Parkes (1991) defined the primary task of the driver as maintenance of safe control over the vehicle. One of the major subtasks in vehicle control is lateral position control. Therefore, a measure of driving deviations from the centre of the lane is a good means to assess primary-task performance in car driving. Lateral deviation, or more specifically the SD of the Lateral Position (SDLP), has been shown to be a sensitive performance measure (e.g., Hicks & Wierwille, 1979, O'Hanlon et al., 1982, O'Hanlon, 1984, Brookhuis et al., 1985b, Green et al., 1993b). The task of keeping a vehicle between the lines of a lane is largely a psychomotor task involving eye-hand coordination. The term `tracking-ability' is sometimes applied to it (e.g., Stein et al., 1987), stressing the strong resemblance to the laboratory task.
Standard Deviation of the Lateral Position
In figure 6 the average (right-hand lane) SDLP in baseline and load conditions is displayed, while in figure 7
the change in SD of the lateral position compared with baseline is shown. This relative measure was
added to neutralize the differences in baselines between studies, which are likely to have been caused by
differences between roads/road segments, season (weather) and so on. The absolute value of the SDLP in the
Tutoring experiment is omitted from figure 6 because in the experiment the road width and test environment
were very different from the on-the-road tests. In both figures the critical impairment levels
(Brookhuis, 1995a, 1995b) are indicated, while in figure 7 the impairment in lateral position control found in an `alcohol calibration study' (Louwerens et al., 1987) is included.
An increase in SDLP, i.e. an increase in swerving, was found near the noise barrier (but only in the condition without eye-movement measurement), and as a result of alcohol (Alc(mw)) and prolonged driving (Fat(mw)). A decrease in the SDLP in the mental load condition was found in conditions in which subjects handled a car-phone (Prr and Pmw), when the enforcement system was switched on (Tut), and on the experimental road-layout (Wr and Mr). In some cases, the relative short section that was selected as load condition could have had an effect on SDLP. Near the noise barrier, for instance, the average lateral position on the road moved to the left. In the road-layout experiment the road surface and effective road width had been reduced, forcing drivers into more accurate lane-keeping. The effect of lane width on tracking performance was also found in a pilot study performed in a driving simulator (Green et al., 1993b), they found an increase in SDLP with increases in lane width. Taking these factors into account leaves only primary-task performance decrements on the SDLP measure as a result of alcohol and prolonged driving. In the Tutoring and Car-phone experiment primary-task performance under mental load, as measured by SDLP, even improved, while the sedative drug Triprolidine and driving on the Weaving Section did not lead to a significant increase in SDLP.
Figure 6. Standard deviation of the lateral position under baseline and mental load conditions. The studies from which the conditions were selected are listed in table 3. The indicated absolute threshold indicates driver impairment (see Brookhuis, 1995ab). The 95% confidence interval is also indicated.
Figure 7. Change in standard deviation of the lateral position under mental load compared with baseline measurements. The studies from which the conditions were selected are listed in table 3. The indicated relative threshold denotes driver impairment (see Brookhuis, 1995ab). The indicated BAC (Blood Alcohol Concentration) values are impairment levels as found in an `alcohol calibration study' (see Louwerens et al., 1987)
Standard Deviation of the Steering wheel movements
Related to the SDLP, but closer to one of the main sources of swerving, is the driver's steering behaviour. Due to relatively low-attentional driving demands, or due to attentional demands of additional tasks, drivers do not pay continuous attention to the lane-tracking (steering) task. This results in steering `holds', i.e. periods without steering-wheel movements (see Macdonald & Hoffmann, 1980, Godthelp et al., 1984). Several steering measures have been developed, from relatively simple measures, such as the number of zero-degree crossings of the steering-wheel or steering-reversal rate (McLean & Hoffmann, 1975), to more complex measures involving frequency analyses (McLean & Hoffmann, 1971, Blaauw, 1984) and compound functions (Fairclough, 1994). Steering-reversal rate (McLean & Hoffmann, 1975, Macdonald & Hoffmann, 1980) and the SD of the steering-wheel movements, always measured on straight road segments, are frequently used performance measures that are not complicated to calculate. In the figures 8 and 9 the (delta) SD of the steering-wheel movements (SDSTW), on sections with hardly any or no curvature is shown. Again the critical impairment level (Brookhuis, 1995ab) is displayed in both figures. In three studies the SDSTW increased in the load condition, in two studies to a level above the absolute impairment criterium.
