The influence of attention allocation and age on intersection accidents

Transportation Research Part F 43 (2016) 1–14

Contents lists available at ScienceDirect

Transportation Research Part F journal homepage: www.elsevier.com/locate/trf

The influence of attention allocation and age on intersection accidents Juela Kazazi ⇑, Susann Winkler 1, Mark Vollrath 1 Technische Universität Braunschweig, Department of Traffic and Engineering Psychology, Gaußstraße 23, 38106 Braunschweig, Germany

a r t i c l e

i n f o

Article history: Received 19 May 2015 Received in revised form 14 July 2016 Accepted 3 September 2016

Keywords: Driving simulation Left-turn intersection accidents Driver age Gaze behaviour

a b s t r a c t Many severe accidents occur in urban areas. As part of the research project UR:BAN, this study investigated the causes of driver errors (e.g., inadequate attention allocation) in urban areas when turning left at intersections. As intersection accidents are especially difficult for older drivers, differences between older and younger drivers were examined as well. In a first step, accident protocols of left turn crashes with pedestrians and bicyclists were analysed in detail, since they are the most dangerous ones. Characteristics of the oncoming traffic and the location of crossing bicyclists and pedestrians were identified as possible causes. Accordingly, critical scenarios were implemented in a static driving simulator, varying the characteristics of the oncoming traffic, the direction and location of crossing vulnerable road users. These factors were examined in a within-subject design, with two different aged groups of participants (12 aged 20–35 y, 12 aged 65+ y; between-subjects factor). The results revealed that the presence of the oncoming traffic, which was assumed to capture the drivers’ attention, did not lead to more accidents with vulnerable road users. However, this may be because many drivers waited until the oncoming traffic had passed. Unexpectedly, older drivers had fewer accidents. This may be explained by the more cautious behaviour of older drivers, who drove significantly slower and waited significantly longer at the stop line before turning. Further analyses showed that a more cautious behaviour, independently of the age, predicted accident avoidance better than attention allocation. From these results, warning systems for younger and older drivers, especially for those not driving cautious, need to be developed. This idea will be tested in future studies introducing different warning concepts. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction In contrast to highways, the complexity and density of events in urban areas is much higher and each intersection might require a different reaction of the driver. For example, while making a turn the driver often has to keep track of many traffic elements simultaneously (e.g., oncoming traffic, vulnerable road users), continually process new information and make proper decisions (Lord, Smiley, & Haroun, 1998). Furthermore, when looking at German traffic accidents in urban areas from

⇑ Corresponding author. Fax: +49 (0) 531 391 8181. E-mail addresses: [email protected] (J. Kazazi), [email protected] (S. Winkler), [email protected] (M. Vollrath). 1 Fax: +49 (0) 531 391 8181. http://dx.doi.org/10.1016/j.trf.2016.09.010 1369-8478/Ó 2016 Elsevier Ltd. All rights reserved.

2

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

2011, accidents with pedestrians crossing comprise 27% of all fatal accidents. These accidents are the most frequent and most severe ones (high death rate) in urban areas (Morgenroth et al., 2009; Robertson & Carter, 1984; Statistisches Bundesamt, 2012). Especially left turns at intersections represent a considerable safety problem to vulnerable road users. These manoeuvres are more challenging for drivers, especially older ones, since they demand extra attention and thus increase the mental workload and the detection of other road users (Fu et al., 2011; Harms, 1991; Lord et al., 1998). Consequently, as the presence of vulnerable road users at intersections is very high, these situations are relevant if one wants to prevent accidents in urban areas. One aim of the research project UR:BAN (www.urban-online.org) is to prevent these kinds of accidents by developing advanced driver assistance systems (ADAS), which are able to warn the driver early. The current study was conducted within this research project. In order to develop an effective warning strategy, it is helpful to understand how and why these accidents happen, with special regard to driver errors involved in the causation of accidents. The first step was an analysis of urban intersection accidents with vulnerable road users, in order to define relevant scenarios and derive hypotheses about why these accidents occurred. The complexity of the primary driving task can influence the peripheral detection of vulnerable road. For instance, a driver who is approaching an intersection has to concentrate on the intersection (primary task) and has to detect potential critical situations or objects (peripheral detection). Additionally, a variety of subtasks while driving constitutes to the driver’s workload (Lord et al., 1998). Harms (1991) examined drivers’ cognitive load when they were performing three manoeuvres at rural junctions in Sweden: straight, left turn and right turn. The results showed that the mental load was greatest when turning left. Correspondingly, accidents while turning left, especially with vulnerable road users travelling in the same direction as the driver, are quite frequent. The reason for this may be that vulnerable road users travelling in this direction are harder to detect than oncoming pedestrians or bicyclists (Insurance Institute for Highway Safety, 2000). Moreover, as drivers have to yield to oncoming traffic, their attention is directed in this direction. Collisions may then occur when relevant objects (bicyclist, pedestrian) appear in the periphery of sight and are not detected, since the attention of drivers is focused somewhere else. Thus, inadequate responses (Caird & Hancock, 2002) occur, caused by an inadequate allocation of attention (Caird & Chugh, 1997), visual search difficulties (McDowd & Shaw, 2000) and inappropriate selective attention (Owsley et al., 1998). Consequently, driver attention problems and late detection of traffic conflicts (Rumar, 1990) are generally cited as causal factors in a large proportion of crashes. This contribution of attention allocation to overseeing road users, are already revealed in previous studies (Larsen & Kines, 2002; Räsänen & Summala, 1998; Werneke & Vollrath, 2012). For example, studies showed that the attention of participants while making a right turn was directed more towards the left side of the intersection, compared to the right side (Summala, Pasanen, Räsänen, & Sievänen, 1996; Werneke & Vollrath, 2012), since possible traffic might rather appear from the left side. Accordingly, in this situation it is difficult to react to vehicles or obstacles appearing at the right side of an intersection. Thus, attention plays a crucial role in the causation of intersection accidents, especially when turning left. Presumably, accidents should be more likely if relevant critical objects, like pedestrians or bicyclists, appear at locations where the attention is not focused on. For example, when drivers turn left, a bicyclist travelling in the wrong direction, at least in Germany (same direction as the driver), should be much harder to be notice in time, than a bicyclist travelling in the right direction (same direction as oncoming traffic). Moreover, in demanding situations, the attention is focused more on the oncoming traffic, thus the overseeing of oversee relevant objects becomes more likely. Investigating intersection accidents is especially relevant for older drivers as they are overrepresented in crashes at intersections and in left turn gap-acceptance crashes (Caird & Hancock, 2002; Mayhew, Simpson, & Ferguson, 2006; McGwin & Brown, 1999; OECD, 2001). Garber and Srinivasan (1991) revealed that the accident involvement ratios of older drivers to younger ones, for right and left turnings, are significantly higher than for straight-through movements, the ratio for left turning being the highest. The higher accident involvement of older drivers is explained by various changes with age, including narrowing of the visual field, increased time required to change the focus, problems with depth perception and slower decision-making (Dewar, 1995; Tarawneh, McCoy, Bishu, & Ballard, 1993). Additionally, Anstey, Wood, Lord, and Walker (2005) pointed out that age-related changes in various aspects of visual attention, including selective attention, divided attention and sustained attention (i.e., vigilance) are relevant for older drivers’ accident occurrence. Yet, other researchers doubt that the accident risk of older drivers is truly higher. When controlling for exposure, older drivers with a low yearly mileage, as compared to younger drivers with a similar low mileage, have even a lower accident risk (Hakamies-Blomqvist, O’neill, & Raitanen, 2002; Hanson & Hildebrand, 2011; Janke, 1991). This is referred to as the ‘‘low-mileage bias”, a concept explored and demonstrated in research by e.g. Hakamies-Blomqvist et al. (2002). If the larger crash risk of older drivers is due to the larger exposition to urban areas, as revealed by Janke (1991), then there should be no differences in the crash risk when examining them in the same situations as younger drivers. Additionally, older drivers may compensate for their cognitive impairments by simplifying their driving task, for example by driving slower, accelerating at a slower pace or crossing the intersection with a larger gap size of the oncoming traffic, as compared to younger drivers (Case, Hulbert, & Beers, 1970; Fofanova, Maciej, & Vollrath, 2011; Hakamies-Blomqvist, Siren, & Davidse, 2004; Rackoff, 1974; Reed, Kinnear, & Weaver, 2012; Vollrath, Maciej, Howe, & Briest, 2009). Yan, Radwan, and Guo (2007) conducted a driving simulator experiment for left turn gap acceptance at a stop-controlled intersection. This study examined the effects of traffic speed, driver age and gender on gap acceptance behaviour, driver’s acceleration rate and steering action. A total of 63 participants were divided into three age groups (young: 20–30 years; mid aged: 31–55 years; old: 56–83 years). The participants had to make a left turn with slow (40.2 km/h) as well as fast

