Journal of Occupational Accidents, 12 (1990) 207-222 207

Elsevier

Behavioural Epidemiology as a Tool for Accident

Research

ANN WILLIAMSON and ANNE-MARIE FEYER

National Institute of Occupational Health and Safety, P.O. Box 58, Sydney 2001, Australia

ABSTRACT

Williamson, A. and Feyer, A.-M., 1990. Behavioural epidemiology as a tool for accident research.
Journal of Occupational Accidents, 12: 207-222.

In this study the incidence and nature of behavioural events preceding work-related fatalities
were investigated. All occupational facilities were examined for the years 1982 to 1984. The events
preceding the fatality and any pre-existing factors that contributed to the fatality were coded. The
coding focussed on such aspects as the environment, equipment design and upkeep, as well as
human factors like work practice, supervision, training, medical precursors and task errors. The
primary events and contributory factors were ranked and weighted in order to establish the casual
relationships between them. Multivariate analysis was used to determine the nature of these re-
lationships across all fatalities.
As this study covers so many types of accidents in such a wide variety of occupational groups,
the results provide more specific information about human behavioural epidemiological method
to evaluate the impact of different aspects of human error on accident causation. This method
will allow highlighting of the most suitable strategies for accident prevention.

INTRODUCTION

Although one of the most well-studied areas of accident analysis, human

factors are among the most difficult to isolate when looking for strategies to
improve accident prevention. There is very little real-world information other
than for specific occupations on what aspect of the human factor warrants
modification. The aviation industry, for example, has provided an often emu-
lated model for accident investigation and analysis. It is often stated that well
over half of aviation accidents are caused by human factors. Similar propor-
tions are quoted for other industries.
Unfortunately such analyses of human factors involvement often do not take
the field much further. Aspects of human factors such as the nature and timing

*Presented at the International Conference on Strategies for Occupational Accident Prevention,

Stockholm, Sweden, 21-22 September 1989.

03786349/90/$03.50 0 1990-Elsevier Science Publishers B.V.