Figure 8. Standard deviation of the steering-wheel movements under baseline and mental load conditions. The studies from which the conditions were selected are listed in table 3. The critical threshold level indicates driver impairment (see Brookhuis, 1995ab). If available the 95% confidence interval is displayed.
Figure 9. Change in standard deviation of the steering wheel movements under mental load compared with baseline measurements. The studies from which the conditions were selected from are listed in table 3. The indicated relative threshold indicates driver impairment (see Brookhuis, 1995ab).
The elevated SDSTW at the experimental road-layout (Mr) was unexpected. However, the two selected road segments may have differed slightly in curvature. The experimental road section was somewhat more curved. Road curvature may have had a similar effect on the SDSTW in the experiment that focused on the Weaving Section. In the other experiments completely straight or even the same motorway sections were compared with each other, which in general is to be preferred. In the simulator (Tut) and Noise Barrier study a significant decrease in SDSTW was found. A decrease in SDSTW may be indicative of increased steering effort, and thereby of more accurate steering, e.g., as a result of road environmental demands.
A combined statistical test
The statistical power of the individual tests can be increased by combined testing of the effects found in the different experiments. If it is assumed that it is the same parameter that is affected in the different studies (and that that parameter is mental workload), then the effects found in the studies can be tested in combination by (Snijders, 1995):
with
- k = the number of experiments
-
i = the estimated effect in experiment i
- SEi = standard error in experiment i
-
i = weight of experiment i.
= 0. This test was applied to the SDLP and SDSTW measures. The following results were found:
SD of the lateral position:
Complexitya,b: z = -0.29, NS
Complexity (environment)a,c: z = +1.87, p < 0.05
Complexity (task)b: z = -2.63, p < 0.005
Statec: z = +4.51, p < 0.0005
a = Without Wr and Mr due to reduced road width.
b = Weighted, = 1 for Prr and Pmw, = 2 for Tut.
c = All conditions equal weights.
The road layout experiment was excluded from the tests as changes in SDLP cannot be solely attributed to changes in mental workload, but are combined with effects of reduced road width. In the Complexity tests the two car-phone conditions were weighted to balance effects with the single condition of the tutoring experiment.
Increased complexity in terms of a change in environment as opposed to additional tasks have dissimilar effects on SDLP. Increased task complexity concur with reduced SDLP, while increased environmental demands coincide with an increase in SDLP. Tested together as effect of `complexity' levels out effects and renders a nonsignificant result. These results will be further discussed in chapter 5.5.
SD of the steering wheel movements
Complexitya: z = +3.86, p < 0.0005
Complexity (environment)d: z = +8.35, p < 0.0005
Complexity (environment)a: z = +4.93, p < 0.0005
Complexity (task)e: -
State: z = +16.45, p < 0.0005
a = Without Mr due to reduced road width in load condition.
d = Mr is included in this test with Mr= 2, while Weav = Weav(c) =
= NoiseB = NoiseB(c) = 1.
e = Not tested, only standard error information from one study (Tut) available
An increase in complexity of the environment and a decreased driver state both lead to a significant increase in the SD of the steering wheel movements. Increased task complexity reduces the SD of the steering wheel movements. These results will, together with the effects on SDLP, be discussed in chapter 5.5.