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

3

oncoming traffic (88.5 km/h). The results showed significant age effects with respect to the gap acceptance. Older drivers tended to accept larger gaps than younger drivers did. The gap size acceptance decreased with the speed of the oncoming traffic for all age groups. There was an interaction between age and the speed of the oncoming traffic. During the slow oncoming traffic, older drivers selected significantly larger gaps and accelerated at a slower pace, while no age differences were found for the fast moving traffic. These results show that older drivers have a more defensive driving behaviour, possibly to compensate for their cognitive impairments (Yan et al., 2007). These results show that comparing older and younger drivers in left-turn situations at intersections is a very relevant field of studies with regard to traffic safety. On the one hand, problems with divided attention as causes of accidents could be larger for older drivers. Yet, on the other hand, older drivers’ more cautious driving behaviour and life-long experience could prevent accidents. The following study was conducted in a static driving simulator as it enabled an easy manipulation of situation characteristics and a presentation of identical situations to the different age groups of drivers. Moreover, as critical situations were expected to happen, subjects could not have been harmed even if a (simulated) accident would have occurred, making this study suited for the simulator. This study examined older drivers’ performance at intersections in an experimental setting. The objective was to find out if they cause more accidents and if they try to compensate for their perceived deficits, for example by driving slower. As attention allocation problems (looking at oncoming traffic and missing to see vulnerable road users crossing) may be an important factor in the causation of accidents when turning left, attention was investigated by varying the characteristics of the oncoming traffic (complexity and presence) as well as the location and travelling direction of a crossing vulnerable road user. In this study, attention allocation was defined as the gaze percentage on the incoming traffic, since it was expected that the complexity of the traffic would cause an inadequate attention allocation. It was assumed that drivers would allocate their attention primarily to certain areas of the intersection (especially on complex oncoming traffic) but neglect others (vulnerable road users). Additionally, it was expected that more collisions would occur with vulnerable road users travelling from an unexpected location. To sum it up, the following hypotheses are examined in the following study: 1. Older drivers show a different driving and gaze behaviour compared to younger drivers. 2. Different complexities of oncoming traffic (complex, normal, without) have an effect on the driving and gaze behaviour of older and younger drivers. 3. The location (expected vs. unexpected) of a vulnerable road user has an effect on the driving and gaze behaviour of older and younger drivers. 4. The crossing direction (opposite direction vs. same direction as the ego) of a bicyclist has an effect on the driving and gaze behaviour of older and younger drivers. Variations within this study will be described in more detail in Section 2.3. 2. Method 2.1. Participants A total of 24 drivers (14 male, 10 female) participated in the following study. There were two groups of participants, 12 younger drivers and 12 older drivers. The younger drivers were aged between 20 and 35 years (six females and six males), with a mean age of 25.8 years (SD = 4.1 years), owning their driver’s license on average for 7.8 years (SD = 3.8 years). The group of older drivers was aged above 65 years (four females and eight males), with a mean age of 70.6 years (SD = 3.5 years), owning their driver’s license on average for 51.3 years (SD = 2.9 years). Participants were recruited from the mailing lists of the Technische Universität Braunschweig and via radio as well as newspaper announcements within the region of Braunschweig. The recruiting took place in the time of November 2012 to January 2013. All drivers were trained in the driving simulator in order to ensure that they were familiar and felt comfortable driving in the simulator, as well as to avoid simulation sickness, which is especially probable for older drivers. The training contained three consecutive driving scenarios. In the first driving scenario, participants drove on a straight rural road. In the second scenario, acceleration and deceleration were trained. Finally, in the third scenario participants practiced right and left turning manoeuvres at different intersections. Participants who developed simulation sickness had to be excluded from the study (n = 23 out of overall 47 trained participants, mainly older drivers). All remaining participants (n = 24) had normal or corrected-to-normal vision. They were compensated for their time either by receiving course credits (students at the Technische Universität Braunschweig) or eight Euros per hour for successful participation. 2.2. Experimental design and driving task In order to analyse how the age of the drivers (Age), different characteristics of the oncoming traffic, as well as different types of vulnerable road users and their behaviour influence accident occurrence and the attention allocation of drivers, a

