by James Reason
The three basic error types may be distinguished along a variety of task, representational and processing dimensions. Different dimensions yield different lines of demarcation; the case for three basic error types can only be made on the basis of the total pattern of distinctions.
Table 3.1. Relating the three basic error types to Rasmussen’s three performance levels.
Performance level
Error type
Skill-based level
Slips and lapses
Rule-based level
RB mistakes
Knowledge-based level
KB mistakes
2.1. Type of activity
A key distinction, based upon Rasmussen’s performance levels, is the question of whether or not an individual is engaged in problem solving at the time an error occurred. Behaviour at the SB level “represents sensorimotor performance during acts or activities that, after a statement of an intention, take place without conscious control as smooth, automated, and highly integrated patterns of behaviour” (Rasmussen, 1986, p. 100). Although it can also be invoked during problem solving to achieve local goals (since all three levels of performance can, and often do, occur concurrently), such behaviour is primarily a way of dealing with routine and nbnproblematic activities in familiar situations.
Both RB and KB performance, on the other hand, are only called into play after the individual has become conscious of a problem, that is, the unanticipated occurrence of some externally or internally produced event or observation that demands a deviation from the current plan. In this sense, SB slips generally precede the detection of a problem, while RB and KB mistakes arise during subsequent attempts to find a solution. Thus, a defining condition for both RB and KB mistakes is an awareness that a problem exists.
2.2. Focus of attention
A necessary condition for the occurrence of a slip of action is the presence of attentional ‘capture’ associated with either distraction or preoccupation (Reason, 1979, 1984a). This means that wherever else the limited attentional resource is being directed at that moment, it will not be focused on the routine task in hand. But in the case of both RB and KB mistakes, we can be reasonably sure that the limited attentional focus will not have strayed far from some feature of the problem configuration.
2.3. Control mode
Both SB slips and RB mistakes share a predominant mode of control that is absent from KB mistakes. Performance at both the SB and RB levels is characterised by feedforward control emanating from stored knowledge structures (motor programs, schemata, rules). Rasmussen summarises this feature of the SB level as follows: “performance is based on feed-forward control and depends upon a very flexible and efficient dynamic internal world model” (Rasmussen, 1986, p. 101). Comparable control mechanisms operate at the RB level: “performance is goal-oriented, but structured by feed-forward control through a stored rule. Very often, the goal is not even explicitly formulated, but is found implicitly in the situation releasing the stored rules. The control is teleologic in the sense that the rule or control is selected from previous successful experiences. The control evolves by the survival of the fittest rule” (Rasmussen, 1986, p. 102).
Control at the KB level, however, is primarily of the feedback kind. This is necessary because the problem solver has exhausted his or her stock of stored problem-solving routines, and is forced to work ‘on-line’, using slow, sequential, laborious and resource-limited conscious processing. The focus of this effortful functional reasoning will be some internalised mental model of the problem space. This proceeds by setting local goals, initiating actions to achieve them, observing the extent to which the actions are successful and then modifying them to minimise the discrepancy between the present position and the desired state. It is, in essence, error-driven.
Another way of characterising these differences is in relation to the cognitive structures identified in Chapter 2. Errors at the SB and RB levels occur while behaviour is under the control of largely automatic units within the knowledge base. KB errors, on the other hand, happen when the individual has ‘run out’ of applicable problem-solving routines and is forced to resort to attentional processing within the conscious workspace.
2.4. Expertise and the predictability of error types
There is a good deal of evidence (see Sections 4 and 5 of this chapter) to show that at both the SB and the RB levels, errors are likely to take the form of ‘strong-but-wrong’ routines. At the SB level, the guidance of action tends to be snatched by the most active motor schema in the vicinity’ of the node at which an attentional check is omitted or mistimed. Similarly, the most probable error at the RB level involves the inappropriate matching of environmental signs to the situational component of well-tried ‘troubleshooting’ rules. In both cases, the error forms are already available within the individual’s stored repertoire of knowledge structures. But the same is not true for errors at the KB level. When the problem space is largely uncharted territory, it is less easy to specify in advance the short-cuts that might be taken in error. Thus, because they arise from a complex interaction between ‘bounded rationality’ and incomplete or inaccurate mental models, KB mistakes will be less predictable in their forms. At best, it is only possible to forecast the general cognitive and situational factors that will conspire to create KB mistakes.
Mistakes at the KB level have hit-and-miss qualities not dissimilar to the errors of beginners. No matter how expert people are at coping with familiar problems, their performance will begin to approximate that of novices once their repertoire of rules has been exhausted by the demands of a novel situation. The important differences between the novice and the expert are to be found at the SB and RB levels. Expertise consists of having a large stock of appropriate routines to deal with a wide variety of contingencies.
