Cardiac Measures of Driver Workload during Simulated Driving with and without Visual Occlusion
Backs, Richard W., Lenneman, John K., Wetzel, Jacob M., Green, Paul, Human Factors
Psychophysiology has a long tradition within human factors (Boucsein & Backs, 2000) and has especially contributed to the present understanding of mental workload (Gaillard & Kramer, 2000). Mental workload is often considered to reflect the costs associated with a person's expenditure of limited-capacity information-processing resources to keep task performance within specification and thus is a function of both the person's abilities and the task demands on his or her abilities (Gopher & Donchin, 1986). Many methods of assessing mental work-load have been proposed, each with its own advantages and disadvantages (O'Donnell & Eggemeier, 1986; Tsang & Wilson, 1997). Psychophysiological methods have been used to assess mental workload in domains such as aviation because measurement of responses such as heart rate usually does not interfere with task performance and because psychophysiological responses are sometimes more sensitive to task demands than are performance measures (Kramer & Weber, 2000). The present study examined whether a new approach to the analysis and interpretation of cardiac psychophysiological responses that has been useful for assessing mental workload in the aviation domain is also useful for simulated driving.
By far, the most popular cardiac response used by the human factors community is heart rate (Wilson & Eggemeier, 1991). The heart is dually innervated by the sympathetic and parasympathetic branches of the autonomic nervous system (ANS), and these two branches have opposing effects on heart rate: sympathetic activation increases heart rate, whereas parasympathetic activation decreases heart rate. In the classic model of ANS function (e.g., Cannon, 1959), sympathetic and parasympathetic activity are reciprocally coupled--that is, sympathetic activation occurs concomitantly with parasympathetic inhibition and vice versa. According to the classic model, heart rate change in response to varying task demands would always be the result of some unknown combination of reciprocal change in both autonomic branches.
However, Backs (1995) reviewed studies from the aviation domain that used heart rate as an index of pilot mental workload and found many instances in which heart rate did not change in a manner consistent with the classic ANS model. He suggested that a newer model of ANS function proposed by Berntson, Cacioppo, and Quigley (1991, 1995) could better account for the observed heart rate results. The Berntson et al. model of autonomic space subsumes the classic model of ANS function and instead proposes multiple "modes of autonomic control" (1991, p. 459)." The modes of autonomic control can be represented as a two-dimensional "autonomic space" (see Figure 1), which can be illustrated by axes plotting parasympathetic activity on the ordinate and sympathetic activity on the abscissa. Vectors on the positive diagonal represent the classic coupled reciprocal modes of autonomic control (sympathetic activation with parasympathetic inhibition, or the reverse). Vectors on the negative diagonal represent coupled nonreciprocal modes of control (coactivation and coinhibition), in which the sympathetic and parasympathetic branches increase or decrease together. Vectors parallel to one axis represent uncoupled modes of control, in which activity in one branch changes but activity in the other branch does not. Table 1 presents the autonomic control mode taxonomy and the effects of change along each mode of autonomic control on heart rate.
[FIGURE 1 OMITTED]
Backs (1995) described two important limitations in using heart rate to make inferences about mental workload, which are evident in Table 1. Both limitations exist because heart rate alone is uninformative about the psychological-physiological mapping responsible for the response. The first is that sensitivity can be limited because heart rate may not change with varying task demands, even though sympathetic and parasympathetic activity may change greatly. The second limitation is that equivalent heart rate changes can occur as the result of many different patterns of neural input, which therefore limits the diagnosticity of heart rate in localizing the effects of task demands to specific information-processing resources.
Knowledge of the underlying mode of autonomic control can potentially overcome both of the inferential limitations with heart rate because different control modes may reflect distinct psychological-physiological mappings (Berntson et al., 1991). Sensitivity can increase because change in task demand that fails to significantly elicit heart rate change may be detectable in autonomic space. This increase in sensitivity will especially occur when the mode of autonomic control is one of the coupled nonreciprocal modes, coactivation or coinhibition.
Similarly, diagnosticity can increase because a given heart rate response to changes in task demands may be attributable to control modes that reflect the engagement of specific processing resources. Increased heart rate can occur as a result of reciprocally coupled sympathetic activation, uncoupled sympathetic activation, uncoupled parasympathetic inhibition, coactivation (in which sympathetic activation exceeds parasympathetic activation), of coinhibition (in which parasympathetic inhibition exceeds sympathetic inhibition). Research suggests that different task manipulations elicit different modes of autonomic control for heart rate (Backs, 2001). For example, increasing task difficulty in a visual-manual tracking task by increasing the order of control or by increasing the memory load of a secondary memory task will both result in increased heart rate. However, the two manipulations elicit different modes of control as determined by factor analysis of multiple cardiovascular measures: Increasing the order of control from velocity to acceleration elicits uncoupled parasympathetic inhibition, whereas increasing the memory load of the secondary task elicits uncoupled sympathetic activation (Backs, 1995, 1998).
However, noninvasive measures of the underlying sympathetic and parasympathetic activity are needed to identify the autonomic control modes for heart rate. Pre-ejection period (PEP) obtained from the impedance cardiogram and respiratory sinus arrhythmia (RSA) obtained from heart rate variability have been validated as measures of sympathetic and parasympathetic cardiac innervation, respectively, in dual pharmacological blockade studies (Cacioppo et al., 1994). PEP and RSA were used in the present study to assess cardiac ANS activity.
