Heinrich's accident triangle postulates a roughly fixed ratio between the number of no-injury events and minor incidents and the number of serious or fatal accidents, and thereby licenses the use of minor-incident counts as a proxy for major-accident risk in safety management and regulation. Later literature has empirically rejected the universality of fixed ratios — the severity mix varies with sector, region, and hazard type, and in general-aviation and mining-company panels no significant minor-to-major association is found — while large establishment-level panels in US mining do show a predictive association that depends on how severity is delineated and is confounded by endogenous reporting (near-miss counts rise after fatal events and with reporting-culture improvements). What remains open is a validated characterization of the conditions — sector, hazard type, aggregation level, severity-delineation convention, and reporting regime — under which the count-based proxy retains predictive value, together with an identified treatment of the reporting process that generates the counts. The problem is to establish that characterization across contexts.

Background

In occupational safety, the accident triangle (or safety pyramid) is the model, introduced by H. W. Heinrich in his 1931 book Industrial Accident Prevention, that accidents of different severity occur in a roughly fixed proportion — classically about 300 no-injury events and 29 minor injuries for every 1 major injury — with a variant by F. Bird using the ratio 1-10-30-600. A near miss is a precursor event that could have escalated to a serious accident but did not. Because serious and fatal accidents are rare, the triangle lets managers and regulators monitor the abundant base of the pyramid — counts of minor incidents and near misses, aggregated in indicators such as the total recordable injury rate (TRIR, recordable injuries per 200,000 work hours) and lost-time injuries (LTI, injuries causing absence from work) — and treat changes at the base as early warning for the apex, where serious injuries and fatalities (SIFs) sit. In the indicator taxonomy of safety science, such outcome counts are lagging indicators (measuring harm after the fact), and the pyramid makes them usable as leading indicators (predictors of future harm); this presumed predictive role underpins corporate safety-management systems, behavior-based safety programs (safety-management schemes built around observing and correcting workers' on-the-job behavior), and regulatory exposure metrics worldwide. Two tenets carry the argument: a roughly constant severity ratio across contexts, and an assumption that low- and high-severity events share underlying causes. Both tenets have been contested. Manuele's critique (in On the Practice of Safety) attacks Heinrich's evidentiary foundations; Anderson and Denkl (2010) conclude from over 1.2 million incident reports in an oilfield-services company that the fixed ratios hold only in large aggregated collections of heterogeneous hazards, that LTI-to-fatality ratios vary widely across accident types, and that some high-risk activities have 'narrow-based' triangles with few or no minor precursors before catastrophic outcomes; Cooper (2019) shows UK fatality and serious-injury rates were essentially flat for three decades while minor-injury rates fell by about two thirds. On the validation side, Marshall, Hirmas and Singer (2018) built a Bayesian two-part statistical model (a Poisson-gamma model of accident counts combined with a multinomial-Dirichlet model of the severity mix — standard Bayesian hierarchical components for, respectively, accident frequencies and the proportions of each severity level) for more than 50,000 Chilean firms and found the constant-ratio hypothesis statistically invalid across economic sectors and regions, though with discrepancies small enough that minor-accident counts remained a practically useful signal; Yorio and Moore (2018), using site-specific data from more than 25,000 US establishments over 13 years, found lower-severity incident counts significantly associated with subsequent fatality probability, but with the triangular form dependent on how severity is delineated. Domain refutations exist as well: Onwuka (2020) found no statistically significant relationship between total and fatal general-aviation accidents in NTSB data (2004–2013), and Rocha et al. (2025) found no significant correlation between frequent less-severe and rare severe events across four mining companies. A further complication is reporting endogeneity: the counts are produced by reporting processes, not by hazards alone. Yorio et al. (2020) found fatal events increase subsequently reported near misses (a cyclic relationship), Bakshi and Peura (2022) modeled formally how economic incentives make agents underreport near misses and firms decline to record them, and Hossain and Rahman (2026) documented civil-aviation databases where incident counts rise with reporting improvements while non-fatal events remain underreported. Bayona, Hallowell and Bhandari (2024) report key differences in the proximal causes of low- and high-severity construction injuries, challenging the common-cause tenet. The unresolved target that emerges from this record is a general, validated characterization of when the count-based proxy actually works.

Problem Statement

Determine the conditions under which the number of reported near misses and minor incidents predicts the frequency or likelihood of serious and fatal accidents. The characterization has four linked parts: (i) a model of the joint incident-generation and reporting process in which the severity distribution of incidents may vary with context — economic sector, hazard or accident type, aggregation level (establishment, firm, industry, national), severity-delineation convention, and reporting regime — rather than assuming a fixed ratio; (ii) how much predictive information minor-incident and near-miss counts carry, with quantified uncertainty, about subsequent serious or fatal events in each regime, including whether the mapping weakens, flattens, or reverses in identified contexts; (iii) quantification of the reporting channel — underreporting, incentive-driven suppression, and post-accident reporting spikes — so that observed count-based associations are decomposed into hazard signal and reporting artifact; and (iv) criteria stating for which sectors, hazard classes, aggregation levels, and reporting conditions the count-based proxy is valid, weakened, or invalid. These four components are linked parts of one objective — a decision-usable characterization of the accident-triangle proxy's validity conditions — and none of them individually closes the problem: a model without cross-context validation, an empirical association without a reporting-process treatment, or a single-domain study each leave the general conditions question open.