of errors and the involvement of poor work practices are sometimes dealt with
but very few studies have investigated the network of relationships between
human and other factors or within the category of human error. Consequently
accident epidemiology has not fully realised its potential value for accident
prevention.
In the development of strategies for prevention of accidents, it is essential
to gain an understanding of how and why accidents occur. Statistics are gath-
ered on accidents and deaths in the workplace as an acknowledged starting
point for preventive strategies. They usually provide information about the
time and place of the accident, the location and nature of the injury and some
categorisation of the nature of the event directly leading to the injury (Harri-
son et al., 1989; Oleske et al., 1989). Unfortunately, the kind of data available
to be collected, often provides only limited information on the causes of the
accident.
Questions about how and why accidents occur are difficult to answer. Data
is not always available that allows these factors to be determined. Even where
many of the events and circumstances surrounding the accident are known,
reports of the accident can be biased by the subjective and memory-dependent
nature of any reporting system.
This is especially true for the determination of the extent of human involve-
ment in accident causation. Studies of the role of human error in accident
causation reflect this problem. Some studies have merely estimated the extent
of the contribution of behavioural factors to the accident (e.g., Schuckburgh,
1975), thereby providing little to the development of preventive strategies.
Other studies take the opposite tack and develop complex models of the causes
of either specific accidents (Corlett and Gilbank, 1978), or for specific occu-
pational settings such as the aviation industry (Feggetter, 1982; Gerbert and
Kemmler, 1986). These more complex approaches constitute attempts to ex-
amine the influence of human error in more detail and in the context in which
they occur. Their limitations, however, arise from relying heavily on subjec-
tively collected information and from their narrow scope on the problem of
clarification of the nature of human involvement in accident causation.
This study was an attempt to use a mixture of simple factual data and com-
plex (subjective) reports collected from the Coroner’s records on all occupa-
tional fatalities occurring in Australia over a three year period. The study aimed
to gain an understanding of the role of human behaviour in accident causation,
particularly human error.
Of all the aspects of accident analysis, the assessment of human error is
probably the most prone to subjective bias. However the use of Coroner’s rec-
ords considerably reduces subjectivity, since they record accounts of the acci-
dent from as many sources as possible.
Reductions in subjectivity, however, also reduce the information available
to classify error. This presents something of a problem. Without information
regarding “intention to act”, which of course is notoriously subjective, it is
difficult to determine which component(s) of behaviour was faulty. Error clas-
sification systems have been developed which recognise this problem and focus
on other more context-related aspects such as the task or the general behaviour
required by the task. There are a number of reviews of error classification
systems (Hale and Glendon, 1987; Miller and Swain, 1987 ) .
Practical application of classification systems can be a problem. The amount
of information available about the circumstances surrounding an accident will
govern the level and type of error classification assessments possible. For ex-
ample, post hoc assessments of decision-making errors can only be made with
certainty when there is evidence of conscious thought about the behaviour in
error. As this information is not usually available except through interview
with the accident victim (which introduces subjective bias) the classification
of decision errors can become very unreliable.
TWO well-known error classifications were used in this study in order to
assess the ease with which they can be used for the kind of data available in
this study and their ability to reflect aspects of human behaviour in a form that
can be used to direct preventive action. One classification focussed on task-
related errors (Swain, 1963) and the other on errors occurring during different
levels of behavioural function (Rasmussen, 1982). Task-related errors in-
cluded omissions and commissions. The classification divides further into five
types of commission, however to code them requires more specific knowledge
of the task than is available in most studies or general accidents.
The behaviour-based error classification is seen as a continuum ranging from
skill-based in which behaviour is automatic and does not require conscious
thought or control to rule-based in which behaviour is generated through the
application of memorised rules to knowledge-based behaviour which is gener-
ated by problem solving and interpretation of pieces of existing information.
This classification has overlap with other classifications. Other behaviour-based
classifications have named the error rather than the “level” of behaviour at
which it occurs. A number of authors (Reason, 1979; Norman, 1981) have
classified errors during skill-based behaviour as slips and those occurring dur-
ing rule and knowledge-based behaviour as mistakes.

METHOD

The study population was based on all work-related traumatic deaths occur-
ring throughout Australia in the years 1982-1984. The study population was a
subset of fatalities identified by the Study of Work-related Fatalities con-
ducted at the National Institute of Occupational Health and Safety. The data
were collected from Coroner’s records as described by Harrison et al. ( 1989 ) .
Of the 1738 fatalities judged to be work-related by Harrison et al. (1989)) 1020
210

were included in the current study. Cases (718) were excluded on the following
grounds:
(i ) case not in active work at the time of the fatality;
(ii) case was not of working age, that is, below 15 years or above 65 years of
age;
(iii) death occurred during journeys between home and workplace;
(iv) deaths due to road traffic accidents when the person’s normal workplace
was not in or on the vehicle itself;
(v) death occurred when person was on recess from active work or was a
bystander to work.
In addition, 130 cases were excluded because they did not contain enough
information to be coded.
Reports from the Coroner’s records were relatively comprehensive. They
contained a written account of the circumstances of the accident and subse-
quent fatality as well as demographic information about the person, medical
information regarding the nature of the injuries and cause of death and details
of all investigations into the causes of death.
A coding instrument was developed to establish from the Coroner’s records
the sequence of events immediately leading to the fatality. The instrument was
designed to allow coding of up to three temporally contiguous events immedi-
ately preceding the accident which led to the fatality as well as any further
factor occurring at an earlier time that made a direct contribution to the oc-
currence of the accident or fatality. These were titled precursor events (abbre-
viated to PEl, PE2 or PE3, depending on location in the sequence with respect
to the accident) and contributing factors, respectively. For purposes of coding,
time before the accident referred to the relative sequential order of the events
leading up to the fatality. Conventional units of time were therefore variable.
In other words, PEl was closest to the accident and PE2 occurred earlier than
PEl in the accident sequence but the time separating the events could have
varied from seconds to minutes to hours or perhaps even days (see Fig. 1).
The nature of each of the three possible precursor events was coded in one
of the following categories:
(i) Environmental - events resulting from the location of the accident and
are events that could not have been changed at that point in time, e.g., the roof
of a mine falling in.
(ii) Equipment - events resulting from breakages or malfunctions of ma-
chinery or tools that occur at that point in time.
(iii) Medical - events resulting from the person’s current state of physical
well-being.
(iv) Behauioural - events resulting from direct human involvement.
A category was coded in terms of the certainty of evidence for its existence,
with 1 = yes, certainly; 2 = yes, probably; 3 = yes, possible; 4 = indeterminant;
5=no.
 211