4

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

mixed design was used. The potential differences in driving performance of younger and older drivers were examined as a between-subjects factor (Age). The type of variation in the oncoming traffic, location of vulnerable road use and crossing direction of bicyclist were treated as a within-subject factor. All scenarios (in total five) are described in detail in Section 2.3. Since it was not possible to completely randomise the order of the five scenarios, as this would lead to too many combinations, four different orders were used to at least partly control for time effects. Subjects were informed of possible difficult situations during the drive. Thus, the results are not representative for completely unexpected events in everyday traffic. This will be taken into account in the discussion of the results. However, in everyday traffic at busy intersections drivers also have to be cautious, since it is likely that something might happen. However, the situations presented (see below) are typical scenarios that occur in everyday traffic quite often. 2.3. Simulator and driving scenarios The study was conducted in the fixed base driving simulator of the Department of Engineering and Traffic Psychology at the Technische Universität Braunschweig. It consists of a seat box with a steering wheel with force feedback, accelerator and brake pedals and two LCD screens serving as rear-view mirrors. The virtual scenery is projected onto three screens (left, ahead, right), providing the drivers with a 180° field of view at about 2.1 m distance from the driver’s seat. The different scenarios, as can be seen in Table 1, were created using the driving simulation software SILAB (Krüger, Grein, Kaussner, & Mark, 2005). In each scenario the ego vehicle, driven by the participant, travelled through a simulated urban road and had to turn when indicated by a voice output and an arrow in the area of the speedometer. In order to ensure that every driver halted at the stop line before turning, five oncoming vehicles (starting vehicles) were implemented with small gaps of 2.5 s (35 m) between them. This was essential since the vulnerable road user crossing the road was triggered to start moving when the participant drove over the stop line. By having the participant stop at the stop line, it was ensured that the vulnerable road user would be triggered at the time the participant crossed the stop line. Altogether there were five scenarios. In the first three scenarios, the complexity of oncoming traffic was varied in three steps, in order to achieve a stronger or lesser shift of attention towards this direction and to discover possible effects of the complexity of oncoming traffic on attention allocation. The most complex scenario (S5) comprised varying gap sizes (gap sizes of 2.5–3.0–3.5–3.0–3.5–4.0–3.5–4.0–4.5–4.0–4.5–5.0–4.5–5.0 s). It was assumed that this variation in gap size would require the most attention from the participants and should thus lead to the largest number of collisions with the crossing vulnerable road user. The slow increase was chosen to enable the drivers to finally turn. The second condition included large gaps of 5 s (normal), which required some attention in order to be able to use these gaps to turn safely. The third condition did not have any oncoming traffic (without) after the five starting vehicles had passed. In this case, it was expected that the drivers could easily shift their attention towards the left part of the road where the vulnerable road user would cross, right after the oncoming traffic had passed. These three scenarios included a bicyclist travelling from the opposite direction of the ego and crossing the drivers’ route at the intersection (see Fig. 1). In order to perceive this bicyclist, attention had to be shifted away from the oncoming traffic. In a fourth scenario a bicyclist was travelling in the same direction as the ego vehicle without oncoming traffic, in order to discover possible effects of the crossing direction of a bicyclist. Here the attention had to be shifted away from the oncoming traffic to the very left of the road. This was expected to be very difficult and thus being the reason for not having oncoming traffic in this situation. Finally, in a fifth scenario a pedestrian was implemented with normal oncoming traffic. In order to find possible effects of the location of vulnerable road users on attention allocation, the pedestrian crossed from the end of the pedestrian crosswalk. In this case, the attention had to be shifted somewhat more to the left than for the oncoming bicyclist, but not as much as for the bicyclist travelling in the same direction as the ego vehicle To summarise, comparing the first three scenarios with the different oncoming traffic conditions it is analysed whether (1) the different oncoming traffic conditions (complex, normal, without) influence the attention allocation on the oncoming traffic and (2) if this changes the way drivers are able to cope with the suddenly crossing bicyclist. The more the attention is focused on the oncoming traffic, the less drivers should be able to react adequately upon the bicyclist (effects of oncoming traffic). By comparing the scenario of a bicyclist crossing (S4) to the scenario with a pedestrian crossing (S5), both scenarios with normal oncoming traffic, the effect of the different locations to which the attention has to be shift to, is examined (effects of the location of vulnerable road users). It was expected that the pedestrian should be harder to notice, as it is further away from the oncoming traffic. Note that this is confounded with the fact that the pedestrian might be harder to distinguish from the background as also other pedestrians were sojourning in both conditions. However, it did not seem plausible to let a bicyclist cross at this point behind the crosswalk, since there was no bicycle path at this location. Finally, comparing the bicyclist travelling from the opposite direction (S3) without oncoming traffic to the one travelling in the same direction as the ego vehicle (S4), a further shift of attention is examined (effects of crossing direction of the bicyclist). It was anticipated that it is be more difficult to perceive the bicyclist travelling from the same direction, than the one travelling from the opposite direction, quite near to the oncoming traffic, since the participants attention is directed towards the oncoming traffic. In order to keep expectations somewhat lower that something will permanently happen when turning, additional scenarios (without critical situations), altogether 13, with left and right turns were inserted between these in a randomised order.