The present research examined whether the autonomic control mode approach assessed driver mental workload during simulated driving better than did heart rate. Driving difficulty was manipulated by varying the curvature of the road driven (sharper curves have a greater workload) and whether or not the road was occluded (greater workload) while driving. Performance, subjective, and physiological measures have all been found to differ across curves of different radii (Godthelp, 1986; Richter, Wagner, Heger, & Weise, 1998; Tsimhoni & Green, in press; Tsimhoni, Yoo, & Green, 1999). Driving performance deteriorates, subjective difficulty increases, and heart rate generally increases as the degree of curvature increases.
Visual occlusion is a technique used to assess the visual demand of driving (Senders, Kristofferson, Levison, Dietrich, & Ward, 1967). Briefly, the logic of the visual occlusion technique is that the more difficult the driving situation (the sharper the curve, the narrower the road, etc.), the more the driver needs to look at the road. There are at least 10 ways this can be achieved (Green, 2001), which involve various combinations of drivers closing their eyes, blocking their view of the scene with goggles, of, in a simulator, blanking the forward scene. For ease of implementation, recent research (e.g., Tsimhoni et al., 1999) had drivers press a button every time they wanted a 0.5-s glimpse of the road, a typical glance duration to the road. The proportion of time the road is visible, ranging from 0.0 to 1.0, is a measure of visual demand in which a visual demand of 1.0 means the driver needs to look at the road all of the time to drive safely.
Use of the visual occlusion technique as a task difficulty manipulation in the present study was intended as an analog for driving with the eyes off the road. Tsimhoni et al. (1999), who used the same simulator as in the present study, found that compared with a baseline driving condition, the deterioration in simulated driving performance was very similar for driving with visual occlusion and driving while performing a secondary electronic map task. Further, use of the visual occlusion technique in the present study permitted direct comparison of visual demand and cardiac measures of driver workload.
Fifteen male university students (age range 18-30 years, median 21 years) who were in good health and free of medications that affect the cardiovascular system were paid $20 for their participation. All participants were licensed drivers who reported driving between 4800 and 35 400 km per year (M = 20 000 km). Only 1 participant had previous experience in a driving simulator.
The electrocardiogram (ECG) and the first derivative of pulsatile changes in transthoracic impedance (dZ/dt) were obtained from a Minnesota Impedance Cardiograph Model 504B using the Biopac physiological data acquisition system. A tetrapolar montage of Mylar band electrodes was used for acquisition of the ECG and impedance cardiogram (ICG: Sherwood et al., 1990).
Data were collected in the University of Michigan Transportation Research Institute Driver Interface Research Simulator (Olson & Green, 1997). The driving simulator consisted of a Macintosh computer system projecting images of the driving landscape, an A-to-B pillar mockup of a car, a projection screen, a torque motor connected to the steering wheel, and a sound system providing engine, drive train, and wind noise. The projection screen for the driving landscape was 7.3 m in front of the driver and provided a 33[degrees] horizontal by 23[degrees] vertical field of view. The driving environment depicted in the simulation was a two-lane winding road with stationary oncoming cars, traffic signs, road edge posts, and no traffic ahead. The width of the lanes was 3.66 m, and the width of the vehicle was 1.83 m. Auditory and tactile feedback were given to the driver if the vehicle exceeded the centerline of right edge line. When the left tire of the vehicle reached the centerline of the road, there was a pulse to the steering wheel and a sound for each dashed centerline stripe, simulating driving over reflective markers. When the right tire reached the right edge of the road, there was a pull on the steering wheel (simulating the tire dropping off of the edge), and then the wheel vibrated as if driving over gravel (and the sound of driving over gravel was presented).
The participants drove a course through a series of straightaways and curves at a fixed (cruise-controlled) speed of 72.4 km/hr. Curves of three radii were driven: 582, 291, and 194 m (3, 6, and 9 degrees of curvature, respectively, for which the degree of curvature was defined as the angle subtended by an arc of 30.5 m). Each curve occurred twice (once to the left and once to the right) during the course in a fixed order: left 3 degrees; right 9 degrees; left 6 degrees; right 6 degrees; left 9 degrees; right 3 degrees. Each 5-min run through the course began with a 40-s ramp-up to speed on a straightaway followed by the sequence of curves separated by straight sections (tangents). Each separating straight took 15 s to drive. Unlike most driving simulations, in this simulation each curve lasted for a constant duration of 30 s to permit the acquisition of sufficient physiological data to compute the cardiac measure used to estimate parasympathetic nervous system activity.
Participants drove the course seven times: one practice run without visual occlusion; two practice runs with visual occlusion; two test runs with visual occlusion; and two test runs without visual occlusion. The test runs with visual occlusion always followed the practice runs with visual occlusion, forming a block of four successive occlusion runs to facilitate stabilization of the visual demand scores. Psychophysiological data were collected during all seven runs and during a 120-s resting baseline period that occurred before the first simulation run but while participants were seated in the simulator. Data from only the four test runs are presented here.
During simulation runs with visual occlusion, the road scene was normally blank. As in previous studies, pressing a switch mounted on the steering wheel caused the road scene to appear for 0.5 s. Holding the switch down did not extend the duration that the road scene was presented, even if it was held down after the 0.5-s period had elapsed. However, the road scene was presented if a depression occurred after the 0.5-s period had elapsed.
Performance. Driving performance data were collected at 30 Hz. Three measures of driving performance (the standard deviation of lateral position in meters, the standard deviation of steering wheel angle in radians, and the frequency of lane excursions per minute) were recorded during the same 30-s curve epochs used to score the physiological measures. The frequency of lane excursions per minute was defined as the time during which the centerline or right edge line was exceeded by the center of the vehicle deviating from the center of the lane by more than [+ or -] 0.92 m ([lane width--vehicle width)/2]).