Contributinq ~IRO~~
Factor EQUIPHEZNT ACCIDENT
WORK PRACTICE t MEDICAL

Contributing
Factor
TASK ERROR t

Contributinq
Factor
OTHER t
I I

Fig. 1.Schematic representation ofeventsequence used for coding accidents.

Behavioural events were coded further in terms of who performed the be-
haviour (the deceased or another person), and whether the behaviour consti-
tuted an error (i.e., incorrect performance of standard operating procedure 1.
Errors were coded further in terms ofi
 (i) Swain’s (1963) task-related error classification of the following:
(a) omissions - omitting a step in a task
(b) commissions - doing the task incorrectly (includes selection errors, time
errors, qualitative errors).
We have called this classification error type A.
(ii) Rasmussen’s model (1982) of levels of behaviour, which were coded as:
(a) Skill-based behaviour - errors occurring when stimuli are assigned to
responses in a rapid automatic mode; e.g. highly practiced routines (also
called slips by Reason (1979) ) .
(b) Rule-based behaviour - errors occurring when actions are selected by
making decisions on appropriate behaviour from a hierarchy of “learned
rules”, e.g., “If X occurs, then do Y”.
(c) Knowledge-based behaviour - errors occurring when entirely new prob-
lems are encountered for which neither rules nor automatic mappings
exist, e.g., trouble shooting strategies.
We have called errors occurring during these types of behaviours, error type B.
(iii) Decision-making errors - occurring during conscious evaluation of the
situation, i.e., rule or knowledged based behaviour. Coded according to cer-
tainty of evidence as described above.
(iv) Skill or training errors - occurring from inadequate training of person-
nel. Coded according to the certainty of evidence as above.
The nature of the contributing factors was coded in the following categories:
(i) Environmental - factors resulting from the location of the accident oc-
curring at an earlier time.
(ii) Equipment - factors associated with the design of machinery, tools, per-
sonal protective equipment or safety equipment.
(iii) Work practice - factors involving poor or risky standard operating pro-
cedures accepted by management and/or personnel. This included separate
categories of poor upkeep or misuse of equipment.
(iv) Supervision - factors relating to inadequate charge of workers.
(v) Training - factors relating to inadequate training of workers.
(vi) Task error - factors relating to incorrect performance of study.
(vii) Medical - factors involving physical well-being at an earlier time.
(viii) Other - factors not coded in the above categories, including alcohol/
drug involvement, delays in receiving medical treatment and social factors.
Coding for contributing factors was again in terms of the quality of evidence
for the existence of the contributing factor.
The contributing factors and events leading up to the fatality were ranked
in terms of their significance for the occurrence of the fatality. The chain (or
chains) of events and factors were ranked such that the root cause was deemed
as the most significant (scored a “1”) and occurrences following directly from
the root were ranked by diminishing degrees as significance decreased (i.e., 2,
3,4, etc.).
 213

Information was also collected on demographic details, details of the per-

son’s task at time of death, the nature of their usual job (if different), the use
of personal protective and other safety equipment and the degree of forewarn-
ing of the accident. These results will not be discussed in this paper, but will
be the subject of subsequent ones.
The coding instrument was developed by 3 coders test-coding a random sam-
ple of 30 cases. The results were then discussed and the coding instrument was
modified to balance the scope of information collected with the quality of in-
formation available. Once the instrument was finalised, a single coder coded
all cases.
The accuracy of coding was checked by two additional coders coding 40 ran-
domly selected cases. The reliability of coding was checked by the main coder
recording 40 cases, 6 months after the original coding. These results were dis-
cussed in Feyer and Williamson (1989 ) .

