Can Complete-Novice E-Bike Riders Be Trained to Detect Unmaterialized Traffic Hazards in the Urban Environment? An Exploratory Study

Abstract

Although hazard perception is an important skill found to be crucial for negotiating traffic among various types of road users, few studies have systematically investigated e-bike riders’ ability to perceive potential unmaterialized hazardous situations or aimed to enhance these abilities through training. The present study explored the formation of two hazard perception training interventions based upon exposing young complete-novice e-bike riders to a vast array of materialized traffic hazards, with the aim of evaluating their effectiveness in enriching the ability to anticipate unmaterialized hazards. Young complete-novice e-bike riders were allocated into one of two intervention modes (‘Act and Anticipate Training’ or ‘Predictive and Commentary Training’) or a control group (ten in each group). AAT members underwent a theoretical tutorial, then observed clips depicting real-time hazardous situations footage taken from an e-bike rider’s perspective and were asked to perform a hazard detection task. PCT members underwent a theoretical tutorial, then a ‘what might happen next?’ task, followed by observation of video footage with expert commentary. A week later, participants were requested to complete a hazard perception test, during which they viewed ten videos and pressed a response button whenever they identified a hazardous situation. Overall, participants in both interventions were more aware of potential unmaterialized hazards compared to the control in both their response sensitivity and verbal descriptions. Trainees were responsive to the developed training interventions. Thus, actively detecting materialized hazards may produce effective training that enriches these road users’ hazard perception skills and allows them to safely negotiate traffic. Advantages of each of the training methodologies along with implications for intervention strategies, licensing, and policy development are discussed.

1. Introduction

Electric powered bicycles (e-bikes) are an emerging form of micro-mobility (i.e., self-driven lightweight vehicles operating at relatively low speeds) that aims to provide a feasible and environmentally sustainable mobility solution to first- and last-mile travel problems [1,2,3,4,5]. In recent years, they are becoming a common and popular mode of transportation worldwide, mostly in urban surroundings [6,7]. Due to their many inherent advantages [8,9], such as convenience of use and affordability [10,11], they are becoming an efficient means of personal commuting and are greatly changing the urban mobility landscape [12,13]. As a result of the growth in their prevalence [14,15,16,17], e-bike riders are becoming increasingly involved in road crashes [18,19,20]. For instance, in Israel there has been a 15-fold increase in the number of hospitalized patients associated with e-bike incidents between 2013 and 2020, reaching a total of 695 in 2020 [21]. It has been suggested that e-bike riders are more likely to be involved in severe crashes than mechanical bicycle riders [22], as their users are able to travel faster while exerting the same physical effort [9,23]. For instance, early research [15] showed that the utilization of e-bikes was associated with an increased risk of being treated at an emergency department due to a crash. Accordingly, it has been suggested that, while in comparison to mechanical bicycle riders e-bike riders have a lower casualty rate, these casualties are more severely injured and tend to utilize more hospital resources [11].

In various countries worldwide, e-bikes are emerging as a transport mode for youngsters [24,25,26]. For instance, in a recent study [27] it has been suggested that the majority (approximately 86%) of adult e-bike users in Israel are 19–34-year-old youngsters. In a similar manner, e-bikes have become a chief form of transportation in Vietnam, especially among youngsters [25], as they do not require a designated license [26]. Notably, a recent study conducted in the Netherlands [12] indicated that e-bikes have become increasingly popular among the younger age groups, who are adopting them as a commute mode, hence shifting the traditional image of the e-bike in the country as a mode of transport used mostly by the elderly.

As many of these road users are youngsters, who may be inexperienced in negotiating road safety, they are inclined to put themselves as well as other vulnerable road users (e.g., inexperienced drivers, pedestrians, and micro-mobility riders) in jeopardy [9]. Indeed, research findings have demonstrated that, compared to experienced adults, young novice road users are less skilled at negotiating complex traffic situations [28,29,30,31], their visual search strategies tend to be inadequate, and they are more inclined to demonstrate subpar performance in evaluating hazardous road traffic situations [32,33]. Hence, it is crucial to explore their skill set further, and in particular their ability to identify, holistically appraise, and predict traffic events [34,35,36].

The ability to detect, assess, and predict traffic situations from a holistic perspective, that is, to be aware of hazardous road situations or to ‘read’ roads in anticipation of forthcoming events, is often termed ‘hazard perception’ [35,36,37]. This ability has been identified as an underlying element in safety and crash prevention [38]. It has been thoroughly researched for a variety of road users, including automobile drivers [28,29], mechanical bicycle users [39], and pedestrians [40,41]. Evidence suggests that hazard perception, one of the few traffic negotiating skills that correlates with traffic collision involvement [38,42,43,44], is an essential capability for road users, with drivers with better hazard perception abilities more likely to have a reduced crash risk [43].

In order to examine hazard perception skills, researchers tend to present participants with traffic scene movies or still pictures taken from a road user’s perspective, then request that they press a response button each time they identify a hazard [35,45,46]. These hazard instigators (i.e., the sources of danger) are distinguished in the literature [40,47,48,49,50] as being of two types: (a) materialized hazards, traffic occurrences where a failure to engage in an immediate evasive action may cause harm to a road user or damage to property [33,41]; and (b) potential unmaterialized hazards, a source of danger that, while it does not require engaging in an immediate action, should be detected and monitored so that the precursor does not develop into a materialized hazard [51,52].

Using a variety of video-based materials, research has shown that potential unmaterialized road hazard instigators are more discriminative than materialized hazards [28,48,51]. Namely, the divergence between road users’ experience can be reflected by their ability to identify and respond to different types of traffic hazards [40,48]. For instance, young novice drivers are considered to have deficient hazard perception skills compared to experienced drivers [35,43,53]. That is, while younger inexperienced drivers tend to mostly identify materialized hazards, experienced drivers tend to identify both materialized and unmaterialized hazards [48]. These differences are apparent when examining participants’ verbal descriptions of detected hazards as well. For example, examining participants’ verbal descriptions of detected hazards, a pattern emerged in which 7–9-year-olds referred much less to potential unmaterialized hazardous situations in their verbal descriptions compared to 9–10-year-olds, 10–13-year-olds, and experienced adult pedestrians [40]. Notably, although only few studies have thoroughly and methodically examined hazard perception abilities among e-bike riders, [49] explored these skills and indicated that e-bike riders’ hazard perception performance tends to be enhanced with additional experience. Specifically, the authors showed that experienced e-bike riders are more inclined to identify potential unmaterialized hazards and to provide more verbal descriptions regarding potential unmaterialized hazardous situations.

