In general, experimental studies have suggested that the sensitivity decrement occurs more frequently and has greater magnitude when stimuli are presented at a high rate in successive discrimination tasks (Parasuraman, 1979; Parasuraman & Davies, 1977), in simultaneous discrimination tasks in which furthermore the signal/noise is very low (Parasuraman & Mouloua, 1987), and in sensory tasks (tasks in which the difference between signal and noise stimuli is a change in some physical characteristic) (Koelega, Brinkman, Hendriks, & Verbaten, 1989). See, Howe, Warm, & Dember (1995) carried out a meta-analysis of 42 vigilance studies in an attempt to identify the task characteristics related to the sensitivity decrement. Their analysis confirmed that these task characteristics (type of discrimination, event rate, and type of stimuli) and the average level of sensitivity associated with the task were the only significant task predictors of the sensitivity decrement.

The most widely accepted explanation of this pattern of results suggests that the crucial factor in determining the sensitivity decrement is the excessive processing demand imposed by a combination of time pressure (or high processing rate) and memory and perceptual loads (Fisk & Schneider, 1981; Parasuraman, 1979, 1985; Parasuraman & Davies, 1977; Parasuraman & Mouloua, 1987; Temple, Warm, Dember, Jones, LaGrange, & Matthews, 2000). The theory proposes that, when attentional resource demands imposed on the participant by the vigilance task are low, sensitivity is high and the decrement is small, whereas when demands are high, sensitivity is low and tends to decrease over time to an even lower level. The theory is based on the three typical assumptions of capacity models of attention (Parasuraman & Davies, 1984): 1 / there is a limited pool of processing resources that can be allocated to a processing task; 2 / discrimination level is a function of the amount of resources allocated to the task, although it can be also data-limited (Norman & Bobrow, 1975); 3 / when task demands exceed the available supply, performance efficiency drops and it will be more likely that this occurs as task difficulty increases. To explain specifically the sensitivity decrement in vigilance tasks, Parasuraman (1979, 1985) adds an additional assumption: the amount of available resources decreases progressively over time on task.

This article is related to a recent study (Blanco & Leirós, 2000) in which we investigated the extent to which the sensitivity decrement can be affected by temporal variations in stimulus luminance. First, we confirmed that, under normal conditions, the luminance of the stimuli displayed on the screen of a cathode-ray-tube (CRT) monitor decreases significantly during the first hour after switching the monitor on and then tends to remain stable (Cowan, 1991; Metha, Vingrys, & Badcock, 1993). Second, Blanco and Leirós carried out an experiment in which participants performed a visual vigilance task at different time periods after switching the monitor, in which the luminance of the stimuli displayed on the screen of the CRT either decreased (by about 13%, approximately) or remained stable from the beginning to the end of the task. It was expected that, according to the limited capacity theory (Fisk & Schneider, 1981; Parasuraman, 1979, 1985; Parasuraman & Davies, 1977; Parasuraman & Mouloua, 1987), performance would decline over time on task when the stimulus luminance diminished, due to progressive impairment of visual conditions. However, this did not occur. On the contrary, analysis of the data showed a significant decrement when the luminance remained stable and no decrement when the luminance decreased. The posthoc explanation given by Blanco and Leirós (2000) was that, when luminance decreased, the participants had attempted to compensate for the supposed deterioration in vision quality, perhaps by devoting more attentional resources to the task. Furthermore, if this compensation process exists, it would presumably be automatic, because the self-reports indicated that participants were unaware of the luminance variations during the task.

In this paper, we report two experiments designed to examine whether vigilance participants can compensate for small changes in the luminance of the stimuli displayed on a CRT. The central idea is that, during a vigilance task, the participant monitors his/her cognitive state or activity and compares his/her performance level with a reference level defined at the beginning of the task and which would act as a “goal state” that the observer attempts to reach. If a discrepancy is detected, adjustments are made in order to match his/her cognitive state and that goal-state. This adjustment might consist of a change in the amount of attentional resources (e.g., visual resolution, use of working memory, speed pro- cessing, etc.) that he/she dedicates to the vigilance task until his/her performance level is stabilized within limits that he/she considers acceptable in relation to the target state.

