Why fatal-fall records need more than a frequency table
Work-at-height accidents are often described using broad categories such as fall, unsafe act, missing protection or worker error. These labels are useful for reporting, but they are too shallow for prevention unless we can connect them to the work activity, occupation, site conditions, management system and equipment involved.
My research programme used Malaysian Department of Occupational Safety and Health accident records to move through three analytical levels: comparative contributing-factor analysis, fault-tree risk assessment, and pattern or machine-learning analysis across a larger set of fatal-fall scenarios. Each level answers a different question.
- Descriptive analysis asks which categories appear most often.
- Fault tree analysis asks how causes can combine into plausible fatal pathways.
- Pattern and predictive analysis asks which variables are associated and which factors help distinguish accident outcomes.
System weaknesses were more prominent than isolated worker behaviour
In the fault-tree study, several high-occurrence factors demanded immediate attention. Lack of supervision and leadership occurred in 89.84% of the analysed cases, failure to use personal protective equipment in 85.93%, and absent or incorrectly followed work standards in 85.15%.
The practical message is not simply “workers must wear PPE.” The repeated coexistence of supervision, leadership and work-standard problems shows that the control system surrounding the worker must also be repaired.
PPE remains important, but it is the last line of defence. An effective prevention strategy starts earlier: eliminate unnecessary work at height, select safer access methods, design collective protection, define the sequence of work, verify competence, supervise critical steps and stop work when conditions change.
The most vulnerable occupations were not evenly distributed
A later pattern study structured 3,321 fatal-fall accident scenarios recorded between 2010 and 2020. General workers were the most vulnerable category, accounting for a 32% yearly average. Roofers recorded 15.5%, electricians 12%, and supervisors were the least represented group at 4%.
These differences matter because a generic work-at-height campaign may not address the conditions faced by each group. Roofers encounter fragile surfaces, edges, changing weather and material handling. Electricians may work near openings, temporary access systems and energised equipment. General workers can be assigned varied tasks without the specialist controls, experience or authority needed to challenge unsafe arrangements.
Weak associations are still useful when they prevent overclaiming
The pattern analysis found relationships ranging from negligible to strong. Dates and some contributing factors showed stronger associations, while direct and root-cause categories often had weak or negligible relationships with other variables. That is not a failed result. It warns us that administrative accident labels may be inconsistent, too broad or too incomplete to support confident causal conclusions.
Good safety analytics should make these limitations visible. When the source record is short, a model cannot reconstruct every organisational decision, barrier failure or human interaction that preceded the event. The result should therefore support prioritisation and further investigation, not pretend to replace a full investigation.
Five prevention priorities supported by the evidence
- Design out work at height wherever possible. Prevention through design is more reliable than depending on behaviour at the edge.
- Treat supervision as a critical control. Define who verifies the access system, work method, weather, anchorage and changing site conditions.
- Use collective protection before personal protection. Guardrails, working platforms and engineered access reduce dependence on individual compliance.
- Make work standards task-specific. A generic procedure is not enough for roofing, electrical work, scaffold use or temporary openings.
- Improve the quality of incident records. Consistent information about task, equipment, immediate mechanism, control failures and organisational conditions produces better learning.
Related research
This article summarises findings from the following peer-reviewed publications. It is an explanatory overview, not a substitute for the methods, limitations and complete results in the papers.