Road lighting and distraction whilst driving: Establishing the significant types of distraction

There is a body of research showing how changes in lighting conditions affect hazard detection in the context of driving after dark. There is a separate body of research showing that driving is impaired by distraction. The two have yet to be integrated: this is critical for lighting design recommendations because giving consideration to distraction may affect the optimal conditions established in lighting studies. A first step in that process is establishing the critical type(s) of distraction that might then be simulated in lighting research. This paper reviews evidence for the prevalence of driving distractions as recorded by two methods: interviews with drivers following collision and observation of drivers on real roads. These data suggest that auditory distractions such as conversation with passengers and listening to music are prevalent distractions, and are therefore appropriate distractions to incorporate in further research of lighting and hazard detection experiments.


Introduction
Hazard visibility is a contributory factor to road traffic collisions (RTCs) [Owens and Sivak, 1996]. After dark, there is a deterioration in visual performance, for example, sensitivity to contrast is diminished and motion-based perception is impaired [Plainis et al. 2006]. Road lighting is installed to offset this deterioration in vision and improve a driver's ability to detect and recognise potential hazards not otherwise revealed by vehicle headlights.
Previous studies have therefore examined how changes in lighting conditions affect detection and identification performance. Some studies are laboratory based and use abstract targets such as Landolt rings or discs presented on screens Walkey et al. 2007, He et al. 1997. It is usually found in peripheral detection experiments, for example, that increases in luminance and scotopic/photopic (S/P) luminance ratio lead to increases in detection probability and reductions in reaction time (RT) to detection following onset of the target. At some point the benefit of these increases reach a plateau, with further increase in luminance and/or S/P ratio bringing negligible improvement in visual performance, and this point provides a useful estimate of optimal lighting conditions.
In these studies, the test participant is required to focus on the specific task as instructed by the experimenter. For example, to look towards a static fixation mark and press a button when a target is detected at a peripheral location. This does not resemble driving in natural situations. A driver's point of visual fixation is not static [Winter et al. 2017] reflecting the need to deal with varied taskssteering, speed control, monitoring other road users and giving attention to road signs and other sources of information [Land 2006]. Furthermore, drivers are frequently distracted from their primary task of driving. Despite driving being a highly complex and responsible task, with mistakes or risk-taking having potentially fatal consequences [Drews, Pasupathi and Strayer, 2008], drivers regularly engage in supplementary non-driving related activities [Young and Regan 2007;Laurier, 2004]. There are many forms of distraction, including manual distraction such as operating the radio, visual distraction such as advertising hoardings, cognitive distraction such as thoughts and auditory distraction such as passenger conversation [Young et al. 2009]. Driver distraction is recognised as a contributing factor for up to 25% of all RTCs in EU countries [European Commission, 2018] and in the USA, 6% of all fatal RTCs are attributed to distraction [NHTSA, 2017]. Due to under-reporting, these may be underestimates of the effect of distraction [Haworth, 2003]. Driving distraction is therefore a growing international road safety concern [Regan et al. 2008].
The ecological validity of lighting research has been increased in some studies by having test participants engage in a driving video game [Bullough and Rea 2000;Lingard and Rea 2002] or driving simulator Alferdinck, 2006] whilst recording target detection in parallel. In a small number of studies, the test participant is actually driving [Akashi, Rea and Bullough 2007;Várady et al. 2007]. These studies confirm that for detection of peripheral targets, lighting of higher S/P ratio reduces RT and increases detection probability. However, they do not account for distraction by tasks that are not related to driving. For example, driving along a closed road and being accompanied by the experimenter means there were no other vehicles or pedestrians to negotiate and the participant-driver would be less likely to engage in conversation with passengers or to use their mobile phone.
Driving distraction is defined as the situation when drivers' cognitive resources are not sufficient for them to adequately or safely perform the driving task [Salvucci, 2002]. An interaction between distraction and the benefits of road lighting after dark is expected. For example, auditory distractions cause an increase of gaze concentration towards the road centre [Kountouriotis and Merat, 2016] and thus away from hazards in the peripheral field.
Road lighting of higher luminance and S/P ratio may be able to offset the detrimental effect of auditory distraction by enhancing peripheral detection.
The validity of road lighting design recommendations therefore requires evidence of the degree to which those recommendations are influenced by driving distractions. To do that first requires an understanding of the critical form(s) of distraction. The aim of the current review is, therefore, to identify the types of distractions taking place during driving. Two sorts of data are employed: interviews with drivers following collision and observation of drivers on real roads. The article also compares the prevalence of real world distractions with drivers' perceptions of risk.