2. Training Intervention: A Tool for Enhancing Hazard Perception

Because the hazard perception skill has been found to be an essential component of road users’ abilities [35], and as research has shown that hazard perception skills tend to improve with experience [28,35,49,54], traffic safety researchers have engaged in various attempts to construct training intervention programs in the hope of enhancing young novice road users’ hazard perception skills [41,47,55,56]. A variety of hazard perception training interventions have been developed over the years to account for automobile drivers, pedestrians, and mechanical bicycle users’ skills [29,36,41,55,57,58,59,60]. Indeed, evidence has shown that hazard perception is susceptible to training [29,36,41,58,59], as training in hazard perception has been found to benefit novice road users [57]. However, there has been a shortage of studies systematically investigating this skill with regard to e-bikes, mapping e-bike riders’ abilities to perceive potential unmaterialized hazardous situations [49], or aiming to create designated training interventions. Specifically, there has been only one study that aimed to develop a hazard anticipation training program for e-bike riders and to evaluate its effectiveness [61]. Nonetheless, it focused on experienced (as opposed to novice or complete-novice) older adults (as opposed to youngsters) and did not focus on potential unmaterialized hazards. Hence, the present study is aimed at constructing and evaluating hazard perception training interventions for young complete-novice e-bike riders’ ability, focusing on their ability to detect potential unmaterialized hazard instigators prior to their materialization.

Particularly, this study attempts to enhance these road users’ attentiveness towards potential unmaterialized hazards by applying a conceptual approach taken from the hazard perception training realm and based on theoretical and methodological grounds that have been found effective in enhancing young novice drivers’ [28,29] and child-pedestrians’ hazard perception abilities [41].

Particularly, an attempt was made to utilize a similar set of theoretical and methodological tools which have been found highly efficient in training young novice drivers’ and child-pedestrians’ hazard perception skills [40,41] in order to lay the groundwork for producing an efficient mechanism to enhance complete-novice e-bike riders’ hazard perception skills. Specifically, [29] constructed a hazard perception training methodology for automobile drivers called the Act and Anticipate Hazard Perception Training (AAHPT), in which young novice drivers practiced one of three hazard perception training modes (i.e., ‘Active’, ‘Instructional’, or ‘Hybrid’), and compared them to a control group. The AAHPT was developed based on the underlying notion that exposing trainees to a vast array of materialized road hazards would be an effective tool towards enhancing their ability to anticipate potential unmaterialized hazards during testing. The ‘Active’ training mode had participants observe real-world traffic scenes and actively search and respond to hazardous situations by pressing a response button each time they identified a hazard. The ‘Instructional’ training mode had participants undergo a tutorial that included both written material and video-based examples from real-world settings regarding the hazard perception skill. The ‘Hybrid’ training mode had participants receive a combination of both in which participants were presented with a concise theoretical component followed by a condensed active component. In comparison to the untrained control group members, the ‘Active’ and ‘Hybrid’ trainees (i.e., those who were exposed to the two active-practical intervention modes) were more attentive to potential unmaterialized traffic occurrences. Furthermore, ref. [41] used this paradigm to train young child-pedestrians’ (completely inexperienced as road users, as they were not allowed to cross roadways unaccompanied) hazard perception skills in simulated settings by exposing them to simulated typical urban scenarios from a pedestrian’s point of view and having them engage in an active hazard detection task (i.e., pressing a response button each time they identified a hazard). Control group members were found to be less able to identify potential hazard instigators relative to trainees; they were mostly challenged by traffic scenes where the field of view was restricted.

Another paradigm in developing hazard perception programs is represented by [57], in which four modes of video-based training interventions were created: (a) a ‘what might happen next?’ intervention based on [62]; (b) an expert commentary training intervention; (c) a hybrid commentary training intervention (i.e., with both an expert and self-generated commentaries); and (d) a ‘full training package’ intervention consisting of both a ‘what might happen next?’ intervention and hybrid commentary training. Although all of the training modes presented significant results compared with the untrained groups’ results, the ‘full training package’ resulted in the greatest improvement. Importantly, adding self-generated commentaries to the expert commentary exercise in the hybrid commentary condition failed to significantly benefit participant performance. Furthermore, all of the training effects decayed considerably after one week except for the effect of the ‘full training package’, which remained significant. Notably, although no benefit was found when adding self-generated commentaries to expert commentaries, it has been suggested that combining ‘what might happen next?’ exercises with expert commentary training may provide an additional benefit [57].

3. The Present Study

Because e-bikes have only entered common use in recent years, there are substantial gaps in research regarding the skills and performance of e-bike riders [8]. Correspondingly, the present study examined the possibility of enhancing young complete-novice e-bike riders’ ability to identify unmaterialized road hazards in the traffic environment through training. Studies have shown that there are differences between e-bikes and mechanical bicycles [13,63], as the duration and distance of trips performed by e-bike riders are longer compared to the trips performed on mechanical bicycles [13]. In [11], the authors found differences between these two types of bicycles with regard to riding-related injuries; e-bike riders were found to have a higher risk of head injury compared to those riding mechanical bicycles. Furthermore, [64] has shown that e-bike use and injury patterns tend to differ from those of mechanical bicycle users. According to [18], e-bikes tend to be ridden faster on average and their riders are more inclined to accelerate faster than mechanical bicycle riders; in addition, these vehicles interact differently with other road users (i.e., motorized vehicles are more hazardous for e-bike riders, while vulnerable road users are more hazardous for mechanical bicycle riders) and is subject to specific types of hazards. As such, it has been argued that the design of countermeasures for accident prevention should differ between these two types of vehicles. Similarly, [11] advocated that e-bikes have specific characteristics that should be studied and further addressed in an attempt to produce training interventions that can enrich appropriate skills for safely riding this type of vehicle. It should be noted that in Israel [65], as in other countries worldwide, road users who hold a driving license do not need a riding license, insurance, or e-bike license, and can ride e-bikes on the same bicycle trails as mechanical bicycle users. Thus, there is a pressing need to address these road users’ ability to identify hazardous situations in the traffic environment. Aiming to simulate this typical situation, in which any driver’s license holder can ride an e-bike without any (or much) prior experience (for instance, by grabbing one of the e-bikes that are widely scattered and accessible for use at any moment in many urban areas), participants in the present study were specifically chosen as complete novices, i.e., those who have zero experience in riding e-bikes. Furthermore, as prior studies have noted that road users who have experience using multiple transportation modes may demonstrate improved hazard perception abilities as a result of their broad experiences [37,66], it was important to examine whether these complete-novice e-bike riders might benefit from designated hazard perception training intervention in light of their experience as drivers.

