1 Introduction

Mobility is evolving in many regions throughout the world, especially in large cities where bicycles seem to be gaining more and more appeal. Following this trend, the current development of e-bikes, notably in Europe, is based on a number of personal and collective benefits. Whether bicycle-style or scooter style, e-bikes are attractive in terms of mobility, because they allow for covering longer distances than traditional bikes with a savings of time and effort [19, 20, 28].

Thereafter, a clarity evaluation of the inventory and the entire survey questionnaire was undertaken among a panel of 13 male and seven female two-wheeler users on the campus of Tongji University in Shanghai. The questionnaire was evaluated as appropriately worded and satisfactorily comprehensible, and hence no further modifications were made.

A total of 16 items were constructed for the study. Although they were presented to the participants in Mandarin Chinese, they are presented in English for convenience in Table 1. These items provide information about various potentially hazardous riding behaviors: overtaking, progressing under potentially dangerous circumstances, riding between two lanes, risky behavior at a red traffic signal, risky lane changes, risky riding behavior at intersections, and risky riding behavior while approaching pedestrian crossings. To assess the propensity toward taking risks while riding in these situations, respondents were requested to state how frequently they had adopted each of the listed riding behaviors over the last six months (0-never, 1-almost never, 2-occasionally, 3-often, 4-very often, 5-almost all the time).

Table 1 The 16 items of the A-TRIBE inventory before reduction procedure
Full size table

3.1.3 Procedure and participants

A self-administered questionnaire was used for collecting data from 400 local panelists recruited through a private Internet survey company based in Shanghai.Footnote 1 Prior Institutional Review Board (IRB) approval was obtained from both of the authors’ institutions. The participants completed the questionnaire without monetary compensation. Answers were automatically and anonymously recorded, as guaranteed to the participants. The 400 participants were recruited with an equal distribution of 80 individuals within each of the five following categories: currently being solely or mainly a rider of either a bike (human-propelled; 50% male riders), a bicycle-style e-bike (55% males), a scooter-style e-bike (62.5% males), a non-electric engine scooter (LPG scooter, gasoline scooter; 50% males), or a motorcycle (72.5% males). Participants were between 18 and 65 years of age (M = 31.7 years, MD = 31 years, SD = 7.9) and were mainly from the working population (92%).

3.1.4 Measurements

The research questionnaire also incorporated measured variables generally associated with driving risk. Relationships with the variables introduced below are relevant to inform criterion-related validity of the A-TRIBE inventory, and more precisely its concurrent validity [39].

Past accidents Respondents were asked to provide information about how many accidents they had been involved in over the last three years while riding their primary vehicle. Accidents were defined as crashes, collisions, or just falls, with or without serious injury, whether the accident was the rider’s responsibility or not. With the intention of gathering information about the risky nature of some riding behaviors, a significant relationship was expected between the A-TRIBE inventory sum-scores and self-reported past accidents.

Primary two-wheeler Participants provided information about whether they were only or mainly a rider of a bicycle, a bicycle-type e-bike, a scooter-type e-bike, a non-electric engine scooter (LPG scooter, gasoline scooter), or a motorcycle. The A-TRIBE inventory sum- scores were expected to have a significant relationship with the type of two-wheeler ridden, referred to here as the “primary two-wheeler.”

Maximum speed Respondents who rode a motorized vehicle were asked to provide information about the highest speed that their primary two-wheeler could attain: maximum speed below 20 km/h, between 20 and 50 km/h, between 51 and 70 km/h, and over 70 km/h.

Riding/driving experience variables Experience as a rider was assessed using five different measures: length of use (less than 6 months, 6 months to 2 years, 2 to 5 years, 5 to 10 years, or more than 10 years), number of riding and driving licenses obtained, switching back and forth between one type of two-wheeler and another, prior experience with an alternative type of two-wheeler (within the 6 months prior to use of the current two-wheeler), and current and past car driving experience.

Socio-demographic variables: Correlations between A-TRIBE sum-scores and gender, age (classified as 18–25, 26–30, 31–35, and 36–65 years old), and occupation were explored.

There was a need to ensure that a latent factor from the A-TRIBE inventory (adopting risky riding behaviors with two-wheelers) had relationships with the external markers listed (discriminant validity). Overall, significant correlations between the A-TRIBE inventory’s item sum-scores and these measurements were expected.

