What the licence record could not say

The obvious question is how many restaurants closed. We cannot answer it, and neither can anyone else working from this record. A closure is not visible in Chicago’s licence data for roughly three years.

2 of 215 is not evidence that those restaurants survived. It is evidence that nothing about them is observable yet. Any count of recent closures would be a measurement of how slowly closures become visible, wearing the costume of a finding.

So the question was narrowed to the half the data can answer without an asterisk: openings. A new licence appears in the record the day it starts.

We needed a way to say which neighbourhoods are Latino. The quick approach used what we already hold: the share of a neighbourhood’s restaurants serving Mexican and Latin American food. On that definition the result was strong. Winter openings across the 11 highest-share neighbourhoods came in at 18 against the 33 the previous fifteen winters imply, at p = 0.0031.

Cuisine mix does not record who lives somewhere. It records who used to.

Restaurants outlive the demographics that opened them. 3 of the 11 neighbourhoods in that cohort are not majority Hispanic at all: Albany Park at 32%, East Side at 39% and Lower West Side at 47%. A cohort we had described as the city’s most Latino neighbourhoods was substantially made of neighbourhoods that are not. The sentence was wrong before the statistics were.

Banded properly, on the American Community Survey table B03002 at ZIP level, a neighbourhood counts when it is majority Hispanic. That is 7 ZIPs: 60609, 60623, 60629, 60632, 60638, 60639, 60641. The drop is still there. It is about half the size, and it no longer clears the bar.

On the window we chose before looking at anything, the same test gives 11 against 16.2 expected, p = 0.12. Choosing a window after seeing where a dip sits is the commonest way a weak result is made to look strong, so both are reported.

Across the sixteen winters from 2010, the cohort’s share of citywide openings has run between 13.6% and 22.7%. The season under test is 12.7%, against a long-run 17.7%. It is the lowest in the series, and not far outside it.

A weak result that survived no scrutiny would not be worth publishing. These are the tests designed to make it fail.

• A placebo cohort. The identical test on majority-Black ZIPs, where no comparable change was reported. Openings there went up, 49 against 32.4 expected. The method does not manufacture a collapse wherever it is pointed.
• Place, or cuisine. If Mexican and Latin restaurants had simply stopped opening citywide, this would be a story about a cuisine rather than a place. Outside the cohort they ran at 15, inside a 12 to 33 range. The effect stays with the place.
• Every threshold, not ours. A real effect usually strengthens as the cohort tightens. This one does not move in one direction.

Openings in Chicago’s majority-Hispanic ZIPs ran below their long-run rate last winter, by roughly a third, and came back afterwards: 46 in the spring and summer of 2026 against 38 a year earlier, a 15.8% share against a 17.3% norm. That is consistent with what the restaurateurs in the article described.

We will not tell you why. A licence record contains no reason. It can date a change and size it. It cannot attribute one to any cause, and we are not going to imply one by placing a chart next to a headline.

So the honest summary is the unsatisfying one. Real in direction, about half the size we first reported, and short of significance on the only window we did not choose after the fact.

The reason we publish it anyway is that the correction is the useful part. A cohort definition that reads as reasonable, produces a clean number, and describes different places than its own name is not a rare mistake. It is the ordinary one.

Every figure on this page is generated rather than typed, including the retracted one. It comes from scripts/latino-hood-shock.ts, which recomputes both cohort definitions on every run and writes the file this page imports. Re-running it against the current licence record moves these numbers, and the page moves with them. Data measured 2026-08-15.

Restaurant licences are filtered to actual restaurants. Chicago’s Retail Food Establishment licence also covers groceries, pharmacies, hotels and school cafeterias, which is a little under half the holders in the record. Demographics are American Community Survey five-year estimates, table B03002, at ZIP level. How the score itself is built is a separate document: see the methodology.

The counts above come from 235 restaurants opened four years ago, 235 opened a year before the window, and 215 opened within it.