Distraction at the time of collision
One approach to identifying critical distractions is to identify the distraction (if any) occurring at the moment of a collision. This has been established using three methods in past studies;  [Dingus et al. 2006;Dingus et al. 2016]; and secondary data such as police roadside reports of RTCs [Gordon, 2005]. The top three most prevalent driving distractions in each of six studies are shown in Table 1.
Conversing with passengers was the most prevalent reported distraction associated with RTCs, this being the most frequent distractor in five studies and the second most frequent in the remaining study. This is followed by cognitive distractions (such as a lack of concentration and feeling stressed), and distractors which include visual and physical distraction (using a mobile phone and adjusting vehicle/radio systems). That conversation with passengers is the most prevalent distraction at the time of an RTC confirms the earlier review of Young and Salmon [Young and Salmon 2012] who considered only three of the studies in Table 1 [ McEvoy et al. 2007, Gordon 2005, Dingus et al. 2006].
This finding is in contrast with the driving distraction literature which primarily focuses on the use of mobile phones. For example, the recent review of Huemer et al [Huemer et al. 2018] included 41 observation studies which purposefully sought to record the use of mobile phones whilst driving and did not record instances of other types of distraction.
It is possible that Table 1 under reports the frequency of mobile-phone related distractions.
Three of the studies in Table 1  In the two studies by Dingus et al [Dingus et al. 2006, Dingus et al. 2016] RTCs were captured using cameras which were installed in the participants own vehicle to monitor their behaviour. An advantage of this method is that no instructions are imposed on the driver which might affect their behaviour. In contrast, the installation of a video camera could cause experimenter-demand effects, resulting in changes in drivers' behaviour due to what they think constitutes appropriate behaviour [Zizzo, 2010]. However, when a camera is installed in a driver's vehicle, the study usually runs for an extended period of time, and asked them to recall whether they were engaged in any of these activities before the crash, this list was more extensive, with 31 activities (including listening to music), and drivers could comment on as many as they wanted from the list.
One explanation as to why passenger conversation and listening to music are associated with a high proportion of reported RTCs could be due to increased exposure to this activity [Pöysti et al. 2005]. The greater the time spent on the road (annual mileage), the greater the number of traffic incidents experienced [e.g. Lemaire et al. 2016]. Therefore, if drivers are engaging more often in passenger conversations or listening to music than in other distracting activities, the increased exposure may explain the increase in reported RTCs.

Observation of driving behaviour
An alternative approach to establish the prevalence and willingness to engage in distracting tasks is to observe drivers' behaviour in their natural driving environment, using roadside observation of passing vehicles. In effect, these data measure exposure. Table 2 shows the top three most prevalent distractions as reported in each of seven road observations studies.
Conversing with passengers again appears to be the most prevalent distraction, being the most frequent distraction in six studies and third most frequent in the seventh. This confirms the conclusion drawn from post-collision data (Table 1). Listening to music does not appear in Table 2. One explanation for this is that it is not possible to observe whether a driver is listening to music.
The studies in Tables 1 and 2 report conversing when a passenger was present, but do not report whether they were with single or multiple passengers. In one study, consideration was given to conversation when there were no passengers [Kidd et al. 2016]. This being a roadside observation study, it is not known whether that was talking to oneself, a hands-free telephone conversation, or other activity. Kidd  distinguished between hand-held and hands-free mobile phone use (this distinction was not made in the other studies of Table 1). The percentage of reported RTCs associated with the use of a hands-free kit (0.59%) was slightly higher than hand-held mobile phone use (0.47%) however, this self-report of mobile phone use at the time of an RTC was low in both cases.
Two roadside observation studies ( Note: * Self-report: Participants were interviewed when attending hospital after the RTC.

Study Sample period
Location Sample (n)