5

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14 Table 1 Results of the effects of the complexity of oncoming traffic. Between-subject age (A) F Effects of the complexity of oncoming traffic Speed while approaching 25.2 Waiting time at stop line 4.7 Gaze percentage on oncoming traffic 3.7 Distance to vulnerable road user 36.3

A ⁄ OTC

Within-subject Oncoming Traffic Complexity (OTC)

df

p

g2

F

df

p

g2

F

df

p

g2

1,66 1,66 1,66 1,66

.000 .033 .068 .000

.277 .067 .050 .355

0.4 17.0 1.8 5.1

2,66 2,66 2,66 2,66

.700 .000 .170 .009

.011 .340 .052 .133

0.1 1.6 0.0 4.6

2,66 2,66 1,44 2,66

.870 .209 .875 .013

.004 .046 .001 .123

Fig. 1. Scenarios and comparisons.

2.4. Procedure After reading and signing a consent form, participants were instructed in written form about the aims and the procedure of the experiment. Afterward, drivers completed the training drive. The eye-tracking system was mounted and a calibration was carried out. During the whole experiment the investigator was seated in a separate room, having the opportunity to communicate with the participant via a microphone. The test drive lasted 40–45 min. Afterwards the subjects were compensated and thanked for their participation. The overall duration of the trial was about two hours. 2.5. Data The simulation software SILAB (Krüger et al., 2005) recorded the driving behaviour of participants. In order to analyse the general driving style, the speed when approaching the intersection (about 83 m before the stop line) was noted. To describe the turning behaviour, the waiting time at the stop line was examined. For the critical events, the distance to the vulnerable road user when the vehicle had stopped (in case of an accident, this was set to ‘‘0”) and the occurrence of accidents with the pedestrian or bicyclist were recorded. Additionally, the eye-tracking system Dikablis (Lange, 2005) was used to measure the gaze behaviour during the waiting time at the stop line. This was expected to give insight in the effect of the oncoming traffic on the gaze behaviour directly before turning. This included the gaze percentage, measuring the durations of all glances on the oncoming traffic in seconds. Fig. 1 gives an overview of the data collection process on a SILAB section (see Fig. 2). 2.6. Data analysis For data analysis, IBM SPSS Statistics 20 was used. When applicable, a general linear model, with a within-subject factor (oncoming traffic, location of vulnerable road user, crossing direction of bicyclist) and one between-subjects factor (Age) was used. For accident occurrence, chi-squared tests were conducted.

6

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

Fig. 2. Analysed data.

In order to examine the different influencing factors within the scenarios three analyses were employed. The first analysis included scenarios with a bicyclist travelling from the opposite direction to the ego vehicle, facing different types of oncoming traffic (complex (S1), normal (S2), without (S3)). The second analysis included scenarios with a pedestrian (S4) vs. bicyclist, both in scenarios with large gap size oncoming traffic. The third analysis examined the different directions of the bicyclist, travelling either in the same direction as the ego vehicle, or in the opposite direction to the ego vehicle, both in scenarios without oncoming traffic. If sphericity was violated, degrees of freedom were corrected using Greenhouse-Geisser. For significant results, g2 is given as a measure of effect size. A significance level of alpha p = .05 was adopted for all statistical tests. 3. Results 3.1. Effects of the complexity of oncoming traffic Table 2 gives an overview of the results of the comparison of the three scenarios (S1-S3) with a bicyclist crossing from the opposite direction of the ego vehicle and different oncoming traffic conditions. The three different oncoming traffic conditions are referred to as ‘‘without” for no oncoming traffic, ‘‘normal” for large gap sizes between oncoming vehicles and ‘‘complex” for varying gap sizes between oncoming vehicles. The speed while approaching the intersection showed a significant main effect of the factor Age (see Table 1). Older drivers approached the intersection with a significantly lower velocity (M = 38.7 km/h, SD = 4.7 km/h) compared to younger drivers (M = 44.7 km/h, SD = 5.4 km/h, see Fig. 3). Neither the main effect of complexity of the oncoming traffic nor the interaction between Age and this factor was significant for the speed while approaching. For the waiting time at the stop line, there were significant main effects of complexity of the oncoming traffic and Age but not a significant interaction (see Table 1). Bonferroni post hoc tests showed that drivers waited significantly longer in the scenario with complex oncoming traffic (S1), compared to the scenarios with normal oncoming traffic (S2; p = .008) and scenarios without oncoming traffic (S3; p < .001). The waiting time was also longer in the scenario with normal oncoming traffic compared to the scenario without oncoming traffic (p = .003). In addition, concerning the age effect, older drivers waited longer in all scenarios (M = 32.0 s, SD = 22.4 s) compared to younger drivers (M = 23.6 s, SD = 17.3 s, see Fig. 4). Although the interaction was not significant, Fig. 4 shows that there was no difference between the two age groups without oncoming traffic (S3). When examining the gaze behaviour, there was a moderate trend toward significance (p = .068) for the factor Age, in the percentage of gazes towards the oncoming traffic. Older drivers had a higher gaze percentage towards the oncoming traffic in all scenarios, compared to younger drivers (older drivers: M = 56.2%, SD = 23.2% vs. younger drivers: M = 45.8%, SD = 24.2%, also see Table 2 and Fig. 5). The gaze percentages on the vulnerable road user were quite low, being the reason not included the results (see Fig. 6). When looking at the distance to the vulnerable road user, there was a significant interaction of the factors complexity of the oncoming traffic and Age, as well as for both main effects. As Fig. 5 shows, in both age groups the distance to the vulnerable road user was smaller with large gap size, as compared to the other types of oncoming traffic. Overall, older drivers managed to keep a larger distance than younger drivers did. However, this difference was smallest in the condition without oncoming traffic (S3), where these two age groups had almost the identical distance. In other words, older drivers had a sig-

7

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14 Table 2 Results of the effects of the location of vulnerable road users. Between-subject age (A) F Effects of locations of vulnerable road users Speed while approaching 11.5 Waiting time at stop line 7.9 Gaze percentage on oncoming traffic 3.8 Distance to vulnerable road user 11.9

Within-subject Location Vulnerable Road User (LVRU)

