Errors: An Inherent Part of Human-System Performance - Foundations for Designing User-Centered Systems - page 293

Crash at
T-junction

To p
Event

p = .001

1 in 10
1 in 100
1 in 1,000
1 in 10,000
1 in 100,000
1 in 1M

Frequent
Probable
Occasional
Remote
Improbable
Extremely
Remote

AND

Car on
main road

Car on side road fails
to stop

p = .01

p = .131

OR

Car on side road
did not stop

Car on side road
could not stop

OR

OR

p = .12 p = .011

Driver's
vision
blocked

Car
tyres
worn

Road
too
slippery

Car
brakes
failed

Driving
too fast

Driver
too ill

p = .1 p = .01 p = .01 p = .01 p = .001 p = .0001

Fig. 10.4 A quantified fault tree showing the likelihood of a crash occurring at a given road
junction (reproduced with permission from the Institution for Engineering and Technology)

antecedent for the specific consequent (rather than the error mode). As the process
continues, a structure that looks like a tree of antecedents and consequents is built
up. Note that these are really the possible antecedents and possible consequents.
Some of them may be ruled out by what really happened during the event.
Using the CREAM to analyze errors can become quite involved, and is beyond
the scope of this topic. Hollnagel's ( 1998 ) book includes a good example of an
analysis of a railway accident that occurred in New York in 1995 which is worth
looking at, for those who are interested. Although the CREAM has not been
widely adopted—partly because of a lack of tool support 1 —it does offer a nice
illustration of how taking a different view of errors can generate new insights.

10.4.4 THEA

The Technique for Human Error Assessment (Pocock et al. 2001 ) assumes, like the
CREAM, that the context in which actions are performed is one of the major
influences on human performance. Like the CREAM, THEA is an iterative

1

Although there is a browser based tool at http://www.ews.uiuc.edu/*serwy/cream/ .