CPWR Construction Safety & Health Data

CPWR Construction Safety & Health Data is the statistical memory of the American building site. CPWR - The Center for Construction Research and Training has run its Data Center since the mid-1990s, and today it bundles four products: the Construction Chart Book, now in its 7th edition (published 2025), organized as short single-topic chapters on economics, demographics, employment and safety; Interactive Data Dashboards anchored to the Chart Book since 2021, spanning more than three dozen topics in six groups; Data Reports and Data Bulletins arriving roughly six times a year; and the Construction FACE Database, a raw research file coding 768 construction fatalities from NIOSH and State FACE investigation reports across 71 columns.

That combination is rare. Federal enforcement records tell you what happened after an inspection; this archive preserves what investigators concluded would have prevented each death, alongside workforce and employment context no enforcement ledger carries.

Get a sample of this dataset →

Three rows pulled straight from the FACE extract - exactly as a delivery lands them:

Obs  TYPE  RECORDID  STATE   AGE  GENDER  OCCUP  INDUSTRY
1    I     1982-01   MA      59   1       558    1629
2    I     1982-02   WV      29   1       558    1541
3    I     1982-03   PA      .    1       804    1542

One row per investigated fatality. Read the columns and the coding discipline shows immediately: OCCUP and INDUSTRY carry standard classification codes rather than free-text job titles, so a panel of 768 deaths aggregates cleanly by trade and sector. AGE arrives coded, with . marking missing rather than zero - treat non-numeric cells as unknown, never impute. And the state codes mean every question about where deaths cluster resolves with a group-by, not a re-read of narrative investigation reports.

Get a sample of this dataset and we route rows scoped to your states, trades or hazard codes the same day.

The FACE file runs 71 columns; the nine families below do most of the analytical work and were verified against the parsed file during cataloging - 768 rows and 71 columns confirmed, not described. Each family groups related variables: decedent demographics arrive as four separate columns, injury circumstances as five, safety-program flags as five, and the recommendation scheme as individual R-codes plus derived category counts.

Everything else folds under additional fields on request: employer size-band and ownership variables, the complete two-digit recommendation code lists, and the full value definitions from the companion coding-systems workbook. The complete dictionary ships with every sample, so nothing in your pipeline rests on guesswork.

• Geography: United States nationwide. FACE records carry state identifiers for every investigation, and the Construction Fatality Map dashboard plots deaths by location - so state-level cuts come standard, down to the individual record.
• Time: two clocks running at once. The Chart Book and dashboards reference recent federal data, roughly 2023-2025 at the 7th edition, refreshed on an annual cycle. The FACE Database is fixed as of June 30, 2015, holding investigations back to 1982 - a frozen but deep archive rather than a live feed.
• Granularity: one row per fatality in the FACE file, 768 rows deep. Dashboard aggregates cut by topic, year, state and worker or employer characteristic, which makes the pair useful in both directions: incident-level modeling on the raw side, presentation-ready aggregates on the published side.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the channel and the cadence; the records arrive identical either way - typed against the field dictionary above, with the coding-systems workbook alongside so every category code resolves on arrival. Teams typically validate against the incident-level FACE rows first, then promote the same fields to a scheduled feed or land them beside their own EHS and workforce tables. Because the schema is fixed, the pipeline you build for one state panel scales to all 768 records without a rewrite.

Six uses come up again and again, each owned by a different seat at the table:

1. Fatality-pattern analysis by trade - 768 coded investigations keyed by occupation and industry codes turn narrative reports into countable patterns (Data Scientists & ML Engineers).
2. Prevention-category tallies - the recommendation codes let you quantify how often personal-fall-arrest findings accompany fatal falls, or how often violations co-occur with equipment findings, without reading a single report (Journalists, Academics & Students).
3. Workforce and employment context - Chart Book employment, wage, union-membership and demographic series give injury rates credible denominators (Market Researchers & Consultants).
4. Safety-equipment market sizing - hazard and exposure dashboards show which risks dominate, grounding PPE and training pitches in published statistics (Sales & Growth Teams).
5. EHS product backends - embed FACE records and dashboard aggregates into safety tooling where investigator conclusions carry authority (Developers & Data-Product Builders).
6. Compliance-demand signals - OSHA inspection and citation dashboard trends hint at where enforcement pressure, and therefore compliance spending, heads next (Competitive Intelligence & Product Teams).

See where this dataset slots into each workflow on the data scientists use cases or journalists academics use cases pages.

Journalists, academics and students get the deepest read: investigator-coded prevention conclusions behind every fatality claim, with state identifiers for local reporting. Data scientists and ML engineers get a fixed labeled archive - 768 incidents, 71 columns, spanning 1982 to mid-2015 - ideal for hazard-classification models that need stable labels. Market researchers and consultants get workforce, wage and demographic series that give safety statistics their denominators. Sales and growth teams get hazard-frequency context that turns generic safety pitches into specific ones. Competitive intelligence and product teams read the citation and inspection dashboards for demand signals. Developers building data products get a documented, codebook-backed schema ready to embed.

Within the slice, this dataset and OSHA Enforcement & Establishment Data answer different questions about the same subject. OSHA supplies the primary enforcement record - millions of inspections back to 1972, establishment-level Form 300A summaries for roughly 383,000 workplaces in the 2025 file - and refreshes far faster. CPWR adds what enforcement records never capture: what investigators concluded would have prevented each death, coded into countable categories, plus the workforce context around the people involved. The practical split: use OSHA for scale and currency, use CPWR for prevention logic and published trend analysis.

Against the macro series in the slice - housing starts, permits, construction put in place - there is no overlap at all: those measure output volume, this measures who gets hurt. Where it ranks among its quality-scored neighbors is on the best construction engineering datasets list.

The questions below cover what the data contains and how it behaves - fatality counts, coding schemes, dashboard topics, vintage. Anything about delivery mechanics belongs in a conversation: request a sample and we walk your exact use case.

• Construction & Engineering industry hub - every dataset in the slice, scored and ranked, with the catalog-wide stats behind the rankings.
• OSHA Enforcement & Establishment Data - the primary federal record: inspections, citations and Form 300A summaries that pair naturally with CPWR's investigator conclusions.
• Best construction engineering datasets - where this archive sits among its quality-scored neighbors, including permit ledgers and macro construction series.
• Journalists, academics & students use cases - for teams citing worker-safety statistics in published work.
• Market researchers use cases - for workforce and safety benchmarking built on published series.