If no specific instructions are given it is not clear which task is given priority. In heavy traffic the conversation with a passenger will probably be disrupted to maintain driving performance while in quieter environment and during a very interesting conversation driving performance will be affected (Wickens, 1984). Moreover, while the division between primary and secondary tasks may be very clear-cut for most laboratory tasks, this is not the case in driving. In traffic, behaviour is quite often related to the manoeuvre that is performed. Monitoring of rear traffic can be crucial if an overtaking manoeuvre is planned. In those cases the task of looking into mirrors and over one's shoulder cannot be called `secondary'. Task integration can also blur the transition from primary to secondary task. A good example of dual-task integration is car-following. In heavy traffic this task will be added to the primary task of lateral and longitudinal vehicle control. It is the addition of a task, but the added task is not artificial. The experience of various subtasks as a `single task' is in particular likely if the subtasks are related or coherent (see, e.g., Korteling, 1994ab). Viewed in this way, car-following performance could be an embedded secondary task. However, a condition for a task to be termed embedded is that it is given lower priority than the primary task. It is not certain that car-following is given lower priority than lane-keeping. Perhaps a useful description of a secondary task in traffic research is that the task does not have to be performed continuously. In this way, the primary task remains restricted to speed and lateral vehicle control. Secondary tasks are non-continuous tasks, i.e. headway keeping can only be performed in case a lead vehicle is present and looking into the mirrors is performed at intervals. The definition is weak, but so is the separation of primary and secondary tasks in traffic.
Car-following
At the Traffic Research Centre a car-following task for use in real traffic has been developed (Brookhuis et al., 1994). In the task, a lead car's speed fluctuations have to be followed by the driver of an experimental vehicle. This task is designed to be sensitive to impairment of performance in attention and perception, while lane-tracking is merely sensitive to performance on eye-hand coordination. In terms of the hierarchical model of car driving (Janssen, 1979, Michon, 1985, see also chapter 1) the lane-tracking parameters (SDLP, SDSTW) reflect performance at the control level, while car-following parameters reflect performance at the manoeuvre level. The main parameter in car-following performance is the delay in reaction to speed changes of the lead vehicle. We (Brookhuis et al., 1994) obtain this measure by performing a coherence analysis on the speed signals of the lead and the following car. Apart from delay (calculated as `phase shift' between the two speed signals in the frequency domain) two other parameters are computed, which both give an indication of `how well' the car-following task is performed. Coherence is a measure of the accuracy of car-following performance, while the modulus indicates the amount of overreaction to speed changes by the following car (Porges et al., 1980).
The car-following task was included in the car-phone, DREAM and antihistamine studies. Delay increased in conditions in which a car-phone was used (+23%), after alcohol consumption (+19%), and in the condition in which Triprolidine had been taken (+42%). Time-on-task (Fatigue) did not affect delay, but coherence slightly decreased in this condition.
Mirror checking
Mirror checking is another good example of an embedded secondary task that is specific for car driving. Two variables can be distinguished in mirror checking: frequency and duration. Total duration of mirror checking was measured both in the Weaving Section and the Noise Barrier study. In the Noise Barrier study, however, only data related to the load condition were available. In this condition no more than 2.7% of the total time was spent looking in the mirrors. In the Weaving Section study, the difference in mirror-looking time between load (10.6%) and control (10.2%) was not significantly different. In an in-vehicle navigation study, Fairclough et al. (1993) compared driving performance and visual attention while navigating from map vs. from a text-LCD screen. They found a decrease in duration of fixations in the rear-view mirror in the higher demand (i.e., map) condition. In another study, reported in the same paper, glance frequency (but not glance duration) in the rear-view mirror was decreased in the condition in which internal vehicle `checking behaviour' of a display was higher. The authors' conclusion was that glance duration and glance frequency are representative for different aspects of driver behaviour. Duration appears to be sensitive to difficulty of information intake, while glance frequency represents visual activity in terms of checking behaviour, both inside (e.g., speedometer checking) and outside (e.g., mirror checking) the vehicle.
Figure 10. Frequency of interior and outside mirror checking in the car-phone and antihistamine study. In both studies motorway and ringroad sections were scored. If available the 95% confidence interval is indicated.