RESULTS

Behavioural factors were involved in 91.2% of fatalities. For this analysis,

behavioural factors included cases where either human error or poor work
practices (including poor supervision or poor training) were precursors of the
fatality. In 62.4% of cases (69.5% if probable and possible errors were in-
cluded), some kind of behavioural error was made. In 42.0% of fatalities, un-
safe work practices were a major factor leading to the fatality (42.7% if prob-
able and possible errors were included).
Coding of behavioural events was relatively unambiguous. Where probably
and possible categories were used, they were used most frequently in the de-
scription of errors. Furthermore, error coding was indeterminate in only 6.9%
of cases.
Task-related or error type A classification revealed that commissions were
about three times more frequent than omissions. The error type B or behaviour
based classification system showed that errors during skill-based behaviour
were most common (54.8% of errors) compared to considerably fewer errors
during rule and knowledge-based behaviour (13.6% and 13.7%, respectively).
The ambiguity of these measures was also low, with 0.1% and 9.3% indeter-
minant codings for type A and B classifications, respectively.
Decision errors in contrast were infrequent. Only 16.4% of cases involving
error were coded with certainty as decision errors. A similar percentage (13.7% ),
however, were coded as possible or probable decision errors and a further 59.2%
could not be coded due to insufficient data.
Only 14.3% of cases involving error were directly attributable to the actions
of another worker. Most errors were due to the subject themselves (83.7% ). In
only 1% of cases was it not possible to code this information.
214

Behuvioural involvement as a function of time before the fatality

The number of codable events decreased with increasing time away from the
fatality. While all cases had at least one event, just over half had two codable
precursor events and less than one-fifth of cases involved three events. Most
cases on the other hand, involved at least one type of contributing factor. The
extent of behavioural involvement showed the same pattern. The proportion
of errors, however, remained very similar across events but was markedly lower
for contributing factors (see Table 1) .
Error type varied with time relative to the fatality. Commissions were much
more frequent that omissions and remained so for all events and for contrib-
uting factors. Omissions increased across events away from the fatality. That
is, between precursor event 1 (PEl ), the event immediately preceding the ac-
cident, omissions were about twice as common as they were for both precursor
event 3 (PE3) and errors occurring as contributing factors (X2,,,=2.15,
p<O.OOl) (see Fig. 2).
For the error type B classification (see Table 2) errors occurring during skill-
based behaviour were twice as frequent for the event leading directly to the
fatality compared to earlier events. Errors during rule and knowledge-based
behaviour on the other hand, showed the opposite tendency, occurring much
more frequently earlier in time with respect to the fatality (X2c6)= 145.7,
p < 0.001). The number of errors that could not be classified using error type
B classification also increased with increasing time away from the fatality.
Almost 50% of errors occurring as contributing factors could not be coded with
a reasonable degree of certainty.
Decision-making errors showed a similar trend to rule and knowledge-based
errors. Few occurred in the event most contiguous with the accident leading to
the fatality but the incidence of such errors increased with time away from the
fatality.

TABLE 1

Involvement of behavioural factors (error/no error) by time before fatality

Precursor event Contributing

factors
PE1 PE2 PE3

Behavioural involvement 42.7 35.2 13.3 9.5

expressed as % of total
sample
Errors expressed as % of 79.4 82.0 88.2 16.5
behavioural events at
that time
 215

n OHISSIONS

( m COHMISSIONS 1

9: ERRORS OCCURRING 50
FOR THAT
EVENT/FACTOR 40

20

10 i

0 l- + +
CONTRIBUTING
PEl PEZ PE3
FACTORS

Fig. 2. Results for the task-related classification of errors (error type A) for errors occurring at
each precursor event (PE) and for errors occurring as contributing factors.

