Only driver fatigue has a significant effect on HRV (increase), the total test of reduced driver state is not significant. Finally, spectral energy of heart rate variability in the 0.10 Hz frequency band is consistently and significantly reduced in the increased complexity conditions, but is not significantly affected in the test of the effect of a reduced driver state. This last aspect is very important and supports Mulder's idea (G. Mulder, 1995) that the 0.10 Hz component is sensitive to task-related effort and not to state-related effort.

Other physiological measures used

ElectroEncephaloGram (EEG)
Ongoing EEG was more frequently used as indicator of driver state than as indicator of driver workload. The two are, however, not unrelated. As argued by some authors (Schneider et al., 1984, Kantowitz, 1992a), fatigue, e.g. as a result of the time spent performing a task, will be accompanied by a decreased arousal level and a reduced capacity, or a reduced willingness to spend resources (Meijman, 1991), and may therefore increase mental load. Ingestion of sedative drugs can be expected to result in the same effect. Brookhuis et al. (1985b, 1986) have found major increases in alpha and theta energy that were related to decreased driver activation caused by the use of antidepressant drugs. During prolonged train (Thorsvall & ¸kerstedt, 1987) or truck driving (Kecklund & ¸kerstedt, 1993), the driver's activation level as indicated by energy in the alpha and/or theta band was found to decrease rapidly. We (De Waard & Brookhuis, 1991a, appendix 5) have used the relative energy parameter [(alpha + theta) / beta] as indicator of driver state and found a significant increase on the parameter with time-on-task. When, after two hours of non-stop driving, subjects returned to a busy ringroad and had to follow a lead car, activation level increased again. Clearly, the increased task demands on the ringroad, increased mental load. It seems that EEG frequency analysis are most useful as an indicator of tonic driver activation, and can be included in workload research for these purposes.

Electromyogram (EMG)
Facial EMG of the corrugator supercilii muscle was measured in the road-layout experiment (De Waard et al., 1995). An effect of driving vs. rest, and of the two different road environments was found, while no effect of mental load as a result of the experimental road-layout was found. As the 0.10 Hz component of heart rate variability was sensitive to the (expected) difference in workload between the experimental and control road, and EMG activity of the corrugator was not, it is suggested that these measures may be tapping different dimensions of task load (see De Waard et al., 1995). To my knowledge, no experimental field-studies that further examine the differential sensitivity to workload of these two measures have as yet been performed.

Eye movements
The number and duration of eye fixations on instruments or in the mirrors while driving (see for mirror scanning also the section on secondary-task performance) may well be indicative for driver strategy. Rockwell (1988) found more glances instead of longer glances at a radio that had to be adjusted while driving. The strategy for most of the complex tasks was to take a series of glances of 1.25 s until the task was completed. Only if information could not be extracted in a glance, e.g. due to legibility, drivers could be tempted to increase glance duration. A minority of glances of up to 3 s were found when adjusting the stereo. Rockwell (1988) argues that glances of this duration are a threat to traffic safety, in particular in car-following situations.
In the Noise Screen and Weaving Section studies, fixation time (as proportion of the total looking time) was determined for various categories. Parkes (1991) refers to this measure as `glance allocation'. In the studies initially eye movements were scored in various categories that were later combined into larger categories. Three categories were analysed:

• traffic relevant fixations: looking straight forward, at other traffic, at the blind spot
• traffic irrelevant fixations: fixations on the other carriageway (which is irrelevant for motorway driving), the road environment, noise barriers, in the air
• mirrors & dashboard (`other points of focus')

