CanaryWhistle/us-warn-layoffs

US WARN Act Layoff and Closure Filings

Every US Worker Adjustment and Retraining Notification (WARN) Act filing that has already taken effect, collected from 34 state labor department registries, normalised into a single schema, and resolved so that one employer's filings across many states read as one company.

17,000+ filings · ~2.2 million affected workers · 2005 to present.

Maintained and updated continuously by CanaryWhistle.

What WARN filings are

US employers above certain size thresholds are legally required to notify a state agency roughly 60 days before a mass layoff or plant closure. Those notices are public records. They are the only systematic, primary-source record of US mass layoffs, but they are published in incompatible formats across dozens of separate state government websites, which is why they are rarely used as a dataset.

This is that record, assembled and cleaned.

Why this dataset is different from other layoff data

Most public layoff datasets are compiled from press coverage, which makes them retrospective, skewed heavily toward technology companies, and reliant on whether a journalist wrote something up.

This dataset is:

• Primary source. Every record originates in a government filing. Nothing is estimated, modelled, or scraped from news.
• All sectors. Manufacturing, healthcare, logistics, retail, hospitality, energy, agriculture, education, finance, not just tech.
• Entity-resolved. Legal suffixes, store numbers, site codes and street addresses are stripped so a company's filings across nine states collapse into one identity. Join on employer_canonical.
• Verifiable. Most records link back to the original government filing.

Columns

Column	Type	Description
employer	string	Employer name as published by the state agency, case-normalised
employer_canonical	string	Resolved canonical employer, join on this
city	string	City of the affected site (empty where the state doesn't publish it)
state	string	Two-letter US state code
state_name	string	Full state name
employees	int	Workers affected, as filed
industry	string	Normalised industry bucket (derived; may be empty or "Other")
role_families	string	Inferred role families, semicolon separated (derived)
notice_date	date	Date the notice was filed or recorded (ISO 8601)
notice_date_is_approximate	bool	true where the state publishes no real notice date and a fallback was used
effective_date	date	Date the job losses took effect (ISO 8601)
closure_type	string	Layoff, Closure or Relocation, as filed
source_agency	string	Originating state agency, e.g. CA-EDD
official_filing_url	string	Link to the underlying government filing where available
canarywhistle_url	string	Canonical CanaryWhistle page for this employer

Example uses

• Labour-market and regional-economics research
• Time-series analysis of layoffs by sector, state or metro
• Studying notice-period practice (median notice is 61 days; 36% of filings show under 60)
• Industrial and plant-closure geography
• Grounding factual questions about US layoffs in primary sources

Loading

from datasets import load_dataset
ds = load_dataset("CanaryWhistle/us-warn-layoffs")

Or straight from the source, always current:

import pandas as pd
df = pd.read_csv("https://canarywhistle.com/data/canarywhistle-warn-layoffs.csv")

Limitations, please read before drawing conclusions

• This is not every US layoff. WARN obligations apply only above statutory size thresholds, and coverage and publication practice vary considerably by state. Treat it as the layoffs employers were legally required to declare.
• State coverage is uneven. 34 states are included. Some publish far richer detail than others; a few publish almost nothing.
• Do not compute notice periods where notice_date_is_approximate is true. The notice date there is a sorting fallback, not a filing date.
• industry and role_families are derived from the employer name and filing text. They are conveniences, not filed fields.
• A notice period shorter than 60 days is not evidence of a legal violation. WARN carries statutory exceptions (faltering company, unforeseeable business circumstances, natural disaster), some states publish the date a notice was recorded rather than served, and amendments can reset dates.
• Amended filings can restate headcounts and dates; the latest version is carried.

Coverage note

This dataset is the historical archive: filings already notified and already effective. Filings from the last 30 days and layoffs scheduled but not yet effective are not included here; those are maintained at canarywhistle.com/upcoming.

Licence

Creative Commons Attribution 4.0 International (CC BY 4.0). Free to use, including commercially, provided you credit the source.

Citation

CanaryWhistle, US WARN Act Layoff and Closure Filings, https://canarywhistle.com/data (CC BY 4.0)

Contact

canary@canarywhistle.com, happy to provide custom cuts by state, sector or period.