TABLE 2

Analysis of error type B results for each precursor event and contributing factor showing the
percentage of cases making each kind of error at each point in the accident sequence

Nature of error Precursor event Contributing

factors
1 2 3

Slips 75.8 32.9 35.0 13.4

Rule-based 6.6 27.1 22.5 23.7
Knowledge-based 11.5 21.9 22.5 16.5
Unknown 6.0 17.8 20.0 46.4

Similarly, errors occurring at an earlier time were much more likely to have
been due to another person than errors occurring just before the fatality
(X2 C3)=148.8,p<O.O01).

The influence of other factors on the production of an error

The involvement of earlier events on the production of errors was investi-

gated for errors occurring in each of the precursor events in the accident
sequence.
Figure 3 shows the major patterns of categories of events leading to an error
at PEl. A significant proportion of errors occurring at this stage (85.2%)were
preceded by one of two sequence patterns. The pattern was either, no other
precursor events in the sequence (65.1% ), or the PEl error was preceded by a
 216

\
environmental
(20.3%)
J
\ f >f
’ workpractice
(35.4%) -- no PE3 (65.1%) - no PE2

\ / \ J L

u environmental

[;zce
(21.9%)

fno
I

PE3 (20.1%+vtoUra, j (X]

>
environmental

behavioural ,
(3.3%)

Contributing _ PE 1
PE3 - PE2 -
Factors

Fig. 3. Patterns of event sequences preceding error at precursor event 1 (PEl). The bracketted
number under each PE refers to the percentage of total cases represented by the sequence to that
point. For the contributing factors, the bracketted number refers to the percentage of cases rep-
resented by the PE sequence which also showed involvement of that contributing factor.

single behavioural event at precursor event 2 (PE2, 20.1% ). Both patterns

were most commonly preceded by one or more of three categories of contrib-
uting factors; Environmental, Work practice or Other in about the same pro-
portions. The remaining 14.8% of errors occurring at PEl showed a range of
patterns of precursors.
For errors occurring at PE2 (see Fig. 4) two basic patterns of preceding
events accounted for 85.3% of cases. The most common pattern (75.7% of
 217

environmental

BEHAVIOURAL 1 behavioural 1
ERROR AT PE2 t 1 PEl
I

no PE3 (3.4%) equipment PEl

ERROR AT PE2

Sequence of
events (n=292)

workpractice
(35.2%) no PE3 (49.7%)

Contrlbutlng
Factors - ‘IZ3 -
PE2 - PEl

Fig. 4. Patterns of event sequences preceding error at precursor event 2 (PE2 ). The bracketted
number under each PE refers to the percentage of total cases represented by the sequence to that
point. For the contributing factors, the bracketted number refers to the percentage of cases rep-
resented by the PE sequence which also showed involvement of that contributing factor.

cases) involved PE2 as the final event in the accident sequence but with con-
tributing factors occurring earlier in time. The most common contributing fac-
tors fell again into the Work practice, Other and Environmental categories.
This pattern did, however, divide into two slightly different sequences follow-
218

behavlOurZ3l PEl

sequence Of
even** (n=120) I

Contributing
Factors ___, PE3 ____, PE2 -_+ PEI

Fig. 5. Patterns of event sequences preceding error at precursor event 3 (PE3). The bracketted
number under each PE refers to the percentage of total cases represented by the sequence to that
point. For the contributing factors, the bracketted number refers to the percentage of cases rep-
resented by the PE sequence which also showed involvement of that contributing factor.

ing the error at PEZ. In one sequence it led to another behavourial error (26.0%)
just before the fatality and, in the other, PE2 error led to an environmental
event at PEl (49.7%).
The second main pattern of events preceding an error at PE2 involved an-
 219

