· Planet Money has an episode on Busia “the randomized trial capital of the world” (transcript) and the story of how Carol Nekesa got working on these studies and has made this a career.
· The Econ that matters blog interviews Mushfiq Mobarak and finds he gets some of his paper ideas from his Middle School self. Also his advice for grad students “The big piece of advice is that if you’re stuck, try to take a bite-sized piece of a problem. Don’t get stuck just reading for two years without actually trying out something on your own”
· On Let’s Talk Development, Paul Corral, Heath Henderson and Sandra Segovia look at how well machine learning does in creating poverty maps: “We show that the quality of data used for poverty mapping is as important as the method. ML-based poverty maps with the appropriate data rival the traditional, more data demanding (e.g., requiring access to census microdata) maps done at the World Bank since the turn of the 21st century …However, the quality of data is instrumental to obtaining high-quality poverty maps. Even the top ML algorithms cannot do much with poor data – the publicly available geospatial data used does not yield suitable estimates”
· On Vox, Sean Collins and Izzie Ramirez summarize the evidence from many studies of how Black Americans get treated differently from White Americans in many facets of life – and makes an argument that this not only costs money, opportunity, and dignity, but also results in a big time cost.
· In Nature Climate Change, Richard Tol documents how estimates of the social cost of carbon vary widely and have increased over time: “In the past 10 years, estimates of the social cost of carbon have increased from US$9 per tCO2 to US$40 per tCO2 for a high discount rate and from US$122 per tCO2 to US$525 per tCO2 for a low discount rate”…” The social cost of carbon depends on many things, including the total economic impact of climate change; potential tipping points; the scenarios for population, economy and emissions; changes in vulnerability and relative prices with development; the rate of degradation of carbon dioxide from the atmosphere; the rate and level of global warming; the discount rate; the distribution of impacts and inequity aversion; and the uncertainties about impacts and risk aversion. The estimates used here—5,905 estimates in 207 papers, published before 2022—make different assumptions about all these matters.”
· Scott Cunningham writes about how the gap between synthetic controls and difference-in-differences has narrowed with the recent DiD work, when we might want to entertain using negative weights in synthetic controls, and how to think of econometrics as a book of magic spells: “Causal inference is always magic. It’s either good or bad magic, but don’t ever fool yourself into thinking it’s anything other than pure, childlike, awesome magic. You fit the models on the non-magical parts, sure, but when you’re successful, it always is magic about an unknown, untestable reality. Problem is you never know 100% when you were or were not successful in the spell. You’re pretty sure with randomization, because those spells are tight. But randomization is the first spell you learn. It’s the easiest of all spells. Hermione learned that one in the Sorcerer’s Stone not the Prisoner of Azkaban! I’m sorry but it’s true — we have far more confidence in the RCT because of randomization. Outside of randomization, we are working with very difficult spells with the potential to cause great havoc, but also the potential to move mountains.”
· On the JPAL blog, Eitan Paul, Farah Amalia, and Sarah Kopper have a post on setting up international research collaborations for success, including tips on specific areas to set out expectations in advance around.
· Congrats to Tavneet Suri, who is signing off after 6 years editing VoxDev – “Over the last six years, VoxDev has published 698 columns, 122 videos, 195 podcasts and six VoxDevLits, and we have featured about 1350 different contributing authors” – but fear not, a new editorial board is taking up her mantle and plenty more is to come.
Join the Conversation