In the experiments presented here, the task involved detection of an oriented line that was displayed with stable or decreasing luminance (from the beginning to the end of the task). Thus, this task involved simultaneous discrimination (discriminating a line from the background) and, on the basis of previous research, was expected to be non-susceptible to the vigilance decrement (Parasuraman & Davies, 1977), at least in the stable-luminance condition. In the first experiment, the main comparison was between 1 / performance in this vigilance task when the stimulus luminance was gradually decreased over time on task, and 2 / per- formance in a similar task but with rests between the trial blocks and with participants being informed of the changes in stimulus luminance, in a procedure similar to that used in psychophysics to obtain the psy- chometric function. It was expected that, if the participants in the vigilance task fixed a target state as a goal to reach and if they were unaware of the manipulation of luminance, their performance levels would not be very different from those observed in the alert condition, because they would attempt to compensate for the deterioration of vision (for example, by devoting more effort to the task). In the second experiment, a similar vigilance task in which the stimuli were displayed with a stable luminance level over time was compared with an equivalent task in which the luminance level varied in each trial according to the rules of an adaptive psychophysical procedure (Levitt, 1971) designed to ensure that the participant maintained a stable performance level throughout the task. In this latter condition, because the system (i.e. the computer) adapted the task to the participant, no regulatory activity was expected on the part of the participant, and the vigilance decrement was thus expected to be greater than in the former condition.

Apparatus. A Pentium II computer with an ATI-Wonder Pro graphics card controlled the stimulus displays and responses. The experiment was programmed and run on this computer using the Micro Experimental Laboratory package (MEL; Schneider, 1995). Stimuli were displayed on a high resolution color monitor (Sony Multiscan 20 SE II). Monitor resolution was 1,024×768 pixels and frame rate 60 Hz, permitting display times to be varied in steps of 16.67 msec. The monitor was mounted at eye level at a viewing distance of approximately 70 cm from the seated participant. Luminance and color of the stimuli were measured with a J17 Tektronix photometer equipped with a J1820 sensor head. Responses were entered on a PST response box (see Schneider, 1995, for technical specifications). An adjustable chin rest helped to maintain head position to a distance of 70 cm from the monitor. The experiment was run in a dark room.

Procedure. The task required the identification of an oriented line. Each trial began with the presentation of a fixation point for 750 msec at the center of the screen. The target stimulus was a line 1.15^o in length tilted 9^o to the right or left. The line could appear to the left or to the right at a distance of 4.26º from the fixation point, but the position was irrelevant for the task, because participants had to indicate only the orientation of the line. A line appeared in only 20% of the trials; in the remaining 80%, no stimulus was presented (blank trials). After 50 msec, the stimulus disappeared and two noise masks were displayed at the locations at which the line stimulus could be displayed. These masks were rectangular random-dot patterns with a mean luminance of 12.33 cd/m² and remained on the screen for 700 ms, until the end of the trial. The participant had to identify the orientation of the line (left or right tilt) by pressing a button on the response box; buttons 1 and 5 of the response box, arranged from left to right, were assigned to left and right tilts, respectively. The participant was instructed to respond as accurately as possible within a time window of 700 msec.

The task consisted of five 5-min blocks. Forty signals (twenty lines of each orientation) and 160 blank trials were presented in each block. The task thus had a high event rate (40 trials/min) and a probability of signal occurrence throughout the task of 0.20.

Participants were divided into three groups of 11 participants each. One group, here called the Psychometric Function (PF) group, performed the five blocks of trials in the usual conditions employed in psychophysics to obtain the psychometric function. These participants performed each block of trials separately, with a rest of a few minutes between blocks. In each block, the signal stimuli had a different luminance: 4.162, 4.892, 5.681, 6.543, or 7.482 cd/m². The order of presentation of the five levels was selected at random for each participant. Each block was clearly marked with a message at the center of the screen indicating the luminance level that would be used in that block.

The other two groups performed the five blocks of trials under the usual conditions for vigilance experiments, that is, without interruptions between blocks. One of these groups performed the task with the signal luminance constant throughout the task, 7.482 cd/m², the highest level applied in the PF group. This group is here called the Stable Luminance (SL) group. The other group performed the five blocks of trials likewise without interruptions but with the luminance of the signal stimuli decreasing after each 5 minute block. In this group, the luminances of the signal stimuli in the different blocks were the same as those used in the PF group, but ordered from highest to lowest: 7.482, 6.543, 5.681, 4.892, and 4.162 cd/m². This group is here called the Decreasing Luminance (DL) group.

