OSHA 300A Data for Industrial Target Screening: What It Shows

Executive Summary

The Occupational Safety and Health Administration (OSHA) publishes establishment-level Form 300A injury and illness summaries through its Injury Tracking Application (ITA). As of November 2, 2026, OSHA's download page lists 2025 summary and case detail files covering submissions received from 1/1/2026 through 8/31/2026 ^[1]. Those files are a snapshot of a still-accumulating filing year, not a final count, and they are not "current-year" injury rates. For a private equity origination analyst, the practical question is narrower than "what is the target's injury rate": it is whether a public row belongs to the candidate, whether it justifies one follow-up question, and when the honest answer is "unknown".

Coverage is the first limit. Routine electronic submission applies to establishments with 250 or more employees, to establishments with 20 to 249 employees in listed industries, and, for Form 300 and 301 case detail, to establishments with 100 or more employees in a second list of industries ^[2]. Everyone else submits only if notified. OSHA states that the requirements "do not cover all establishments" ^[1], so a missing row is not proof that a site, or an injury, does not exist.

Identity is the second limit. Records are kept per establishment, while a company may own many establishments and hold several Employer Identification Numbers (EINs) ^{[3] [4]}. OSHA does not standardize addresses or validate company names ^[5], and the establishment identifier is not a reliable indicator of one site over time. Matching therefore needs preserved evidence, not a single fuzzy-name hit.

Interpretation is the third limit. OSHA does not validate employee or injury counts ^[1], and a recorded case does not mean the employer or employee was at fault ^[1]. Rates follow OSHA's formula, cases multiplied by 200,000 and divided by hours worked; the Total Case Rate (TCR) uses columns H, I and J and the Days Away, Restricted or Transferred (DART) rate uses H and I. The Bureau of Labor Statistics (BLS) reported a 2024 private-industry total recordable case rate of 2.3 per 100 full-time workers ^[6], and its table lists 2.7 for manufacturing ^[7], but those national estimates and a single establishment's rate are different kinds of numbers.

This report supplies an eligibility decision tree, a thirteen-field row-level worksheet, and four closing screen outcomes: confirmed match and question, ambiguous match, outside reporting scope, or no observed submission. Concerns belong with qualified diligence staff; this is pre-contact screening, not a safety rating.

Introduction and Background

OSHA 300A data answers a narrow question well: what did a specific establishment report, for a specific year, about its recordable work-related injuries and illnesses, hours worked and average headcount? It answers a broader question badly: how safe is this company? The distance between those two questions is where screening errors happen, and it is the subject of this report, written for industrial-services origination analysts and research reviewers who need to decide whether a public hit deserves a follow-up question before any contact with the candidate.

The data have a long public history. OSHA says it has published establishment-specific Form 300A information since 2009 ^[8], and the ITA has collected summary data since 2016 and case detail since 2023 ^[1]. The 2023 final rule, 88 FR 47254, took effect on January 1, 2024 ^[2]. It kept the two Form 300A groups and added a third group of establishments that submit case-level Forms 300 and 301 ^[2]. It also required establishments to include a company name in electronic submissions ^[2]. That company-name requirement matters for target screening because it is the field most analysts will try to match first.

Three terms recur throughout. An establishment is a single physical location at which business is conducted, and the Census Bureau notes it is not necessarily identical with a company or enterprise ^[3]. A filing year is the calendar year in which the injuries occurred, not the year of submission. A submitted form is what an eligible establishment actually uploaded, which can differ from what it was required to upload. Keeping these three apart is the whole discipline of the worksheet described below.

Scope notes. This report covers the federal OSHA recordkeeping framework and the ITA public files. Employers in State Plan states are covered by the same electronic submission requirements, though state-specific additions exist and are outside this report. It does not evaluate any company, does not assess compliance, and does not treat any public row as a legal or safety conclusion. The worksheet here is written so any research team can apply it by hand or inside its own tooling.