A ⁄ LVRU

df

p

g2

F

df

p

g2

F

df

p

g2

1,66 1,44 1,44 1,44

.002 .007 .057 .001

.213 .152 .080 .213

0.6 0.0 0.4 0.6

1,44 1,44 1,44 1,44

.458 .909 .529 .458

.013 .000 .009 .013

0.0 0.0 1.9 1.3

1,44 1,44 1,44 1,44

.973 .938 .175 .256

.000 .000 .041 .029

Fig. 3. Mean and standard deviation for the speed while approaching for older and younger drivers in all comparisons and scenarios.

nificantly smaller distance to the vulnerable road user when there was no oncoming traffic, compared to all other oncoming traffic conditions (S1 and S2). Furthermore, the distance to the bicyclist for older drivers increased with the increasing complexity of the oncoming traffic. There was no significant difference in the number of collisions between in the factor complexity of the oncoming traffic, neither for older nor for younger drivers (v2 = 0.57, p = .751). However, when comparing older and younger drivers in each of the three complexities, older drivers had significantly less accidents compared to younger drivers in scenarios with complex oncoming traffic (S1; v2 = 5.04, p = .025) and in scenarios with normal oncoming traffic (S2; v2 = 5.04, p = .025, see Fig. 8). When no oncoming traffic was present (S3), there was no significant difference in the number of collisions between older and younger drivers (v2 = 2.27, p = .132). 3.2. Effects of the location of vulnerable road users By comparing the bicyclist (S2) to the pedestrian (S5), both scenarios with normal oncoming traffic, the effect of the different locations of the vulnerable road user is examined. Since the pedestrian crossed at an unexpected location, it was assumed that it was more demanding to be noticed as it started moving from a farther end. Table 2 gives the results of the ANOVAs. There was no effect of the factor Location Vulnerable Road User, as well as no interaction in any of the four examined parameters (see Table 2). However, there was a significant effect of the factor Age in all parameters except the gaze percentage, where only a tendency was found (p = .057). Older drivers had a significant lower speed while approaching the intersection (older drivers: = 39.1 km/h, SD = 4.8 km/h vs. younger drivers: M = 43.9 km/h, SD = 5.3 km/h, see Fig. 3). They also

8

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

Fig. 4. Mean and standard deviation for the waiting time at stop line for older and younger drivers in all comparisons and scenarios.

Fig. 5. Mean and standard deviation for the gaze percentage on oncoming traffic for older and younger drivers in all comparisons and scenarios.

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

9

Fig. 6. Mean and standard deviation for the gaze percentage on vulnerable road user for older and younger drivers in all comparisons and scenarios.

waited significantly longer at the stop line before making a left turn (older drivers: M = 35.9 s, SD = 14.2 s vs. younger drivers: M = 20.1 s, SD = 22.9 s, see Fig. 4). During the waiting time older drivers had a somewhat higher percentage of gazes towards the oncoming traffic, compared to younger drivers (older drivers: M = 56.2%, SD = 26.4% vs. younger drivers: M = 42.1%, SD = 23.6%, see Fig. 5). Older drivers also had a significantly higher distance to the vulnerable road user (older drivers: M = 18.2 m, SD = 10.6 m vs. younger drivers: M = 8.4 m, SD = 8.9 m, see Fig. 9). When looking at the number of collisions, there was no significant difference for the number of collisions in the two scenarios (v2 = 0.13, p = .721). However, older drivers had significantly less collisions compared to younger drivers in the pedestrian crossing scenario (v2 = 8.22, p = .004) as well as in the bicyclist crossing scenario (v2 = 5.04, p = .025; see Fig. 8). 3.3. Effects of crossing direction of a bicyclist Comparing the bicyclist travelling from the opposite direction of the ego vehicle (S3), to the one travelling in the same direction as the ego vehicle (S4), the effect of having to shift the attention even further towards the bicyclist is examined. Table 3 gives an overview of the results of the ANOVAs. Again, there was neither a significant main effect of crossing direction of bicyclist nor an interaction for any of the parameters. However, there was a significant main effect of Age in the speed while approaching (see Fig. 3), as well as in the gaze percentage (see Fig. 5). The distance to the vulnerable road user showed a tendency to significance (p = .053; see Fig. 7). The waiting time at the stop line was not significantly different between the two age groups (see Fig. 4). Older drivers had a significantly lower speed while approaching the intersection (older drivers: M = 39.6 km/h, SD = 4.9 km/h vs. younger drivers: M = 45.1 km/h, SD = 4.1 km/h, see Fig. 3) and a higher gaze percentage towards the oncoming traffic (older drivers: M = 55.2%, SD = 23.2% vs. younger drivers: M = 39.4%, SD = 25.4%, see Fig. 6). There was also a somewhat larger distance to the bicyclist for the older drivers (older drivers: M = 9.1 m, SD = 6.0 m vs. younger drivers: M = 6.3 m, SD = 3.5 m, see Fig. 7). There was neither a significant effect of the factor crossing direction of bicyclist (v2 = 0.14, p = .712), nor Age on the number of collisions (v2 = 2.27, p = .132, see Fig. 8). 3.4. Explorative analyses: drivers’ behaviour prior to the critical scenarios As described in the analyses above, older drivers had in almost all scenarios significantly fewer accidents compared to younger drivers. What is it about the factor Age that has a positive effect on driving safety (fewer accidents), at least when driving in the simulator? To examine this question, the drivers’ behaviour right before the occurrence of the critical situation (crossing vulnerable road user) was examined, separately for older and younger drivers, with and without an accident. Here,

10

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

Table 3 Results of the effects of the crossing direction of bicyclist. Between-subject age (A)

Effects of crossing direction of bicyclist Speed while approaching Waiting time at stop line Gaze percentage on oncoming traffic Distance to vulnerable road user

A ⁄ CDB

Within-subject Crossing Direction Bicyclist

F

df

p

g2

F

df

p

g2

F

df

p

g2

17.4 1.9 4.9 4.0

1,44 1,44 1,44 1,44

.000 .175 .032 .053

.284 .041 .100 .083

0.0 0.0 0.7 0.0

1,44 1,44 1,44 1,44

.847 .292 .400 .840

.001 .000 .016 .001

0.8 2.3 0.0 1.9

1,44 1,44 1,44 1,44

.497 .279 .882 .086

.485 .588 .001 .771

Fig. 7. Mean and standard deviation for the distance to vulnerable road user for older and younger drivers in all comparisons and scenarios.