First Second Third
Gras et al (2012) [Kidd et al. 2016] conducted roadside observations at different roadway situations which were reported to vary in cognitive demand. Less demanding situations included when a vehicle was stationary (e.g. stopped at traffic lights) and driving along a straight road whereas more demanding situations included navigating around a roundabout or performing a manoeuvre at an intersection. They found that distraction which involved visual and physical components (such as using a hand-held mobile phone, eating, drinking and smoking) were observed more frequently in lower demand situations compared to higher demand situations. The opposite was found for distractors that involve auditory distraction, with talking to a passenger or singing being observed more at roundabouts compared to a straight road. This pattern of results is supported by Huisingh et al [Huisingh et al. 2015] who found that texting/dialling on a mobile phone was more prevalent on low demand urban roads than high demand urban roads, whereas interacting with another passenger was similar across road types. Burns et al [Burns et al. 2002] also found that drivers' reported use of a mobile phone depended on the traffic situation, with all respondents indicating that they would not use their mobile phone during difficult driving conditions. Drivers may adjust their behaviour according to the degree of risk they perceive at a specific location, in other words, behavioural adaptation or risk compensation [Wilde, 2002]. Rather than choosing not to engage in a distracting task of high cognitive load in certain situations, drivers may instead choose to reduce the cognitive demand of driving to compensate for their decision to engage in a distracting task, for example, reducing speed or increasing headway when using a mobile phone [Yannis et al. 2010].
Such risk compensation does not seem to be observed with passenger conversations [Dula et al. 2011], which indicates that passenger conversation is not perceived to be a risky activity. This was found by Burns et al [Burns et al. 2002] in trials using a driving simulator.
Participants were instructed to rate each distraction using a linear rating scale, (the Rating Scale Mental Effort [Zijlstra, 1993]). Conversing with a passenger was considered to be the least distracting task, followed by eating/drinking and tuning the radio: Using a hand-held mobile phone and texting where seen to be the most distracting. This is supported by roadside interviews [Prat et al. 2017]: conversing with passengers was deemed distracting by fewer participants (83%) than using a hand-held phone and manipulating the GPS (100%). Figure 1 shows the contrast between drivers' perception of risk (the results from Burns et al [Burns et al. 2002] and Prat et al [Prat et al. 2017]) and those distractions associated with reported RTCs (Table 1). For Figure 1 the data were converted to z-scores to aid comparison across different types of dependent variable. For perceived risk data [Burns et al. 2002, Prat et al. 2017, perceived risk was evaluated using category rating, and the zscores shown in Figure 1 were determined using only these four distraction categories. For the frequency of association with an RTC, z-scores were determined for the individual studies of Table 1 for the same four distraction categories as perceived risk, and then averaged across those six studies. Although conversing with a passenger was consistently associated with a high proportion of reported RTCs, it was rated the least distracting activity by drivers. There is a dissociation between the perceived risk of conversing with a passenger, and the reported occurrence of RTCs this distraction causes. In contrast, although the use of a hand-held mobile phone was associated with the lowest rate of reported RTCs, this distraction was perceived the riskiest by drivers. Figure 1 does not imply that mobile phone use is not a dangerous task: but instead that the contrast between reported risk and perceived risk demonstrates the importance of investigating passenger conversations.
Drivers are apparently willing to undertake secondary tasks which they perceive as less distracting. According to the Theory of Planned Behaviour [Ajzen, 1991], a person's behaviour is directly influenced by their attitudes about the outcome of that behaviour, their perceptions about how others would view this behaviour, and the extent to which they feel they can perform the behaviour. Therefore, if a driver believes that conversing with a passenger is a commonly accepted behaviour while driving, and that they can complete both tasks adequately, then they are more likely to carry out this activity.  Burns et al. (2002) and Prat et al. (2017), along with an estimate of associated reported RTCs, averaged (mean) from the studies in Table 1. Listening to music was not reported in Burns et al. (2002) and Prat et al. (2017) and therefore is not reported in the figure. These data were standardised by calculating z-scores: Lower z-scores indicate lower ratings of perceived risk and RTC risk, and higher z-scores indicate higher ratings of perceived risk and RTC risk.
This lack of perceived risk may be problematic when considering cognitive resource capacity. Distractions use up cognitive resource and therefore reduce the amount of cognitive attention drivers are able to allocate to the primary task of driving [Lavie 2005]. For example, involvement in an auditory task can affect the allocation of cognitive resources in a visual search task [Boot et al. 2005]. This sharing of attentional resources may explain why auditory distractions (such as conversing with a passenger) are associated with high levels of reported RTCs.

Conclusion
Tables 1 and 2 reveal that conversation with a passenger is a prevalent form of distraction, being associated with a larger proportion of reported RTCs than other types of distraction and being a frequently observed distraction in field observations. Listening to music may also be a significant distraction: it has been excluded from the majority of past RTC studies by experimental design, and by its nature, as it is difficult to observe in field studies of natural driving.
One limitation of this conclusion is that it considers RTC frequency but not RTC severity. For example, whether RTCs attributed to distraction from passenger conversation are more or less severe than those attributed to mobile phone use. This distinction would provide additional context in which to evaluate the risk of distraction tasks. Further work is required to establish this. A second limitation is that the distraction studies (Tables 1 and 2) [Shinar et al. 2005].
An alternative approach is the use of standardised laboratory tasks to approximate the interference of real world distractions: these offer greater consistency between participants [Shinar et al. 2005]. In previous studies a number of standardised tasks have been used to simulate the distraction imposed by passenger conversation including the Continuous Memory Task which requires drivers to maintain a count of a 'target' sound heard randomly in a sequence of non-target sounds [Antila and Louma 2005], the Count Task which involves hearing a three digit number and counting backwards in units of seven [Merat and Jamson 2008], and the Word Association Task which requires drivers to free associate using oneword responses to a stimulus word spoken over a speaker [Atchely and Chan 2011]. We propose that the n-back task is considered in further work [Li et al. 2018]. This is a delayed digit recall task, an auditory distraction in which a series of digits (or letters) are played over a speaker. In the lowest workload, n=0, participants are required to repeat out loud the number just heard. For n=1, participants are required to repeat the number before the one just heard: for n=2, participants are required to repeat out loud the number that was presented two numbers back. Higher values of 'n' require more items to be held in working memory, and thus vary the degree of cognitive distraction [Mehler et al. 2011].
In this work we focus on the benefit of road lighting to mitigate RTCs after dark. By impairing detection, distraction may influence the estimate of optimal lighting characteristics: by enhancing detection of objects in the peripheral field, lighting may counter the effect of auditory distraction in prompting gaze away from the peripheral field [Kountouriotis and RTCs nor distraction. One often proposed solution is driver education, although studies suggest that education has either no effect or is associated with an increase in RTC risk [Vernick et al. 1999, Nasvadi and Vavrik 2007, Robertson and Zador 1978.

Funding
This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) grant reference EP/S004009/1.

Declaration of Interests:
None.