Morgan Kelly writes:
Back in 2013 you had a column in Chance magazine on the Ashraf-Galor “Out of Africa” paper which claims that genetic diversity determines modern income. That paper is part of a much large literature in economics on Persistence or “Deep Origins” that shows how medieval pogroms prefigure Nazi support, adoption of the plough determines women’s rights etc.
However, most papers in that literature combine unusually high t statistics with extreme spatial autocorrelation of residuals and I wanted to see if these things were connected. The basic idea is in the picture below: regress spatial noise series on each other and you get results that look a lot like persistence:
I [Kelly] go on to examine 27 persistence papers in Top Four economics journals and find that, in most cases, the big persistence variable has lower explanatory power than spatial noise but can, at the same time, strongly predict spatial noise.
For more discussion on that “Out of Africa” paper, see the comment threads here (from 2013) and here (from 2018).
Also some general discussion here of the statistical issue of correlated errors in this and similar examples.
This paper seems a bit overblown. Using a spatial HAC estimator with a large enough bandwidth should solve (most) key concerns. See the recent paper on “Inference and Arbitrary Clustering” (https://www.econstor.eu/bitstream/10419/207409/1/dp12584.pdf). The inference revolution in economics has been moving fast over the past few years with people like Alwyn Young and others making important contributions as outsiders/non-practitioners. The paper in this post raises an important point about spatial correlation that is by now well known. However, its tests of unconditional correlations between the persistence variable (X) and noise seem unfair given that all of the 27 papers’ preferred specifications are conditional on observables (ie other spatial X), which, once partialled out from X, break some (though of course not necessarily all) of the conjectured problematic spatial spuriousness.