There is considerable evidence (Adelson, 1984) to show that in skilled problem solving, the crucial differences between experts and novices lie in both the level and the complexity of their knowledge representation and rules. In general, experts represent the problem space at a more abstract level than nonexperts. The latter focus more on the surface features of the problem. The classic result on the abstract representations of experts was obtained by Chase and Simon (1973), who demonstrated the marked superiority of chess masters in reconstructing meaningful midgame boards after a 5-second presentation. They found that chess masters’ recall clusters frequently consisted of pieces that formed attack or defence configurations. Thus, individual chess pieces were ‘chunked’ as integral parts of larger meaningful units. Comparable findings have been obtained for master Go players (Reitman, 1976), physicists (Chi, Glaser & Rees, 1981), mathematicians (Lewis, 1981) and computer programmers (Adelson, 1981,1984).
Experts, then, have a much larger collection of problem-solving rules than novices. They are also formulated at a more abstract level of representation. Taken to an unlikely extreme, this indicates that expertise means never having to resort to the KB mode of problem solving. More realistically, however, it establishes a close relationship between the predictability of error and the degree of expertise; the more skilled an individual is in carrying out a particular task, the more likely it is that his or her errors will take ‘strong-but-wrong’ forms at the SB and RB levels of performance.
2.5. The ratio of error to opportunity
Virtually all adult actions, even when directed by knowledge-based processing, have very substantial skill-based and rule-based components. These are the ready-made routines of everyday life. It therefore follows that if one counts the errors made during a particular sequence of actions, the absolute numbers of SB errors and, to a lesser extent, RB errors will greatly exceed those specifically due to KB failures, simply because of the greater involvement of SB and RB processing in human performance. Viewed in this way, SB and RB errors will be more abundant than KB errors.
This picture reverses, however, if one considers the relative ratios of error numbers to opportunities for error at each of the three levels of perf
ormance. Skill-based and rule-based processing are the hallmarks of expertise. They are the essence of skilled performance. When expressed as proportions of the total number of opportunities for error at each performance level, we expect that the percentage of errors in the SB and RB modes will be very much smaller than at the KB level of processing, even though their absolute numbers are very much greater.
Consider the task of driving a car. Both the numbers of discrete skill-based actions involved in controlling the vehicle and the individual rule-based decisions needed to negotiate the traffic are several orders of magnitude greater than the number of occasions requiring knowledge-based processing. Of course, these proportions vary widely with the nature of the activity. Some tasks are considerably less routinised than others; but it seems a safe generalisation to assert that all activities are likely to involve greater amounts of SB and RB processing. These are the favoured modes, and the ones at which human beings excel. And skill-based actions are needed for the implementation of any control directive, regardless of whether it comes from a prepackaged problem-solving rule or is the product of effortful on-line processing at the KB level. We will discuss this issue further in the context of error detection (Chapter 6).
2.6. The influence of situational factors
It is further evident from the preceding discussion that errors at each of the three levels will vary in the degree to which they are shaped by both intrinsic (cognitive biases, attentional limitations) and extrinsic factors (the structural characteristics of the task, context effects). In SB slips, the primary error-shaping factors are attentional ‘capture’ and the ‘strength’ of the associated action schemata—where ‘strength’ is, in large measure, determined by the relative frequency of successful execution. All that is required to elicit a ‘strong-but-wrong’ action sequence is the omission (or misapplication) of an attentional check in circumstances where some departure from previous routine was intended or necessary. For RB mistakes, the story is much the same. It is reasonable to assume that rules too are arranged in an ordered priority list (see Payne, 1982; Anderson, 1983), where the most available production system is also the one whose conditional components are most frequently satisfied by the prevailing state indications. In this case, however, we need to know more about the nature of the task in order to predict which rule an individual is most likely to apply in error. With RB mistakes, it is necessary to understand what other rules could be satisfied, either wholly or partially, by the current situational cues, and for this a detailed knowledge of both the task and the person’s training is required.
At the KB level, however, mistakes can take a wide variety of forms, none of which is necessarily predictable on the basis of the individual’s past experience and acquired ‘knowledge stock’. Of particular importance here is the way in which both the task and other situational variables direct the limited attentional resource to relevant or nonrelevant areas of the problem space (see Payne, 1982). Most evidence relating to activity at this level stresses the extent to which cognitive performance depends upon “seemingly minor changes in tasks” (Einhorn & Hogarth, 1981). This is hardly surprising since, in the absence of suitable preprogramming, performance must of necessity be shaped primarily by extrinsic factors.
2.7. Detectability
The claim that mistakes are harder to detect than slips (see Chapter 1) receives clear support from studies in which experienced nuclear power plant (NPP) operators were exposed to a number of simulated plant failures. Woods (1984) reviewed the data from 99 test scenarios, using 23 crews in 8 different failures (or events). Nearly two-thirds of all errors went undetected. Whereas half of the execution failures (slips and lapses) were detected by the crews themselves, none of the state identification failures (RB and KB mistakes) were discovered by unaided crews. Mistakes were corrected only through the intervention of some external agent. This observation accords closely with what actually occurred during both the Three Mile Island (Kemeny, 1979) and the Oyster Creek (Pew et al., 1981) NPP emergencies.