In summation, as earlier studies have shown that hazard perception skills tend to improve with experience [36], and have shown that experienced road users [28,33,40] and specifically e-bike riders [49] tend to detect more potential unmaterialized hazards, the objective of this study was to lay foundations for exploring whether developing hazard perception skills might enrich complete-novice e-bike riders’ ability to perceive potentially hazardous situations and predict future hazardous events prior to their materialization. Thus, the formation and evaluation of two designated hazard perception interventions for youngster complete-novice e-bike riders are described here, namely, ‘Act and Anticipate Training’ (AAT) and ‘Predictive and Commentary Training’ (PCT), with the aim of exploring whether they can be beneficial for increasing road safety. It was expected that trainees would be more sensitive than the control group to potential unmaterialized hazards. To investigate whether either of the techniques is preferable to the control or to the other group, the two training modes (AAT and PCT) were set up for comparison and participants’ responses were examined.

4. Materials and Methods

4.1. Participants

Thirty participants were recruited from within our academic institute and volunteered to take part in the study. Participants had normal vision with uncorrected [67,68] static acuity testing of 6/12 (20/40) or better [33,51,69]. Aiming to create homogeneity between the groups, several demographic characteristics were equalized; the process of allocating participants into groups was controlled by accounting for age, sex, automobile driving license experience, socioeconomic status, and sensation-seeking score. The participants’ mean age was 26.1 years (sd = 2.86), 26.1 years (sd = 2.73), and 26.33 years (sd = 1.00) for the AAT, PCT, and control groups, respectively. Accounting for sex (found to be a crucial factor in road users’ crash rate [70,71], the same male to female ratio (4:6, respectively) was kept constant in the three participant groups. Furthermore, as automobile driving license status has been found to play an important role in their performance in a hazard detection task [49], the participants’ driving experience was equalized; their mean automobile driving experience was 8.60 years (sd = 2.63), 8.78 years (sd = 1.29), and 9.54 years (sd = 2.56) for the AAT, PCT, and control groups, respectively. In accordance with prior research noting that the risk of road users’ injury is inversely related to their socioeconomic status [72], the participants were all of a similar socioeconomic level (i.e., middle class, 10,500–17,500 NIS per household). Furthermore, it has been argued [59] that participants’ sensation-seeking (i.e., their tendency to proactively pursue novel experiences and external stimuli aiming to appease their need for sensation [73]) should be considered when examining their hazard perception skills, as higher sensation-seeking scores (on a scale between ‘1’—low sensation seeking tendency, and ‘40’—high tendency) have been found to be associated with unsafe traffic behavior. Hence, participants’ sensation-seeking score was calculated and kept similar between the groups (mean sensation-seeking scale score 15.2, 16.1, and 15.7 for AAT, PCT, and control, respectively). Notably, participants all had the same level of experience with mechanical bicycles, that is, all knew how to ride mechanical bicycles but seldom used them. Lastly, exposure to traffic as e-bike riders was kept constant, as none of the participants had prior e-bike experience (similar to [41]). Study protocols were approved by the academic institute’s ethics board, and informed consent was obtained from all participants involved in the study.

4.2. Materials and Setup

4.2.1. Formation of the Training Interventions

In the present study, two hazard perception training interventions named ‘Act and Anticipate Training’ (AAT) and ‘Predictive and Commentary Training’ (PCT) were constructed according to the principles described below.

(a)

Exposure to a large number of diverse traffic situations:

When faced with unfamiliar conditions, experienced individuals tend to use their past experience in similar circumstances to better decipher the new situation, fill in gaps, and infer missing information, as well as to better predict forthcoming events [74]. Hence, an attempt was made to increase trainees’ exposure to genuine and unstaged real-life traffic situations [75] by presenting them with a large number of diverse road traffic situations, with the aim of enabling them to gain riding experience in various traffic settings in a short period of time and in a safe setting [29,41].

(b)

Scenarios situated in local settings:

Background and contextual cues play a key role in many cognitive processes, including learning [76,77]. In keeping with this argument, it has been suggested that environmental settings should be taken into account when constructing an effective hazard perception training intervention in order to ensure linkage between the unique characteristics of the geographic region and the typical road hazards it produces [28,29,41,51]. Accordingly, the present training interventions were developed in order to measure the Israeli landscape, infrastructure, and e-bike rider characteristics by presenting e-bike riders with an array of hazard perception movie clips depicting common traffic hazards that local e-bike riders tend to encounter.

(c)

Types of hazardous events:

Previous findings have indicated that the most prominent deficiency of young novice road users is identifying typical potential unmaterialized hazards [33,40,48,51]. Furthermore, it has been suggested that in order to better anticipate potential unmaterialized hazardous situations, trainees should first engage with situations where hazard instigators materialize [28,29,41]. In keeping with this assertion, the participants in this study observed more situations in which the hazard source materialized during the training sessions then during the test session, where most of the hazards were unmaterialized hazards. The videos presented to the participants were matched [29] such that similar hazardous events located in similar traffic environments appeared in both the training and the test session, while more potential unmaterialized hazards appeared in the test (Figure 1). For instance, where a training movie would depict a materialized hazardous situation such as riding in close proximity to parked vehicles when a door opens, a matching test movie would depict a potential unmaterialized hazardous situation such as riding in close proximity to parked vehicles in the same traffic environment without a door opening.

(d)

Theoretical–verbal component

A common practice found to be useful in hazard perception training is to present trainees with theoretical declarative knowledge-based material [41,78]. Indeed, it has been demonstrated that adding a theoretical–verbal section to a training intervention can increase the longevity and generalization of learning [79,80]. Accordingly, a theoretical–verbal component was used in this research in both training interventions, aiming to enhance participants’ awareness of road hazards by introducing trainees to the various types of hazards that exist and the locations where they are most likely to appear.

