Plant closings vs mass layoffs, across 48 states

The WARN Act asks one two-valued question of every notice — did the site close, or was it a mass layoff — and 48 state portals answer it in 475 different ways, or not at all. This page normalizes that column across the whole archive and shows its work: 16,674 plant closings and 21,720 mass layoffs from 59,802 notices, with 21,381 rows honestly left as unknown rather than guessed.

Get told when a site you care about closes — $49/year Download the closings CSV

Also published as a Hugging Face dataset with the full method card and both CSVs, same daily rebuild: us-plant-closings-vs-mass-layoffs-warn-act.

Why 64% and not 100%

21,381 of 59,802 rows (36%) are unknown. None of them is folded into either side. The reasons, from the same build:

27 rows say both ('closing and layoff') and are counted as neither. Row accounting is asserted at build time: closings + layoffs + mixed + unknown = 59,802, so no row can be dropped to flatter the share.

Newest plant closings on record (1988–present)

Filter by employer, location or state

300 of 16,674 closings shown, by effective date with the newest first (dates still ahead of today are announced closings not yet in effect); the full list is in the CSV. The last column is the exact string the state used.

FAQ

What is the legal difference?

The federal WARN Act (29 U.S.C. §2101) covers two events: a plant closing — a site or operating unit shuts and 50+ employees lose their jobs — and a mass layoff — the site stays open but 50+ (and a third of the site) or 500+ employees are cut. Several states extend the thresholds. The two matter differently to a recruiter: a closing puts every role on the market on one date; a layoff is partial and often role-specific.

Why is only 64% classified? Isn't the rest just a layoff?

No, and folding it in would manufacture thousands of rows. The largest unknown bucket after blanks is rows where the state's column records which statute applies ('State', 'WARN', a numeric code) rather than what happened. We report those as unknown, with the reason, and keep the raw string on every row so you can re-classify under your own definition.

How was the normalization built?

475 distinct raw strings across 48 states ('Closure', 'Closing *', 'CL', 'LO', 'WR', 'Plant Closing', 'facility closure', 'Mass Layoff - No Recall', …) map through an audited rule table in notice_kind.py with a fatal selftest. Every daily rebuild re-runs it over the whole archive, so a state that changes its export shows up as new unmapped strings the same morning — the kind of maintenance nobody does by hand.

What about 'permanent' vs 'temporary'?

Where a state says so we keep it as a separate permanence column (13,364 permanent, 4,775 temporary). It is not the same question: a temporary closing is still a closing.

Which rows are in the downloadable table?

plant_closings.csv holds every notice classified as a plant closing in the free window (1988–present, 16,674 rows); this page shows the 300 newest by effective date. notice_kind_by_state_year.csv is counts only and runs the whole archive. Both rebuild daily and are CC BY 4.0.