Visual demand. As noted earlier, visual demand (VisD) is an indicator of visual mental workload. As defined by Tsimhoni et al. (1999), visual demand is computed as the viewing time divided by the time between the start of successive viewing periods:
(1) Vis[D.sub.i] = 0.5/[[t.sub.i] - ([t.sub.i] - 1)],
in which Vis[D.sub.1] = visual demand for the ith occlusion period, 0.5 = nonoccluded period (in seconds), [t.sub.i] = time when current nonoccluded period begins, and [t.sub.1] - 1 = time when previous nonoccluded period began.
Visual demand scores could range from 0.0 (always occluded) to 1.0 (never occluded); thus higher scores correspond to greater visual demand. Mean visual demand scores were obtained across the same 30-s epochs as the physiological and performance measures.
Psychophysiological. ECG and ICG data were sampled at 1000 Hz using the Biopac Acknowledge data acquisition system during each run of the simulation. The 30-s epochs of physiological data across the resting baseline and during each curve within the four test runs were scanned for artifact, and six psychophysiological measures were computed using MindWare Technologies psychophysiological data reduction modules. Heart period (the time in milliseconds between successive R waves) has statistical properties (Berntson et al., 1993) superior to those of heart rate (the number of beats per minute), but in the present study there was no difference between the two measures, so for ease of interpretation only heart rate data will be presented. However, heart period was used to obtain RSA, which was calculated as the natural logarithm of the variance in the high-frequency band (0.12-0.40 Hz) of heart rate variability. PEP was calculated as the time in milliseconds between the Q wave of the ECG (onset of ventricular depolarization) and the B wave of dZ/dt (onset of left ventricular ejection into the aorta). Respiration rate in breaths per minute and amplitude in arbitrary units was obtained from the impedance pneumogram (Ernst, Litvak, Lozano, Cacioppo, & Berntson, 1999) to determine whether cardiac change was associated with respiration change.
Heart rate, PEE RSA, and respiration rate and amplitude from each 30-s curve segment from each of the four test simulation runs were converted to change scores from the mean of the four 30-s epochs across the resting baseline. Positive change scores for heart rate indicate faster heart rate during the test runs than during resting baseline, whereas positive change scores for respiration rate and amplitude indicate faster and deeper respiration during the test runs. Negative change scores for RSA indicate suppression of heart rate variability (i.e., parasympathetic inhibition) during the test runs compared with resting baseline, whereas negative PEP change scores from baseline indicate faster ventricular contraction (i.e., sympathetic activation) during the test runs.
Mean change-from-baseline heart rate, PEE RSA, and respiration rate and amplitude were analyzed separately using SPSS for Windows (Green, Salkind, & Akey, 2000). The results from this traditional univariate method of analysis of psychophysiological data were contrasted with ah analysis of the autonomic control modes. Autonomic modes of cardiac control were estimated with multivariate analyses of PEP and RSA using SAS Proc GLM (SAS Institute, Inc., 1989).
A 2 (visual occlusion) x 3 (curve radius) repeated-measures analysis of variance (ANOVA) was conducted for most dependent measures, except for VisD and lane excursions. VisD was available only for visual occlusion conditions, making curve the only repeated measure. A third repeated-measures factor was used for the line excursions measure to examine whether the type of lane excursion (i.e., crossing the edge-line of the centerline of the road) differed across the visual occlusion and curve radius factors. An alpha of .05 was used to determine statistical significance; Huynh-Feldt epsilon-corrected alpha levels were used for all effects involving the curve radius factor (only those epsilons that were less than 1.0 are reported here). Post hoc orthogonal polynomials were used to follow up significant effects involving the curve radius factor for the performance and VisD measures because the levels represent equal changes in degrees of curvature. Eta-squared was used to estimate effect size for all significant effects.
Two preliminary ANOVAs were conducted for each dependent measure to test for order effects. One analysis included the factors of test run (first or second) and curve repetition (first or second), whereas the other included the factors of test run and curve direction (right or left). With one exception, for respiration rate, there were no significant main effects for any of these factors. There were no significant interactions involving the test run factor, but visual demand, steering wheel angle, and RSA differed across the two occurrences of the 582-m curve. However, except for respiration rate, these analyses will not be considered further because the ordinal relation of the measures across curves was not affected.
For the standard deviation of lateral position, significant main effects were found for both occlusion, F(1, 14) = 46.61, p < .0001, MSE = 2.818, [[eta].sup.2] = .76, and curve radius, F(2, 28) = 10.17, p < .001, MSE = .388, [[eta].sup.2] = .42. These data are presented in Figure 2a. The standard deviation of lateral position was greater with visual occlusion (M = 0.64 m) than without (M = 0.29 m). Post hoc tests conducted across the three levels of curve revealed a significant linear trend, which accounted for 98.0% of the variance, and that the standard deviation of lateral position during each curve was significantly different from the others on paired t tests (all ps < .001).
[FIGURE 2 OMITTED]
Similar results were obtained for the standard deviation of steering wheel angle, which can be seen in Figure 2b. A significant main effect of occlusion was present, F(1, 14) = 43.35, p < .0001, MSE = .012, [[eta].sup.2] = .75, in which the standard deviation of steering angle was higher with visual occlusion (M = 0.13 rad) than without (M = 0.01 rad). A main effect of curve radius was again found, F(2, 28) = 19.55, p < .001, MSE = .006, [[eta].sup.2] = .58, in which a significant linear trend across the three curves accounted for 96.2% of the variance. Follow-up t tests revealed that the standard deviation of steering wheel angle during each curve was significantly different from that of the others (all ps < .001), except for the 582- and 291-m curves, which only approached significance, t(14) = 1.95, p = .071.