(e)

Active–practical component

In addition to the theoretical–verbal component, each of the training interventions contained an active–practical component in which participants were required to actively search for and respond to hazardous situations by engaging in action. Indeed, it has been previously suggested [29,81] that hybrid training interventions composed of both theoretical and active components can be more effective than can exclusive use of either component individually. As learning is based upon the link between action and perception [77,82], training processes should advance from action to concept [83], and can be more effective when an individual is required to physically respond to a given situation as opposed to only observing it as a spectator [82]. Hence, the active component in the AAT intervention was aimed at emulating an action–perception loop by requesting participants to produce responses to hazardous situations over a limited time period. It was conjectured that riding experience could be gained, at least to an extent, by producing repeated responses to hazardous situations, and that proactively identifying and responding to hazards might benefit the process of detecting hazards. AAT trainees were first presented with the theoretical–verbal component, aiming to improve awareness of hazards by exposing them to theoretical knowledge regarding which types of hazards exist and where they are most likely to appear. Then, trainees observed the movie footage depicting typical urban hazard perception test scenarios and performed a hazard detection task in which they actively searched and responded to hazardous situations by pressing a response button each time a hazard instigator was detected. At the end of each clip, they were instructed to provide a description of the hazard situation that triggered their response. Then, feedback was provided in the form of an expert pointing out the hazard instigators depicted in the movie.

The PCT intervention was in line with the ‘full training package’ intervention in [57], which was found to improve hazard perception skill among drivers. Specifically, this intervention was based upon the authors’ recommendation to use the ‘full training package’, minus self-generated commentaries, as no benefit was found when adding self-generated commentaries to expert commentaries. Thus, in addition to the theoretical–verbal component, this intervention included a ‘what might happen next?’ exercise and expert commentary training. Trainees were first shown half of the movie clip footage of a hazardous traffic situation that was cut short, occluding as soon as the hazard began to develop, and were required to predict ‘what might happen next?’ [57,62]. Then, they were shown what actually happened when the video resumed. After each movie clip, feedback was provided in the form of an expert pointing out the hazard instigators depicted in the movie. In the third and last part, trainees were required to watch the other half of the movie clip footage of the hazardous traffic situation, which was provided along with expert commentary describing the hazard instigators depicted in each of the movies.

4.2.2. Hazard Perception Traffic Scene Clips

A set of daylight real-world traffic scene movie clips were recorded using a GoPro digital recording device mounted on an e-bike, presenting an e-bike rider’s view while riding an e-bike through the streets. Notably, it has been suggested that each traffic location tends to display its own unique characteristics that dictate a specific array of typical hazards; thus, the utility of a hazard perception traffic scene clip recording depends upon the setting (i.e., region, city, country) in which it is utilized [41]. As it has been shown that urban areas accounted for approximately 75% of e-bike fatalities in Israel between 2015 and 2018 [84], participants were required to observe an array of typical Israeli traffic movie clips filmed in urban Tel Aviv streets [29,75]. Movies were edited into a set of thirty movie clips of 10–40 s long that were utilized in the training session. An additional set of ten movie clips were utilized in the test session (as was the case in [49]). Consistent with prior studies [41,49,51,85,86], each movie clip depicted a different number of hazardous situations.

Aiming to construct intervention programs that are specific for e-bike riding, the movie footage depicted real-world traffic scenes taken from an e-bike rider’s perspective and illustrating a variety hazard types and traffic environments that users would typically encounter, thereby presenting the e-bike rider’s experience of riding fast and accelerating fast [18,87] on designated cycling routes, traffic roads, and pavement. This is in line with previous research indicating that these riders use a mixture of urban facilities and switch between different road layouts [27] in a manner that increases their risk of being involved in conflicts with other road users as compared to mechanical bicycle users, mainly at intersections and near crosswalks [88]. Hazard situations consisted of dynamic (moving automobiles, pedestrians, motorcycles, and buses) as well as static (junctions, curves, vegetation, and other types of restricted field of view) instigators. In line with prior findings [18] suggesting motorized vehicles to be more hazardous for e-bike riders, the hazard situations consisted of 33% automobiles, 29% pedestrians, 19% motorcycles, and 19% buses. The movies presented to trainees were matched [29], with similar situations in the same traffic settings appearing in both the training and the test session; however, the testing movie clips depicted more potential unmaterialized hazards (Figure 1). All movie clips presented to the participants were preserved in their original version and were not amended or retouched.

Each participant was seated at an approximate distance of 60 cm from a 19″ LCD screen with a resolution of 1920 × 1080 pixels, on which the hazard perception movie clips were presented using a designated Visual Basic program. This program was developed in-house to present participants with the experimental materials (instructions, questionnaires, theoretical–verbal component), to play the movie clips, and to record the participants’ responses. The participants’ button presses were captured using a push button.

4.2.3. Questionnaires

The participants completed two computerized questionnaires. The first consisted of demographic and background questions. The second was a computerized Hebrew version [89] of the Sensation-Seeking Scale [90] comprising forty forced-choice items on a scale, designed to evaluate individual differences in sensation-seeking with regard to the four aspects of sensation-seeking: experience-seeking, disinhibition, thrill/adventure-seeking, and boredom susceptibility.

4.3. Procedure

Aiming to create matching and homogenous groups and to ensure to a large extent that the groups differed only by the particular treatment they each received, participants were allocated into one of three conditions (AAT, PCT, or control group) while accounting for age, sex, automobile driving license experience, socioeconomic status, and sensation-seeking score. Trainees underwent a training section, followed by a test section conducted one week later (resembling similar short-term assessments in the hazard perception training literature [29,41,57,91,92,93]), whereas the control group members underwent only the test.

4.3.1. Training Session

Trainees were individually invited into the academic institute’s human factors laboratory for an hour-and-a-half long training intervention session. First, they each provided the experimenter with a signed informed consent form approving their participation in the study. Each trainee was introduced to the laboratory (where temperature and illumination conditions were kept constant throughout the experiment), then underwent the Snellen static acuity test. Trainees were then positioned in front of the laboratory’s computer screen. Next, the experimenter set up and activated the designated software.