2.8. Relationship to change
Change in one guise or another is a regular feature of error-producing situations. It will be argued here that each error type differs from the others in its relationship to change.
In SB slips and lapses, the error-triggering changes generally involve a necessary departure from some well-established routine. They may be occasioned either by an intended deviation from normal practice or by an alteration in the physical circumstances in which a routine is customarily executed. In both cases, these variants are usually known about in advance. Two domestic examples will illustrate the point.
Imagine that you have a visitor who has requested tea, while you only drink coffee. You go to the kitchen intending to prepare both coffee and tea, but return with two cups of coffee. The reason for this slip is clear; you failed to make an attentional check on your plan at the point where the initial common pathway, boiling a kettle, branches into its separate tea- and coffee-making components. As a result, you proceeded along the habitual coffee route. Imagine now that you intended to make coffee simply for yourself, but that in the recent past you had reorganised your kitchen so that the coffee was no longer in its accustomed place. You go to the kitchen and begin searching vainly for the coffee jar in its original location. Only then do you remember the changes that you yourself have made.
In one case, the slip arose from a failure to monitor the current intention; in the other, it was due to a failure to recall earlier situational changes in the kitchen. In both instances, however, the actor possessed knowledge of the precise nature of the change in advance and could, in theory at least, have been forewarned of its slip-making potential. The slips arose because knowledge relating to these changes was not accessed at the appropriate time, due almost invariably to attentional ‘capture’.
In RB mistakes, the situation is subtly different. Here the nature of the likely changes are, in some degree, anticipated, either as the result of past encounters or because they are considered as likely possibilities by instructors or designers. In either case, some contingency routines for handling these troublesome variations will have been established within the individual’s knowledge base or written into his or her operating procedures. What is lacking, however, is adequate knowledge of when such changes will occur and what precise forms they will take.
At the KB level, on the other hand, mistakes result from changes in the world that have neither been prepared for nor anticipated. By definition, errors arise from the fact that the problem solver has encountered a novel situation for which he or she possesses no contingency plans or preprogrammed solutions.
The three error types can therefore be discriminated according to the degree of preparedness that exists prior to the change. At the SB level, the nature and the time of the change are potentially knowable and the actor possesses routines for dealing with them. What is missing is the timely investment of an attentional check to ensure that these alternative routines are called into play. At the RB level, the changes have been anticipated, but the time of their occurrence is not known in advance. Here, the mistake arises from the application of a ‘bad’ rule or the misapplication of a ‘good’ rule. At the KB level, however, the encountered change falls outside the scope of prior experience or forethought and has to be dealt with by error-prone ‘on-line’ reasoning.
A summary of the distinctions made in this section between the three basic error types is given in Table 3.2.
3. A generic error-modelling system (GEMS)
As will become evident, the rule-based and knowledge-based operations of this system owe much to the models of Rasmussen (1986) and Rouse (1981) that were outlined briefly in the previous chapter. The main difference between GEMS and these earlier models lies in its attempt to present an integrated picture of the error mechanisms operating at all three levels of performance: SB as well as RB and KB. It is, in effect, a composite of two sets of error theories: those of Norman (1981) and Reason and Mycielska (1
982) and the General Problem Solver (see Chapter 2) tradition of theorising that has been applied by Rasmussen and Rouse to operator failures in high-risk technologies, most notably in the aircraft and nuclear power industries.
Table 3.2. Summarising the distinctions between skill-based, rule-based and knowledge-based errors.
Its operations divide conveniently into two areas: those that precede the detection of a problem (the SB level) and those that follow it (the RB and KB levels). Errors (slips and lapses) occurring prior to problem detection are seen as being mainly asociated with monitoring failures, while those that appear subsequently (RB and KB mistakes) are subsumed under the general heading of problem-solving failures. The ‘mechanics’ of GEMS are summarized in Figure 3.1.
3.1. Monitoring failures
Well-practised actions carried out by skilled individuals in familiar surroundings comprise segments of preprogrammed behavioural sequences interspersed with attentional checks upon progress. These checks involve bringing the higher levels of the cognitive system (the ‘workspace’) momentarily into the control loop in order to establish (a) whether the actions are running according to plan and (b) more complexly, whether the plan is still adequate to achieve the desired outcome. The former kind of deviation, as has been shown, is detected far more readily than the latter kind.
It is also meaningful to regard routine action sequences as involving a series of nodes or choice points beyond which subsequent actions can take a number of possible routes. Harking back to our earlier example of making a beverage, it can be appreciated that the initial step of boiling a kettle can lead to a variety of outcomes: making tea or coffee, speeding up the cooking of vegetables, making instant soups, filling hot-water bottles, and so forth. For a given individual, these post-nodal routes will vary widely in their frequency and recency of prior employment. In order to ensure that actions are carried out as planned, attentional checks should occur in the region of these choice points, particularly—as we have discussed—when the current intention is not to take the most ‘popular’ post-nodal route.