Good point! We already address this in the Appendix (4.4.1): 1) VA gravestones offer both religious and non-religious symbols (-> no mechanical effect), and more importantly, 2) the effect holds when restricting the sample to non-military graveyards. So we’re less concerned that this drives results.

#QJE Feb 2025, #6, “LinkedOut? A Field Experiment on Discrimination in Job Network Formation,” by Evsyukova (@yuliaevsyukova.bsky.social), Rusche (@felixrusche.bsky.social), and Mill (@econmill.bsky.social): doi.org/10.1093/qje/...

Thank you Sascha!

Thanks Max!

Thanks Henning!

Thanks Igor=)

As we were unsure whether we will ever solve those issues, we treated the project always as a high-risk, fun side project!

3. The process of having the whole experiment fully automated (as we had 400+ fake profiles which we needed to visit every day) was also quite tricky. LinkedIn actively works to prevent scraping, so setting up a fully automated process took time and finesse.

2. We were uncertain if we could generate good-enough high-quality AI photos that would be indistinguishable from real ones and whether we can successfully create a Black-White-morphing procedure.

1. We did not know whether we can successfully create (hundreds of) fake LinkedIn accounts and whether those will survive long enough to have a second stage (this was likely the biggest challenge and much of our design reflects the fear of being blocked)

Thanks!

There were a bunch of challenges in the project