The opportunity to look at, for driving, irrelevant stimuli will increase with decreases in workload (low time-pressure). This is partly comparable to the path-neglect time in TLC (see under primary-task performance measures). A more demanding task environment requires an increase in the time spent looking at the road. In particular, the time spent looking at, for task performance relevant, objects, such as other traffic participants, road signs, road layout, etcetera, will increase. This includes looking in the mirrors. If it is not the road environment that requires additional attention but a device inside the car, it may have the opposite effect. Less time will be spent looking at relevant objects in the traffic environment.
In the Weaving Section study a reduction in time spent looking at the dashboard (speedometer) was found in the mental load condition (see table 11), while in the NoiseBarrier study, only data regarding the load condition were available. It is therefore difficult to draw conclusions on the basis of one study only. In addition to this, fixation time was scored in these analyses, and not fixation frequency, which is additionally required to asses driver strategy (Rockwell, 1988). In the Weaving Section study, fixation frequency on relevant objects increased in the load condition. While fixation time increases significantly with 6%, the number of fixations on traffic relevant objects is elevated with 13.2 fixations, an increase of 56%. Data from this study thus indicate larger sensitivity for fixation frequency compared with fixation duration expressed as proportion looking time. Scanning behaviour in which more glances instead of longer glances are taken (cf. Rockwell, 1988) could account for this difference in measure sensitivity.

Results from other studies

ECG
In other studies found in the literature similar effects of mental load on ECG measures are reported as were found in the studies listed in table 3. Zeier (1979) measured heart rate in heavy city traffic while subjects drove a car with manual transmission, a car with automatic transmission or were just passengers. Both average heart rate and HRV (time domain) differed significantly between the manual-transmission condition and the other two conditions. Driving with automatic transmission or riding as a passenger did not lead to a significant difference in heart rate measures.
Egelund (1982) concluded that the 0.10 Hz component was an indicator of driver fatigue. Although average HR decreased with time-on-task, Egelund found that HR was, just as time-domain-HRV, not sensitive to fatigue in this study. Janssen & Gaillard (1985) concluded that the 0.10 Hz component of heart rate variability was a more sensitive measure in mental load assessment than the P300 amplitude in ERPs in their on-the-road study.
Fairclough et al. (1991) found an effect on HR of car-phone use. Average heart rate while performing a secondary task presented through a hands-free phone was found to be higher compared with the same task presented by an experimenter that accompanied the driver in the passenger seat. The authors give two possible explanations for the effect, either additional effort is required in the phone condition due to lack of cues in conversation, or unfamiliarity with cellular mobile phones aroused the subjects (cf. the practice effects found by Brookhuis et al., 1991). Van Winsum et al. (1989) found an effect of mental load on average HR and on the 0.10 Hz component of HRV. They found navigation based on a map to be more effortful than navigation by vocal messages, as measured by a decrease in power in the 0.10 Hz component band of HRV.
Janssen et al. (1994) did not find significant effects on the 0.10 Hz component in an on-the-road study in which a control group and two groups that received driver support were compared. The trend in the displayed figure, however, indicated decreased variability with driver support, a situation that could be comparable to the DETER Tutoring study. The authors suggested that the measure's insensitivity could be due to sensitivity to `an averaged workload level'. If so-called heart rate (variability) profiles had been determined, a more detailed picture might have emerged in that study.

EMG
One of the facial muscles that has been found to be sensitive to workload, is the frontalis (e.g., Van Boxtel & Jessurun, 1993). Zeier (1979) did not find an effect on EMG frontalis-activity of driving a car with automatic vs. manual gear transmission. However, he did find an effect of driving vs. being a passenger, the latter leading to lower muscle tension. The findings of Zeier (1979) support the idea that facial EMG activity taps a different dimension than (the 0.10 Hz component of) heart rate variability. Both in Zeier's study and the road layout study, EMG and HRV were differentially sensitive to workload. In addition, the two muscles that were measured, corrugator and frontalis, might also differ in selectivity. Jäncke (1994) found that the frontalis is not sensitive to emotional evaluation, while the corrugator is. A practical constraint of measurement of the corrugator in driving are the electrode positions that may interfere with the visual field.

