Denominator explorer
Pick a peer group, see the screened TRIR spread, and place your own numbers on it. Same pipeline, same screen, as the main page.
1. Pick a peer group
NAICS-3 industry and establishment-size band, pooled across 2016-2024. Values are the screened TRIR distribution within that cell, straight from the pipeline's percentile table.
Data table
2. Where would you sit
Enter recordable cases and total hours worked for the reporting period. TRIR = cases × 200,000 ÷ hours. The marker lands on the band above, and the hours value is run through the same plausibility screen the pipeline uses.
Rule: a filing is flagged flag_hours_per_employee_high or flag_hours_per_employee_low when hours ÷ employees falls outside [120, 4,500] per year. This tool does not collect an employee count, so it screens on total hours alone against the same bounds as a rough check. Exact rule and source:
src/ehs_osha/quality.py.
3. The denominator effect, year by year
Share of filings flagged implausible each year (bars), and aggregate TRIR before and after the screen (lines). The 2019 point is the extreme case: one keying error in the hours field put 99.6% of that year's reported hours inside implausible filings.
Data table
4. What this can and cannot tell you
Can
- Show where a screened TRIR sits against a public, self-reported peer distribution for a NAICS-3 industry and size band.
- Show how much a national or industry aggregate rate depends on whether implausible hours values are screened out first.
Cannot
- Stand in for OSHA's own enforcement data or a regulator's determination. This is Form 300A summary data only, not inspection or citation records.
- Guarantee any individual filing is accurate. The screen catches implausible hours values; it cannot verify case counts, classification, or intent.
- Identify a peer group more precisely than NAICS-3 and one of seven size bands. Two establishments in the same cell can differ a great deal in process, geography and workforce.
- Treat a small n as reliable. Check the publishable flag and the n shown for the selected cell before reading percentiles as a fact.