Who Submits Form 300A, and Why a Missing Row Proves Little

The three routine submission groups

The legal text defines the groups by establishment size and industry, measured over the previous calendar year. An establishment with 250 or more employees at any time in that year that must keep injury and illness records submits Form 300A ^[9]. An establishment with 20 to 249 employees in an industry on the Appendix A list also submits ^[9]. An establishment with 100 or more employees in an Appendix B industry also submits the case-level Forms 300 and 301 ^[9]. The regulation says only three categories of employers must routinely submit ^[9].

Headcount here counts every person employed at any time in the year, so a seasonal or temporary workforce can push an establishment over a threshold even if its average staffing is lower. The same logic applies in reverse for diligence: a site with a modest average headcount may still have been in scope.

Appendix A is broad. It lists entire two-digit sectors, including manufacturing (NAICS 31 to 33). The OSHA estimate in the rule preamble was about 463,000 Appendix A establishments with 20 to 249 employees ^[2]. Appendix B is narrower. The preamble describes it as a newly added list of designated industries for establishments with 100 or more employees ^[2].

Establishment tests, company tests, and exemptions

The thresholds are applied to the establishment, not the firm: Part 1904 records must be specific for each individual establishment ^[2]. Two exemptions run on different units, and mixing them up produces wrong scoping calls.

• Size exemption (company level). A company with 10 or fewer employees at all times in the last calendar year is partially exempt from keeping records ^[10], and the test counts employees across the entire company ^[10].
• Industry exemption (establishment level). The partial industry exemption applies to individual business establishments ^[11], so one company can have both exempt and non-exempt sites.
• Partially exempt establishments. A partially exempt establishment does not routinely submit under the main provisions; the rule makes a separate provision for the case where OSHA informs it in writing ^[9].

Table 1 below turns these rules into a screening reference. It is a reading aid, not legal advice; eligibility for a real site is a question for qualified compliance counsel.

Table 1 carries one lesson that the rest of the report depends on. Three of seven rows end in "outside reporting scope", and two others end in "unresolved". A pre-contact screen that treats silence as a clean record is mis-scoped for the lower end of the market, where many candidate sites fall below 20 employees or in non-listed industries. OSHA also cautions that it has not purged submissions from establishments that were not required to file ^[1], so the reverse also happens: a row can exist for a site that was never required to submit.

Summary versus case detail

Form 300A is the annual summary; Forms 300 and 301 are the log and incident report. The rule creates three separate groups ^[2], so many establishments submit summary data but not case detail. Since case rows only exist for the Appendix B group, and only if the establishment had recordable cases, a summary row with no linked case rows is ordinary and does not indicate a data gap. The corporate parent may collect and submit data on behalf of its establishments, so one account can produce many rows, and OSHA's FAQ states that responsibility for completeness and accuracy lies with the employer, even when a third party files.

Obtaining and Reading the Official Files

What is published, and what "current" means

OSHA's ITA page offers a summary file, a case detail file and a data dictionary for each year; the newest listing is the partial 2025 cut described above. OSHA states that case detail data has been collected since 2023. Year labels in these files refer to the filing year, meaning the year the injuries occurred, not the year of submission. The ITA Data Users Guide says each establishment has one record each year and that, where an establishment updates a submission, the most recent record is kept ^[5]. It also says employers may submit after the deadline, so counts for a given year can grow.

Two timing facts are easy to mix up: the legal deadline is March 2 of the year after the covered year ^[2], while the file name carries a "through" date, here 8/31/2026, which is the data cut. A 2024 file listed as the newest in an earlier snapshot is not the newest now, and BLS industry rates for 2024, released January 22, 2026 ^[6], are the matching benchmark vintage for a 2024 establishment row, not for the 2025 file. BLS states the next biennial case characteristics release is for 2025 and 2026 ^[13].

Record the provenance first

Before any matching, write down five provenance items for each row. These make a hit reproducible and let a second reviewer reject it.

• Dataset year and file name. For example, the 2025 summary file with its "through" date.
• Collection date. The date you downloaded it, since OSHA updates files.
• File URL. The exact OSHA link.
• Row key. The record ID and the establishment_ID value.
• Data dictionary version. The version and date printed on the dictionary used, since summary and case detail files have separate dictionaries.

