[gammarator]

The bank of course has much greater incentive to get their siting estimate right—but it’s hard for me to see how the bank-local REAS feedback loop would work in practice. There’s no ability to test a counterfactual, and local branch performance only becomes apparent years after the siting decision, and with many confounders.

In contrast if the professor’s paper a) actually gets noticed and b) the study conclusions are actually sensitive to the details of the distance estimation methodology, there’s a possibility (not guarantee) of someone writing a rebuttal in six months, which is embarrassing. B is important—spherical cows are frequently good enough!

Fundamentally, we shouldn’t be surprised that the methods appropriate for “decide where to spend $20M on a building” and “spend <$100k of researcher time to perform population-level analysis” should differ.