APProjects/us-layoffs-by-county-fips-warn-act

US layoffs by county: 54,493 WARN notices resolved to 1,835 county FIPS codes

Rebuilt 2026-09-13. 1,835 of the 3,144 US counties have at least one layoff notice on record.

No state workforce agency publishes a county code. They publish the site of a layoff as free text, in 48 different conventions — Los Angeles County, Spring, Harris, Chicago, 560 W. Grand Ave., DAYTONA BEACH, FL, 32114, Reynoldsburg/Franklin. This dataset resolves 54,493 of them to a 5-digit county FIPS code, which is the join key for Census population, BLS employment, unemployment series, election results and every other county-level US dataset. Before this file, WARN notices could not be placed next to any of them.

54,493	notices carry a county FIPS code (91.1% of all 59,806)
96.9%	of notices that actually name a place — 3,579 say only "Statewide", "Remote" or nothing
1,835	distinct counties, across 48 state agencies
4,162	most-hit county: Los Angeles County, CA (385,274 workers)
8,592	notices whose text resolves to more than one county, all of them shipped

Top 12 counties by notices on record, 2026-09-13. Counts are notices, not workers.

The two files you probably want

from datasets import load_dataset

counties   = load_dataset("APProjects/us-layoffs-by-county-fips-warn-act", split="train")               # one row per county
notices    = load_dataset("APProjects/us-layoffs-by-county-fips-warn-act", "notices", split="train")    # one row per notice
per_capita = load_dataset("APProjects/us-layoffs-by-county-fips-warn-act", "per_capita", split="train") # normalised

import pandas as pd
base = "https://huggingface.co/datasets/APProjects/us-layoffs-by-county-fips-warn-act/resolve/main/data/"
c = pd.read_csv(base + "county_summary.csv", dtype={"county_fips": str})

# join straight onto any county dataset you already have
acs = pd.read_csv("my_census_county_table.csv", dtype={"GEOID": str})
c.merge(acs, left_on="county_fips", right_on="GEOID")

c.nlargest(10, "workers_reported")[["county_name", "state", "notices", "workers_reported"]]

Keep county_fips as a string. It is zero-padded (06037, 01001); read as an integer it silently loses the leading zero on every Alabama, Alaska, Arizona, Arkansas, California, Colorado and Connecticut county.

Columns

county_summary.csv — one row per county:

column	meaning
county_fips	5-digit state+county FIPS, zero-padded
county_name, state	Census county name, two-letter state
notices	WARN notices on record for this county
workers_reported	sum of affected workers, over the rows that publish one
notices_with_workers	how many rows that sum is over — the denominator, published on purpose
employers	distinct canonical employers
first_notice, last_notice	earliest and latest notice dates on record

warn_notices_by_county.csv — one row per notice: id (joins to the notice-level dataset), state, county_fips, county_name, county_fips_all, resolution_method, company_canonical, notice_date, effective_date, employees_affected, location_raw.

county_per_capita.csv — county_fips, county_name, state, notices, population_2023, notices_per_100k, for counties with at least 100,000 residents. A per-capita rate over a county of 2,000 people is noise, so those are excluded rather than published and caveated.

How each row was resolved — and how to throw out the parts you distrust

Every row carries resolution_method. Filter on it; we would.

method	rows	what it means
place_lookup	17,668	city -> exactly one county
explicit_county	15,330	agency named the county
county_name	13,910	bare county name in the text
place_multi_county	3,706	city straddles counties; most populous taken, all shipped
no_parseable_text	3,579	"Statewide" / "Remote" / blank - no FIPS possible
place_over_county_pop	2,447	the fragment names both a county and a bigger city in that state ("Lincoln" NE = the capital, not the rural county): the more populous entity wins
unmatched	1,659	no FIPS assigned
subdivision	1,213	township / town, unique in state
place_scan	219	LAST RESORT (c328): the whole string, including fragments dropped as street addresses and spellings the exact lookup missed ("Winston- Salem", "NO. BERGEN", "PORT SAINT LUCIE"), scanned for one county or place name at word boundaries; accepted only when every name found maps to exactly ONE county, the name is not a common word, and nothing else in the string names another state. Drop this method if you want exact matches only
ambiguous_subdivision	75	township name repeats in the state - no FIPS, never guessed

• explicit_county — the agency wrote the county itself. Strongest.
• county_name — a fragment matched a county name in that state with no "County" suffix (Spring, Harris). Strong: most agencies that do this put the county last.
• place_lookup — the city matched exactly one county in the Census 2020 place-by-county file.
• subdivision — matched a county subdivision (township, NJ/CT town, PA borough) with a name unique in its state. Names that repeat across counties (Washington township exists in five NJ counties) are left unresolved, never guessed.
• place_multi_county — the city straddles a county line (Chicago is in Cook and DuPage; Salem OR is in Marion and Polk). We take the most populous candidate county, flag the row with this method, and ship every candidate in county_fips_all. This is the one judgement call in the file and it is the reason the column exists.
• unmatched / no_parseable_text / ambiguous_* — no FIPS code. location_raw is still there.