the focus laid on the speed of drivers when entering the intersection in all scenarios. Fig. 9 shows the speed of both older and younger drivers depending on whether a collision occurred, including all scenarios. Drivers having a lower speed when entering the intersection, were less involved in accidents (older drivers: M = 10.3 km/ h, SD = 7.0 km/h; younger drivers: M = 17.8 km/h, SD = 7.3 km/h). When drivers had a higher speed, they were more prone to be involved in an accident (older drivers: M = 23.1 km/h, SD = 4.4 km/h; younger drivers: M = 25.2 km/h, SD = 6.7 km/h). 4. Conclusions The aim of this work was to examine why accidents with vulnerable road users happen while turning left at intersections. It was expected that attention would play a major role in accident occurrence. For this, different types of oncoming traffic (complex, normal, without) were introduced in order to examine possible effects the gaze behaviour. This expectation could not be confirmed by the present study. When examining the different situational factors, the different types of the oncoming traffic did not have an effect on the gazes of participants. All drivers showed the same gaze behaviour in all three oncoming traffic conditions. Older drivers showed a difference between the three different conditions of oncoming traffic when regarding their distance to the vulnerable road user. However, this difference was in the opposite direction as expected, as older drivers had their smallest distance to the bicyclist in the condition with no oncoming traffic, where the largest distance was assumed, since in this scenario they were not distracted by oncoming traffic. Furthermore, the collision frequency in the different oncoming traffic situations did not vary.

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

11

Fig. 8. Number of collisions for older and younger drivers in all comparisons and scenarios.

Fig. 9. Speed of older and younger drivers depending on whether a collision occurred, including all scenarios.

Yet, some of the findings appear to be unusual, since when no oncoming vehicles are present, the drivers performing a left turn only had to keep track of the pedestrians. However, in the present study, the participants still had collisions. It seems that the assessment of the situation as being not critical played a larger role than the attention allocation. Accordingly, Lord (1994) also reported that more conflicts occur when no oncoming traffic is present, since the driving task seems easier. The explanation might be the drivers’ expectancy that a pedestrian will not cross their road any time soon. In a similar manner, the location of the vulnerable road user also did have an effect on the collision frequency as expected. Neither the bicyclist (crossing from the opposite direction of the ego vehicle, but closer to the intersection) nor the pedestrian (at the farther end of the intersection) changed the attention allocation as measured by the driving and gaze behaviour

12

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

or the collision frequency. A reason for this might be that the operationalization of location of vulnerable road users might be confounded, since both conditions do not only differ in the location of the vulnerable road user but also for example in size and speed of these. It may also be that the overall scenery in the simulator is simply not as complex as in the real world. Furthermore, participants expected that critical situations might possibly happen. This might have influenced their visual search behaviour, which would not have been the case if the situations had been completely unexpected, even though the scenarios were shown in a randomised order with 13 additional scenarios. This being a limitation of this study. However, it was found that the occurrence of accidents strongly depended on the way in which the drivers approached the critical situations. When drivers entered the intersection somewhat slower, they were able to avoid the accident rather than those drivers who entered it faster, independently of the drivers’ age. From this point of view, the effect of Age, found in this study, might be caused by the speed of drivers. Yet, this explanation needs further studies, as this was not a main topic of this study and the sample size needs to be larger to be able to determine this effect. As lots of intersection accidents involve older drivers, a comparison of younger and older drivers was carried out in this study. It was expected that older drivers are aware of their cognitive deficits and try to compensate for their cognitive impairments in simplifying their driving task, for example by driving slower (Hakamies-Blomqvist et al., 2004). This expectation was supported by the results of this study. In all scenarios, older drivers had a significantly lower speed when approaching the intersection. They also waited longer before turning compared to younger drivers. Their compensation efforts were mostly successful, as older drivers were able to stop with a larger distance towards the vulnerable road user and had significantly less collisions compared to younger drivers. Accordingly, they had more time to react and to avoid accidents. These results are in line with those of Yan et al. (2007) and Vollrath et al. (2009). This compensatory behaviour of driving rather slowly seems to be crucial to prevent an accident. The results of the study are in contrast to the findings of an overrepresentation of older drivers in intersection accidents. One explanation might be the increased exposure of this age group to intersection situations (e.g., driving shorter routes and driving in urban areas), which are quite dangerous (McKnight & McKnight, 1999). Additionally, it could also be that those older drivers, who do not compensate for their deficits, are those that are more likely to get involved in a crash and do not participate in these kind of research studies. From this point of view, it would be very interesting to do further research on the relationship between driver characteristics and compensatory behaviour in order to better identify the risk of this group. Nevertheless, it is also possible that being in a driving simulator contributes to these findings. On the one hand a selection bias might have occurred, since the old drivers volunteering for the study might be those that are in good shape and feel competent to do the driving tasks. Then again, older drivers might have been especially cautious as the driving simulator was novel to them and they were well aware of being examined. This might also lead to a compensatory behaviour that they may not show in real traffic. In order to examine this more closely field tests would be very helpful. However, a first observational study in real traffic for older drivers turning left, found indications for a similar compensatory behaviour of older drivers (Fofanova et al., 2011) and thus supports the validity of this simulator study. With regard to the factors influencing the compensatory behaviour, there was one interesting result when looking at the effect of the different types of oncoming traffic. The differences between older and younger drivers, for example in the distance to vulnerable road user, were much smaller when no oncoming traffic was present, as compared to the other two traffic conditions. This could indicate that in situations without oncoming traffic, older drivers estimated the situation as being relatively easy and did not adapt their behaviour very much. Accordingly, the distance to the vulnerable road user was very similar to that of younger drivers, as well as the collision frequency. This might demonstrate that older drivers use an individual estimation of the riskiness of the situation in order to adapt or not adapt their behaviour. Thus, one would assume that older drivers are especially at risk of a collision if they think that the current situation is not very difficult or risky and do not adapt their behaviour accordingly. It would be interesting to examine older drivers’ accidents accordingly and derive some kind of measure of the complexity of the intersection. Likewise, Skyving, Berg, and Laflamme (2009) found out that older drivers are over-represented in intersection accidents during weekends, in low-speed areas and during dry road conditions, which they might estimate as easy conditions. Something similar needs to be examined for the complexity of intersections. Besides the limitation due to the not completely unexpected scenarios for participants, further ones need to be mentioned. In general driving simulator studies provide a safe and controlled virtual environment to test driving behaviour, yet, participants often suffer from simulation sickness, frequently occurring while driving on curvy roads and/or in urban areas with many intersections and turns. Simulation sickness may confound data and influence participants’ dropout rates. Older participants have a greater likelihood of simulator sickness than younger participants do. The test drive in this study was limited to an urban context. For the following studies, it would be interesting to have more highways and rural roads between the test drives, since this might also lead to a lower dropout rate of older participants due to simulation sickness. By having more straight roads, older participants could recover from the high requirements of urban roads. Thus, they should have breaks between the drives, combined with less difficult training roads (e.g., highways) as well as test-drives. Furthermore, the loss of many older participants due to simulation sickness may affect the representatives of this group of drivers in the sample, as it may be that those who dropped out may be more at risk for collisions at intersections (Edwards, Creaser, Caird, Lamsdale, & Chrisholm, 2003). Additionally, the lack of significant effects might be due to the low sample size also resulting from the high rate of simulation sickness in older participants. Moreover, it might have been that those older participants passing the simulator training may have driven more cautiously in order to lessen the symptoms of simulator sickness, thus also being another form of compensatory behaviour.