At the beginning of the experimental session, each participant performed a variable number of trials at the highest luminance level (7.482 cd/m²), until his/her performance level stabilized (proportion of correct responses=0.80). This number ranged among participants from 400 to 800 trials. He/she then performed the five blocks of trials.

At the end of the session, the participants who had performed the vigilance tasks (groups SL and DL) were asked to report perceived variation in stimulus luminance over time on task on a three-category scale: “1 - The luminance declined”, “2 - The luminance did not change”, and “3 - The luminance increased”. In order to avoid biases in these self-reports (the question about variation in perceived luminance could have forced participants to guess if they were uncertain), the category scale was included as part of a questionnaire on viewing of characters in computer monitor and visual display ergonomics. Other items (49 in total) made reference to participant characteristics (e.g., ophthalmic problems, age, experience in work with computers), and other aspects of the task and experimental situation (legibility of the stimuli, size of the stimuli, illumination, noise produced by apparatus in the experimental room, temperature, chair, etc.). We expected thus that the goal of the experiment, which could be suggested by the question about the perceived variation in the luminance level, would be masked by the other items of the questionnaire.

Analysis of performance. The number of false alarms (responses “left” or “right” to no stimulus or noise) was very small. Only one participant from the SL group and another from the DL group gave false alarms, with their respective percentages of false alarms being only 0.25% and 0.70%. Consequently, false alarm data were not examined further and a data analysis based on signal detection theory was not included.

Table 1 shows means and standard deviations of proportions of correct detections for each task period (SL and DL groups) or luminance level (PF group). These data were first examined by an analysis of variance (ANOVA) with the factors Group and Period, with repeated measures on this last factor. The analysis showed a significant main effect for Period (F(4,120)=4.20, p<.003) and for the interaction between the two factors (F(8,120)=3.11, p<.003). The effect of group factor was not significant (F(2,30)=1.36, p<.272).

TABLE 1 :
Data from Experiment 1. Mean and standard deviation (in brackets) of proportions of correct detections for each task period (SL and DL groups) and luminance level (PF group). Lum=Luminance; SL=Stable Luminance; DL=Decreasing Luminance; PF=Psychometric Function
Résultats de l’expérience 1. Moyenne et écart type (entre parenthèses) des proportions de détections correctes pour chaque période dans l’exécution de la tâche (groupes SL et DL) et chaque niveau de luminance (groupe PF). Lum=Luminance; SL=Luminance stable; DL=Luminance décroissante; PF=Fonction psychométrique

Fig. 1. — Data from Experiment 1. Mean psychometric function for PF groupRésultats de l’expérience 1. Fonction psychométrique moyenne pour le groupe PF

The data for each group were then analysed separately by ANOVA with Period (SL and DL groups) or Luminance (PF group) as the only factor. ANOVA of the PF data revealed a significant main effect of luminance (F(4,40)=6.24, p<.001). As expected, the proportion of correct detec- tions declined with decreasing luminance (fig. 1). In order to separate the effect of the luminance from the effect of time on task in this group, a new ANOVA was carried out on the data of this group with block number as the only factor. This ANOVA showed that the proportion of correct detections did not vary significantly with block number (F=0.90, p<.500).

ANOVAs of the data for the vigilance groups (SL and DL) did not show significant effects of Period (F=1.14, p<.353, and F=1.80, p<.147, respectively). Figure 2 plots the proportion of correct detections as a function of time on task for each of the two vigilance groups (the figure also shows the proportions observed for each luminance level in the PF group). The figure shows a small decline over time (8.40%) in the proportion of correct detections in the DL group, who per- formed the task with decreasing levels of luminance, but the figure also makes clear that this decline was not parallel to that observed in the PF group.