In the car-phone and antihistamine study, mirror checking frequency was scored from video, in both studies separately for the (quiet) motorway and (busy) ringroad. In the Weaving Section study the CEMRE-condition could be used to assess mirror-scanning frequency. As can be seen in figure 10, frequency of mirror-looking is reduced in the load condition of the car-phone study. The main effect of car-phone was not significant, but the interaction between road type and phone was. The larger effect of load on the motorway may be responsible for this. No effect of load was found in the antihistamine study, only the effect of road type (again ringroad vs. motorway) was significant. In both studies mirror-looking frequency was lower on the more traffic-dense ringroad, where a car-following task had to be performed. In the Weaving Section study a significant increase in frequency of mirror checking was found in the load condition. This is particularly important because no difference in duration, i.e. proportion of the total time, of looking into the mirrors between the load and baseline conditions was found. Again the road environment may be responsible for the increase. The load section of the motorway was a combined entrance/exit with vehicles merging in and out of traffic, while the control section did not contain any entrances or exits. An increase in mirror-checking frequency and `behind traffic monitoring' is important near entrances, even if no change-of-lane is planned, owing to the possible need of an evasive manoeuvre to the left-hand lane.
Rear mirror checking was also affected in the study reported by Van Winsum et al. (1989). In an unfamiliar environment, frequency of looking into the rear view mirror was reduced in the higher workload condition. Frequency of fixations seems most useful for workload assessment, though only if workload demand is not low. Fixation duration may be useful to assess certain aspects of task difficulty, in particular legibility, layout and amount of information (Fairclough et al., 1993).
Additional tasks
An actual additional task that had to be performed simultaneously to driving was the PASAT, the Paced Serial Addition Task (Gronwall & Sampson, 1974). The task itself is a demanding combination of a memory load and an addition test. This secondary task was used in the car-phone study, where the stimuli (digits) were presented over the phone. The task was used to create a fixed, heavy information-processing load on the subjects, more or less comparable to a difficult conversation. There was no control condition in which the task was performed without having to drive a car and/or use the car-phone. No significant differences in performance between the two road classes, motorway and ringroad, were found.
Earlier, at the end of the 60's, Brown et al. (1969) had studied the effects of telephoning on car driving performance by having subjects drive a car and perform gap-acceptance tests which were combined with a reasoning test. Subjects had to judge the correctness of sentences in relation to pairs of letters, e.g. "A follows B, -BA" (answer: True). Any impairment in driving performance could be attributed to divided attention; there was no need for the subjects to manually operate the car-phone. No effects of the additional task on primary-task vehicle-control measures were found, with the exception of an increase in time that was required to complete the circuit. Performance on the secondary task, however, was poorer in the condition in which the task was combined with driving. Both reaction time and the proportion of errors increased. Gap-acceptance performance was also reduced by the additional task.
Verwey (1993b) carried out an experiment in which 48 subjects drove an instrumented vehicle over rural and inner-city roads while as secondary task they performed a visual detection task or an auditory addition task. While driving, subjects were guided by vocal messages issued by the experimenter. The experiment was a between-subject study with as factors: age (young vs. old), secondary task (auditory addition task vs. visual detection task), route familiarity (2 levels) and traffic density (2 levels). Subjects were instructed to give priority to the primary task of driving (Subsidiary Task Paradigm). Single-task performance of the secondary task while standing still was poorest for the elderly (79% opposed to 88% correct for the young). When driving, the older subjects' secondary-task performance (73% correct) was affected, while the younger subjects' performance did not decline (87% correct). Familiarity and traffic density had little effect on performance, while large differences on secondary-task performance were found between road situations. Between similar situations, i.e. between comparable road characteristics, no differences on secondary-task performance were found. In the study primary-task performance was only measured by assessment of speed control. Since different road segments had different speed limits, conclusions regarding primary-task performance are restricted. However, subjects unfamiliar with the road drove slightly slower and may therefore have reduced workload by adapting primary-task performance.
