Weekly links May 13: Nigerian poverty trends, ML skepticism, $2.15 is the new $1, and more…
This page in:
· We’ve featured recent discussions about poverty trends in India, but there has been much less discussion about trends in Nigeria, which is projected to have overtaken India as the country with the largest number of extreme poor in the World. On Let’s Talk Development, Jonathan Lain, Marta Schoch and Tara Vishwanath summarize their work trying to measure trends in poverty in Nigeria over the 2009 to 2019 period. The usual issues of potential changes in survey methodology and lack of regular consumption measurement make it hard to measure trends, and they use survey-to-survey imputation methods and supplementary datasets to investigate poverty trends – they find that “poverty reduction stalled in the decade prior to COVID-19 . Although poverty reduction progressed slowly in the first half of the 2010s, it reversed after the 2016 recession, sparked by the collapse in global oil prices”.
· Also on poverty measurement, on the Voices blog, the old dollar-a-day poverty line is now up to $2.15 a day in 2017 PPP terms – this keeps the real value of the international poverty line constant, it is just reflects price changes (dollar a day was originally based on 1985 prices).
· The Innovation Growth Lab (IGL)’s new Evidence Bites provide an attractive and accessible way of summarizing different evidence on entrepreneurship programs through answering bite-sized questions for policymakers based on the existing evidence. E.g. Should I offer subsidized business consulting? Should I go for group or one-on-one consulting? And a list of lessons on what to avoid when designing these programs.
· In part IV of the MRU series of interviews between Isaiah Andrews and Josh Angrist and Guido Imbens, they discuss the use of machine learning in economics. Josh is a lot more skeptical of the use cases than Guido, pointing out the problems that can arise when using them for IV; the limited economics behind some of the treatment heterogeneity regressions; and the unlikeliness that policies are going to target to the highly non-linear combinations of characteristics identified by random forests. Guido is more positive on the potential upsides.
· Scott Cunningham interviews Larry Katz. There is an interesting discussion about the three different approaches Larry sees to understanding what causes trends in wages and employment over time: a supply and demand-based approach that he has used to look at long-run trends, such as the race between education and technology; an institutions-based approach that tries to look for natural experiments to measure impacts of one particular institution like the minimum wage or unions; and a structural approach that tries to do both – and the pros and cons of each (and why it is hard to do the structural well). He also talks about his work as editor at the QJE, and how they went about establishing a norm for quick turnarounds, what types of topics they emphasized, and the importance of the editor offering an R&R as a clear contract. He also makes a useful point about short vs long papers, saying that the idea that economics has too many long papers relative to some other social sciences is a bit misleading, since sometimes it is better to have all the results self-contained in one paper with a coherent story, than to have them split across multiple short papers as is the case in some other sciences.
· On Let’s Talk Development, Yannick Markhof, Philip Wollburg and Giulia Ponzini discuss issues with survey measures of crop losses due to disasters.
· On the World Bank’s data blog, Daniel Mahler, Umar Serajuddin and Hiroko Maeda use simulations and the World Development Indicators to provide some guidance on the question of when is there enough data to provide a global statistic, even when data from some countries is missing.
Join the Conversation