Citation

Lortie M, Rizzo P. Safety Sci. 1999; 33(1-2): 69-85.

Copyright

(Copyright © 1999, Elsevier Publishing)

DOI

unavailable

PMID

unavailable

Abstract

The main objective of the study was to assess how a classification system designed to code slip, trip and fall accidents (STF) could extract data provided in accident reports and the use of the system by the coder. Its specific aims were to evaluate the importance of loss of balance (LOB), other than that related to a slip-trip-fall event (STF), and the impact on statistics of overlapping LOB events, as well as to characterize the nature of the descriptive material, in particular as concerns the implicit nature of events or circumstances reported. Two databases were constituted from an initial group of 1540 accident reports acquired from a large telecommunications company. The classification system provided three codes to cover S, T, and F accidents, as well as 25 subcodes (S: 7; T: 6; F: 12) for classifying STF context. The first database grouped together 378 accident descriptions coded S, T, or F, while the second grouped together 104 accident descriptions reporting an LOB but not coded ST or F. For the three main STF codes and 25 subcodes, all vocabulary or locutions referring explicitly or implicitly to these codes were identified. In 81% of the descriptions for the first database, an S, T, or F corresponding to its resultant coding was identifiable (S: 98%, T: 73%; F: 72%). The study shows that the coder was very reliable and consistent. However, the system as designed led to significant distortions. In particular, the subcodes led to misclassifications. Correspondence between the descriptive data and the subcodes was low for trip events (51%), average for slipping (75%), and almost perfect for fall events (96%). An LOB other than an STF was identified in 23% of the descriptions. Two different LOB events were reported in 32% of the cases. A review of the second database revealed that only 32% of the descriptions could have suited the proposed classification system. The study showed that implicit data were used by the coder. Overall, the combined analysis of both databases showed that slipping events as well as other LOBs were underestimated and that a good number falling events were falsely identified. The statistics appeared to be seriously flawed. A possible strategy for structuring LOB events is proposed.