In short, here is what appears to have happened:
- Gabaix and Landier make a modelling assumption for purposes of analytic convenience.
- I describe their model and its implications on this blog.
- Wessel quotes part of that description in the Journal.
- Reich reads the Journal and cites me as an authority using the partial quotation.
- As a result, a modelling assumption morphs into an established fact.
This reminds me of Andrew Gelman's description in Of beauty, sex, and power: Statistical challenges in estimating small effects, on a series of papers by Kanazawa: “Big and tall parents have more sons,” “Violent men have more sons,” “Engineers have more sons, nurses have more daughters,” and “Beautiful parents have more daughters”,
... the estimated effect grew during the reporting. As noted above, the 4.7% (and not statistically significant) difference in the data became 8% in Kanazawa’s choice of the largest comparison, which then became 26% when reported as a logistic regression
coefficient (see Gelman, 2007a), and then jumped to 36% for reasons unknown (possibly a typo in a newspaper report).