ERPs
Measurement of Event Related Potentials (ERPs) has mainly been restricted to laboratory experiments. An exception to this are the studies reported by Janssen & Gaillard (1985). In two studies subjects had to drive an instrumented car through three road environments: through the city, over rural primary-roads and over motorways. During these rides they had to perform a secondary, auditory, Sternberg task. EEG was measured and P₃₀₀ amplitude and its latency to task-relevant stimulus presentation (a secondary task) was determined. In the first experiment P₃₀₀ amplitude was decreased and latency increased as a result of task load. City driving caused the largest increase in latency, surprisingly followed by motorway driving. In addition, motorway driving decreased P₃₀₀ amplitude most, while amplitude was equally decreased during city and rural primary road driving, compared with rest measurements. In the second, similar study, city driving was left out. No effects on the P300 were measured in this experiment. The authors report large individual differences and significant variance in the ERP data. They relate the remarkable position of motorway driving compared with the other conditions to the self-pacedness of the driving task. Complexity of the selected motorway section may, however, have had an effect on task demands (e.g., driving of a clover-leaf was included).

EDA
In different studies Electrodermal Activity (EDA) has been related to the traffic environment (for an overview see Fairclough, 1993). Michaels (1962) reports an increase in EDR amplitude with an increase in traffic density, while Brown & Huffman (1972) report an increase in SCL if there is more traffic and there are more traffic lanes. Most in-vehicle studies have been performed in the sixties and focused on the effect of traffic environment on driver's EDA. In the seventies, Zeier (1979) measured EDA with electrodes positioned on the inner side of the left foot. He compared the effect of three conditions on psychophysiological measures, driving a car with manual transmission, with automatic transmission or being a passenger in a car. Effects on Skin Conductance level were not significant, but SCR (Skin Conductance Responses) were most numerous while driving the car with manual transmission. Least SCR were measured in the condition where subjects were passengers.
EDA is not only sensitive to all SNS activation, it might also be susceptible to physical movements. This last aspect is particularly relevant in car driving where EDA generally is measured on the palm of the hand, while both hands have to be used in steering. In mental workload research EDA might be useful to assess overall SNS activation level, but movements artifacts are a possible source of disturbance.

Hormones
There are not many mental load studies that include the evaluation of hormone levels. In general, the measurement of hormone levels is restricted to situations in which the driver's occupation is very demanding. Examples of this type of stress research are the studies regarding city-bus drivers (Mulders et al., 1988) and coach drivers (Raggatt & Morrisay, submitted). One exception to the long-term impact studies was found, in a study reported by Zeier (1979) examining the effects of driving in heavy city traffic were examined. Adrenaline levels were found to be higher when driving a car as opposed to being a passenger. In addition, driving with manual transmission also led to higher adrenaline levels than driving with automatic transmission. No differences were found on noradrenaline levels.

properties of physiological measures

Background EEG is sensitive as an indicator of operator state, hence in region A to D. Average heart rate and heart rate variability in the time-domain are useful indicators of overall operator arousal level, i.e. in region D/B. The 0.10 Hz component however, is sensitive to task-related effort. It seems -as Mulder (1980) supposed- that the measure is sensitive to the Defense Response (Sokolov, 1963). The defense response is associated with a cardiovascular pattern of increased blood pressure, heart rate and stroke volume, decreased blood flow to renal, intestinal, and skin vascular beds, and increased skeletal muscle blood flow (Johnson & Anderson, 1990). The pattern is similar to responses evoked by stressful stimuli producing arousal in preparation for fighting. The defense response is coupled to increased sympathetic and reduced vagal activation, reflecting task-related effort and is accordingly connected to A3-region performance. Sensitivity of eye movements also seems to be highest in case of region A3 performance. Moreover, eye movements are related to visual demand, making it the highest diagnostic measure of table 12. Selectivity of EEG is low, operator state is reflected. The ECG measures differ in selectivity; HR and HRV are affected by many influences (respiration rate, physical effort) while this is less true for the 0.10 Hz component. Background EEG is a highly reliable, between-tests, measure for operator state, but individual differences (e.g., in the production of -waves) weaken this qualification. The many tests in which ECG measures were found to be sensitive to workload result in a reliability that is rated high. Primary-task intrusion when taking EEG and ECG measures is low once the electrodes have been attached. Measurement of eye movements may interfere with primary-task performance if cornea reflection is registered with the aid of a CEMRE. Intrusion is low if the driver's face is registered on video or in case of registration of EOG. Implementation requirements are high for most physiological measures, as special equipment such as sensitive amplifiers are required. For spectral analysis, for example, precise, i.e. 1 ms resolution R-top detection is required (L.J.M.Mulder, 1992). Special software is also needed. Only when average heart rate and HRV are determined, are implementation requirements less stringent. Finally, operator acceptance is inversely related to intrusiveness of measure registration. In table 12 the properties of different physiological measures are summarised.