other behavioural event at PE3 preceded by at least one of the contributing

factors, Work practice, Environmental or Equipment. For both patterns pre-
ceding PE2 errors, the most common contributing factor was Work practice.
The patterns for the remaining 14.6% of cases with PE2 errors consisted
mainly of either 3 behavioural events preceded by one or more of Work prac-
tice, Supervision, Training and Other contributing factors (3.4% ) or an event
involving equipment at PEl following the error at PE2 which was preceded by
one or more of up to 6 contributing factors (3.4% ).
For errors occurring at the beginning of the designated accident sequence
(i.e., PE3) four main patterns were revealed, accounting for 83.4% of cases
(see Fig. 5). In all of these patterns the error at that time led to the fatality
through either further errors or behavioural events immediately preceding the
fatality. There was a wide range of contributing factors preceding the PE3
error. Each of the four patterns had 3 or 4 categories of contributing factors,
with the Work practice and Other categories being common to all and occur-
ring most frequently in all. The environmental category was represented in 3
of the 4 patterns. The categories, Supervision, Training, Equipment and Task
error appeared only once in the patterns. The remaining 16.6% of cases with
PE3 errors showed no particular patterns of occurrence.

DISCUSSION

This study has shown that about two-thirds of occupational fatalities in-
volved human error. This is an important starting point for the development
of preventive strategies, however, it provides only a beginning. To be really
useful for prevention, more information is needed about the nature of the error
and how it was caused. The results of this study provide some insights.
Analysis of error classified in terms of aspects of the task showed that overall
commissions, or incorrect performance of the task, were much more common
than omissions. The frequency of omissions relative to commissions, however,
was greater in the early stages of the accident sequence indicating that things
not done were more likely to exert their influence at a later stage whereas the
incorrect performance of a task was more likely to immediately precipitate a
fatality.
Task-related classifications, however, are not good descriptions of error if
preventive strategies are the aim, since they reveal little about the deficiency
in information processing that produces the error (Hale and Glendon, 1987 ) .
They therefore are best used to describe errors rather than discriminate their
underlying psychological causes, an important prerequisite for predicting when
they might occur.
They can, however, be applied relatively easily. Virtually all errors in this
study could be classified as either commissions or omissions whereas other
classifications, notably decision-errors, were much more difficult to code.
220

The behaviour-based classification system was selected for this study be-
cause it provided some insights into psychological processes and because it was
thought that it could be coded with the type of external information available.
This proved to be the case with fewer than 1 in 10 errors not codable.
This analysis showed that the most common error type was slips or errors
during skill-based behaviour, although their frequency varied with time away
from the fatality. Skill-based errors were more common rule and knowledge-
based errors combined in the event leading directly to the fatality. They be-
come comparatively less frequent with time away from the fatality, such that
errors occurring earlier in time (i.e., contributing factors) were hardly ever
skill-based. The incidence of uncodable errors also increased with time away
from the fatality and particularly for errors occurring as contributing factors.
Not surprisingly less information was available to code errors occurring some
time before the accident.
Clearly the concept of good work design will only correct a certain proportion
of errors as accident precursors. Mistakes, whether occurring during rule or
knowledge-based behaviour, can be designed out of the system. It is possible
to predict, for example, how an untrained operator might make errors or even
how different rules for action could be applied in work situations by trained
operators. It is difficult, however, to predict when a slip might occur. Studies
by Reason (1976, 1984) examining “absentminded” behaviour have shown
that it can occur at any time. Since the next common error type was a slip of
“absentminded” behaviour, it is essential to focus preventive strategy devel-
opment on methods for reducing this type of error.
The most notable characteristic of slips was that they occurred most com-
monly just before the accident itself. It seems that a slip often led most directly
to the accident.
Unfortunately, this is not good news for the development of preventive strat-
egies. That slips often occur immediately before the fatality, thereby allowing
no time for correction or recovery, indicates that a goal for an accident preven-
tion strategy should be the elimination of slips. Their unpredictable nature,
however, makes the elimination of slips unlikely.
Two avenues are available to overcome this problem. One follows the view-
point that errors are an integral part of human behaviour and, as the results
of this study indicate, cannot be designed out of the system (Rasmussen, 1983;
Wickens, 1984). The solution therefore, is to create work environments that
are flexible enough to allow recovery from errors.
The second avenue is to avoid the circumstances and factors that precede
the error. A striking feature of this study was that comparatively few patterns
of causation accounted for most cases involving error. Only seven major pri-
mary events patterns emerged, all of which involved only additional behav-
ioural events and/or events due to the location of the person (environmental).
Even more striking was the finding that for almost all patterns the most com-
 221

