Duration Analysis

Figure 5-1 indicates the cumulated probability of glance duration to the forward area and non-forward focal points for renewal cycles with different number of glances.

Figure 5-1 (a) indicates that the glance duration to the forward area in various types of renewal cycles differs slightly. The 90th percentile glance durations to the forward area are 10.7, 8.2, and 7.2 s for renewal cycles with 2, 3, and 4-or-more focal points, respectively. For the duration of non-forward glances, Figure 5-1(b) indicates that the attention was transited away from the frontal side longer in renewal cycles with more focal points. Using 2.0-s as the threshold, 8% of 2-glance renewal cycles were unsafe.

However, 33% of 3-glance and 74% of 4-or-more-glance renewal cycles might be unsafe since the time away from the forward side was more than 2.0 s. This result suggests that drivers glancing at more non-forward focal points consecutively in a renewal cycle were more likely to encounter insufficient time to respond to harmful changes in front of the vehicle. Therefore, to maintain situational awareness, drivers should avoid looking at too many focal points off-road in a sequence.

(a) Forward focal point (b) Non-forward focal points

Figure 5-1 Cumulated probability of glance duration

This section explores the characteristics of attention allocation when facing different maneuver intentions, distractions, traffic densities, time of day, and weather.

Because of the small sample size of 4-or-more-glance renewal cycles, the following analyses included only 2- and 3-glance renewal cycles.

(1) Maneuver intentions

Different maneuvers create different future trajectories and unique expectations of potential threats. Thus, drivers may distinctly concentrate on focal points and vary the glance duration at non-forward areas.

90 92

67

26

Table 5-1 indicates the descriptive statistics and ANOVA results for glance duration to forward and non-forward areas under different maneuver intentions.

The duration of the forward glance in each renewal cycle was not significantly affected by maneuver intentions (F = 0.93, p = .39), the number of glances (F = 1.15, p = .28), and their interaction (F = 1.98, p = .14). However, for the total duration of glancing at non-forward focal points in a renewal cycle, the ANOVA results indicate that the maneuver intentions (F = 4.16, p = .02), number of glances (F = 81.93, p = .00), and their interaction effects (F = 2.80, p = .06) are significant.

Table 5-1 ANOVA of glance duration under different maneuver intentions.

Sample size / Mean duration on forward area (Std.) / Mean duration on non-forward focal points (Std.) 2-glance renewal cycle 3-glance renewal cycle Total

Driving straight on segments 1039 / 4.08 (5.12) / 0.92 (0.79) 62 / 2.38 (2.92) / 2.11 (1.71) 1101 / 3.99 (5.03) / 0.98 (0.91) Changing lanes on segments 174 / 3.54 (4.64) / 0.81 (0.62) 24 / 4.31 (5.82) / 1.53 (0.86) 198 / 3.63 (4.79) / 0.89 (0.70) Passing through intersections 415 / 4.19 (4.80) / 1.10 (1.27) 36 / 3.48 (5.16) / 1.90 (1.44) 451 / 4.13 (4.83) / 1.16 (1.31)

ANOVA of glance duration at forward and non-forward focal points

Source of variance Sum of square Degree of freedom Mean square F (p-value)

Glance at forward focal point

*: Significant at the level of 0.1

Compared with driving straight on segments, both changing lanes and passing through an intersection may be considered to be mentally demanding tasks. Drivers must pay more attention to surroundings against the increased interaction with other vehicles. The higher mental effort resulted in distinct ways of allocating attention. As in the 2-glance renewal cycles, changing lanes on segments indicated decreased glance duration at both focal points, suggesting quick vision transitions between the forward area and the adjacent lanes. Shorter durations enable higher sampling rates for drivers to gather information more efficiently under pressure. Thus, they can reach a balance of distributing their

caused by a sudden deceleration by the leading vehicle. In addition, drivers would also pay close attention to the intersected roadway, particularly when there were vehicles attempting to enter the intersection. The close attention to both sides of the intersected roadway enabled continued observation to check for possible maneuvers by other vehicles constantly. On the other hand, under such circumstances, drivers may not be able to timely check the status of leading vehicles.

In 3-glance renewal cycles, drivers glanced at the forward area for shorter durations and on non-forward focal point for longer durations when driving straight on segments than for the two other tasks. The significant interaction effect shows that the average duration of each non-forward glance was shorter in 3-glance renewal cycles when changing lanes on road segments and when passing through intersections. This result may imply the uneasiness of shifting attention away to more focal points in these mentally demanding situations. The sample drivers speedily glanced at the non-forward focal points to reduce the possible deterioration of awareness of the leading area. Interestingly, driving straight on segments had the shortest forward glances among all maneuver intentions in 3-glance renewal cycles. On the contrary, it also had the longest non-forward glances among all maneuver intentions in 3-glance renewal cycles and the average duration was longer than the one in 2-glance renewal cycles.

