The concept of vigilance refers to a human observer’s state of alertness in tasks that demand efficient registration and processing of signals. The main characteristics of vigilance tasks are relatively long durations and the requirement to detect infrequent and unpredictable target stimuli (signals) against a background of other stimulus events.
The prototypical task for vigilance research was that of radar operators. Historically, their apparently unsatisfactory performance during the Second World War has been a major impetus for the extensive study of vigilance. Another major task requiring vigilance is an industrial inspection. More generally, all kinds of monitoring tasks which require the detection of relatively infrequent signals embody the risk of failures to detect and to respond to these critical events.
Vigilance tasks make up a heterogeneous set and vary on several dimensions, in spite of their common characteristics. An obviously important dimension is the overall stimulus rate as well as the rate of target stimuli. It is not always possible to define the stimulus rate unambiguously. This is the case in tasks that require the detection of target events against continuously presented background stimuli, as in detecting critical values on a set of dials in a monitoring task. A less obviously important distinction is that between successive-discrimination tasks and simultaneous-discrimination tasks. In simultaneous-discrimination tasks, both target stimuli and background stimuli are present at the same time, while in successive-discrimination tasks one is presented after the other so that some demands on memory are made. Although most vigilance tasks require the detection of visual stimuli, stimuli in other modalities have also been studied. Stimuli can be confined to a single spatial location, or there can be different sources for target stimuli. Target stimuli can differ from background stimuli by physical characteristics, but also by more conceptual ones (like a certain pattern of meter readings that can differ from other patterns). Of course, the conspicuousness of targets can vary: some can be detected easily, while others may be hard to discriminate from background stimuli. Target stimuli can be unique or there can be sets of target stimuli without well-defined boundaries to set them off from background stimuli, as is the case in many industrial inspection tasks. This list of dimensions on which vigilance tasks differ can be expanded, but even this length of the list suffices to emphasize the heterogeneity of vigilance tasks and thus the risks involved in generalizing certain observations across the full set.
Performance Variations and the Vigilance Decrement
The most frequently used performance measure in vigilance tasks is the proportion of target stimuli, for example, faulty products in industrial inspection, that have been detected; this is an estimate of the probability of so-called hits. Those target stimuli that remain unnoticed are called misses. Although the hit rate is a convenient measure, it is somewhat incomplete. There is a trivial strategy that allows one to achieve 100% hits: one only has to classify all stimuli as targets. However, the hit rate of 100% is then accompanied by a false-alarm rate of 100%, that is, not only the target stimuli are correctly detected, but the background stimuli are incorrectly “detected” as well. This line of reasoning makes it quite clear that whenever there are false alarms at all, it is important to know their proportion in addition to the hit rate. Another measure for performance in a vigilance task is the time needed to respond to target stimuli (response time).
Performance in vigilance tasks exhibits two typical attributes. The first one is the low overall level of vigilance performance. It is low in comparison with an ideal situation for the same stimuli (short observation periods, high readiness of the observer for each discrimination, etc.). The second attribute is the so-called vigilance decrement, the decline of performance in the course of the watch which can start within the first few minutes. Both these observations refer to the proportion of hits, but they have also been reported for response times. Although the vigilance decrement is typical of vigilance tasks, it is not universal.
In investigating the causes of poor overall performance and vigilance decrements, a distinction will be made among concepts that are related to the basic characteristics of the task and concepts that are related to organismic and task-unrelated situational factors. Among the task-related factors strategic and non-strategic ones can be distinguished.
Strategic processes in vigilance tasks
The detection of a signal like a faulty product is partly a matter of the observer’s strategy and partly a matter of the signal’s discriminability. This distinction is based on the theory of signal detection (TSD), and some basics of the theory need to be presented in order to highlight the distinction’s importance. Consider a hypothetical variable, defined as “evidence for the presence of a signal”. Whenever a signal is presented, this variable takes on some value, and whenever a background stimulus is presented, it takes on a value that is lower on the average. The value of the evidence variable is assumed to vary across repeated presentations of the signal. Thus it can be characterized by a so-called probability density function as is illustrated in figure 1. Another density function characterizes the values of the evidence variable upon presentation of a background stimulus. When the signals are similar to the background stimuli, the functions will overlap, so that a certain value of the evidence variable can originate either from a signal or from a background stimulus. The particular shape of the density functions of figure 1 is not essential for the argument.
The detection response of the observer is based on the evidence variable. It is assumed that a threshold is set so that a detection response is given whenever the value of the evidence variable is above the threshold. As is illustrated in figure 1, the areas under the density functions to the right of the threshold correspond to the probabilities of hits and false alarms. In practice, estimates of the separation of the two functions and the location of the threshold can be derived. The separation of the two density functions characterizes the discriminability of the target stimuli from the background stimuli, while the location of the threshold characterizes the observer’s strategy. Variation of the threshold produces a joint variation of the proportions of hits and false alarms. With a high threshold, the proportions of hits and false alarms will be small, while with a low threshold the proportions will be large. Thus, the selection of a strategy (placement of the threshold) essentially is the selection of a certain combination of hit rate and false-alarm rate among the combinations that are possible for a certain discriminability.