The number of lane excursions per minute was analyzed across occlusion conditions, curve radii, and line excursion types. A significant main effect was obtained for occlusion, F(1, 14) = 95.94, p < .0001, MSE = 2.260, [[eta].sup.2] = .87, in which more excursions per minute were made with visual occlusion (M = 2.82) than without (M = 0.65). A main effect for curve radius was also obtained, F(2, 28) = 22.67, p < .0001, MSE = 1.090, [[eta].sup.2] = .61, in which a significant linear trend across the three curves accounted for 98.7% of the variance and t tests revealed that the number of lane excursions per minute during each curve was significantly different from that of the others (all ps < .01). A significant main effect of line excursion type also occurred, F(1, 14) = 51.01, p < .0001, MSE = 3.994, [[eta].sup.2] = .78, in which more centerline (M = 2.80) than edge line (M = 0.67) excursions were made per minute.
Unlike the other performance measures, the Occlusion x Curve Radius interaction was significant for excursions per minute, F(2, 28) = 4.99, p < 02, MSE = 19.689, [[eta].sup.2] = .26. Further, the Curve Radius x Excursion Type, F(2, 28) = 13.65, p < .0001, MSE = 21.539, [[eta].sup.2] = .49, and Occlusion x Excursion Type, F(1, 14) = 27.91, p < .0001, MSE = .697, [[eta].sup.2] = .66, interactions were also significant, and the three-way interaction approached significance, F(2, 28) = 5.46, p < .06, MSE = 22.272, [[eta].sup.2] = .19, [epsilon] = .79. These effects are presented in Figure 3, in which it can be seen that centerline excursions per minute increased as curve radius decreased, both with, F(2, 28) = 25.52, p < .0001, and without visual occlusion, F(2, 28) = 4.34, p < .03, but that edge-line excursions per minute did not change significantly across curves. The linear trend was significant for centerline excursions with and without visual occlusion; however, whereas all paired t tests across curves were significant with occlusion, only the 582- and 194-m curves differed without occlusion (all ps < .03). Finally, the Occlusion x Excursion Type interaction revealed that the bias toward centerline excursions was significantly greater when the road scene was occluded.
[FIGURE 3 OMITTED]
VisD increased significantly as curve radius decreased, F(2, 28) = 52.17, p < .0001, MSE = .002, [[eta].sup.2] = .78, in which a significant linear trend accounted for 99.8% of the variance. Follow-up t tests revealed that VisD during each curve differed significantly from that of the others: 582 m (M = .278, SD = .083) < 291 m (M = .320, SD = .086) < 194 m (M = .369, SD = .083).
Unlike performance, none of the physiological measures were affected by visual occlusion of the road scene. However, all cardiac measures were affected by curve radius, but not in the same way (see Table 2 for means). The main effect of curve radius was significant for heart rate change, F(2, 28) = 5.36, p = .017, MSE = 8.955, [[eta].sup.2] = .277, [epsilon] = .817, for RSA change, F(2, 28) = 4.58, p = .022, MSE = .006, [[eta].sup.2] = .258, and for PEP change, F(2, 28) = 4.04, p = .029, MSE = 7.454, [[eta].sup.2] = .224. Follow-up t tests revealed that heart rate during the 582- and 291-m curves was significantly faster than during the 194-m curve, whereas PEP was significantly shorter during the 291-m curve than during the 582- or 194-m curves. Similarly, RSA was significantly more suppressed during the 291-m curve than during the 582-m curve and was marginally more suppressed during the 582-m curve than during the 194-m curve (p = .077). The cardiac changes were not attributable to changes in respiration. There were no significant effects of visual occlusion or curve radius for respiration rate or amplitude. Although neither respiration measure was significantly different from baseline, respiration rate change differed significantly between the first (M = 0.76 breaths/min) and second (M = -0.40 breaths/min) repetition of the curves, F(1, 14) = 4.788, p = .046, MSE = 25.29, [[eta].sup.2] =. 17.
Autonomic Modes of Cardiac Control
A 2 (visual occlusion) x 3 (curve radius) repeated-measures multivariate analysis of variance (MANOVA) was conducted with RSA and PEP as dependent variables to identify the autonomic modes of cardiac control during simulated driving. However, before these two dependent variables were submitted to the MANOVA, they were converted to standard deviation units so that they would have a common scale for plotting purposes. Each variable was divided by its standard deviation, which preserved the meaning of the sign of the variable (i.e., negative values representing sympathetic activation for PEP and parasympathetic inhibition for RSA). With this scaling, a value of [+ or -] 1.0 means that the measure differed from baseline by one standard deviation. Statistical significance in the MANOVAs was determined by F ratios computed from Wilks's lambda that had a probability of .05 or less. Similar to the univariate analyses, the MANOVA revealed that the modes of control for heart rate differed with curve radius but not with visual occlusion or their interaction.
Modes of autonomic control for heart rate change elicited during simulated driving of the three curve radii were determined from the vectors in an autonomic space defined by axes plotting the standardized RSA (parasympathetic activity) on the ordinate and the standardized PEP (sympathetic activity) on the abscissa. Because RSA and PEP were computed from baseline-corrected data, the origin represents no change from resting baseline. Negative values of the axes correspond to autonomic activity that elicits faster heart rate (i.e., sympathetic activation or parasympathetic inhibition). Figure 4 illustrates the modes of autonomic control for the curve radius main effect, in which the vector for each curve radius is plotted using the degree of curvature as the plotting symbol.