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

13

In this study, participants were informed about difficult situations occurring during the test drive, thus making the results not representative for completely unexpected events in everyday traffic. In further studies, participants should not be informed about possible difficult situations, although they will learn to expect them after the first experience anyways. Moreover, the gap size of the oncoming traffic used in this study may have been problematic. Many participants waited until no oncoming traffic was present before they turned left, especially older drivers. This may also have contributed to the fact, that the oncoming traffic did not influence gaze behaviour as expected. Using a pretest, the gap sizes in the present study were chosen to be between 2.5 and 5.0 s. However, the AASHTO manual (2001) states 5.5 s as a critical gap accepted by participants turning left (Yan et al., 2007). Thus, for further studies the gap size should be larger than 5.0 s. Additionally, in the present study the intersection was not controlled by traffic lights. It would be interesting to test the same scenarios, yet at a traffic light controlled intersection, since a driver who has stopped at the intersection can look for pedestrians who are waiting to cross, and prepare for the proper turning manoeuvres as well as reaction (Lord et al., 1998). Comparing a traffic light controlled intersection to an intersection without a traffic light might have given an insight if participants indeed use the time at the light to look for vulnerable road users. Furthermore, the present study compared younger drivers with a mean age of 25 years to older drivers with a mean age of 71 years. The larger number of accidents for the younger group might also, at least to some extent, be due to their young age and the lack of driving experience, as well as underestimation of their collision risk. A study of Matthews and Moran (1986) compared the drivers’ perception of the accident risk of younger (aged 18–25) and older drivers (aged 35–50). This study showed that younger drivers perceived older drivers as being significantly higher at risk and having poorer abilities than themselves. Young drivers were more confident in their driving abilities than older drivers were. Evidence is provided to suggest that younger drivers view themselves as invulnerable from the effects of higher risk levels and ascribe a higher risk to their peers. The younger drivers participated in the present study might have also underestimated their own accident risks. Additionally, it would be beneficial to introduce an additional group of experienced, middle-aged drivers, as these comprise the majority of the drivers on the road. The consideration of middle-aged drivers might consequently allow to better transfer the results to the overall driving population. Finally, the scenarios developed here resulted in quite a number of accidents and thus can be used to examine the effect of warning systems as described in the introduction. For example, different types of warnings (generic vs. specific warnings; action vs. attention oriented warnings) could be compared or different warning modalities could be used (e.g., visual or auditory). Anyway, warning systems for younger drivers should be developed in order to prevent accidents. Yet, older drivers working at the limits of their compensation strategies might also profit from warning systems especially in situations where they do not expect critical events. In order to increase the safety at urban intersections for all aged drivers, these questions will be examined in coming studies. References Anstey, K. J., Wood, J., Lord, S., & Walker, J. G. (2005). Cognitive, sensory and physical factors enabling driving safety in older adults. Clinical Psychology Review, 25, 45–65. Caird, J. K., & Chugh, J. (1997). The time cost of head-up displays for older drivers: Critical event onset, task location, and display type. In Proceedings of the human factors and ergonomics society 41st annual meeting (pp. 1008–1012). Santa Monica, CA: Human Factors and Ergonomics Society. Caird, J. K., & Hancock, P. A. (2002). Left turn and gap acceptance accidents. In R. E. Dewar & R. Olson (Eds.), Human factors in traffic safety (pp. 591–640). Tucson, AZ: Lawyers & Judges. Case, H. W., Hulbert, S., & Beers, J. (1970). Driving ability as affected by age Final Report No. 70-18. Los Angeles: Institute of Transportation and Traffic Engineering, University of California. Dewar, R. (1995). Intersection design for older driver and pedestrian safety. In 65th Annual meeting of compendium of technical papers. Institute of Transportation Engineers (pp. 605–608). Edwards, C. J., Creaser, J. I., Caird, J. K., Lamsdale, A. M., & Chrisholm, S. L. (2003). Older and younger driver performance at complex intersections; implications for using perception-response time and driving simulation. In Proceedings of the second international driving symposium on human factors in driver assessment, training and vehicle design. Park City, Utah, 21–24, July 2003 (pp. 33–38). Fofanova, J., Maciej, J., & Vollrath, M. (2011). Ältere Autofahrer beim Linksabbiegen: Eine Beobachtungsstudie im Realverkehr. Zeitschrift für Verkehrssicherheit, 57(4), 176–180. Fu, R., Yingshi, G., Yang, C., Wie, Y., Young, M., Jinshua, P., . . . Chang, W. (2011). Research on heart rate and eye movement as indicators of drivers’ mental workload. In 3rd International conference on road safety and simulation. Indianapolis, USA. Garber, N. J., & Srinivasan, R. (1991). Characteristics of accidents involving elderly drivers at intersections. In Transportation research record 1325 (pp. 8–16). Washington, DC: TRB, National Research Council. Hakamies-Blomqvist, L., O’neill, D., & Raitanen, T. (2002). Driver ageing does not cause higher accident rates per km. Transportation Research Part F: Traffic Psychology and Behavior, 5(4), 271–274. Hakamies-Blomqvist, L., Siren, A., & Davidse, R. (2004). Older drivers – A review VTI rapport 497A. Swedish National Road and Transport Research Institute. Hanson, T. R., & Hildebrand, E. D. (2011). Are rural older drivers subject to low-mileage bias? Accident Analysis and Prevention, 43(5), 1872–1877. Harms, L. (1991). Variation in drivers’ cognitive load. Effects of driving through village areas and rural junctions. Ergonomics, 34(2), 51–160. Insurance Institute for Highway Safety (2000). Q & A: Pedestrians, December. Janke, M. K. (1991). Accidents, mileage, and the exaggeration of risk. Accident Analysis and Prevention, 23, 183–188. Krüger, H.-P., Grein, M., Kaussner, A., & Mark, C. (2005). SILAB-A task-oriented driving simulation. In Proceedings of the driving simulator conference (DSC) (pp. 232–331). Iowa City: University of Iowa, USA. Lange, C. (2005). The development and usage of Dikablis (digital wireless gaze tracking system). In Abstracts of the 13th European conference on eye movements ECEM13 (p. 50). Bern. Larsen, L., & Kines, P. (2002). Multidisciplinary in-depth investigations of head-on and left turn road collisions. Accident Analysis and Prevention, 34, 367–380. Lord, D. (1994). Pedestrian conflicts and left-turning traffic at signalized intersections M.A.Sc. Thesis. Ont., Canada: Department of Civil Engineering, University of Toronto. Lord, D., Smiley, A., & Haroun, A. (1998). Pedestrian accidents with left- turning traffic at signalized intersections: Characteristics, human factors and unconsidered issues. In Paper presented at the 77th annual transportation research board meeting. Washington, DC.