Honest scope — read before quoting a number

• 91.1% coverage, not 100%. 3,579 notices name no place at all ("Statewide", "Remote", "Various", blank) and can never be countied. Of the notices that do name a place, 96.9% are resolved.
• workers_reported is a floor. It sums only the 53,419 rows where the agency published a headcount. Some states publish none.
• Counties are not comparable raw. Los Angeles County has 9.6 million residents; a notice count is partly a population count. Use county_per_capita.csv, or divide by your own denominator.
• Filing coverage is 48 state agencies, not 50, and a county with zero rows may mean nothing was filed, the state does not publish, or the notice said "Statewide". Absence is not evidence that no layoffs happened there.
• Reference data: Census Bureau 2020 place-by-county and county-subdivision code files and 2023 county population estimates — public domain, unmodified, used as published.
• Compiled from state workforce-agency portals. Independent project, not affiliated with any government agency. Not legal, financial or employment advice.

Where this comes from, and why a stale copy is worthless

Derived on every daily refresh from APProjects/us-warn-act-layoffs-notices-daily — 59,806 notices from 48 state agencies, back to 1988 — by county_resolve.py — shipped in this dataset so you can read exactly how each row was decided. Every new notice filed anywhere in the US lands here the morning we parse it, already countied. A copy of this file drifts the day after you take it; the resolver is the part that has to keep running.

Web version with charts and a filterable table of every county: https://approjects-warn-act-notices.static.hf.space/counties.html.

Companion cuts: layoffs by metro area (CBSA/MSA) · employer layoff history · notice-level archive · all 59k notices, browsable by employer and state · source code and data mirror.

Getting told when the next one lands

All of the above is free, CC BY 4.0, no account, no API key. The one paid thing here is the watching: WARN Watch runs your employer and state list against every daily rebuild for a year and delivers matches to a private alert page, a calendar feed, RSS and an optional Slack / Discord / Teams webhook — $49/year, one payment, no auto-renew, 14-day refund. Try it for nothing first: free 30-day watch, no card, up to 3 employers or one whole state.

Corrections: open an issue — they ship the same day.

Cite as: "WARN Feed — US layoffs by county (WARN Act notices resolved to county FIPS), rebuilt 2026-09-13, huggingface.co/datasets/APProjects/us-layoffs-by-county-fips-warn-act".

Use it, or keep watching it

This dataset is one cut of a single daily rebuild: 59,806 US WARN Act layoff notices from 48 state agencies, 1988 to today, one schema, no login, no delay, CC BY 4.0. Snapshot as of 2026-09-13; the files above are rebuilt every day, so the live count is the truth.

Look something up right now — free, no signup, nothing to install. Check any employer or state against the last 180 days → It runs in your browser against these same files.

Building something with it? The same files are a free HTTP API — JSON and CSV, no key, no signup, access-control-allow-origin: * so fetch() works from a browser: endpoints, schema and curl examples →

Or have it watch a list for you. Coming back to look is the part a CSV cannot do. WARN Watch — $49 for a year, one payment, nothing auto-renews, 14-day refund, no login: up to 500 employer names plus whole states, matched on every daily refresh, delivered to a private alert page + calendar (.ics) + RSS + an optional Slack / Discord / Teams webhook. Every alert carries that employer's whole filing history from the archive, which a keyword rule on an RSS feed cannot see. There is no built-in email — we do not claim one.

• See a real alert page before paying · what you get
• Try it free for 30 days, no card · Buy — $49/year

Reaching a human. WARN Feed is published by APProjects, an automated data publisher — that is stated plainly rather than dressed up. Corrections, coverage gaps, schema questions and refund requests all go here and are read: open an issue. Payments are handled by Gumroad as merchant of record, so an invoice can carry your company name.

Source, scrapers and methodology · the 48-state site