Table 12. Summary of properties of physiological workload measures.
	Measure
Property	EEG Background	ECG: HR	ECG: HRV	ECG: .10 Hz	Eye movements (fixations/minute)
sensitivity (Region)	D-A2	D, B	D, B	A3	A3(?)
diagnosticity	low	low	low	low	high
selectivity	low	low	low-moderate	moderate-high	?
Reliability	moderate-high	high	high	high	?
primary-task intrusion	low	low	low	low	1
implementation requirements	high	moderate	moderate	high	high
operator acceptance	moderate	high	high	high	high-moderate1

¹ depends upon measurement technique

5.5 Discussion

Driving a vehicle is a task that demands continuous adaptation to a changing environment. A large part of the subtasks that have to be performed, such as lateral position control and speed maintenance, are tasks that are largely performed automatically at the control level, with hardly any driver effort. Representatives of performance measures at this level are the SDLP and steering wheel measures. At irregular intervals the control-level tasks are extended to include manoeuvre tasks, such as overtaking of other vehicles and following of leading cars. These tasks are not automated and require the driver's attention. Indicative measures of performance at this level are delay in car-following and the frequency of mirror checking.
A deteriorated driver state has been separated from increased task complexity as sources of increased workload. The effect of a deteriorated driver state and the increase in task complexity on primary-task performance might, however, appear to be the same. The primary-task parameter SDSTW changes in conditions of increased task complexity (e.g., Weaving Section study) and as a result of time-on-task. However, in combination with self-report ratings and physiology, a more differentiated picture emerges.
The pattern of measure sensitivity that emerges from the key studies (listed in table 3) is as follows: increased complexity, both in environment and in task, has an effect on the self-report scale RSME, and on the ECG. Task complexity vs. increases in environmental complexity seem to differentially affect the SDLP and SDSTW. Additional tasks lead to a decrease in SDLP and SDSTW, while an increase in complexity of the environment increases both measures. An affected driver state resulting from the consumption of alcohol or sedative drugs does not affect heart rate variability as much as increases in complexity do. Time-on-task mainly affects the average heart rate level and the driver's EEG. Ratings on the self-report scale RSME and activation scale are more sensitive to changes in driver state. Secondary-task performance, in particular the embedded task of car-following, is sensitive to both sources of increased workload.
Region of performance remains a very important factor, as an increase in a primary-task parameter such as the SDLP can be the result of being overloaded as well as of driver deactivation. It seems that all deviations from optimal performance, both as a result of increased and decreased demand, can be traced by the combination of performance parameters and self-report and/or physiological indices. The moment task demands increase and the driver has to try harder, i.e. has to invest effort, heart rate variability in the 0.10 Hz band will decrease. The 0.10 Hz component is in particular sensitive to the defense response when task demands increase, and the driver exerts task-related effort. The changes on this parameter as a result of state-related effort and driver deactivation are less conclusive. Though the effects are large in terms of size, they fail to reach the 5% level of significance. Only Egelund (1982) reports significant changes on this parameter as a result of fatigue. The self-report scale RSME has more general sensitivity to driver effort, irrespective of whether it concerns state-related effort or task-related effort. It seems that these two measures, in combination with a primary-task performance measure, are the most useful to assess mental workload in the complete A region.
In most of the experiments listed in table 3, peak loads (Verwey & Veltman, 1995) play only a limited role. Workload during the car-phone conversation, while driving over the Weaving Section or over the adapted road layout; in all three conditions overall workload was increased. Only driving with the tutoring device could lead to peak loads at the moment messages are issued. However, on the basis of conversations with subjects after completion of the experiment it seems that the increase in mental workload in this experiment is more related to continuously intensified monitoring of the road environment and speedometer, than to information processing peaks at the moment of warnings. In sum, sensitivity of measures as reported above is sensitivity to overall workload, but no conclusions with respect to sensitivity to peak loads can be made on the basis of these experiments.

