment of treatment. Cluster SE are way too big, and robust SE are close.

1. The conventional explanation for food deserts—that these places are too poor or too rural to generate enough spending on groceries, or too Black to overcome racist corporate redlining — fail to grapple with a key fact: food deserts didn’t used to exist. My new piece in The Atlantic.

Mapping out tangible ways economists can use language models in their daily workflow, from ideation to publication, from Anton Korinek https://www.nber.org/papers/w33198

P-EBT, which gave grocery vouchers worth about $300 to students who lost access to free meals during COVID, reduced food hardship and moms' mental health problems, from Lauren L. Bauer, Krista J. Ruffini, and Diane Whitmore Schanzenbach https://www.nber.org/papers/w33199

Two-thirds of Illinoisans live in the six-county Chicago region. 41% live in Cook County.

Oh, forgot to include the "policy didn't cause inflation" chart 5/

Consider the simplest DiD setup with panel data. T = 2, controls X(i) don't change over time. D(i) is the "ever treated" indicator, f2(t) the second period time dummy. W(i,t) the time-varying treatment W(i,t) = D(i)*f2(t). The X(i) appear flexibly to allow selection and heterogeneous trends.