The fields that matter for a screen

The summary data dictionary lists identity fields (establishment name, establishment_ID, EIN, company name, street address, city, state, ZIP code, NAICS code and year, industry description, establishment type, size), denominators (annual average employees and total hours worked), case counts (deaths, days away, job transfer or restriction, other recordable, plus day counts and illness categories), and administrative fields (a change reason, a creation timestamp and the filing year). The establishment_ID is a unique number for each establishment and is the variable that links summary data to case detail ^{[14] [1]}.

Field cautions that have a direct screening consequence:

• EIN. The dictionary defines an EIN field and each reporting establishment must provide the EIN it uses ^[9]. Whether a given extract populates or masks that field should be confirmed in the downloaded file; this report treats it as unknown until checked, and recommends not republishing EINs.
• Company name. A submission may use codes as the establishment name but must include a legal company name ^[9], a requirement the 2023 rule added ^[2]. OSHA does not validate company names.
• Establishment_ID. Not a reliable indicator of a single establishment over time ^[5], so do not join years on it.
• NAICS. Filers choose their own codes in many federal programs ^[15]. Records may use 2012, 2017 or 2022 vintages, and more than 100 six-digit codes were affected by the 2022 revision ^[16].
• Counts. OSHA publishes establishment name, address, hours and counts, but states it does not validate them ^[1].

Matching a Public Row to a Candidate

Why the company and the establishment are different units

A candidate in an origination list is usually a legal entity or a brand, while an ITA row is an establishment. BLS likewise defines an establishment as a single economic unit, such as a farm, a mine, a factory, or a store ^[17], and the Census Bureau defines it as a single physical location at which business is conducted ^[3]. The Census Bureau states that one multi-unit company may have several EINs, that EIN and establishment are not equivalent for multi-unit companies ^[4], and that only one company can use a given EIN ^[4]. The IRS describes an EIN as a federal tax ID number for businesses and other entities ^[18], and says a business entity should have only one EIN ^[18]. Combined, those statements yield a useful but bounded rule: a matching EIN is strong evidence that two records belong to the same legal entity, but a non-matching EIN is weak evidence of the opposite, because a group can operate under several entities.

Industry classification adds a second layer. NAICS classifies establishments, not companies ^[19], and each is assigned to a six-digit code based on the majority of its activity ^[15].

A layered evidence ladder

Match on layers and record each one in the worksheet. Do not collapse them into a single score.

1. Address evidence. Street address, city, state and ZIP code of the site. OSHA does not standardize addresses, so normalize suite numbers, abbreviations and highway names before comparison, and retain the original strings.
2. Name evidence. The company name, establishment name and any known trade names. Keep conflicting names in separate cells rather than choosing one.
3. Tax identity evidence. EIN, where legitimately available to the research team from the candidate or a lawful source. Never publish it. A business entity should have only one EIN ^[18], but a group may hold several entities.
4. Industry evidence. NAICS plausibility against what the candidate does at that site.
5. Scale evidence. Annual average employees against the candidate's known site headcount, with the caution in the next section.
6. Ownership and timing evidence. Whether the establishment changed hands during the filing year. OSHA's FAQ says only the current owner submits, for the portion of the year it owned the site.

Decision rules for ambiguous matches

• One strong layer is not enough. An exact address with a different legal name is an ambiguous match, not a confirmed one, until ownership history explains it.
• Preserve the conflict. If a reviewer cannot decide, record both names, both addresses and the reason, and keep the decision as "ambiguous".

OSHA's Data Users Guide warns that the establishment_ID is not a reliable indicator of one site over time, and its comparison document adds that variation in how establishments report industry and company structure can make cross-company comparison difficult. Both statements argue for human review of every hit that will be relied on.

Reading Cases, Hours and Rates

What the counts mean, and what they do not

The 300A reports counts of recordable cases. A recordable case is one involving death, days away from work, restricted work or transfer, medical treatment beyond first aid, or loss of consciousness ^[20]. A recorded case does not mean the employer or employee was at fault ^[1], and that statement applies equally to a high count. Nothing in the public row supports a conclusion about compliance, fault or a rule breach.