References

  1. H. W. Heinrich, Industrial Accident Prevention: A Scientific Approach, McGraw-Hill Book Company, New York, 1931 (first edition)
  2. F. A. Manuele, 'Heinrich Revisited: Truisms or Myths', chapter 7, pp. 122–146, in On the Practice of Safety, 2nd edition, Wiley, 2003. DOI: 10.1002/0471721697.ch7. https://doi.org/10.1002/0471721697.ch7
  3. A. Saloniemi and H. Oksanen, 'Accidents and fatal accidents—some paradoxes', Safety Science 29(1):59–66, 1998. DOI: 10.1016/S0925-7535(98)00016-2. https://doi.org/10.1016/S0925-7535(98)00016-2
  4. M. Anderson and M. Denkl, 'The Heinrich Accident Triangle—Too Simplistic A Model for HSE Management in the 21st Century?', SPE International Conference on Health, Safety and Environment in Oil and Gas Exploration and Production, SPE-126661-MS, 2010. DOI: 10.2118/126661-MS. https://doi.org/10.2118/126661-ms
  5. M. D. Cooper, 'The efficacy of industrial safety science constructs for addressing serious injuries & fatalities (SIFs)', Safety Science 120:164–178, 2019. DOI: 10.1016/j.ssci.2019.06.038. https://doi.org/10.1016/j.ssci.2019.06.038
  6. P. L. Yorio and S. M. Moore, 'Examining Factors that Influence the Existence of Heinrich's Safety Triangle Using Site-Specific H&S Data from More than 25,000 Establishments', Risk Analysis 38(4):839–852, 2018. DOI: 10.1111/risa.12869. https://doi.org/10.1111/risa.12869
  7. P. Marshall, A. Hirmas and M. Singer, 'Heinrich's pyramid and occupational safety: A statistical validation methodology', Safety Science 101:180–189, 2018. DOI: 10.1016/j.ssci.2017.09.005. https://doi.org/10.1016/j.ssci.2017.09.005
  8. C. Onwuka, 'A Quantitative Test of the Predictive Validity of Heinrich's Accident Pyramid', African Journal of Health, Safety and Environment 1(1):1–10, 2020. DOI: 10.52417/ajhse.v1i1.18. https://doi.org/10.52417/AJHSE.V1I1.18
  9. S. M. Moore, P. L. Yorio, E. J. Haas, J. L. Bell and L. A. Greenawald, 'Heinrich Revisited: a New Data-Driven Examination of the Safety Pyramid', Mining, Metallurgy & Exploration 37:1857–1863, 2020. DOI: 10.1007/s42461-020-00263-0. https://doi.org/10.1007/s42461-020-00263-0
  10. P. L. Yorio, E. J. Haas, J. L. Bell, S. M. Moore and L. A. Greenawald, 'Lagging or leading? Exploring the temporal relationship among lagging indicators in mining establishments 2006–2017', Journal of Safety Research 74:179–185, 2020. DOI: 10.1016/j.jsr.2020.06.018. https://doi.org/10.1016/j.jsr.2020.06.018
  11. N. Bakshi and H. Peura, 'Prevent or Report? Managing Near Misses for Safer Operations', Manufacturing & Service Operations Management 24(4):2064–2080, 2022. DOI: 10.1287/msom.2022.1100. https://doi.org/10.1287/msom.2022.1100
  12. Z. Šimić, M. Peinador Veira and R. Banov, 'Correlation between events with different safety significance in nuclear power plants', Nuclear Engineering and Technology 54(7):2510–2518, 2022. DOI: 10.1016/j.net.2022.01.034. https://doi.org/10.1016/j.net.2022.01.034
  13. S. Cho, D. Kim, H. Khan and C. K. Lee, 'Heinrich's Law for Traffic Incidents – Using the Digital Tachograph Data to Identify Traffic Accident Hotspots', Promet – Traffic&Transportation 35(6):829–837, 2023. DOI: 10.7307/ptt.v35i6.207. https://doi.org/10.7307/ptt.v35i6.207
  14. A. Bayona, M. R. Hallowell and S. Bhandari, 'The Things That Hurt People Are Not the Same as the Things That Kill People: Key Differences in the Proximal Causes of Low- and High-Severity Construction Injuries', Journal of Construction Engineering and Management 150(8), article 04024089, 2024. DOI: 10.1061/JCEMD4.COENG-14545. https://doi.org/10.1061/jcemd4.coeng-14545
  15. R. Rocha, E. P. H. Diniz, M. Parreiras and I. Almeida, 'The illusion of safety indicators in the mining industry', International Journal of Occupational Safety and Ergonomics, published online 24 November 2025, pp. 1–13. DOI: 10.1080/10803548.2025.2580782. https://doi.org/10.1080/10803548.2025.2580782
  16. S. Hossain and T. Rahman, 'Evaluating the Safety Pyramid in Civil Aviation: A Data-driven Assessment of Occurrence Trends across Three South Asian Countries (1990–2023)', Journal of Aviation, advance online publication, 2026. DOI: 10.30518/jav.1799258. https://doi.org/10.30518/jav.1799258