Fig. 2.
Data from Experiment 1. Proportion of correct detections as a function of time for each of the two vigilance groups. The figure also shows the mean proportions observed for each luminance level in the Psychometric Function group
Résultats de l’expérience 1. Proportions de détections correctes en fonction du temps pour les deux groupes de vigilance. La figure présente également les proportions moyennes observées pour chaque niveau de luminance dans le groupe PF

In the present experiment, variations in luminance did not affect vigilance performance, although these same variations had signifi- cant effects on a psychometric function obtained under the usual psychophysical conditions. Under these conditions, with participants being informed of the changes in stimulus luminance and with brief rests between blocks, sensitivity decreased with decreasing luminance. Although our psychometric function was not fitted to any psychophysical model, because this was not the main purpose of the experiment, visual inspection of the data (fig. 2) suggests that it could be well fitted by a normal cumulative probability function, as in numerous previous psychophysical studies (e.g., Baird & Noma, 1978; Gescheider, 1985).

The main finding of this experiment is that, using the same luminance levels as used for the psychometric function, a decrement in performance is not observed in the vigilance task; rather, performance is maintained at the same level as that observed when luminance was unchanged through the task. There was an approximate difference of 17% in performance between the highest and the lowest luminance in the PF group, but this difference was only about 8% and not statistically significant in the DL group over the same luminance range. This pattern of results is consistent with the possibility that participants paid more attention to the task in order to compensate for the visual deterioration produced by the reduced luminance and in order to maintain a stable performance level.

Adaptative procedures were introduced in vigilance research by Wiener (1973), who used a distance discrimination task. Two separated dots were intermittently displayed every 1 second; the signal was a display of two dots slightly more separated, and the adaptive task parameter the width or separation of the two signal dots. Wiener (1973) examined various adaptive methods, each varying in the number of previous trials used to modify the signal width. If the number of detections over the previous trials exceeded a given criterion, the signal width on the next trial was reduced and the signal was thus made less discriminable; if the number of detections did not exceed the criterion, signal width did not change or was increased. Wiener compared the standard measure of vigilance decrement (based on level of accuracy through the task) with an adaptive measure (based on changes in the adaptive task-parameter, i.e. width, over time on task). Wiener found that the two measures gave similar results. In a group of participants who performed the task with the adaptive procedure, the adaptive variable increased over time on task, whereas the group that performed the task with fixed parameters, in the usual way, showed a significant performance decrement.

Procedure. The task (programmed and run with Micro Experimental Laboratory software) was very similar to that used in the previous experiment. Each trial began with the presentation of a fixation point for 750 ms at the center of the screen. The target stimulus was a line 1.15^o in length tilted 45^o to the right or left. The line could appear to the left or to the right, at a distance of 4.26^o, from the fixation point. Each stimulus occurred with a probability of 0.50. After 50 ms, the stimulus disappeared and two noise masks were displayed at the same locations at which the line stimulus could be displayed. These masks were rectangular random dot patterns with a mean luminance of 12.33 cd/m². The participant had to respond to lines tilted to the right by pressing the right button of the response box, as accurately as possible, within a time window of 700 ms, with no response to lines tilted to the left.

The task lasted for 25 min, comprising five continuous 5 min periods. One hundred lines oriented 45^o to the left and other one hundred lines oriented 45^o to the right were presented in each period. The task had a high event rate (40 events/min) and a probability of signal occurrence throughout the task of 0.50. This probability changed relative to the first experiment, in which it was 0.20, in order to prevent response biases, because the adaptative procedure used in this experiment was originally designed for conditions in which the decision or response criterion is symmetrical. It is well known that in two alternative-forced-choice (2AFC) tasks the response criterion is symmetrical when both response alternatives have the same probability. This change relative to the first experiment does not seem important, because the signal probability is not statistically related to the sensitivity decrement observed in vigilance tasks (See et al., 1995).

The participants were divided into two groups of nine. In one group, which is here called the Adaptive Luminance group (AL), the luminance of the signal stimulus was adjusted in each trial using a transformed up-down method (Levitt, 1971). The luminance was adjusted in each trial to obtain a proportion of correct identifications of 0.794 (i.e. the point on the psychometric function obtained by decreasing the luminance level by one step after three correct responses and by increasing it by one step in the remaining cases). Luminance was controlled with the MEL sentence Set_Vga_Palette (color code, R, G, B). This sentence specifies the luminance of the stimulus by a number, for each of the three primary colors, from 0 to 63. The luminance was increased or decreased by adding or subtracting 1 to/from the last value in this sentence. Using this procedure, it was not possible to obtain a constant step in luminance units; the step in fact varied from 0.41 to 0.99 cd/m². However, subsequent analysis of the data revealed that the proportion of correct responses converged effectively to the point on the psychometric function selected a priori. The dependent variable was the mean luminance in each time period.