14

J. Kazazi et al. / Transportation Research Part F 43 (2016) 1–14

Matthews, M., & Moran, A. (1986). Age differences in male drivers’ perception of accident risk: The role of perceived driving ability. Accident Analysis and Prevention, 18, 299–313. Mayhew, D. R., Simpson, H. M., & Ferguson, S. A. (2006). Collisions involving senior drivers: High-risk conditions and locations. Traffic Injury Prevention, 7, 117–124. McDowd, J. M., & Shaw, R. J. (2000). Attention and aging: A functional perspective. In F. I. M. Craik & T. A. Salthouse (Eds.), The handbook of aging and cognition (pp. 221–292). Mahwah, NJ: Erlbaum. McGwin, G., Jr., & Brown, D. B. (1999). Characteristics of traffic crashes among young, middle-aged, and older drivers. Accident Analysis and Prevention, 31, 181–198. McKnight, A. J., & McKnight, A. S. (1999). Multivariate analysis of age-related driver ability and performance deficits. Accident Analysis and Prevention, 31, 445–454. Morgenroth, J., Wolf, L., Macht, S., Sasse, A., Hecker, P., Schack, M., et al (2009). Improving the safety of pedestrians by using a cooperative system. In Proceedings of the 9th International conference on intelligent transportation systems (ITS) telecommunications. Lille, France. . OECD Organisation for Economic Co-operation and Development (2001). Ageing and transport: Mobility needs and safety issues. Paris, France: OECD Scientific Expert Group. Owsley, C., Ball, K., McGwin, G., Jr., Sloane, M. E., Roenker, D. L., White, M. F., & Todd Overley, E. (1998). Visual processing impairment and risk of motor vehicle crash among older adults. Journal of the American Medical Association, 279, 1083–1088. Rackoff, N. (1974). An investigation of age related changes in driver’s visual search patterns and driving performance and the relation to tests of basic functional capacities Ph.D. dissertation. Columbus, Ohio, USA: The Ohio State University. Räsänen, M., & Summala, H. (1998). Attention and expectation problems in bicycle–car collisions: An in-depth study. Accident Analysis and Prevention, 30(5), 657–666. Reed, N., Kinnear, N., & Weaver, L. (2012). Why do older drivers have more ‘failed to look’ crashes? Berkshire, UK: TRL. Robertson, H. D., & Carter, E. C. (1984). The safety, operation, and cost impacts of pedestrian indications at signalized intersections. In Transportation research record 959 (pp. 1–7). Washington, DC: TRB, National Research Council. Rumar, K. (1990). The basic driver error: Late detection. Ergonomics, 33(10–11), 1281–1290. Skyving, M., Berg, H.-Y., & Laflamme, L. (2009). A pattern analysis of traffic crashes to older drivers. Accident Analysis and Prevention, 41, 253–258. Statistisches Bundesamt (2012). Verkehr. Verkehrsunfälle 2011. Wiesbaden: Statistisches Bundesamt. Summala, H., Pasanen, E., Räsänen, M., & Sievänen, J. (1996). Bicycle accidents and drivers’ visual search at left and right turns. Accident Analysis and Prevention, 28(2), 147–153. Tarawneh, M. S., McCoy, P. T., Bishu, R. R., & Ballard, J. L. (1993). Factors associated with driving performance of older drivers. In Transportation research record 1405 (pp. 65–71). Washington, DC: TRB, National Research Council. Vollrath, M., Maciej, J., Howe, J., & Briest, S. (2009). Fahren Ältere situationsbewusster als Jüngere? Zeitschrift für Arbeitswissenschaften, 1(63), 23–31. Werneke, J., & Vollrath, M. (2012). What does the driver look at? – The influence of intersection characteristics on attention allocation and driving behavior. Accident Analysis and Prevention, 45, 610–619. Yan, X., Radwan, E., & Guo, D. (2007). Effects of major-road vehicle speed and driver age and gender on left turn gap acceptance. Accident Analysis and Prevention, 39, 843–852.