[FIGURE 4 OMITTED]
Physiologically, the vectors in the autonomic space illustrated in Figure 4 provide information about the underlying sympathetic and parasympathetic innervation responsible for heart rate change during each curve. Because heart rate was significantly faster than baseline during all three curves--t(14) ranged from 2.46 to 3.85, all ps < .03--each vector was first tested to see if it was significantly different from zero using Hotelling's [T.sup.2] (Littell, Freund, & Spector, 1991). Consistent with heart rate, all three vectors differed significantly from zero (resting baseline), in which [T.sup.2](14) ranged from 76.32 to 277.67, all ps < .0001. Follow-up univariate t tests of standardized RSA and PEP were then conducted to see if they varied in a manner consistent with increased heart rate from baseline. The suppression in standardized RSA was significant for all three vectors, for which t(14) ranged from -8.48 to -15.42, all ps < .0001; however, the decrease in standardized PEP was significant only for the vector for the 291-m curve, t(14) = -3.13, p < .007. Thus all three vectors within autonomic space are in directions that indicate increased heart rate, but they are not in the same direction.
The direction of the vector within the space indicates the mode of autonomic control responsible for heart rate change. The multivariate main effect of curve radius, F(4, 54) = 3.78, p = .009, means that the direction of the three vectors differed significantly. Therefore, three follow-up 2 (visual occlusion) x 2 (curve radius) simple MANOVAs were conducted to compare all pairs of curve radii to find out how they differed. Similar to the findings from the univariate analyses, standardized PEP and RSA during the 291-m radius curve differed significantly from those for the 582-m curve, F(2, 13) = 7.21, p = .008, and the 191-m curve, F(2, 13) = 7.24, p = .029, but did not differ between the 582- and 191-m curves. Figure 4 illustrates the modes of autonomic control during each curve. The vector for the 291-m curve (6 degrees of curvature) was classified as reciprocally coupled sympathetic activation and parasympathetic inhibition because the decrease from baseline along both the PEP and RSA axes was statistically significant. The vectors for the 582-m (3 degrees of curvature) and 194-m (9 degrees of curvature) curves were classified as uncoupled parasympathetic inhibition because only the decrease along the RSA axis was statistically significant.
Correlations between Visual Demand and Cardiac Measures
Pearson product-moment correlation coefficients between VisD and the three cardiac measures were computed to better understand the relation between visual demand and cardiac change across the curve radius manipulation. Table 3 presents the correlation coefficients between mean VisD and the mean change-from-baseline scores for heart rate, RSA, and PEP collapsed across the two simulation runs with visual occlusion. As can be seen, only heart rate showed a consistently significant association with VisD. The negative correlation between heart rate and visual demand indicates an inverse relation between the two measures; participants with the highest VisD had the smallest heart rate increase from baseline. In summary, heart rate decreased as the need for visual information increased, but RSA and PEP were not associated with the intake of visual information.
The purpose of the present study was twofold: to evaluate whether cardiac modes of autonomic control had better sensitivity and diagnosticity than heart rate to well-validated manipulations of simulated driving task difficulty, and to compare visual demand and cardiac measures of driver workload. The various measures examined differed in their sensitivity to the two task manipulations. Performance, visual demand, and the cardiac measures were affected by the radius of the curves to be negotiated within the simulation, but only performance was affected by occlusion of the road scene. Although all measures were sensitive to the curve radius manipulation, three distinct patterns of effects emerged. The first pattern was evident in the driving performance and visual demand measures, whereas the other two patterns were evident in the cardiac measures. Because dissociations were observed from performance and visual demand and from each other for both task manipulations, the cardiac measures have the potential to be diagnostic of the information-processing demands of simulated driving (Brookhuis & De Waard, 2001).
The first pattern was the decrease in driving performance and increase in visual demand as curve radius decreased. Driving performance and VisD very closely replicated previous studies using the same task manipulations (Tsimhoni & Green, in press; Tsimhoni et al., 1999). Secondary-task performance (Kantowitz, 1995) and subjective (Tsimhoni & Green) measures of driver workload have also been consistently observed to follow this pattern during simulated driving. Therefore, the dissociations observed with the cardiac measures need to be interpreted against these findings.
The second pattern was the slower heart rate during the 194-m radius curve than during the 582- and 291-m curves, a pattern that was also observed in a field study conducted by Richter et al. (1998). In their study, heart rate increased from baseline over curves from very low to moderate degrees of curvature but then decreased for curves with the highest degrees of curvature during on-road driving. Heart rate has long been known to decrease during the intake and processing of perceptual information (Lacey, 1969). The visual demand measure verified that drivers in the present study needed more visual information to negotiate the 194-m curve as compared with the 582- and 291-m curves. Further, the correlation between visual demand and heart rate indicated that drivers who required the most visual information to negotiate the curves (i.e., had the highest VisD) also had the lowest heart rate change from baseline. Thus the lower heart rate in the 194-m radius curve would seem to be indicative of the perceptual processing demands of driving the simulation.
The third pattern was the more suppressed RSA and shorter PEP during the 291-m radius curve than during the other curves. The explanation for this pattern was evident in the autonomic space analysis; decreased PEP with suppressed RSA is indicative of the coupled reciprocal sympathetic activation and parasympathetic inhibition mode of autonomic control. To our knowledge this is the first study to examine these cardiac measures during simulated driving, but laboratory studies using single-axis visual-manual compensatory tracking as an analog for vehicular control (Ash & Backs, 2000; Backs, Knowles, & Short, 2001; Lenneman & Backs, 2000) can aid in the interpretation of this pattern. Those studies found that task manipulations that increase the perceptual/ central information-processing demands of tracking elicit an uncoupled parasympathetic inhibition mode of control, whereas tracking task manipulations that increase the motor response information processing demands of tracking do not. Dual-task studies have found that adding working memory or mental arithmetic tasks that are performed concurrently with tracking elicit change along uncoupled sympathetic activation modes of control (Backs, 1995, 1998).