The other group of participants, here called the Stable luminance group (SL), was run only after all participants of the other group. Participants of this group performed the same task, but now stimulus luminance was maintained constant through the task at 5.790 cd/m², the mean of the luminances obtained by the adjustment procedure in the AL group during the first task period.

Each experimental session lasted approximately one hour. At the beginning of the session, each participant performed a variable number of trials until his/her performance level has stabilized. In the AL group, stimulus luminance was adjusted for each person to obtain a proportion of correct identifications of 0.794. The mean of the last few values was then used during the task. The SL group performed the practice trials with a constant luminance level. The mean number of practice trials was 600. The experiment was conducted with the same equipment as used in the first experiment.

TABLE 2 :
Data from Experiment 2. CR=Mean proportion of correct responses; AL=Adjusted Luminance (adaptative procedure); SL=Stable Luminance (traditional procedure). The values in brackets are standard deviations
Résultats de l’expérience 2 CR=Proportion moyenne de réponses correctes; AL=Luminance ajustée (procédure adaptative); SL=Luminance stable (procédure classique); Les valeurs entre parenthèses sont les écarts types

Second, the mean stimulus luminance across trials was calculated for each person and task period. Table 2 also shows the group means. These data were also analysed by an ANOVA with Period as the only factor. This ANOVA revealed a significant increment (6.18%) in stimulus luminance (F(4,32)=3.52, p<0.017). This result must be interpreted as a vigilance decrement, because luminance here is the inverse of sensitivity. Figure 3 represents the performance of one person who is representative of the behaviour of this group. The figure shows the stimulus luminance in each trial. Throughout the task, there was an increasing trend in luminance, although there were peaks and troughs. These data indicate a vigilance decrement: an increment in the task parameter (luminance) in order to maintain fixed performance over time on task (probability of correct detections=0.794). Figure 4 represents the mean luminance for all participants as a function of the task period.

Analysis of SL performance. Table 2 shows the mean proportion of correct responses for each task period. A small decline (3.34%) over time on task is apparent, but an ANOVA with Period as the only factor revealed that this decline was not statistically significant (F(4,32)=1.20, p<.328).

Lastly, proportions of correct responses during the first trial block in each experimental group were compared with a Student’s T. This statistical test showed that there was not a significant difference in performance level between both experimental groups (t(16)=0.33, p>0.05)

The signal probability used in this experiment deserves a commentary. One of the key characteristics of vigilance experiments is the low signal probability, which is not generally greater than 0.25. One well established fact in vigilance research is that, because of this low signal probability, the participant does not fix a symmetrical decision criterion; rather, he/she uses a stricter or more cautious response or decision criterion to report the signal stimuli, and, given certain conditions, this criterion is made more cautious with time-on-task (Davies & Parasuraman, 1982). But recognizing that signal probability is crucial for defining the vigilance phenomenon, it seems that this variable is independent of sensitivity, because it has no impact on the physical signal that stimulates the sensory receptors. In their meta-analysis of recent vigilance studies, See et al. (1995) examined 138 experimental conditions with varying levels of signal probability and did not find any statistical relation with sensitivity decrement. With the object of preventing possible effects of signal probability on sensitivity, we used a signal probability of 0.50, because it is well known that in two alternative-forced-choice (2AFC) tasks and yes-no tasks response biases are not observed with this probability. Another solution would have been to use an adaptive procedure that modifies the stimulus parameter on the basis of a performance measure independent of the decision criterion; however, these procedures are not yet sufficiently developed (Green, 1993; Saberi & Green, 1996).

Another methodological aspect which also deserves a commentary is the difficulty of making a direct comparison between vigilance performance in both groups because we had to use different measures with each procedure: an accuracy measure in the group with stable luminance and the mean stimulus luminance required to maintain a stable performance in the group with variable luminance. However, the fact that the proportion of correct responses was statistically similar in both groups during the first trial block suggests that the difference observed in the vigilance decrement in both groups was due to the procedure used and not to the measures on which ANOVAs were carried out.