Other primary-task performance measures
Time-to-line crossing
While the SDLP and SDSTW mainly reflect performance at the control level, one level higher, at the manoeuvring level of performance, the Time-to-Line Crossing (TLC, Godthelp 1984) is a measure of driver primary-task performance. TLC is a continuous measure that represents the time required for the vehicle to reach either the centre or edge line of the driving lane if no further corrective steering-wheel movements are executed. TLC reflects the amount of time drivers can neglect path errors. Due to the measure's skewness, in general minimum, median or 15% TLC values are calculated (Godthelp et al., 1984, Godthelp, 1988). TLC is expected to reflect driving strategy and in particular occlusion strategy (time spent not looking at the road). With increases in mental-load, smaller TLC values can be expected; a more demanding task is likely to decrease the amount of time spent looking at the road.
In table 7 median and minimum TLC, as well as the change in TLC relative to baseline are depicted for the DREAM (for TLC see De Waard & Brookhuis, 1991b) and road-layout study. In the vigilance condition a decrease in TLC was found. This is in accord with the increase in number of steering-wheel holds that was found as a result of time-on-task (De Waard & Brookhuis, 1991b). In the road-layout study the layout of the road had been changed significantly. Drivers were more or less forced to drive close to the centre line and as a result the left-hand TLC decreased, while the right-hand TLC increased. This measure actually reflects the time required to reach an imaginary edge line, as the line had been removed! As a result, interpretation of the TLC measures in terms of mental load measures is not useful with data of the road-layout study.
Results with respect to TLC, SDLP and SDSTW from other author's studies
Riedel (1991) also used the TLC measure in a drug study in which subjects performed a driving task on the road. He found a maximum increase in median TLC (undifferentiated to line) of 0.15 s in the Triazolam condition, while baseline median TLC on the motorway was 4.69 s. SDLP in the same condition increased with 6.6 cm to 30.7 cm. On the basis of these data, he concluded that SDLP is the most sensitive measure for driver impairment.
The effect of Blood Alcohol Concentrations on SDLP as found by Louwerens et al. (1987) have been indicated in figure 6. The sensitivity of the measure to an affected driver state as a result of the use of hypnotics are summarized in Brookhuis (1995b). Significant increases in SDLP starting at 2.5 up to 7 cm are reported.
Van Winsum et al. (1989) compared steering-wheel movements of drivers who navigated from a map with the steering-wheel movements of drivers who were guided vocally. They found no effect on steering-wheel movements in the more demanding map condition. This result may be related to the urban road environment. It is likely that the use of most primary-task control indices (SDLP, SDSTW) is confined to non-urban environments. In urban traffic most steering-wheel movements will be related to longitudinal and lateral tracking demands (Wildervanck et al., 1978).
Green et al. (1993a) compared driving behaviour and self-report ratings of difficulty of route guidance messages using three different interfaces. Only slight differences in SD of steering wheel movements were found, the largest SD of steering wheel movements were measured when the information was displayed in the instrument panel (1.1 degrees), followed by a simulated Head-up display (1.0 degrees). The SD of the steering wheel movements were smallest (0.9 degrees) for auditory presented information. Ratings of difficulty of use of the route guidance information while driving that were given after the test rides (Green et al, 1993a, p.82) followed the same pattern, the lowest difficulty rating being given for the auditory information. However, memory load in the case of auditory route guidance was largest. In all three conditions route guidance information was additionally combined with information regarding vehicle state and traffic information that was presented to the driver in the instrument panel at a different location. This additional information could have interacted with the route guidance messages and therefore a relation between type of interface and mental load is hard to assess accurately.
Fairclough (1994) measured steering-wheel movements in a study in which subjects drove under the influence of low amounts of alcohol, and under placebo conditions. Just as in the DREAM study (see figure 8) he found an increase in the standard deviation of steering-wheel movements of drivers with a BAC up to 0.5 ‰.