Two modes of intervention were designed: AAT and PCT (Figure 2). The video database that these two training interventions were based upon was the same (i.e., participants were exposed to the same footage), however, there were conceptual differences between the two types of interventions in the way that they disclosed the training material to the trainees. Next, participants proceeded according to the specific intervention mode as described in the following sections.

Act and Anticipate Training (AAT)

Participants were read the experimental instructions and provided with an exhaustive explanation of the experimental task. AAT trainees were then presented with the theoretical–verbal component administered in the form of a twenty-minute power point presentation [85], with the aim of improving their awareness of hazards by exposing them to theoretical knowledge regarding hazard perception and its importance to driving safety. Trainees were exposed to a variety of traffic environments that e-bike users would typically encounter (designated cycling routes, traffic roads, and pavement), demonstrating the types of hazards (materialized, unmaterialized) and hazard instigators (moving automobiles, pedestrians, motorcycles, buses) that exist and the locations where they are most likely to appear.

Next, the trainees were instructed to observe the hazard perception clip database (i.e., thirty movie clips depicting typical urban hazard perception test scenarios taken from the perspective of an e-bike rider) and to perform a hazard detection task (i.e., to actively search and identify hazards and to press a designated response button each time they detect a hazard instigator). In line with prior studies such as [28,29,33,49,51], a hazard instigator was characterized according to the following definition adopted from [94], p. 3: an “object, situation, occurrence or combination of these that introduce[s] the possibility of the individual road user experiencing harm”. Participants were requested to respond as quickly as possible when they identified a hazard, and were further instructed to press the button only once for each of the instigators. Participants observed the scenarios in a random order. Aiming to resolve ambiguities regarding the reason for the initiation of each of the participants’ responses [29,95,96], at the end of each clip, participants were presented with blank fields in equal number to their responses to the clip and were instructed to fill them in with a description of the hazard instigator that triggered their response. Notably, although the entries were open-ended, most participants in almost all movies did not indicate any recall issues, and tended to recall the hazard instigators they identified. After each movie clip, feedback was provided in the form of an expert pointing out the hazard instigators depicted in the movie. Two practice videos were used to allow the participants to familiarize themselves with the experimental task.

Predictive and Commentary Training (PCT)

The PCT intervention resembled the ‘full training package’ intervention in [57], which was found to improve hazard perception skill among drivers. It should be noted that, based on the authors’ recommendations, self-generated commentaries were not used in this study, as no benefit was found when adding self-generated commentaries to expert commentaries.

Participants were read the experimental instructions and provided with an exhaustive explanation of the experimental task. PCT trainees were then presented with the theoretical–verbal component administered in the form of a twenty-minute power point presentation [85], aiming to improve awareness of hazards by exposing them to theoretical knowledge regarding hazard perception and its importance to driving safety. Trainees were exposed to a variety of traffic environments that e-bike users would typically encounter (designated cycling routes, traffic roads, and pavement), demonstrating the types of hazards (materialized, unmaterialized) and hazard instigators (moving automobiles, pedestrians, motorcycles, buses) that exist and the locations where they are most likely to appear.

Next, trainees were shown half of the movie clip footage (i.e., 15 videos) consisting of hazardous traffic situations that were cut short, occluding as soon as the hazard began to develop, and were requested to predict ‘what might happen next?’ [57,62] by filling in blank spaces. Next, the video resumed and trainees were shown what actually happened. After each movie clip, feedback was provided in the form of an expert pointing out the hazard instigators depicted in the movie.

In the third and last part, PCT trainees were required to watch the other half of the movie clip footage of a hazardous traffic situation (i.e., 15 videos), provided along with running verbal commentary from an expert describing the traffic environment and the hazard instigators depicted in the movie, aiming to direct trainees’ attention to precursors that should be detected and monitored.

4.3.2. Test Session

The test session took place one week after training [29,41,57,91,92,93]. During this session, each of the participants was instructed to observe ten typical urban hazard perception test scenarios taken from an e-bike rider’s point of view and to engage in a hazard detection task (i.e., to press a response button each time a hazard was identified). As previously mentioned, the hazard instigator definition was adopted from [94]. Notably, these testing scenarios were used to examine e-bike riders’ perception of hazards utilizing a hazard detection task, and found differences between e-bike riders with varied levels of e-bike riding experience [49]. Aiming to resolve ambiguities regarding the reason for the initiation of each of the participants’ responses [29,52,95,96], at the end of each clip the participants were presented with blank fields in equal number to their responses to the clip and were instructed to fill them in with a description of the hazard instigator that triggered their response. Although the entries were open-ended, most participants did not indicate any recall issues for almost all of the movies, and tended to recall the hazard instigators they identified. Participants were requested to respond as quickly as possible when a hazard was detected, and further instructed to press the button only once for each hazard instigator. Two practice videos were used to ensure that all of the participants were familiarized with the experimental task. After ensuring that they each understood the task and were ready to continue, the test began; participants observed the scenarios in a random order without knowing the exact number of scenarios they were about to see.

Lastly, it should be noted that the AAT was based upon the notion that repetitive exposure to traffic scene movies can be an effective tool for training hazard perception. As the hazard perception test itself necessitated exposure to traffic scenes, the mere fact of exposure to the test (i.e., use of pre-test) could have altered the results by increasing the participants’ sensitivity to the format and potentially modifying their performance. Hence, this reactive nature of the AAT intervention required the use of a post-test-only control design employing an array of matching participants, thereby creating equivalence between the trainee and control groups.

4.4. Data Preparation and Analyses

A summary file was generated for each of the participants depicting their button presses alongside the related time stamp and their matching verbal descriptions of each hazardous event. In accordance with the previous literature [29,48,49], a hazardous event was defined as the interval between the first and the last press made by participants, given that at least two of the participants responded at approximately the same time and referred to the same hazard instigator. An event onset was set as the earliest press made by any of the participants (assuming that the hazard instigator was indeed present in the scene), and its endpoint was set to the latest button press made by any of the participants (from the point when the hazard was first present until it was no longer visible). A button press was classified as a response to a hazard instigator only if both the verbal description of the hazard and the temporal location of the button press matched the attributes of that instigator. Participants’ self-reported incorrect responses (i.e., a response button that was mistakenly pressed) were excluded.