This unique pattern may be related to leisurely driving, or sometimes probably to distracted activities. Because drivers spent almost half of the time not looking at the front area, such an attention allocation might expose the driver to dangerous risks of conflict with leading traffic.

Figure 5-2 indicates the cumulated probability of the glance duration at other focal points for various maneuver intentions. When changing lanes, only 5% of 2-glance renewal cycles and 29% of 3-glance renewal cycles transit vision away for more than 2.0 s. These results indicated compensatory behavior for fewer abnormal glances being undertaken, probably because of their nervousness in potential conflicts when changing lanes than other maneuver intentions.

However, 3-glance renewal cycles accounted for 11% of the renewal cycles in changing lanes on segments, comparing to 5.6%–8.0% for the other two maneuver intentions. Higher percentages of 3-glance renewal cycles when changing lanes, probably more areas needed to be glanced, suggests more opportunities to lose attention to leading traffic. Fortunately, as stated, drivers in general tried to shorten their glance durations for compensation. As for the maneuver of passing through intersections, Figure 5-2 (c) indicates the highest

probability of 2-glance renewal cycles exceeding the 2.0-s safety margin occurs when passing through intersections. Accordingly, when passing through intersections, compared to other tasks, drivers are more likely to miss information ahead owing to the observation of intersected road.

(a) Driving straight on

Figure 5-2 Cumulated probability of glance duration at non-forward focal points under different maneuver

(2) Distraction

Based on the ANOVA in Table 5-2, both distraction and number of glances in a renewal cycle significantly affected the duration of glancing at the forward area (F = 4.87, p = .03; F = 2.84, p = .09), and on non-forward focal points (F = 10.15, p = .00; F = 121.16, p = .00). However, there is no significant interaction effect of distraction and the number of forward and non-forward glance durations (F = 1.42, p = .23; F = 1.96, p = .16). As indicated in Table 5-2, the glance duration to the forward area with distractions was shorter than that of driving without distractions. In-vehicle distractions or clutter located on the roadsides increased the duration of shifting vision away from the front.

Table 5-2 ANOVA of glance duration under different distraction conditions

Sample size / Mean duration on forward area (Std.) / Mean duration on non-forward focal points (Std.) 2-glance renewal cycle 3-glance renewal cycle Total

Driving with distraction 522 / 3.65 (4.85) / 1.08 (0.96) 38 / 2.29 (3.75) / 2.24 (1.74) 560 / 3.56 (4.80) / 1.16 (1.07) Driving without distraction 1525 / 4.13 (4.91) / 0.91 (0.92) 124 / 3.90 (5.33) / 1.81 (1.30) 1649 / 4.11 (4.94) / 0.98 (0.98)

ANOVA of glance duration at forward and non-forward focal points

Source of variance Sum of square Degree of freedom Mean square F (p-value)

Glance at forward focal point

Figure 5-3 is the cumulated probability distribution of non-forward glance duration associated with and without distractions. When driving with distraction, 11% of 2-glance renewal cycles and 37% of 3-glance renewal cycles transited vision away from the forward area for more than 2.0 s. Specifically, for those 3-glance renewal cycles that exceeded the 2.0-s threshold, a large portion of them were quite long. There were around 5% of 3-glance renewal cycles that even exceeded 7 s. In addition, approximately a quarter of sample drivers in this study were affected by distractions. Undoubtedly, it is a challenging issue deserving of more attention.

(a) Driving without distraction

(b) Driving with distraction

Figure 5-3 Cumulated probability of glance duration at non-forward focal points with and without distractions

(3) Traffic density

Drivers may suffer more complex tasks from frequent interactions with other vehicles under heavy traffic conditions than in free flow conditions. Table 5-3 indicates that traffic density (F = 2.33, p = .07) and its interaction with the number of glances (F = 3.29, p = .02) significantly affected the glance duration to the forward area. However, the duration of forward glances were not significantly affected by the number of glances (F = 0.01, p = .90). The total duration of non-forward glances in a renewal cycle was significantly affected by traffic density (F = 3.43, p = .02), number of glances (F = 148.43, p = .00), and their interactions (F = 2.40, p = .07).