The summary columns line up as follows. Field G is deaths, Field H is cases with days away from work, Field I is cases with job transfer or restriction, and Field J is other recordable cases. A company executive must certify the 300A, which tells you who attested to the numbers but not that anyone outside the company checked them.

The two rate formulas

OSHA's formula is the number of injuries and illnesses multiplied by 200,000 and divided by employee hours worked ^[1]. The 200,000 figure is the equivalent of 100 employees working 40 hours per week for 50 weeks, according to BLS ^[21]. Hours are the total hours worked by all employees covered by the log, and BLS defines the same denominator as total hours worked by all employees during the calendar year ^[7].

• Total Case Rate (TCR). Columns H plus I plus J, times 200,000, divided by hours worked ^[1]. The National Safety Council calls the comparable measure the total recordable incident rate (TRIR) and describes it as injuries and illnesses per 100 full-time workers ^[22], based on OSHA reporting guidelines ^[22]. OSHA's own label for the ITA calculation is TCR.
• DART rate. Columns H plus I, times 200,000, divided by hours worked. BLS defines DART cases as the sum of days-away cases and job transfer or restriction cases ^[6].

(Hypothetical Example) The figures below are fictional and exist only to show the arithmetic. Suppose a fictional machine-shop establishment reports H = 2, I = 3, J = 5, average employees 240 and total hours worked 480,000.

• TCR = (2 + 3 + 5) x 200,000 / 480,000 = 4.17.
• DART = (2 + 3) x 200,000 / 480,000 = 2.08.
• Hours per average employee = 480,000 / 240 = 2,000, which is a plausible full-time year.

Now suppose a second fictional site with the same 10 cases but only 120,000 hours (about 60 full-time equivalents). Its TCR is 10 x 200,000 / 120,000 = 16.67, and a single additional case moves it to 18.33. The same extra case moves the first site only from 4.17 to 4.58. The fictional pair shows why OSHA's rule preamble records the concern that a small number of injuries or illnesses could have a disproportionate effect on incidence rates ^[8]. NIOSH, quoted in the same rule record, warned that some data users might draw unwarranted conclusions about the overall safety record of establishments ^[8].

Denominator plausibility checks

Because OSHA does not validate counts, reviewers should test the denominator before computing any rate. These are analyst heuristics, not OSHA rules.

• Hours missing or zero. Do not compute a rate. Record "denominator unknown".
• Hours per average employee far from a full-time year. A commenter quoted in the rule record cited an employer with 150 employees reporting 24 million hours ^[8], which equals 160,000 hours per employee and is not credible. OSHA's response points to built-in ITA edits that warn users of potential data errors ^[8].
• Temporary labor. Employees supervised day to day by the establishment are recorded by it ^[23], so a staffing-heavy site may carry hours that a payroll-only view would miss.

Comparison warnings

Never rank or label establishments from these rates alone. OSHA states the data may not be generalizable to the worker population, and a single establishment's rate is not a safety rating. BLS describes incidence rates as a way to show the relative level of injuries and illnesses ^[21], which is a comparison tool and not a verdict. Compare only like establishments in the same industry and year, with similar hours, and flag any row where hours are zero or missing, industry codes differ, or the numerators are small enough that one case changes the rate materially.

Table 2 shows the national BLS benchmarks most often used for industrial screens. They are national industry estimates from a BLS survey that is not administered by OSHA ^[24]. They are a reference point for sanity checks, not a pass or fail line for a single site.

Source for every row: BLS Table 1, injury and illness rates by industry, 2024 national ^[7]. BLS notes that, because of rounding, components may not add to totals ^[7].

Table 2 shows spread inside manufacturing: sub-industry total recordable rates run from 0.9 for computer and electronic products to 4.2 for wood products, a more than four-fold range. A single 2.7 manufacturing benchmark would therefore flatter some sub-sectors and penalize others. BLS industry estimates use the 2017 NAICS manual ^[6], while ITA records can carry 2012, 2017 or 2022 codes, so a concordance step belongs in any benchmark comparison. The TRC and DART figures are 2024 values from BLS's January 22, 2026 release and should be stamped with that year.

Implementation: The Row-Level Worksheet and Decision Tree

The eligibility decision tree

Run these questions in order for each candidate site. Stop at the first outcome that applies.