An important characteristic of the individual at work is his/her ability to adapt to changing conditions in the workplace, or in the system that he/she operates. In the visual domain, this adaptive behaviour can be clearly observed in office environments, particularly in workplaces that have a high proportion of natural lighting. During a working day, the lighting can change, either decreasing or increasing, compelling the worker to accommodate his sensitivity and cognitive activity in order to maintain a performance level and to reach the necessary goals. For example, although the source of lighting in my office is mostly artificial, changes in natural lighting entering through a window produce variations in the overall illumination of approximately 25% (minimum lighting=350 lux, maximum =450 lux) during a working day. Also, the changing level of lighting can affect the visibility of the stimuli displayed on a CRT monitor, for example by producing more or less reflections or glare on the screen. If the changes in the visibility of the stimuli displayed on the monitor screen are slow, such changes can go unnoticed for the worker; in these cases, he/she will probably not attempt to improve the visibility by adjusting the corresponding control. Rather, he/she may attempt to compensate for the effects of these changes on vision quality by dedicating more attentional resources to the task.

One of the most salient aspects of vigilance research is the difficulty of finding examples of a performance decrement in occupational settings (see Parasuraman, 1986). This finding contrasts with the ease with which that decrement is observed in laboratory studies. The possible critical differences between laboratory and “real” tasks (e.g., practice, task simplicity/ complexity, signal probability) do not seem to explain this. For example, laboratory studies employing simulations of operational tasks that require complex monitoring, as occurs in many “real” tasks, have found significant vigilance decrements (Parasuraman, 1986). Rather, the reason for the lack of vigilance decrements in occupational settings may concern the strategies employed during task performance. In two well known experiments carried out by Nachreiner (1977), the participants performed the same vigilance task either under normal experimental conditions or believing that they were taking part in a selection test for a well paid job. The results showed that the decrement occurred under the normal experimental conditions, but not when the participants believed that the experiment was part of a selection test. These data suggest that the paucity of evidence for vigilance decrements in occupational settings might be related not so much to the task type used in laboratory studies, but rather to the strategies used during task performance.

The results of the experiments presented here also suggest that the effectiveness of compensatory mechanisms may account for the stability of performance that is observed in many vigilance experiments, particularly in real settings. According to this hypothesis, maintaining vigilance in an information processing task would involve more than just a simple state of perceptual readiness, as the usual explanations of vigilance decrement, including the limited capacity theory, suggest (see Davies & Parasuraman, 1982; Parasuraman, 1986). Specifically, maintenance of vigilance would also involve a certain-degree of control and regulatory activity by the participant.

This proposal is not new in the vigilance literature. It is closely related to the state control model developed by Hockey (1993; Hockey & Tattersall, 1988) to account for effects of stress and environmental factors on human performance. This model considers that maintenance of vigilance is basically a question of preserving control over relevant information processing variables. This model assumes that the participant sets up in memory a target state, which constitutes a plan or schema that guides his/her behaviour during the task, and includes a) specification of optimal functional levels of all relevant information processing resources, b) activation of certain cognitive skills, and c) definition of error tolerance limits. The person would then monitor possible discrepancies between this target state and his/her current information processing state; discrepancies could either be ignored, and no control activity initiated, or could induce a regulatory response, involving modification of the information processing state (e.g., dedicating more attentional resources to the task), changing the target state, or even changing the task environment (e.g., increasing the level of luminance of a CRT monitor).

In brief, the results of the present research, in line with previous models of vigilance (Hockey, 1993; Hockey & Tattersall, 1988), suggest that maintaining vigilance involves regulation of a cognitive or information processing state in order to adapt to changes in task and environmental demands, and not only a state of perceptual readiness, as traditional studies of vigilance suggest. Future research on this issue must solve certain conceptual and methodological problems, however. Goal state, strategy, and regulatory activity are some concepts that require more precise operational definitions. Also, it is necessary to design new experimental procedures to investigate sustained attention that permit investigation of the process during vigilance performance and not simply to assess decrements in general sensitivity as it has been usual in traditional vigilance research.

Paper received: January 2002.

Accepted by J. Patrick after revision: October 2002.