The objectives of this paper were to document how accident data are usually
classified, whether this system makes it possible to classify all the data
contained in the accident reports, and to examine the classification probl
ems encountered. The first part reviews the variables retained and descript
ors used by the accident studies published over the past 10 years. This syn
opsis showed that the types of data considered and the manner in which they
were classified varied greatly between the studies. Data on the accident c
ircumstances (e.g. activity and incidents) were seldom considered, while ac
cident and injury data were extensively analyzed. The second part analyzes
the vocabulary and data reported by injured handlers in 580 accident descri
ptions. Possible grouping vocabulary strategies were explored and the impor
tance of the implicit nature of data was evaluated. This revealed that the
vocabulary used by the injured was both rich and variable. For example, ove
r 80 terms were used to describe one activity. While some grouping strategi
es to classify data could be developed for the worksite or incidents, it wa
s particularly difficult to identify a logic for grouping activity data. Al
so, the analysis showed that many important data are of an implicit nature.
A literal or automatic classification of terms may, therefore, lead to sig
nificant biases. Furthermore, although data on incidents were frequently re
ported, this type of data is generally disregarded by most accident studies
. Finally, the paper discusses various classification problems that emerged
. (C) 1999 Elsevier:Science Ltd. All rights reserved.