Fined, Not Collected: $2,033,906,471 of OSHA Penalties Went Away

“OSHA fines get negotiated down” is a familiar claim and it is true in aggregate. It is also the least interesting thing in the data. The useful finding is how unevenly it happens.

• Most fines are never touched. 2,569,042 of 4,178,731 violations (61.5%) end at exactly the amount assessed. Only 1,502,388 (36%) are reduced — and 107,301 (2.6%) end up higher than they started.
• When a fine is cut, it is cut nearly in half. Among the violations that are reduced at all, 49.5% of the assessed dollars disappear. The 35.95% aggregate is the product of a minority of cases moving a very long way, not of everyone getting a modest discount.
• The bigger the fine, the bigger the share that disappears. Violations assessed under $500 lose 22.2% of their value. Violations of $50,000 and up lose 42.7%. The gradient is monotonic across every band in between — a small citation is a fixed cost, a large one is an opening position.
• Contesting is associated with roughly a third more discount. Contested violations end 45.3% below assessment against 31.9% for uncontested, and 54.3% of them move at all against 33.3%. Employers who contest are self-selected, so this is not proof that contesting causes the discount. What is notable is the other side: a third of uncontested citations are reduced anyway, through OSHA’s informal settlement conference.
• One percent of citations vanish entirely. 41,215 violations (0.99%) fall to exactly $0 — vacated, not negotiated. Those alone account for 7% of the entire write-down, which is why they are reported separately rather than folded into an average discount.
• State enforcement outcomes differ by more than 20 points. Of the 25 states with the most penalty dollars, WA retains the most (21.6% written down) and IN the least (48.3%). Roughly half the states run their own OSHA-approved plans, which is the obvious place to look for the cause — this data does not identify it.

An OSHA penalty is not final when it is written. Informal settlement, contest and Review Commission proceedings can run for years, and the published current_penalty reflects wherever that process has reached today. Recent years therefore look less discounted than they will end up being — not because those employers negotiated worse, but because they have not finished.

The last column is the second reason for a window. The share of violations showing any recorded adjustment climbs steadily from 2.1% in the 1970s to 44.6% in the 2010s. We cannot tell from this data whether older records simply were not updated in place or whether enforcement practice changed, so we do not average the eras together. The headline uses 2000–2020; the full series is published either way.

Source. The U.S. Department of Labor / OSHA enforcement data release — 12,805,474 cited violations going back to 1972, each carrying both the penalty initially assessed and the penalty currently standing. Almost no enforcement dataset publishes both sides; nothing on this page is estimated or imputed.

Cleaning. Two contaminations were removed before analysis. A CSV header row had been imported as data — it is why the maximum issuance date in the raw file reads as the literal string issuance_date. And 452,615 rows carry OSHA’s delete_flag, marking records the agency withdrew; including them would have inflated assessed penalties by $983 million.

Unit of analysis. One row is one cited violation — not one inspection and not one employer. A single inspection routinely produces several violations. Employer, state and industry come from the inspection record joined on the activity number, because the violation record carries no employer field. Employer names are as OSHA recorded them and are not entity-resolved, so the same company can appear under several spellings.

What “written down” means here. It means less is owed on paper. OSHA’s release records the penalty standing after settlement and review; it does not record collection. Nothing on this page should be read as a measure of what employers actually paid.

One number we could not reproduce. A widely quoted internal estimate put this at “5,996,029 violations, 38.0% discount, 47.3% of violations reduced.” The dollar totals are close to the uncleaned figures, but neither the row count nor the 47.3% share reproduces under any filter we tested against the cleaned corpus. The figures published here are the cleaned ones; the two sets should not be mixed.

The aggregated data is published as open CSVs under CC BY 4.0. Analysis by Settlement Insight Research.