2.2 Human Error
2.2.1 Action Models
Actions are performed by identifying an intention, which is broken down into individual acts. The acts form a sequence that begins and ends as required to complete the action. While an action is underway, activity is monitored, and feedback is assessed to determine if intentions are being met. When an action deviates from an intention, an error has occurred (Norman, 1981).
This description of action comes from Norman’s Action-Trigger-Sequence system (Norman, 1981), one of a cluster of models (Norman, 1981; Norman & Shallice, 1986; Reason, 1984) formulated to explain how “systematic” or “predictable” varieties of human error arise (Reason, 1990, p. 36). The models were developed by interpreting and compar- ing accounts of everyday activity.
For example, in categorising slips of action, Norman analysed a thousand incidents that included his own collection of accounts, and a compilation of similar incidents from other researchers. The incidents used by him in the analysis were recorded immediately after the occurrence, either by the person who made the error or by an observer. Reason’s work with slips of action resulted in a behavioural classification of error categories, a theoretical action model, and a set of hypotheses about the cognitive mechanisms that fail when action slips occur (Reason, 1984). His data comprised 625 slips of action compiled out of catalogues developed in two studies. One study collected sixty-three diaries over seven days that included information about what happened when deviations in action were discovered and the completion of a set of standard questions that contextualised individual occurrences.
The Action-Trigger-Sequence system depicts action as a linear, horizontal sequence. It represents how people perform well-learned, habitual actions using pieces of stored knowledge stored that “direct the flow of control” of motor activity (Norman, 1981, p. 4). Norman is referring in this description to schema, a term made familiar in computer science through the work of Minsky (Brewer, n.d.). Reason situates schema within psychology as higher-order, generic cognitive structures underlying all aspects of human knowledge and skill. Their workings are not consciously experienced, but they “lend structure” to perceptual experience and to the information that is stored or retrieved from memory (Reason, 1990, p. 35).
A person selects and activates an action schema when the current state matches the conditions under which it should be activated, but this is dependent on the perceived quality, or “goodness” of the match (Norman, 1981, p. 14). Actions may be initiated by environmental input, previously activated sequences or by thoughts, memories, and competing aims. Slips of action, or errors, occur because multiple sources of activation are possible and conditions are variable.
Intention
Intentions define actions. Without intention, there can be no selection of acts, no corre- sponding activity, and no assessment of completeness or correctness. Intentions are the result of "many considerations", including personal goals, decision-making and problem- solving (Norman, 1981, p. 5). Naturally, some errors arise in forming intentions. Norman's analysis of slips considered only errors, and by extension actions, for which an intention was stated. However, his scheme also represented errors of intent, such as performing a reasonable action in the wrong environment or forming the wrong intention because of incomplete information.
An intention has two components: the expression of the desired “end-state”, and indications of how it is to be achieved (Reason, 1990, p. 5). Different actions require differently specified intentions. Small everyday actions become routine over time and do not require explicit specification. By contrast, a novel or ill-learned action requires greater intentional specificity until it too is repeated enough to become routine. In assessing activity, actions that did not meet prior intentions or were not properly executed are erroneous.
Stated again, actions are at times so well understood and familiar that they can be performed automatically (Norman & Shallice, 1986). They are routine, habitual. They arise out of intentions that can be clearly stated and broken down into a series of physical
acts (Norman, 1981). Their familiarity “invokes” well-specified expectations (Sellen, 1994, p. 486).
Attention
Attention is paid to ensure that intentions are being met. This is done by comparing original intent —what one meant to do— with information or feedback. Comparisons are made between information and expectations, that is what one expects to happen. (Norman, 1981; Reason, 1990). Information thus may come from internal sources, as in statements of intent or expectation, or external sources, as in the effects or outcomes that are produced when activities are undertaken.
Attention is variously described as leading to error, as preventing error, as necessary for diagnosis and for forming intention. Paying attention too closely to simple tasks can lead to errors, as can paying too little attention at key moments (Reason, 1984).
Conscious Control
Periodic attention is used to monitor routine tasks, however, it is not always sufficient. At other times attention is commanded, it must be “close and labored”, so that consequences of actions can be assessed (Reason, 1984, p. 516). Activities that command attention are often novel. This may be because they are not as well understood by the performer (Norman & Shallice, 1986), or arise out of “new” circumstances or unfamiliar sequences that generate unpredictability (Sellen, 1994, p. 486).
There are other special conditions in which “heightened awareness” or conscious control is required: when plans must be made, decisions taken or errors must be corrected. As noted, it is needed for tasks that are not well-learned or have novel sequences, but also for those deemed to be difficult or dangerous, or for actions that counter strong habitual responses (Norman & Shallice, 1986, pp. 2, 8).
Norman and Shallice accounted for conscious command within the Attention to Action model (1986). In this model, the supervisory attention system manages activity by drawing on multiples sources and types of information, including past and present states of the environment, of intentions, and awareness of prior actions and outcomes. It depends upon will. Will must be exerted to meet intentions, even if it means performing actions that one does not want to do. The exertion of will requires attention, but also “conscious knowledge” of the particular end to be met. Norman and Shallice suggest that this knowledge must be formed before conscious control is exerted.
The model of human action given by Reason likewise gives emphasis to the force of
needs in regulating action. Needs are the “motivational springs” of human action. In
agreement with Norman and Shallice, Reason argued that attention or deliberate control must be exerted differently when intentions are in danger of not being met. For Reason, such moments are those in which intention assumes control as the “chief executive”, responsible for organising plans, monitoring and guiding activity (Reason, 1984, p. 533).
The following sub-sections present typologies developed using theories of performance that are action-based. The first reiterates slips of action developed by Norman and Reason. The next sub-section describes Rasmussen’s skill-rules-knowledge performance framework. Two models developed by Reason that combine slips of action with Ras- mussen’s levels of performance are described in Section 2.2.3.3. Finally, a model of action and error developed using Action Theory to describe organisational practice is summarised.