1. Is the unit a site or a legal entity? If only an entity is known, list its sites first. The data are filed per establishment ^[2].
2. Could the site have been in scope in the filing year? Test headcount (250 or more, or 20 to 249), industry (Appendix A), and exemptions using Table 1. If scope cannot be tested because headcount or NAICS is unknown, the outcome is "scope unknown", not "outside scope".
3. Was there ownership change in the year? If yes, expect the submission under the owner at the time and widen the entity search.
4. Is there a row that matches on at least two independent evidence layers? If yes, go to question 5. If only one layer matches, label "ambiguous match".
5. Do the denominators pass the checks above? If not, record the exception and withhold the rate.
6. Is a case-detail join warranted? Join summary to case detail only where the site is in the Appendix B group and the establishment_ID is the same in both files.
7. Assign a screen outcome (see below).

The worksheet

Table 3 lists the thirteen worksheet columns, what to record in each, and where the content comes from. One row is created per candidate site and filing year.

Table 3 keeps raw published values separate from calculated values and from the reviewer's decision, so a second reviewer can rebuild the decision from the first six columns. It also guards against re-keying numbers, which loses what OSHA actually published, and merging conflicting names, which hides the evidence for an ambiguous match.

The four screen outcomes

• Confirmed match and question. At least two independent evidence layers agree, the denominators pass, and a specific, neutral follow-up question exists, such as "can you provide your most recent year's hours worked and recordable case counts for this site?"
• Ambiguous match. One strong layer or conflicting layers. Record both candidates and do not compute a rate for reporting.
• Outside reporting scope. Table 1 supports non-submission. Injury performance is unknown, not clean.
• No observed submission. The site could have been in scope and no row was found. This is a research gap, to be pursued through other lawful sources, not a negative finding.

Send concerns to qualified diligence staff. The worksheet records what a public source reports and what could not be established; it does not rate safety, assess compliance or substitute for transaction diligence.

Data Analysis and Evidence

How large is the public dataset?

OSHA's comparison document reports that in 2024 it collected summary data from over 370,000 establishments and case details for more than 850,000 cases ^[25]. By the rule preamble's estimate, about 463,000 establishments with 20 to 249 employees in Appendix A industries would submit ^[2]. Those two numbers are not directly comparable (one is a realized collection, the other a projection of a particular group), and the public files include some establishments that were not required to submit. They do show the order of magnitude: the data cover hundreds of thousands of establishments, which is large enough for site-level screening but is not a census of employers.

How good are the numbers?

Evidence on completeness is mixed, and the different sources measure different things.

• OSHA on representativeness. OSHA states that the data may not be generalizable to the worker population ^[1].
• Undercount research. A BLS Monthly Labor Review article summarizing research reports estimates of an undercount ranging from 20 percent to 70 percent of cases ^[26], while noting other research concluded the size of the undercount is small ^[26]. That literature concerns the BLS survey and employer logs, not specifically ITA, and no ITA-specific completeness study was identified.
• Verification capacity. A Government Accountability Office report found OSHA audited about 250 of roughly 130,000 worksites in high hazard industries ^[27]; it is a 2009 finding and describes capacity at that time.
• Reporting barriers. GAO reported that workers may not report an injury because they fear job loss ^[27], and a peer-reviewed Washington State study attributes underreporting partly to insufficient recordkeeping systems and limited knowledge of reporting requirements ^[28].

OSHA's preamble itself treats underreporting as a recognized possibility ^[8], and BLS states that non-sampling errors will always occur when data are gathered ^[29]. These sources agree on direction, that recorded counts can understate injuries, and disagree on magnitude. A low reported rate does not establish a low true rate, and a high rate may reflect more complete recording; neither supports a ranking.

Benchmarks and arithmetic

BLS reported a 2024 private-industry total recordable case rate of 2.3 per 100 full-time workers, down from 2.4 in 2023 ^[6]. Using Table 2, the fictional site in the worked example (TCR 4.17, DART 2.08) is above the 2.7 and 1.7 manufacturing figures on both measures, but the comparison is unsafe: a probability-survey industry average includes establishments of all sizes and the fictional site is a single data point whose denominators are unvalidated. The correct screening use is a question ("is the hours figure consistent with your payroll?") rather than a verdict.

