March 29, 2026
Sort your city, bro
Moretti replication published in AER
A coding oops deflates ‘big city = more inventions’; econ Twitter loses it
TLDR: A new AER comment says design and coding errors erase the claimed causal link between big cities and more inventions. The internet split fast: replication diehards cheered the nulls, city boosters urged caution, and everyone memed the sorting bug as a reminder that tiny code choices can tilt policy debates.
Economics just got messy: a replication comment accepted at the American Economic Review says the blockbuster claim that bigger tech cities make inventors produce more may be correlation, not causation. The author says the “movers” test didn’t actually use moving data and an instrument (a proxy built from other‑city coworkers) was coded wrong; fixing both yields no clear boost. Comment threads exploded: open‑science folks cheered “replication FTW,” city boosters cried foul, and policy wonks side‑eyed that this paper helped steer housing money via Coefficient Giving.
Then came the meme parade: “log(y+0.00001)gate,” “sort your city, bro,” and “many‑to‑many merges = multiverse.” The replication claims patent quality falls as clusters grow, flipping a celebrated result, and finds big clusters drive quantity but maybe not quality. Defenders of the original study called this “one data pass” and pointed to other research on clusters; skeptics clapped back that tests and code are the whole ballgame. Neutral voices asked AER to require code audits and pre‑release repos. The vibe: a lesson in how tiny code choices can swing billion‑dollar policy narratives—and how fast the internet will roast you for a stray zero.
Key Points
- •A replication comment on Moretti (2021) has been accepted by the American Economic Review, identifying ten issues in the original study.
- •The replication finds the original event study does not correctly use movers’ variation; a proper mover event study yields a null effect on patenting.
- •A coding error in the instrumental variable construction (unsorted by city) led to a faulty instrument; correcting it yields null results.
- •Data and measurement problems include unreproducible results due to many-to-many merges and a misapplied log transformation for citations that reverses the sign of patent quality effects.
- •Correcting a misspecified heterogeneity analysis shows substantially larger patenting elasticities in larger clusters, contrary to M21’s limited heterogeneity finding.