Two major factors that influence the location of the threshold are payoffs and signal frequency. The threshold will be set to lower values when there is much to gain from a hit and little to lose from a false alarm, and it will be set to higher values when false alarms are costly and the benefits from hits are small. A low threshold setting can also be induced by a high proportion of signals, while a low proportion of signals tends to induce higher threshold settings. The effect of signal frequency on threshold settings is a major factor for the low overall performance in terms of the proportion of hits in vigilance tasks and for the vigilance decrement.
An account of the vigilance decrement in terms of strategic changes (threshold changes) requires that the reduction of the proportion of hits in the course of the watch is accompanied by a reduction of the proportion of false alarms. This is, in fact, the case in many studies, and it is likely that the overall poor performance in vigilance tasks (in comparison with the optimal situation) does also result, at least partly, from a threshold adjustment. In the course of a watch, the relative frequency of detection responses comes to match the relative frequency of targets, and this adjustment implies a high threshold with a relatively small proportion of hits and a relatively small proportion of false alarms as well. Nevertheless, there are vigilance decrements that result from changes in discriminability rather than from changes in threshold settings. These have been observed mainly in successive-discrimination tasks with a relatively high rate of stimulus events.
Nonstrategic processes in vigilance tasks
Although part of the overall poor performance in vigilance tasks and many instances of the vigilance decrement can be accounted for in terms of strategic adjustments of the detection threshold to low signal rates, such an account is not complete. There are changes in the observer during a watch that can reduce the discriminability of stimuli or result in apparent threshold shifts that cannot be considered as an adaptation to the task characteristics. In the more than 40 years of vigilance research, a number of nonstrategic factors that contribute to poor overall performance and to the vigilance decrement have been identified.
A correct response to a target in a vigilance task requires a sufficiently precise sensory registration, an appropriate threshold location, and a link between the perceptual processes and the associated response-related processes. During the watch the observers have to maintain a certain task set, a certain readiness to respond to target stimuli in a certain way. This is a nontrivial requirement because without a particular task set no observer would respond to target stimuli in the way required. Two major sources of failures are thus inaccurate sensory registration and lapses in the readiness to respond to target stimuli. Major hypotheses to account for such failures will be briefly reviewed.
Detection and identification of a stimulus are faster when there is no temporal or spatial uncertainty about its appearance. Temporal and/or spatial uncertainty is likely to reduce vigilance performance. This is the essential prediction of expectancy theory. Optimal preparedness of the observer requires temporal and spatial certainty; obviously vigilance tasks are less than optimal in this respect. Although the major focus of expectancy theory is on the overall low performance, it can also serve to account for parts of the vigilance decrement. With infrequent signals at random intervals, high levels of preparedness might initially exist at times when no signal is presented; in addition, signals will be presented at low levels of preparedness. This discourages occasional high levels of preparedness in general so that whatever benefits accrue from them will vanish in the course of a watch.
Expectancy theory has a close relation to attentional theories. Variants of attentional theories of vigilance, of course, are related to dominant theories of attention in general. Consider a view of attention as “selection for processing” or “selection for action”. According to this view, stimuli are selected from the environment and processed with high efficiency whenever they serve the currently dominant action plan or task set. As already said, the selection will benefit from precise expectations about when and where such stimuli will occur. But stimuli will only be selected if the action plan—the task set—is active. (Drivers of cars, for example, respond to traffic lights, other traffic, etc.; passengers don’t do so normally, although both are in almost the same situation. The critical difference is that between the task sets of the two: only the driver’s task set requires responses to traffic lights.)
The selection of stimuli for processing will suffer when the action plan is temporarily deactivated, that is when the task set is temporarily absent. Vigilance tasks embody a number of features that discourage continuous maintenance of the task set, like short cycle times for processing stimuli, lack of feedback and little motivational challenge by apparent task difficulty. So-called blockings can be observed in almost all simple cognitive tasks with short cycle times like simple mental arithmetic or rapid serial responses to simple signals. Similar blockings occur in the maintenance of the task set in a vigilance task as well. They are not immediately recognizable as delayed responses because responses are infrequent and targets that are presented during a period of absent task set may no longer be there when the absence is over so that a miss will be observed instead of a delayed response. Blockings become more frequent with time spent on the task. This can give rise to the vigilance decrement. There may be additional reasons for temporary lapses in the availability of the appropriate task set, for example, distraction.
Certain stimuli are not selected in the service of the current action plan, but by virtue of their own characteristics. These are stimuli that are intense, novel, moving toward the observer, have an abrupt onset or for any other reason might require immediate action no matter what the current action plan of the observer is. There is little risk of not detecting such stimuli. They attract attention automatically, as is indicated, for example, by the orienting response, which includes a shift of the direction of the gaze toward the stimulus source. However, answering an alarm bell is not normally considered a vigilance task. In addition to stimuli that attract attention by their own characteristics, there are stimuli that are processed automatically as a consequence of the practice. They seem to “pop out” from the environment. This kind of automatic processing requires extended practice with a so-called consistent mapping, that is, a consistent assignment of responses to stimuli. The vigilance decrement is likely to be small or even absent once automatic processing of stimuli has been developed.