mon contributing factors were from the Work Practice, Other and Environ-
ment categories. Unsafe work practices were the most frequent and influential
factors for virtually all patterns. The Other category was the next most com-
mon but was not as influential as safe work practices. The major component
of this factor was alcohol and drug use. It appears that the same basic set of
contributing factors are associated with all types of errors occurring at any
place in the accident sequence. While this stage of the analysis of this data set
does not allow precise determination of causal relationships, it does indicate a
useful starting point for prevention. Knowing that unsafe work practices and
to a considerably lesser extent, alcohol and drug use, are the strongly associ-
ated precursors of errors provides more clearly defined targets for prevention.
This study has provided a broad view of the role of human error in accident
causation. Further analysis will aim to examine specific aspects in detail such
as the nature and effects of error in different industries and in different indus-
trial processes.

ACKNOWLEDGEMENTS

We are very grateful to Debbie Worthington for her work in coding fatalities.
We would like to thank Dr. Michael Frommer and Dr. James Harrison for
allowing access to the study of Work-related Fatalities data.

REFERENCES

Corlett, E.N. and Gilbank, G., 1978. A systemic technique for accident analysis. J. Occup. Accid.,
2: 25-38.
Feggetter, A.J., 1982. A method for investigating human factor aspects of aircraft accidents and
incidents. Ergonomics, 25: 1065-1075.
Feyer, A-M. and Williamson, A.M., 1989. An accident classification system for use in preventive
strategies. Paper presented at the 1st World Conference on Accident and Injury Prevention,
Stockholm.
Gerbert, K. and Kemmler, R., 1986. The causes of causes: Determinants and background variables
of human factor incidents and accidents. Ergonomics, 29( 11): 1439-1452.
Hale, A.R. and Glendon, AI., 1987. Individual Behaviour in the Control of Danger. Elsevier,
Amsterdam.
Harrison, J.E., Frommer, MS., Ruck, E.A. and Blyth, F.M., 1989. Deaths as a result of work-
related injury in Australia 1983-1984. Med. J. Aust., 150: 118-125.
Miller, D.P. and Swain, A.D., 1987. Human factors and human reliability. In: G. Salvendy (Ed.),
Handbook of Human Factors. Wiley, New York.
Norman, D.A., 1981. Categorization of action slips. Psychol. Rev., 88(l): 1-15.
Oleske, D.M., Brewer, R.D., Doan, P. and Hahn, J., 1989. An epidemiologic evaluation of the
injury experience of a cohort of automotive parts workers: A model for surveillance in small
industries. J. Occup. Accid., 10: 239-253.
Rasmussen, J., 1982. Human errors. A taxonomy for describing human malfunction in industrial
installations. J. Occup. Accid., 4: 311-333.
222

Rasmussen, J., 1983. Skills, rules and knowledge; signals, signs and symbols; and other distinctions
in human performance models. IEEE Trans. Syst., Man Cybern., SMC-13: 257-266.
Reason, J.T., 1976. Absent minds. New Society, November 4,244-245.
Reason, J.T., 1979. Actions not as planned: The price of automation. In: G. Underwood and R.
Stevens (Eds.), Aspects of Consciousness, Vol. 1, Academic Press, London.
Reason, J.T., 1984. Lapses of attention in everyday life. In: R. Parasuraman and D.R. Davies
(Eds.), Varieties of Attention. Academic Press, New York.
Schuckburgh, J.S., 1975. Accident statistics and the human-factors element. Aviat., Space Envi-
ron. Med., 46: 76-79.
Swain, A.D., 1963. A method for performing a human factors reliability analysis. Monograph
SCR-685, Schandia National Laboratories, Albuquerque, NM.
Wickens, C.D., 1984. Engineering Psychology and Human Performance. Charles Merrill, Colum-
bus, OH.