In total, thirty-nine hazardous situations were identified. It should be noted that thirty-three hazardous situations were predefined by the researchers. Further, six additional hazards were defined post hoc, where a new hazardous situation was added post hoc to the database only if at least 30% of the all participants identified it as hazardous [29]. Following prior research [51,53,85], these hazard situations were classified into two main hazard types: seventeen materialized hazards (i.e., a traffic occurrence where a failure to engage in an immediate evasive action may cause harm) and twenty-two potential unmaterialized hazards (i.e., sources of danger that do not require road users to engage in an immediate action but which should be detected and monitored). These hazard types were taken into account in the analysis process as detailed in the following paragraphs.

As the study was aimed at examining the training interventions’ effectiveness in enhancing trainees’ ability to identify hazardous situations, two types of analyses were applied: (a) response sensitivity, aiming to explore whether a specific participant group tended to report hazardous situations more often than the others; and (b) verbal description analysis regarding participants’ descriptions of the reasons they provided for identifying an instigator as hazardous, aiming to decipher which of the traffic hazards were detected by the different participant groups.

4.4.1. Response Sensitivity

Participants’ response sensitivity refers to the probability of a participant identifying a hazard and pressing the response button during the hazard instigator’s allotted time window. For this analysis, the dependent variable is binary distributed, indicating whether a participant pressed (‘1′) the response button within the allotted time window of the hazard or did not (‘0′). The analysis included two independent variables; the first independent variable was the participant group (AAT, PCT, control) and the second was the type of hazard (either a materialized or potential unmaterialized road hazard).

In order to examine whether any of the e-bike rider groups identified more hazards than the others, a binary logistic regression with a logit-link function within the framework of GLMM (generalized linear mixed models) was utilized. To account for individual differences, participants were included as a random effect. Sequential Bonferroni correction [97] was used for all post hoc analyses. Main effects and second order interaction of the fixed effects were included in the model.

4.4.2. Analysis of Verbal Descriptions

Participants’ verbal descriptions regarding the reasons they provided for identifying a particular instigator as hazardous were examined. Participants’ verbal descriptions were coded into two hazard types: materialized (e.g., “the car I tailgated suddenly braked”; “as I was riding, a pedestrian emerged and burst into my lane”) or potential unmaterialized (e.g., “my field of view was restricted due to the trashcan on the left”; “the e-bike I am riding is moving at a high speed”). Participants’ verbal descriptions were matched to specific hazardous events based on their depiction of the instigator they identified and the relevant time stamp. In line with prior studies [40,41,51], certain verbal descriptions were counted several times, as they mentioned more than one hazard instigator at a specific time-point due to the complexity of the real-world situations (e.g., “my field of view was blocked while a vehicle approached from the left”, which is both a materialized and a potential unmaterialized hazard). As the number of verbal descriptions was Poisson distributed, a Poisson regression with a log link function within the framework of GLMM was utilized. The between-group fixed effect included was participant group (AAT, PCT, or control). Hazard type (either a materialized or unmaterialized road hazard) was referred to as the within-group fixed effect. The dependent variable for analysis was defined as the number of verbal descriptions of hazard instigators articulated by the participants. Post hoc pairwise multiple comparisons were calculated using Sequential Bonferroni correction [97]. The main effects and second order interaction of the fixed effects were included in the model. To account for individual differences, participants were included as a random effect.

5. Results

5.1. Response Sensitivity

5.1.1. Participant Group

Applying the GLMM revealed a significant main effect of the participant group (F_(2,1122) = 10.34, p < 0.001). Sequential Bonferroni post hoc pairwise comparisons revealed that control group members were significantly less likely to respond to a hazard instigator (estimated mean = 0.41, standard error = 0.06) compared to the PCT and AAT members (0.62, 0.05, P_adj < 0.05; 0.76, 0.04 P_adj < 0.001, respectively). Moreover, PCT members were significantly (P_adj < 0.05) less likely to respond to a hazard instigator than the AAT members (

Table 1. A summary of the participants’ response sensitivity results.

Source	F	Estimated Means (SE)	Pairwise Comparisons
Participant group	10.34 ***	AAT = 0.76 (0.04) PCT = 0.62 (0.05) C = 0.41 (0.06)	C < PCT (Padj < 0.05) C < AAT (Padj < 0.001) PCT < AAT (Padj < 0.05)
Hazard type	204.99 ***	Materialized = 0.83 (0.02) Ummaterialized = 0.33 (0.03)	Unmaterialized < Materialized (Padj < 0.001)
Participant group * Hazard type	1.30	Materialized: AAT = 0.92 (0.03) PCT = 0.81 (0.04) C = 0.69 (0.06)	Materialized: C < AAT (Padj < 0.001)
		Unmaterialized: AAT = 0.47 (0.06) PCT = 0.39 (0.06) C = 0.18 (0.04)	Unmaterialized: C < AAT (Padj < 0.001) C < PCT (Padj < 0.01) PCT = AAT (N.S.)

Note that AAT, PCT, and C represent ‘Act and Anticipate Training’ intervention, ‘Predictive and Commentary Training’ intervention, and ‘Control’ group members, respectively. Additionally, *** denotes a significance level of 1%, and * denotes a significance level of 10%.

).

5.1.2. Hazard Type

A significant main effect of the hazard type was revealed (F_(1,1122) = 204.99, p < 0.001). Sequential Bonferroni post hoc pairwise comparisons revealed that participants were significantly less likely to respond to a potential hazard instigator (estimated mean = 0.33, standard error = 0.03) than a materialized hazard instigator (0.83, 0.02).

5.1.3. Participant Group and Hazard Type

Although the model did not find a significant interaction between participant group and hazard type (F_(1,1122) = 1.30, N.S.). Sequential Bonferroni post hoc pairwise comparisons revealed that control group members were significantly less likely to respond to potential unmaterialized situations (estimated mean = 0.18, standard error = 0.04) than both training group members (0.39, 0.06, P_adj < 0.01; 0.47, 0.06, P_adj < 0.001 for PCT and AAT members, respectively).

Fewer differences were found between the participants’ responses to materialized hazard instigators, where control group members were significantly less likely to respond to materialized situations (estimated mean = 0.69, standard error = 0.06) than AAT group members (0.92, 0.03, P_adj < 0.001); however, no significant difference was found between the responses of control group members and PCT members (0.81, 0.04, N.S.) to materialized situations (Figure 3).