Finally, vigilance performance suffers from a lack of arousal. This concept refers in a rather global manner to the intensity of neural activity, ranging from sleep through normal wakefulness to high excitement. One of the factors that is thought to affect arousal is external stimulation, and this is fairly low and uniform in most vigilance tasks. Thus, the intensity of central nervous system activity can decline overall over the course of a watch. An important aspect of arousal theory is that it links vigilance performance to various task-unrelated situational factors and factors related to the organism.
The Influence of Situational and Organismic Factors
Low arousal contributes to poor performance in vigilance tasks. Thus performance can be enhanced by situational factors that tend to enhance arousal, and it can be reduced by all measures that reduce the level of arousal. On balance, this generalization is mostly correct for the overall performance level in vigilance tasks, but the effects on the vigilance decrement are absent or less reliably observed across different kinds of manipulation of arousal.
One way to raise the level of arousal is the introduction of additional noise. However, the vigilance decrement is generally unaffected, and with respect to overall performance the results are inconsistent: enhanced, unchanged and reduced performance levels have all been observed. Perhaps the complex nature of noise is relevant. For example, it can be affectively neutral or annoying; it cannot only be arousing, but also be distracting. More consistent are the effects of sleep deprivation, which is “de-arousing”. It generally reduces vigilance performance and has sometimes been seen to enhance the vigilance decrement. Appropriate changes of vigilance performance have also been observed with depressant drugs like benzodiazepines or alcohol and stimulant drugs like amphetamine, caffeine or nicotine.
Individual differences are a conspicuous feature of performance in vigilance tasks. Although individual differences are not consistent across all sorts of vigilance tasks, they are fairly consistent across similar ones. There is only little or no effect of sex and general intelligence. With respect to age, vigilance performance increases during childhood and tends to decline beyond the age of sixty. In addition there is a good chance that introverts will show better performance than extroverts.
The Enhancement of Vigilance Performance
The existing theories and data suggest some means to enhance vigilance performance. Depending on the amount of specificity of the suggestions, it is not difficult to compile lists of various lengths. Some rather broad suggestions are given below that have to be fitted to specific task requirements. They are related to the ease of perceptual discriminations, the appropriate strategic adjustments, the reduction of uncertainty, the avoidance of the effects of attentional lapses and the maintenance of arousal.
Vigilance tasks require discriminations under non-optimal conditions. Thus one is well advised in making the discriminations as easy as possible, or the signals as conspicuous as possible. Measures related to this general goal can be straightforward (like appropriate lighting or longer inspection times per product) or more sophisticated, including special devices to enhance the conspicuousness of targets. Simultaneous comparisons are easier than successive ones, so the availability of a reference standard can be helpful. By means of technical devices, it is sometimes possible to present the standard and the object to be examined in rapid alternation, so that differences will appear as motions in the display or other changes for which the visual system is particularly sensitive.
To counteract the strategic changes of the threshold that lead to a relatively low proportion of correct detections of targets (and for making the task less boring in terms of the frequency of actions to be taken) the suggestion has been made to introduce fake targets. However, this seems not to be a good recommendation. Fake targets will increase the proportion of hits overall but at the cost of more frequent false alarms. In addition, the proportion of undetected targets among all stimuli that are not responded to (the outgoing faulty material in an industrial inspection task) will not necessarily be reduced. Better suited seems to be explicit knowledge about the relative importance of hits and false alarms and perhaps other measures to obtain an appropriate placement of the threshold for deciding between “good” and “bad”.
Temporal and spatial uncertainty are important determinants of poor vigilance performance. For some tasks, spatial uncertainty can be reduced by way of defining a certain position of the object to be inspected. However, little can be done about temporal uncertainty: the observer would be unnecessary in a vigilance task if the occurrence of a target could be signaled in advance of its presentation. One thing that can be done in principle, however, is to mix objects to be inspected if faults tend to occur in bunches; this serves to avoid very long intervals without targets as well as very short intervals.
There are some obvious suggestions for the reduction of attentional lapses or at least their impact on performance. By proper training, some kind of automatic processing of targets can perhaps be obtained provided that the background and target stimuli are not too variable. The requirement for sustained maintenance of the task set can be avoided by means of frequent short breaks, job rotation, job enlargement or job enrichment. Introduction of variety can be as simple as having the inspector himself or herself getting the material to be inspected from a box or other location. This also introduces self-pacing, which may help in avoiding signal presentations during temporary deactivations of the task set. Sustained maintenance of task set can be supported by means of feedback, indicated interest by supervisors and operator’s awareness of the importance of the task. Of course, accurate feedback of performance level is not possible in typical vigilance tasks; however, even inaccurate or incomplete feedback can be helpful as far as the observer’s motivation is concerned.
There are some measures that can be taken to maintain a sufficient level of arousal. Continuous use of drugs may exist in practice but is never found among recommendations. Some background music can be useful, but can also have an opposite effect. Social isolation during vigilance tasks should mostly be avoided, and during times of day with low levels of arousal like the late hours of the night, supportive measures such as short watches are particularly important.