One interpretation of the results of those studies is that parasympathetic inhibition reflects the perceptual processing demands of tracking and that sympathetic activation reflects the central processing demands of tracking. Although not supported by other evidence in the present study, this interpretation would suggest that more central processing was required for driving the 291-m radius curve than for the others because significant sympathetic activation was observed only during this curve. However, a more definitive conclusion on the relation between autonomic modes of control and driver workload will require additional research, especially dual-task studies, using more cognitively demanding driving situations.
Performance and the cardiac measures also dissociated for the visual occlusion manipulation. Several possible reasons for the absence of visual occlusion effects on the cardiac measures can be posited. We believe that the most likely explanation is that this manipulation limited the quality of the input data available for processing but did not affect the quantity of processing resources devoted to driving (Norman & Bobrow, 1975) and that the cardiac measures reflect the effortful expenditure of processing resources (Mulder, 1986). That is, there was no need for participants to invest more processing resources when driving with visual occlusion than without visual occlusion because performance could not be sustained in the absence of high-quality visual data.
However, just as looking at a visual stimulus during simulated driving does not guarantee that it will be processed when attention is directed elsewhere (Strayer & Johnston, 2001), the absence of a visual stimulus does not mean that it is not still being processed. Although it seems likely that the visual occlusion technique has multiple influences on the information-processing demands made on the driver (e.g., increasing the demands on central processing to maintain the road scene in memory), these effects were minimized in the present study. Memory demands were low because the participants' task and the simulation course and road scene were fairly simple. Further, because participants were provided with auditory and tactile feedback when the vehicle exceeded the centerline or edge line, they could still drive (albeit less precisely) by using input from the auditory and tactile modalities instead of the visual modality. Thus there was no change in the cardiac measures with visual occlusion of the road scene, even though driving performance deteriorated, because the participants' effort invested in driving was the same with and without visual occlusion.
An alternative explanation for the absence of visual occlusion effects on the cardiac measures may be that task demands on different memory systems (Baddeley, 1986) have different physiological consequences. That heart rate increases and heart rate variability decreases as task demands on verbal working memory increase has been well established (e.g., Backs & Seljos, 1994; Mulder & Mulder, 1981), but whether the same relations hold as task demands on spatial working memory increase is unclear. Presumably, the cardiac measures would vary As a function of the effort invested, regardless of the type of memory employed, but this presumption needs to be explicitly tested.
In summary, the patterns of dissociation found in the present study indicate that although the extraction of the modes of cardiac control did not improve sensitivity over heart rate alone, information about the mode of control can improve diagnosticity. The relation between visual demand and heart rate, along with the patterns of measure dissociation, seemed to isolate the perceptual processing demands of driving from the central of motor processing demands. Backs, Lenneman, and Sicard (1999) suggested how knowledge of autonomic control modes may be used to classify mental workload in the aviation domain. They proposed a tentative mental workload hierarchy of autonomic control modes for heart rate during flight simulation: coupled reciprocal control during low workload; uncoupled sympathetic activation with high but manageable workload; and coactivation to situations having critical consequences that must be immediately addressed.
However, the tentative mental workload hierarchy of autonomic control modes proposed by Backs et al. (1999) for simulated flight may need modification for extension to simulated driving. The present results suggest that uncoupled parasympathetic inhibition may be a lower level of the hierarchy, one that was not observed during simulated flight. Additional research on the autonomic modes of control is clearly needed to establish the validity this approach to the psychophysiological assessment of mental workload.
The present study confirms the value of physiological assessment as a research tool in the driving domain and illustrates that no single measure will suffice for the assessment of driver mental workload. However, a carefully selected set of physiological measures can provide diagnostic information about driver workload that cannot otherwise be readily obtained. One implication of the present study is that the combined application of psychophysiological and visual occlusion methodologies may be a powerful research tool to assess performance and processing resource cost trade-offs associated with using alternative modalities of information presentation in advanced technologies such as in-vehicle navigation and communication systems.
Funds for the support of the research were provided to the first author by a Faculty Research and Creative Endeavors Award from Central Michigan University and by General Motors Corporation.
TABLE 1: The Berntson, Cacioppo, and Quigley (1991) Mode of Autonomic Control Taxonomy and the Effect of Each Control Mode on Heart Rate Sympathetic Parasympathetic Control Mode Input Input Reciprocally Coupled Modes Sympathetic activation/ Increase Decrease parasympathetic inhibition Parasympathetic activation/ Decrease Increase sympathetic inhibition Nonreciprocally Coupled Modes Coactivation Increase Increase Coinhibition Decrease Decrease Uncoupled Modes Sympathetic activation Increase -- Sympathetic inhibition Decrease -- Parasympathetic activation -- Increase Parasympathetic inhibition -- Decrease Heart Rate Control Mode Response Sympathetic activation/ Increase parasympathetic inhibition Parasympathetic activation/ Decrease sympathetic inhibition Coactivation Increase, decrease, or no change (a) Coinhibition Increase, decrease, or no change (a) Sympathetic activation Increase Sympathetic inhibition Decrease Parasympathetic activation Decrease Parasympathetic inhibition Increase (a) Depending on the relative changes in the two branches. TABLE 2: Means and Standard Deviations for the Physiological Changes from Baseline across Curves Curve Radius 582 m 291 m Physiological Measure Mean SD Mean SD Heart rate (beats/min) 2.443 (a) 2.457 2.603 (a) 2.804 RSA (n[[ms.sup.2]]) -0.655 (a) 0.521 -0.900 (b) 0.522 PEP (ms) 0.100 (a) 7.306 -0.683 (b) 6.983 Respiration rate 0.455 (a) 3.097 -0.098 (a) 3.890 (breaths/min) Curve Radius 194 m Physiological Measure Mean SD Heart rate (beats/min) 1.542 (b) 2.419 RSA (n[[ms.sup.2]]) -0.719 (a,b) 0.686 PEP (ms) 0.259 (a) 7.340 Respiration rate 0.179 (a) 4.386 (breaths/min) Note: Means with different subscripts are significantly different from each other on paired t tests (all ps < .025); N = 15. TABLE 3: Correlations between Cardiac Measures and Visual Demand at Each Curve Radius 582 m 291 m 194 m Heart rate -.67 ** -.58 * -.52 * RSA -.00 .10 .35 PEP -.48 -.36 -.40 * p < .05; ** p < .01. N = 15.