5.2. Analysis of Verbal Descriptions

5.2.1. Participant Group

Applying the GLMM revealed a significant main effect of the participant group (F_(2,52) = 6.34, p < 0.01). Sequential Bonferroni post hoc pairwise comparisons indicated that control group members provided significantly fewer verbal descriptions of hazard instigators (estimated mean = 8.68, standard error = 1.34) than were provided by AAT members (15.55, 1.40, P_adj < 0.01); however, no significant difference was found between control and PCT group members (12.61, 1.35, N.S.) (

Table 2. Summary of the participants’ verbal description results.

Source	F	Estimated Means (SE)	Pairwise Comparisons
Participant group	6.34 **	AAT = 15.55 (1.40) PCT = 12.61 (1.35) C = 8.68 (1.34)	C < AAT (Padj < 0.01) C = PCT (N.S.)
Hazard type	16.45 ***	Materialized = 14.09 (0.94) Ummaterialized = 10.47 (0.87)	Unmaterialized < Materialized (Padj < 0.001)
Participant group * Hazard type	4.61 *	Materialized: AAT = 15.98 (1.66) PCT = 14.04 (1.60) C = 12.24 (1.63)	Materialized: N.S.
		Unmaterialized: AAT = 15.11 (1.63) PCT = 11.18 (1.50) C = 5.12 (1.36)	Unmaterialized: C < AAT (Padj < 0.001) C < PCT (Padj < 0.01) PCT = AAT (N.S.)

Note that the AAT, PCT, and C represent ‘Act and Anticipate Training’ intervention, ‘Predictive and Commentary Training’ intervention, and ‘Control’ group members, respectively. Additionally, *** denotes a significance level of 1%, ** denotes a significance level of 5%, and * denotes a significance level of 10%.

).

5.2.2. Hazard Type

A significant main effect of hazard type was revealed (F_(1,52) = 16.45, p < 0.001). Sequential Bonferroni post hoc pairwise comparisons revealed that participants provided significantly fewer verbal descriptions of potential hazard instigators (estimated mean = 10.47, standard error = 0.87) than were provided for materialized hazard instigators (14.09, 0.94).

5.2.3. Participant Group and Hazard Type

The model revealed a significant interaction between participant group and hazard type (F_(2,52) = 4.61, p < 0.01). Sequential Bonferroni post hoc pairwise comparisons revealed that control group members provided significantly fewer verbal descriptions of potential unmaterialized situations (estimated mean = 5.12, standard error = 1.36) than both training groups members (11.18, 1.50, P_adj < 0.01; 15.11, 1.63, P_adj < 0.001, for PCT and AAT members, respectively). However, no differences were found between participants’ responses to materialized hazard instigators (estimated mean = 12.24, standard error = 1.63; 14.04, 1.60; 15.98, 1.66 for control, PCT, and AAT, respectively) (Figure 4).

6. Discussion

In the past decade, e-bikes have become an increasingly popular means of transport for sustainable, affordable, and efficient personal commuting [6,8,9,98,99,100]. Previous research has found that e-bike riders are more likely to become involved in traffic crashes [18,19]. Although hazard perception capabilities are imperative in the process of safely negotiating traffic [28,29,33,34,38,39,41], and even though prior studies have demonstrated that hazard perception training interventions may be an effective tool towards enhancing skills among road users [38,40,41,101], research regarding training in the context of hazard perception skills among young e-bike riders in general, and among young complete-novice e-bike riders in particular, is lacking. The current study was aimed at developing and evaluating two modes of hazard perception training intervention that aspire to enhance young complete-novice e-bike riders’ awareness towards potential, unmaterialized hazards in the traffic environment. As experienced road users in general [33,48] and experienced e-bike riders in particular [49] have been found to identify more potential unmaterialized hazard instigators compared to novices [28,48,53] and complete novices [41], trainees were expected to be more attentive towards potential unmaterialized hazards compared to control and to mention more unmaterialized hazard instigators in their verbal descriptions. As conjectured [29,41], trainees in both intervention programs showed enhanced attentiveness towards potential unmaterialized hazards, and tended to perform better compared to control group members in both their response sensitivity and verbal descriptions.

Specifically, the response sensitivity analysis revealed a significant main effect of the participant group, with control group members less likely to respond to a hazard instigator than both trainee groups; furthermore, PCT group members were less likely to respond to a hazard instigator than the AAT members. Post hoc pairwise comparisons revealed that control group members were significantly less inclined to identify unmaterialized road hazards than members of both training groups. In addition, fewer differences were found between the participants’ responses to materialized hazard instigators, with control group members significantly less likely to respond to materialized situations than AAT group members. Regarding the verbal descriptions, analysis indicated that control group members provided fewer verbal descriptions of the hazard instigators than were provided by AAT members; however, no significant difference was found between control and PCT group members. Moreover, an interaction was revealed between the participant group and hazard type, with control group members providing fewer verbal descriptions of potential unmaterialized situations than members of either training group. However, no differences were found between participants’ responses to materialized hazard instigators.

These results are consistent with the research hypotheses, indicating that training hazard perception may significantly improve complete-novice e-bike riders’ chances of identifying potential unmaterialized hazards. These findings are in line with prior literature regarding road users exemplifying that hazard perception is dependent on prior experience and is susceptible to training [28,33,36,40,101,102]. Notably, considering the main effect of the participant group in the response sensitivity results, in which both PCT members and control were less likely to respond to a hazard instigator than AAT members, it can be suggested that the AAT intervention was more beneficial than the PCT training. Notably, however, this difference disappeared when the type of hazard was taken into account, as no differences were found between AAT and PCT trainees regarding unmaterialized hazards. Because the main focus of this study was participants’ sensitivity towards potential unmaterialized hazards, these results support the idea that both interventions were similarly effective.