The insurance measure often mentioned alongside these rates is different. The workers' compensation experience modifier compares an employer's insurance experience with that of other employers in its class ^[30], generally using the latest three years of data ^[30]. The modifier is based on individual payroll and loss records rather than on OSHA logs, so it is a different measure from the incident rates above, and neither is derivable from the other.

Implications and Future Directions

For screening practice. The most defensible default is to separate three claims that sources often blur: that a row exists, that it belongs to the candidate, and that its numbers mean something. Each claim needs its own evidence. A team that keeps "scope unknown" and "no observed submission" as distinct labels will produce fewer false negatives than one that records a blank.

For the lower-middle market. Table 1 implies a coverage ceiling. Many sub-20-employee sites, small companies and non-listed industries have no routine submission obligation, so the dataset is structurally better at describing mid-size and large industrial establishments. For small targets, absence of a row is expected.

For vintage control. Several things are moving. The newest ITA files are partial-year cuts through 8/31/2026. BLS's tables page still lists 2024 as the latest annual set ^[13], and its next case characteristics release covers 2025 and 2026 ^[13]. OSHA states it plans quarterly updates. Rule status can also change: this report found no verified pending change to the electronic submission requirements, but a reviewer should re-check the eCFR and OSHA's recordkeeping pages before relying on thresholds.

Open questions. No ITA-specific completeness study was found, EIN population in public extracts must be checked in the file, and the handling of zero or missing hours is not documented in the dictionary or users guide reviewed here.

Frequently Asked Questions

Which employers must submit OSHA 300A electronically?

Establishments with 250 or more employees that must keep records, and establishments with 20 to 249 employees in Appendix A industries ^[9]. A third group, 100 or more employees in Appendix B industries, also submits case-level Forms 300 and 301 ^[2].

What is OSHA 300A electronic submission data?

It is the annual summary of work-related injuries and illnesses, submitted through the ITA by establishment, and published by OSHA with company name, address, average employees and hours worked ^[8]; the 2023 rule took effect January 1, 2024 ^[2].

Where can the injury tracking application data be downloaded?

On OSHA's ITA data page, which lists summary and case detail files by year and the data dictionaries.

Can OSHA 300A data support due diligence for a private equity deal?

It can support a pre-contact question when identity and scope are established. It cannot rate safety, show fault, or replace transaction diligence, since a recorded case does not mean fault.

How do I use OSHA data to screen an acquisition target?

Establish identity on several evidence layers, test scope with Table 1, check denominators, compute rates only when the checks pass, and record one of four outcomes in the worksheet.

How are TRIR and DART calculated from 300A data?

TCR (OSHA's term, similar to the TRIR that the National Safety Council describes ^[22]) is H plus I plus J cases times 200,000 divided by hours worked, and DART is H plus I cases on the same basis.

What are the limits of establishment-specific injury and illness data?

OSHA does not validate counts, data may not be generalizable, and company names and addresses are not standardized.

What does workplace safety due diligence look like for a manufacturing acquisition?

That is a transaction-stage activity for qualified staff and outside this pre-contact screen. The public row is at most one input.

Conclusion

OSHA 300A data are a useful and limited public signal. They report self-submitted, unvalidated counts and denominators by establishment, for a filing year, for the subset of establishments that the rules cover or that chose to file. As of November 2, 2026, the newest file covers 2025 injuries and is a partial-year cut through 8/31/2026, while the BLS benchmark vintage is 2024.

The screening discipline that follows is simple to state. Define the unit before searching. Test whether the site could have been in scope. Match identity on several layers and keep conflicting names. Check hours and headcount before computing the TCR or DART rate. Compare only like establishments and years. Finish with one of four outcomes, and let unknown stay unknown. Treating a missing row as a clean record, or a recorded case as a finding, converts a public data limit into a false conclusion.

Reviewers should re-check the latest year, thresholds, dictionary and any correction notices before use. Questions about what a row means for a transaction belong with qualified diligence and legal staff.