The foundations of normal accident theory were
laid by Perrow (1984) and consolidated by Sagan (1993).
The theory holds that
accidents are a normal consequence of interactive complexity and close coupling
of an organizational system. The measure of interactive complexity is the
number of ways in which components of the system can interact. It represents
the number of variables in the system, the number of relationships between the
variables and the number of feedback loops through which the variables
interact. Typically, interactive complexity increases with the technology
incorporated into the system. The measure of close coupling is the speed at
which a change in one variable cascades through the system to cause changes in
other system variables.
Close coupling represents
tightness in the process, which is influenced by such things as component
redundancy, resource buffers/slack, and process flexibility. The idea behind
normal accident theory is that some of the system responses to change are
unforeseen, are causes of incidents, and can potentially lead to catastrophes.
Using the analogy of safety defenses being like slices of Swiss cheese (Reason,
1997), normal accident theory would say that no matter how high you stack the
slices it is inevitable that organizational.
juggling will cause a set of
holes to line up eventually and the defenses will be breached.
High-reliability theory is a
competing organizational theory of accidents whose proponents such as La Porte
and Consolini (1991), Roberts and Bea (2001), and Weick and Sutcliffe (2001)
believe that, while accidents may be normal, serious ones can be prevented by
implementing certain organizational practices. For example, Weick
and Sutcliffe (2001) suggest that high-reliability organizations implement
business processes to instill “mindfulness” qualities into the organization,
which include preoccupation with failure, reluctance to simplify, sensitivity
to operations, commitment to resilience, and deference to expertise.
Sagan (1993) distils
high-reliability theory down to four essential elements for success: high
management priority on safety and reliability; redundancy and backup for people
and equipment; decentralized organization with a strong culture and commitment
to training; and organizational learning through trial and error, supported by
anticipation and simulation. From the perspective
of normal accident theory, he
argues that the organizational learning required for the success of
high-reliability theory will be restricted for several reasons. These include
ambiguity about incident causation, the politicized environments in which
incident investigation takes place, the human tendency to cover up mistakes,
and the secrecy both within and between competing organizations.
Thus, to promote the
necessary learning, it seems clear that a formal organizational system for
learning from incidents is required. The theory of incident learning relies on
the observation made by Turner (1978) that disasters have long incubation
periods during which warning signals (or incidents) are not detected or are
ignored. Thus, while the occurrence of incidents may be normal, an organization
with an effective incident learning system can respond to these incidents to
prevent serious accidents from occurring in the future.
Incident learning is not
unlike the continuous improvement cycle described by Repenning and Sterman
(2001). An organization effectively implementing a formal incident learning
system may evolve into a high-reliability organization over time.
The theory of incident
learning:
To help understand why
incidents happen, and why we need to learn from them, it is useful to introduce
the concept of a risk system. As shown in Figure 1, it is inseparable
from the business system that generates the useful outputs of the organization.
However, we can gain valuable insights from thinking of them as distinct
systems. Although incidents are actually unwanted outputs of the
business system, it is instructive to view them as outputs of the risk
system. The risk system may be hidden from view, but its outputs are real
enough.
No comments:
Post a Comment