Errors: An Inherent Part of Human-System Performance - Foundations for Designing User-Centered Systems - page 287

there were problems. The operators either knew or could guess the purpose of the
study. As a result they had a higher than normal expectation that an abnormal
situation would arise during a simulator session. The sessions were also usually
limited to 1 h rather than the duration of a normal shift.
If you are designing an interface for a nuclear power plant control system, for
example, you may regard ecological validity as being of paramount importance,
and therefore decide against the use of laboratory experiments. Similarly, the hit
and miss nature of using field observation for relatively infrequent events mitigates
against its use, except perhaps in longitudinal studies or where unusual resources
are available. Archive data, on the other hand, naturally have high ecological
validity—a judicious choice of data source should make it possible to gain access
to the hundreds of instances of human error required for analysis (Baxter 2000 ).
Ultimately your choice of method will depend on the weights you give to the
pros and cons of the individual methods. In addition, however, you should make
sure that you take into account the availability and access to the resources you will
need to carry out your study.
10.3 Error Taxonomies
Most scientific study is underpinned by a well-defined classification system or
taxonomy of the relevant phenomena. The taxonomy provides a frame of reference
for the study, and enables other researchers to evaluate the results of that study.
Although several human error taxonomies have been developed, there is no uni-
versal taxonomy that serves all the various purposes of error research (Senders and
Moray 1991 ). Below we consider three well known examples.
The critical factor in generating a taxonomy of erroneous behavior is the choice
of level of abstraction to use for categorization. Determining an appropriate level
of abstraction requires that the purpose of the research be clearly defined. A useful
rule of thumb is to model behavior at a level that allows remedies to be generated
to facilitate the avoidance of a repetition of the same type of error in similar
circumstances in the future.
10.3.1 The Technique for Human Error Rate Prediction
The taxonomy used in the Technique for Human Error Rate Prediction (THERP,
Swain and Guttman 1983 ) is based around the commonly used notions of errors of
omission and commission. In this taxonomy, actions can either be:
• Correct.
• Errors of omission: actions that are omitted. It may be difficult, however, to
determine whether an action has been omitted or has just been delayed for a long
time.