It should be noted that the AAT intervention was based on the notion that exposing young complete-novice e-bike riders to a vast array of materialized hazardous situations during the training intervention would enable them to anticipate more potential unmaterialized hazards during testing. Indeed, repeated exposure to specific hazards has been shown to improve hazard perception skills among drivers [29,101]. These results may provide initial evidence for the notion that the more trainees respond to events and experience riding, as was the case in the AAT mode, the more they accumulate information and experience while learning how to better estimate the likelihood that particular hazards will appear in specific environments [29]. It is suggested that in this case such experience enabled these trainees to integrate elements in the environment and predict future events, as was the case for inexperienced drivers [28] and child-pedestrians [41] in previous research. It can be further suggested that the differences found between the two training interventions with regard to the main effects stemmed from the fact that AAT trainees were required to be more proactive. Namely, these participants had to choose when to push the button, and were thus required to detect and identify hazards and to decide to act, while the PCT trainees were asked to predict ‘what might happen next?’, and were more reactive in the sense that they were required to predict, not identify, a situation as hazardous, and only had to predict future events at a specifically chosen point in time.

Additionally, these findings suggest that including an active–practical component in training intervention programs can be an effective technique for enhancing complete-novice e-bike riders’ ability to perceive potentially hazardous situations. Thus, it is beneficial to promote the idea of actively responding to displayed hazard perception traffic scenes as an off-road tool for training young complete-novice e-bike riders to identify road hazards and to predict hazardous situations prior to their materialization. Utilization of this hazard perception training paradigm developed to produce a coherent intervention (as in the cases of [29,41]) can be useful for creating a future constructive intervention component aimed at reducing involvement in road crashes.

Lastly, this study presents early evidence that a hazard perception training package of the type used here may produce a training effect in young complete-novice e-bike riders that can persist even after a week. This can be regarded as an important first step in establishing that such training interventions have an enduring effect, although further work is needed to determine their longevity. Exploring and demonstrating complete-novice e-bike riders’ knowledge deficiencies in evaluating potential hazard instigators may contribute to the effort of producing public education programs as a major road safety technique.

7. Conclusions

The hazard perception skill has been extensively researched in various road safety domains [28,29,35,39,101] and found to be vital for road users, one of the few traffic negotiating skills to correlate with traffic crash involvement [38,42,43,44]. Furthermore, several countries worldwide utilize a video-clip-based hazard perception test as an integral part of the licensing process for automobile drivers [36,43]. However, only limited research exists on enhancing the hazard perception skill through training among e-bike riders in general and among young complete-novice e-bike riders in particular. This study reports the development and evaluation of two hazard perception interventions. Findings suggest that young complete-novice e-bike riders may benefit from training. In this regard, the findings may serve as a support tool for gaining better comprehension of the hazard perception skill in the hope of laying the initial foundations for designing a hazard perception training intervention for young complete-novice e-bike riders.

As the use of sustainable commuting may be beneficial for the environment, promoting e-bike use can contribute to a more environmentally friendly transport system. For e-bikes to achieve their full potential and become a sustainable and safe mode of transportation, policies should aim to improve these riders’ safety. Indeed, the current exploratory study’s findings suggest that a designated training intervention can be an effective tool towards this goal and play a role in the effort to enhance traffic safety. This study can be regarded as a starting point towards producing proper guidelines for future hazard perception training intervention programs for young complete-novice e-bike riders.

Lastly, as part of the design and implementation of public policies, consideration should be given to the development and distribution of programs that promote e-bike riding as a sustainable means of transportation while focusing on improving potential users’ perceptions of unmaterialized potential traffic hazards. By aiming to increase ease of use while promoting safety, these programs should be delivered to the general public in the most suitable and accessible way. For instance, future training interventions could be carried out either individually (e.g., as a self-administered application) or in a group setting following collaboration with various community institutions while providing designated incentives for the use of this ecological transportation mode and for taking part in the hazard perception training program aiming to enhance safety.

Finally, several limitations should be addressed. First, participants completed the hazard detection task in a controlled laboratory setting. Taking a more ecological approach by comparing their performances to those displayed while similar tasks are being carried out in real-world settings may provide further support for these findings. Moreover, a longer-term evaluation of the training intervention programs is needed in order to ascertain their contribution to riders’ safety and whether it can be sustained over a longer time period. Second, it should be noted that the reactive nature of the AAT intervention prevented the use of a pre-test/post-test control design, as is used in the case of training novice automobile drivers [29] and pedestrians [41]. The AAT was based upon the notion that repetitive exposure to traffic scene movies may be an effective tool for training hazard perception. As the hazard perception test itself necessitates exposure to traffic scenes, mere exposure to the test (i.e., use of pre-test) could alter the results by increasing the participants’ sensitivity and potentially modifying their performance. Thus, it was necessary to employ a post-test-only control design and to use an array of matching participants by creating equivalence between trainees and control and aiming to create groups that differed only in the specific treatment they each received. In addition, it might be argued that the AAT trainees could have performed better because they were familiarized with the hazard detection task. Nevertheless, an early examination conducted prior to the experiment determined that two practice videos were sufficient for participants to comprehend, acclimatize, and seamlessly carry out the experimental task.

Third, sample size might be considered a limitation of the present study. Notwithstanding this, in setting a predetermined commonly used α = 5%, the probability of finding a false positive was determined a priori. Hence, the sample size affected only the power of the procedures and the probability of true positives; that is, although the sample size did affect the probability of missing correct responses, it did not affect the probability of false alarms. Furthermore, the fact that significant effects were revealed strengthens the findings. Additional effects might have been missed as a result of low power following the small sample size.

Fourth, the sample of young complete-novice e-bike riders might not have been representative of the entire user population. Indeed, the process of allocating participants into groups by accounting for age, sex, automobile driving license status, socioeconomic status, and sensation-seeking score was only partially and not fully controlled. Hence, replicating the results with a larger and more diverse sample would further support these research findings. In addition, in the present study the trainees were chosen as complete novices in riding e-bikes, aiming to simulate a typical situation both in Israel [65] and in other countries worldwide, where any driver’s license holder can allow users to ride an e-bike without any prior experience. Future studies may benefit from exploring the effects of training on young inexperienced e-bike riders (e.g., novice riders with e-bike riding experience [49]) beyond zero-experience complete novices.

Lastly, as e-bikes are becoming increasingly popular among youngsters [9,24,25,26,27,99], the present research focused on riders in this age group. Yet, it is worth noting that in several regions, such as in Australia and the USA [5,8], the prevalence of use is rather higher among older and middle-aged individuals. Thus, it is suggested that further research should direct attention toward investigating and improving the hazard perception abilities of these e-bike riders.