Tracking changes in aviation safety narratives: A text mining study of NTSB reports (2008–2025)

Accident investigation narratives provide insights into causal factors not fully captured by structured data fields. This study applied text mining, statistical comparison, and sentiment analysis to 27,000 National Transportation Safety Board (NTSB) accident and incident narratives from 2008 to 2025. The narratives were coded into seven themes: human/pilot error, mechanical/system issues, weather/environmental conditions, fuel management, loss of control/stall, runway excursion/handling events, and wildlife strikes. Comparisons between two periods (2008–2014 and 2015–2025) were conducted using chi-square tests, with linear regression applied to assess temporal trends. Correlations and network analysis were used to examine co-occurrence, while sentiment was evaluated with TextBlob on a stratified sample of 5,000 narratives. Results showed human/pilot error remained the most frequent (>90%) but declined slightly over time, while mechanical/system issues increased and weather references decreased significantly. Runway excursion mentions increased modestly, whereas fuel, stall, and wildlife remained stable. Co-occurrence analysis revealed strong associations between human and system factors, as well as secondary links between weather, stall events, and runway excursions. Sentiment analysis confirmed consistent neutrality (polarity M = 0.02–0.03; subjectivity M ≈ 0.35). The findings demonstrate both persistence and gradual shifts in narrative framing, providing evidence of evolving investigative emphases and establishing a foundation for applying natural language processing in aviation safety research.