Ash, I., & Backs, R. W. (2000, August). Cardiac measures of attentional resource demands during continuous manual tracking. Paper presented at the Third International Conference on Psychophysiology in Ergonomics, San Diego, CA.
Backs, R. W. (1995). Going beyond heart rate: Modes of autonomic control in the cardiovascular assessment of mental workload. International Journal of Aviation Psychology, 5, 25-48.
Backs, R. W. (1998). A comparison of factor analytic methods of obtaining cardiovascular autonomic components for the assessment of mental workload. Ergonomics, 41,733-745.
Backs, R. W. (2001). An autonomic space approach to the psychophysiological assessment of mental workload. In P. A. Hancock & R A. Desmond (Eds.), Stress, workload, and fatigue (pp. 279-289). Mahwah, NJ: Erlbaum.
Backs, R. W., Knowles, J. M., & Short, T. (2001). Cardiac and respiratory measures of practice effects and processing resource demand during continuous manual tracking [Abstract]. Psychophysiology, 38(Suppl. 1), S22.
Backs, R. W., Lenneman, J. K., & Sicard, J. L. (1999). The use of autonomic components to improve cardiovascular assessment of mental workload in flight simulation, International Journal of Aviation Psychology, 9, 33-7.
Backs, R. W., & Seljos, K. A. (1994). Metabolic and cardiorespiratory measures of mental effort: The effects of level of difficulty in a working memory task. International Journal of Psychophysiology, 16, 57-68.
Baddeley, A. D. (1986). Working memory. Oxford, UK: Oxford University Press.
Berntson, G. G., Cacioppo, J. T., & Quigley, K. S. (1991). Autonomic determinism: The modes of autonomic control the doctrine of autonomic space, and the laws of autonomic constraint. Psychological Review, 98, 459-487.
Berntson, G. G., Cacioppo, J. T., & Quigley, K. S. (1993). Cardiac psychophysiology and autonomic space in humans: Empirical perspectives and conceptual implications. Psychological Bulletin, 114, 296-322.
Boucsein, W., & Backs, R. W. (2000). Engineering psychophysiology as a discipline: Historical and theoretical aspects. In R. W. Backs & W. Boucsein (Eds.), Engineering psychophysiology: Issues and applications (pp. 3-29). Mahwah, NJ: Erlbaum.
Brookhuis, K. A., & De Waard, D. (2001). Assessment of drivers' workload: Performance and subjective and physiological indexes. In P. A. Hancock & P. A. Desmond (Eds.), Stress, workload, and fatigue (pp. 321-333). Mahwah, NJ: Erlbaum.
Cacioppo, J. T., Berntson, G. G., Blinkley, P. F., Quigley, K. S., Uchino, B. N., & Fieldstone, A. (1994). Autonomic cardiac control: II. Noninvasive indices and basal response as revealed by autonomic blockade. Psychophysiology, 31, 586-598.
Cannon, W. B. (1939). The wisdom of the body. New York: Norton.
Ernst, J. M., Litvak, D. A., Lozano, D. L., Cacioppo, J. T., & Berntson, G. G. (1999). Impedance pneumography: Noise as signal in impedance cardiography. Psychophysiology, 36, 335-338.
Gaillard, A. W. K., & Kramer, A. F. (2000). Theoretical and methodological issues in psychophysiological research. In R. W. Backs & W. Boucsein (Eds.), Engineering psychophysiology: Issues and applications (pp. 31-58). Mahwah, NJ: Erlbaum.
Godthelp, H. (1986). Vehicle control during curve driving. Human Factors, 28, 211-221.
Gopher, D., & Donchin, E. (1986). Workload: An examination of the concept. In K. R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Handbook of perception and human performance: Vol. II. Cognitive processes and performance (pp. 41/1-41/49). New York: Wiley Interscience.
Green, P. (2001, November). Visual occlusion to assess the demands of driving and tasks: The literature. Paper presented at Exploring the Occlusion Technique: Progress in Recent Research and Applications, Turin, Italy. Retrieved December 17, 2003, from http://www.umich.edu/~driving/occlusionworkshop2001
Green, S. B., Salkind, N. J., & Akey, T. M. (2000). Using SPSS for Windows: Analyses and understanding data (2nd ed.). Upper Saddle River, NJ: Prentice Hall.
Kantowitz, B. H. (1995). Simulator evaluation of heavy-vehicle driver workload. In Proceedings of the Human Factors and Ergonomics Society 39th Annual Meeting (pp. 1107-1111). Santa Monica, CA: Human Factors and Ergonomics Society.
Kramer, A. F., & Weber, T. (2000). Applications of psychophysiology to human factors. In J. T Cacioppo, L. G. Tassinary, & G. G. Berntson (Eds.), Handbook of psychophysiology (2nd ed., pp. 794-814). New York: Cambridge University Press.