93

69

89

63

Table 5-3 ANOVA of glance duration under different Levels of Service (LOS)

Sample size / Mean duration on forward area (Std.) / Mean duration on non-forward focal points (Std.) 2-glance renewal cycle 3-glance renewal cycle Total

Level of service A 536 / 3.91 (4.68) / 0.94 (0.84) 47 / 3.66 (4.89) / 1.83 (1.06) 583 / 3.89 (4.70) / 1.02 (0.89) Level of service B 737 / 3.96 (4.72) / 0.93 (0.86) 65 / 2.45 (3.63) / 1.70 (1.03) 802 / 3.84 (4.66) / 0.99 (0.90) Level of service C 505 / 4.39 (5.33) / 1.02 (1.16) 25 / 3.67 (4.63) / 2.19 (2.25) 530 / 4.36 (5.30) / 1.08 (1.25) Level of service D/E 269 / 3.62 (4.94) / 0.95 (0.79) 25 / 5.87 (7.72) / 2.31 (1.77) 294 / 3.80 (5.26) / 1.06 (0.98)

ANOVA of glance duration at forward and non-forward focal points

Source of variance Sum of square Degree of freedom Mean square F (p-value)

Glance at forward focal point increased with traffic density from LOS A to C. Thereafter, it decreased in LOS D/E, probably owing to the heavy traffic, which seriously restricts drivers’

maneuvers. Drivers in these conditions may check the lead traffic quickly, and transit to surrounding areas frequently for possible lane changes or for relaxation.

By contrast, the 3-glance renewal cycles indicated that glance duration on both forward and non-forward focal points were the longest in LOS D/E. The interaction effect between numbers of glances and density shows increased average duration non-forward glances in 3-glance renewal cycle under LOS D/E.

In contrast with the 2-glance renewal cycles, drivers applying the 3-glance renewal cycles in LOS D/E might drive in a relaxed way without attempting to change lanes. Therefore, they transit vision at a slower pace without urgency.

However, long glances away from the front may increase the chances of drivers failing to notice changes in leading traffic. Moreover, the behavior in the 3-glance renewal cycles in LOS B indicated that drivers glanced shortest at both forward and non-forward focal points. This phenomenon suggests that drivers in moderate traffic conditions may drive more speedily and cautiously by increasing the sampling rate of collecting information from various focal points.

Figure 5-4 indicates the cumulated probability of the duration of non-forward glances under various traffic densities. The 3-glance renewal cycles in LOS D/E contained more long non-forward glances than those in LOS A/B density levels. Fifty-two percent of non-forward glances in 3-glance renewal cycles were longer than 2.0 s in LOS D/E, compared to 30%, 28%, and 32% in LOS A, B, and C, respectively. Moreover, 8.5% renewal cycles in LOS D/E were 3-glance, which was approximately the same level as in LOS A/B, and higher

than the percentage in LOS C (4.7%). This result suggests that drivers in this study did not decrease their percentage of 3-glance renewal cycles to compensate for the increasing traffic conflicts in LOS D/E. This negligence may result in more incidents in heavy traffic than in other traffic densities.

(a) LOS A (b) LOS B (c) LOS C (d) LOS D/E

Figure 5-4 Cumulated probability of glance durations at non-forward focal points under different levels of traffic density

Table 5-4 ANOVA of glance duration at different times of day

Sample size / Mean duration on forward area (Std.) / Mean duration on non-forward focal points (Std.) 2-glance renewal cycle 3-glance renewal cycle Total

Day time 1563 / 3.99 (4.87) / 0.97 (0.94) 132 / 3.50 (5.08) / 1.88 (1.42) 1695 / 3.95 (4.89) / 1.04 (1.02) Night time (with light) 393 / 3.70 (4.53) / 0.91 (0.89) 25 / 3.65 (5.35) / 2.12 (1.55) 418 / 3.70 (4.58) / 0.99 (0.98)

ANOVA of glance duration at forward and non-forward focal points

Source of variance Sum of square Degree of freedom Mean square F (p-value)

Glance at forward focal point

The ANOVA results in Table 5-5 indicate that the durations of both forward and non-forward glances were not affected by weather conditions, which was unexpected. Combining drivers’ similar behavior during the day and night time with light, the results suggest that the sample drivers were slightly aggressive.

91

Table 5-5 ANOVA of glance duration under different weather conditions

Sample size / Mean duration on forward area (Std.) / Mean duration on non-forward focal points (Std.) 2-glance renewal cycle 3-glance renewal cycle Total

Clear 1477 / 4.19 (4.97) / 0.96 (0.95) 111 / 3.54 (5.14) / 1.87 (1.19) 1588 / 4.14 (4.98) / 1.02 (1.00) Poor 570 / 3.55 (4.69) / 0.96 (0.86) 51 / 3.48 (4.89) / 1.99 (1.83) 621 / 3.54 (4.70) / 1.04 (1.03)

ANOVA of glance duration at forward and non-forward focal points

Source of variance Sum of square Degree of freedom Mean square F (p-value)

Glance at forward focal point