- John Horton provides simulation evidence to show that you don’t need to cluster your standard errors when you randomize at the individual level (despite what referees might try to tell you). And talking of clustering, David Roodman has new Stata code for the wild bootstrap for dealing with clustered standard errors.
- Rachel Glennerster picks up on my blog post from earlier this week with “so you want to do an RCT with a government: things you should know”
- Pacific Standard profiles the new World Bank unit on behavioral science (GINI) by highlighting five studies that show how behavioral science can help with development.
- On VoxEU, Tessa Bold discusses mystery shopping work looking at the quality of fertilizer and seeds for sale in markets in Africa, the claim being that the low adoption of hybrid seeds that perform well in controlled experiments may be in part due to the low and varied quality of the hybrid seeds and fertilizer actually for sale in markets.
- From NPREd, new research showing how course evaluations are in fact negatively correlated with how much students learn in those courses! (h/t Dina Pomeranz).
- In 538, discussion of how the recent evidence on microfinance has not filtered through to FINCA fundraising or operations – with the following amazing quote: “Rupert Scofield, who runs FINCA, one of the largest and most prominent microlenders, said randomized controlled trials work well in medicine but don’t make sense in evaluating development aid. He said that for years, his response to claims like Duflo’s was to say he didn’t need data when he could see the effect firsthand. “I would just say dismissively to them, ‘I don’t have time for that,’” Scofield said. “It’s perfectly obvious what our impact is … I don’t need proof. As long as the clients are coming to us, that’s all I need to know.”
- Some more conference calls for papers: BREAD at Georgetown, APPAM at LSE on inequality, PACDEV at Stanford
Join the Conversation