Brouwer et al. (1991) and Van Wolffelaar et al. (1990) have used an elegant `driving-simulation' task. It was not the task environment that was elegant, but the way in which the level of primary-task performance was adapted to individual capability. By individually adapting the level of single task performance they succeeded in obtaining an equal task difficulty for all subjects. The primary task was a compensatory lane-tracking task. Added to this task was a visual analysis task. Van Wolffelaar (1990) added a third task to these, subjects had to respond to visual stimuli presented in the periphery. Although the simulator and the tasks that were used are more similar to laboratory tasks than to actual driving, the advantage of equal single-task difficulty for all is that divided attention problems can be studied taking into account differences in individual capability and allocation strategy. Results show that elderly are less successful in dividing attention in dual-task performance.
Properties of secondary-task measures
If it is assumed that performance of a secondary task uses up `spare capacity', then secondary-task measures could be performance measures that are sensitive in the A region. However, most secondary tasks interfere (to a varying extent) with primary-task performance and task instruction alone cannot determine which task receives priority. Embedded tasks are regarded as the best secondary tasks. Even though it still is not certain what priority the embedded task receives, at least primary-task intrusion is low. In car driving, measurement of car-following performance and mirror checking can supply embedded task measures. Delay in car-following was found to be a sensitive measure in the sedative antihistamine, alcohol and car-phone conditions. Sensitivity of this measure can accordingly be expected in the D/A1 and A3/B regions of performance. Frequency of mirror checking was found to be sensitive in the Weaving Section study, while the measure also differed between motorway and ringroad-driving. This measure is sensitive in the A3/B regions, while the frequency was not affected in the antihistamine study, and sensitivity in the A1/D regions requires further examination. Duration of glances in the mirror was not sensitive in the Weaving Section study, and no conclusions with respect to sensitivity of this measure in regions of performance can be drawn.
Both delay in car-following and mirror checking can reflect performance at the manoeuvre level. Diagnosticity of the latter measure to visual demand is moderate to high (Fairclough et al., 1993). Delay in following a lead vehicle was found to be sensitive in car-following conditions in all studies and seems to be a sensitive and reliable measure. Mirror checking frequency showed a similar sensitivity in the motorway and ringroad conditions of the antihistamine and car-phone studies, and reliability is accordingly rated high. Primary-task intrusion when using embedded secondary tasks is low. However, when studying car-following behaviour and more or less natural variations from a lead car have to be followed, task priorities may become somewhat blurred. Primary-task intrusion and operator acceptance when registering mirror-checking behaviour depends upon measurement technique. The CEMRE is an intrusive device while video registrations made by small cameras can remain completely unnoticed by the subjects. Implementation requirements in terms of instrumentation and time/equipment required for analysis are high for all measures. An overview of secondary-task performance measures' properties is presented in table 9. Mirror checking measures are based on a limited number of studies, and were measured with different techniques.
Apart from quantification of task performance in measures such as the SDLP or the frequency of mirror scanning, task performance could be rated by an observer. This method is sometimes used, but suffers from other methodological problems, such as training of the experimenter. If applied correctly, and if observers are well trained, results could add to the previously discussed primary and secondary task-performance measures. Critical incidents, law violations and lateral position errors are measures of driving performance and have been used as such in task-performance assessment (e.g., Pohlmann & Traenkle, 1994). In particular, complex behaviour, such as the occurrence of critical incidents, or behaviour in a complex driving environment can be easier, or more accurately, detected and judged by an observer than captured in a single performance measure.
| Table 9. Summary of properties of secondary-task workload measures. | |||
|---|---|---|---|
| Measure | |||
| Property | Delay in car following | Mirror checking (duration) | Mirror checking (frequency) |
| sensitivity (Region) | D-A1, A3-B | ? | A3-B |
| diagnosticity | low-moderate | moderate (?) | moderate-high |
| selectivity | moderate (?) | ? | moderate (?) |
| Reliability | high | ? | high |
| primary-task intrusion | low-moderate | low | low |
| implementation requirements | high | high | high |
| operator acceptance | high | high | high |
to chapter 5.4 Physiology and discussion
back to chapter 5.1 Self-report measures
back to chapter 5.2 Primary-task performance measures
back to chapter 5
back to thesis summary
I like to hear from you, so if you find this information useful, a short message is very much appreciated.
For more information you can also contact me.
© Dick de Waard 1996
You may only use (parts) of this thesis if you quote the source:
De Waard, D. (1996). The measurement of drivers' mental workload. PhD thesis, University of Groningen. Haren, The Netherlands: University of Groningen, Traffic Research Centre.
Back to my HomePage