Task interpretation, goal-setting
In the Car-phone study, Road layout and Tutoring experiments, an improvement on one of the primary-task measures, the SDLP, was found in the load condition. Since the effect of load in the three studies should be positioned in optimal performance section of the inverted-U, in the A3 region, no effect on primary-task parameters is expected. The task environment may imply that higher performance is required and the improvement in performance may be the result of increased effort (as measured by a reduction in 0.10 Hz heart rate variability and an increased RSME score). In principle, the primary-task measure could therefore also be used for the assessment of workload in the A3 (and possibly also the A1) region. The best description of performance measures in these regions would then be `no change or improvement in primary-task performance measures'. Finding an improvement in primary-task performance is paradoxical. Optimal performance is defined as the best performance, so no improvement is expected. In many laboratory tasks this is reasonable; in the field, however, conditions exist that allow for inaccuracies in primary-task performance during performance in the A-region. Unless subjects are given the strict instruction to drive in the centre of a lane and to try to steer as accurately as possible, improvement in primary-task performance can occur. A wide motorway lane, or the wide lanes used in the simulator experiment, do not necessitate accurate steering. Goal setting or Task interpretation is an important factor and the need to perform at the highest level possible is in general absent in driving and in field experiments. An improvement in performance was also found on the SDSTW-measure, in the load conditions of the Noise barrier and Tutoring studies. A similar explanation could be given for the improvement in lane-keeping performance, namely increased effort as indicated by physiological and self-report measures in both conditions results in increased primary-task performance.

Predicting the effects of tasks on driver mental workload is very difficult. Firstly, there are individual differences in goal setting and these differences vary from route choice to steering accuracy. Driving is to a large extent a self-paced task. If demands are too high, a slower driving speed can be chosen so as to be better able to deal with these demands. An elderly driver may prefer to make a detour so that he or she can drive over familiar roads thus facilitating the task environment. Once the task goals have been set, the task that has to be performed -the task demands- determine task complexity. How difficult a task is, however, depends upon capability (which may be lower for the elderly driver as just described), state and context. A novice driver will require more effort for vehicle control than an experienced driver. Driving performance itself can be related to externally set performance margins, critical levels, such as the margins proposed by Brookhuis (1995ab). Nevertheless only relative measures can give a further indication of mental workload. Strictly speaking, workload can only be determined per individual. It is always task X performed by individual Y (who is in a certain state) that leads to performance in Region Z. However, not all individuals are all that different and people often use similar strategies for performance of the same tasks. So, even though not all individuals set exactly the same goal, there are margins that are considered acceptable. Heavy swerving and leaving the motorway lane is not considered acceptable by most drivers. Task demands can accordingly be defined in terms of maintaining the vehicle between the lines of the driving lane. For experienced young drivers it is not likely that there is much difference in (e.g. self-reported) effort required for the basic task of lateral and longitudinal vehicle control. This makes a link between a certain task and a region of task performance possible. In table 3 expectations about the region of performance for the different driving tasks have been specified.
Nevertheless, the most important factor in the measurement of workload is to assess changes in mental workload. Performance with the use of any device, in any environment or state under investigation, should be compared with baseline performance, driving without the use of the device, under `normal' or standard conditions or while being sober. Changes in mental workload (measures) give a clear indication of what the effects of the changed demands are, incorporating at the same time changes in strategy or altered goals. This is, after all, the way people deal with changes in task demands in real life.

to chapter 6 (Conclusions)
back to chapter 5.1 Self-report measures
back to chapter 5.2 Primary-task performance measures
back to chapter 5.3 Secondary-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