Lacey, J. I. (1969). Autonomic indices of attention, readiness, and rejection of the external environment. In D. P. Kimble (Ed.), Readiness to remember (pp. xxx-xxx). New York: Gordon and Breach.
Lenneman, J. K., & Backs, R. W. (2000). The validity of factor analytically derived cardiac autonomic components for mental workload assessment. In R. W. Backs & W. Boucsein (Eds.), Engineering psychophysiology: Issues and applications (pp. 161-174). Mahwah, NJ: Erlbaum.
Littell, R. C., Freund, R. J., & Spector, P. C. (1991). SAS system for linear models (3rd ed). Cary, NC: SAS Institute.
Mulder, G. (1986). The concept and measurement of mental effort. In G. R. J. Hockey, A. W. K. Gaillard, & M. G. H. Coles (Eds.), Energetics and human information processing (pp. 175-198). Dordecht, Netherlands: Martinus Nijhoff.
Mulder, G., & Mulder, L. M. J. (1981). Information processing and cardiovascular control. Psychophysiology, 18, 392-402.
Norman, D. A., & Bobrow, D. G. (1975). On data-limited and resource-limited processes. Cognitive Psychology, 7, 44-64.
O'Donnell, R. D., & Eggemeier, F. T. (1986). Workload assessment methodology. In K. R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Handbook of perception and human performance: Vol. II. Cognitive processes and performance (pp. 42/1-42/49). New York: Wiley Interscience.
Olson, A., & Green, E A. (1997). A description of the UMTRI driving simulator architecture and alternatives (Tech. Report UMTRI-97-15), Ann Arbor, MI: University of Michigan Transportation Research Institute. Retrieved December 17, 2003, from http://www.umich.edu/~driving/abstracts/umtri_97_15.html
Richter, P., Wagner, T., Heger, R., & Weise, G. (1998). Psychophysiological analysis of mental load during driving on rural roads--A quasi-experimental field study. Ergonomics, 41, 593-609.
SAS Institute, Inc. (1989). SAS/STAT user's guide (Version 6, 4th ed., Vols. 1 and 2). Cary. NC: Author.
Senders, J. W., Kristofferson, A. B., Levison, W. H., Dietrich, C. W., & Ward, J. L. (1967). The attentional demand of automobile driving. Highway Research Record, 195, 15-33.
Sherwood, A., Allen, M. T., Fahrenberg, J., Kelsey, R. M., Lovallo, W. R., & van Dooran, L. J. P. (1990). Methodological guidelines for impedance cardiography. Psychophysiology, 27, 1-23.
Strayer, D. L., & Johnston, W. A. (2001). Driven to distraction: Dual-task studies of simulated driving and conversing on a cellular telephone. Psychological Science, 12, 462-466.
Tsang, P. S., & Wilson, G. H. (1997). Mental workload. In G. Salvendy (Ed.), Handbook of human factors and ergonomics (2nd ed., pp. 417-449). New York: Wiley Interscience.
Tsimhoni, O., & Green, P. A. (in press). Visual demand of driving curves as determined by visual occlusion. In A. G. Gale, I. D. Brown, C. M. Haselgrave, & S. P. Taylor (Eds.), Vision in vehicles VIII. Amsterdam: Elsevier Science.
Tsimhoni, O., Yoo, H., & Green, P. (1999). Effects of workload and task complexity on driving and task performance for in-vehicle displays as assessed by visual occlusion (Tech. Report UMTRI-99-37). Ann Arbor, MI: University of Michigan Transportation Research Institute.
Wilson, G. F., & Eggemeier, F. T. (1991). Psychophysiological assessment of workload in multi-task environments. In D. L. Damos (Ed.), Multiple-task performance (pp. 217-278). London: Taylor & Francis.
Richard W. Backs is an associate professor of psychology at Central Michigan University. He received his Ph.D. in experimental psychology from the University of Southern California in 1984.
John K. Lenneman is a research scientist at General Motors Research & Development and Planning in Warren, Michigan. He received an M.S. in experimental psychology from Central Michigan University in 1998 and an M.S. in industrial and operations engineering from the University of Michigan in 1999.
Jacob M. Wetzel is a doctoral student at Central Michigan University, where he received his M.S. in experimental psychology in 1999.
Paul Green is a senior research scientist at the University of Michigan Transportation Research Institute and an adjunct associate professor of industrial and operations engineering and mechanical engineering at the University of Michigan, where he received a joint Ph.D. in industrial and operations engineering and psychology in 1979.
Date received: February 28, 2002
Date accepted: April 7, 2005
Richard W. Backs, Central Michigan University, Mount Pleasant, Michigan, John K. Lenneman, General Motors Corporation, Warren, Michigan, Jacob M. Wetzel, Central Michigan University, Mount Pleasant, Michigan, and Paul Green, University of Michigan, Ann Arbor, Michigan
Address correspondence to Richard W. Backs, Department of Psychology, Central Michigan University, Mount Pleasant, MI 48859; firstname.lastname@example.org. HUMAN FACTORS, Vol. 45, No. 4, Winter 2003, pp. 525-558. Copyright [c] 2005, Human Factors and Ergonomics Society. All rights reserved.…
Questia, a part of Gale, Cengage Learning. www.questia.com
Publication information: Article title: Cardiac Measures of Driver Workload during Simulated Driving with and without Visual Occlusion. Contributors: Backs, Richard W. - Author, Lenneman, John K. - Author, Wetzel, Jacob M. - Author, Green, Paul - Author. Journal title: Human Factors. Volume: 45. Issue: 4 Publication date: Winter 2003. Page number: 525+. © 2002 Human Factors and Ergonomics Society. COPYRIGHT 2003 Gale Group.