I bring you a weekly bite-sized chunk of the science behind helicopter human factors and CRM in practice, simplifying the complex and distilling a helicopter related study into a summary of less than 500 words.
TITLE:
Tracking changes in aviation safety narratives: A text-mining study of NTSB reports (2008–2025).
WHAT?
Study examining how narrative themes appearing in around 27,000 NTSB accident and incident reports have changed over nearly two decades. The author used AI driven text-mining techniques to study how investigators describe accidents and if narrative emphasis has shifted over time.
WHERE?
U.S. Washington State Department of Transportation.
WHEN?
Published in Aerospace Traffic and Safety in 2025, using reports spanning 2008–2025.
WHY?
Accident narratives often contain richer contextual information than structured databases and could reveal how safety thinking itself has evolved. The study sought to determine whether investigators continue to frame accidents primarily in terms of individual pilot error or whether broader systemic factors are receiving greater attention.
HOW?
The study applied Natural Language Processing (NLP) and text-mining methods to combine and analyse the narrative sections of approximately 27,000 NTSB reports. Seven themes were identified:
- Human/pilot error
- Mechanical and system issues
- Weather and environmental factors
- Fuel management
- Loss of control/stall
- Runway excursions and handling events
- Wildlife strikes
Reports from 2008–2014 were compared with those from 2015–2025 using statistical tests. The author used regression analysis to identify trends over time and network analysis to examine how themes appeared together within the same reports. Sentiment analysis assessed whether the tone of narratives changed across the study period.
FINDINGS:
Human error remained the most frequently cited factor, appearing in over 90% of narratives. However, references to human error declined slightly over time, while references to mechanical and system factors increased significantly. Weather-related references showed a marked decline, and runway excursion themes increased modestly. Fuel management, loss of control and wildlife strikes remained relatively stable.
Network analysis showed which themes frequently occurred together. The strongest association was between human error and mechanical/system issues, demonstrating that accidents are rarely consequence of a single cause. Weather also showed important connections with loss-of-control events and other factors.
SO WHAT?
The study is significant in demonstrating the potential of artificial intelligence and NLP tools to analyse large bodies of safety data that would be otherwise impossible to review manually. As these techniques mature, they may provide safety organisations with practical methods for identifying emerging risks, monitoring changes in safety culture, and supporting proactive Safety Management Systems.
Although the significance of this paper lies less in the specific accident categories and more in what they reveal about the evolution of safety thinking, the findings did suggest a gradual movement away from viewing accidents through the lens of human error toward a more systems-oriented perspective consistent with frameworks such as HFACS, Threat and Error Management (TEM), and resilience engineering. Investigators increasingly appear to recognise the interaction between technical, environmental and human factors rather than treating human performance in isolation.
REFERENCE:
Ison, D. (2025). Tracking changes in aviation safety narratives: A text mining study of NTSB reports (2008–2025). Aerospace Traffic and Safety, 2, 196–202. https://doi.org/10.1016/j.aets.2025.12.006
competentaviators.com 