Other primary-task measures in other studies
Apart from the above-mentioned accuracy measures in vehicle control, sometimes speed measures are used in the assessment of primary-task performance. An example of a speed measure is the time that is required to finish a route. Both Jordan & Johnson (1993) and Fairclough et al. (1991) found the time required to complete a route to be significantly longer in the load condition in which subjects had to adjust a stereo or had a conversation, compared with normal driving along the same route. The measure can be indicative of a strategic choice for a lower driving speed to compensate for high information load, and accordingly lead to a decrease in mental load. Similar compensatory strategies are reported for slower decision making and slower action performance in elderly drivers (Brouwer & Ponds, 1994). Brown et al. (1969) also found an increase in time required to finish driving a circuit as a result of the use of a car-phone, while Van Winsum et al. (1989) found the same effect -an overall lower driving speed- when they compared map navigation with vocal-route guidance. However, the application of the measure is rather rough, and in non-controlled environments, e.g. in on-the-road studies, the measure is susceptible to disturbance factors such as traffic density. The use of speed measures as a sensitive indicator of increased mental load seems, therefore, to be the most reliable in laboratory and simulator experiments.
Properties of primary-task performance measures
Lane-keeping in experienced drivers is to a large extent determined by automatic, control-level processing. Consequently, measures of accuracy in lane-keeping, such as the SDLP and SDSTW, would not be expected to be sensitive to variations in mental workload in the A-region. The different experiments, however, show that this is not the case, both SDLP and SDSTW being sensitive measures. A likely explanation for this is that there is no `pure' automatic and controlled behaviour, but that aspects of automatic behaviour remain influenced by controlled processing (Schneider and Fisk, 1983). Strategy sets performance margins and the inaccuracies that are allowed. This also clarifies why improvement on these primary-task performance measures is possible. Increased task demands can lead to increased driver effort, which increase primary-task performance if under baseline conditions inaccuracies are allowed. This issue will be further discussed in chapter 5.5.
Although improvement in primary-task performance measures is possible, in general, affected task performance implies reduced task performance, and this is the case in the D, B and C regions. As task performance is at a minimum level in the C-region, performance measures will no longer vary with changes in demand in that region. Sensitivity of the SDLP and SDSTW is highest in the B and D regions. In studies in which driver state was reduced, a decrease in SDLP and SDSTW was found. The same is true for the increased environmental demand studies. Diagnosticity of the measures is low, although the difference in direction of the effect as found between Complexity environment vs. Complexity task may be an indication of differential sensitivity. Selectivity is hard to assess on the basis of the driving studies reported above. Hardly any physical effort is required in driving, and emotional stress, for instance, was not tested. It is quite possible that the measures are affected by these factors and therefore selectivity is expected to be relatively low. Sensitivity to mental workload as found in the different tests results in a `high' rating for reliability. The implementation requirements for the measurement of steering wheel movements are low. A potentiometer mounted on the steering wheel column with a measurement range of 90 degrees (± 45 degrees) and a resolution of 0.1 degrees is adequate for accurate measurement of movements on noncurved road sections. For the measurement of the vehicle's lateral position more complex equipment is required. A useful device is the so-called `lane tracker', which resembles a camera but the interior consists of an array of diodes that are sensitive to differences in light intensity. The camera is directed towards the road delineation. A relatively cheap but labour-intensive solution is to make video registrations of the road scene (De Waard & Steyvers, 1995). The advantage of the latter technique is that it can also be applied on roads without delineation. In the future progress in camera techniques will probably facilitate automatic detection of road delineation or road shoulder. Operator acceptance of the measures is high because registration is unobtrusive. Table 8 provides an overview of primary-task measures' properties.
| Table 8. Summary of properties of primary-task workload measures. | ||
|---|---|---|
| Measure | ||
| Property | SDLP | SDSTW |
| sensitivity (Region) | D, B | D, B |
| diagnosticity | low | low |
| selectivity | (low) | (low) |
| Reliability | high | high |
| primary-task intrusion | none | none |
| implementation requirements | high | low |
| operator acceptance | high | high |
to chapter 5.3 Secondary-task performance measures
to chapter 5.4 Physiology and discussion
back to chapter 5.1 Self-report 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