4 Data analysis

The data analyses were performed using SPSS and AMOS software.

First, the A-TRIBE inventory responses were analyzed to determine the A-TRIBE’s construct and internal validity. Principal axis factoring (PAF) was employed along with split-half test analysis and computation of internal consistency (Cronbach’s alpha value, 1951) for item reduction to the final 12-item structure. A Bartlett sphericity test and a Kaiser–Meyer–Olkin test of sampling adequacy ensured data factorability. Subsequently, confirmatory factor analysis (CFA) was conducted using the unweighted least squares (ULS) estimation method (Browne, 1982, cited in [22, 31].

Next, we assessed the relationship between the A-TRIBE inventory items’ sum-scores and external variables (criteria) as listed in the “Measurements” section using the eta (η) correlation ratio [44]. After checking to ensure that assumptions of linearity, normality, and homogeneity [7] were met, F-ratios (or adjusted F-ratios when homogeneity was not met [26], were computed.

5 Results

5.1 Intrinsic psychometric properties

5.1.1 Exploratory factor analysis

Principal axis factoring (PAF) resulted in the extraction of items 1, 3, 4, and 5. The remaining 12 items explained 52.81% of the variance, with high factor loadings between 0.575 and 0.853, greater than the cutoff threshold of 0.40 (Table 2).

Table 2 Final Principal Axis Factoring (PAF)’s factor loadings and communalities of the final 12- item structure of the inventory (n = 400) after reduction procedure
Full size table

Final communalities were high (> 0.60) to moderate (> 0.40). Only items 6 and 7 contributed slightly less to the single factor, but their communalities were still above the threshold of 0.30, which makes them acceptable. Furthermore, as the size of the sample (n = 400) was quite good for running factor analysis [23] 53 and the other values were consistent (internal reliability test), both of these considerations supported kee** the 12-item structure of the inventory. In addition, data factorability was supported by a significant Bartlett sphericity test (χ2 = 2903.929; p < 0.001 and a high Kaiser -Meyer-Olkin test value of sampling adequacy (KMO = 0.941) exceeding the minimum recommended value of 0.60 [46].

5.2 Internal reliability

The Cronbach’s alpha value of the 16 items was 0.93 on the 12 remaining items. No further increases would have been achieved by eliminating more items. Similarly, split-half reliability tests were satisfactory: a correlation of 0.83 between the forms, a Spearman-Brown coefficient of 0.90, and a Guttman split-half coefficient 0.90.

5.3 Confirmatory factor analysis

Using the ULS estimation method, we tested the 12 items as a single-factor model of the A-TRIBE inventory (Fig. 1). This model yielded satisfactory fit indexes (see Table 3). GFI and AGFI had values close to 1, indicating a perfect fit; PGFI exceeded 0.60, and PRATIO was also greater than 0.60, which is generally considered satisfactory [22]. Lastly, the RMR value was close to zero, which suggests a good fit [40].

Fig. 1
figure 1

CFA single-factor model testing on the 12 final items of the A-TRIBE inventory (unstandardised loadings)

Full size image
Table 3 Fit indexes
Full size table

5.4 Criterion-related validity analysis

The correlations of A-TRIBE inventory sum-scores were assessed with each criterion-related variable (see “Measurements” section). Thus, Table 4 provides information on the F-ratio (or adjusted F-ratio when homogeneity was not met), and eta (η) correlation ratio, with the criterion variable as either an independent (iv) or dependent variable (dv), or the computed eta (η), and lastly eta squared (η2) to indicate effect size (Sir Arthur Lewis Institute of Social and Economic Studies, 2012; Sydney, 1956; [37].

Table 4 F-ratio (or adjusted F-ratio), significance, eta (η) correlation ratio, and effect size (η2) of A-TRIBE inventory with criterion variables (N = 400)
Full size table

Here, results indicated mostly significant (p) but low correlations (η) of small to medium effect sizes (η2; [32, 46]. There were significant correlations as expected, even if very low, with riding/driving licenses (η = 0.18), current experience with a different type of two-wheeler (current switching between different types of two-wheeler (η = 0.17) and prior experience with an alternative type of two- wheeler η = 0.16), and current car driving experience, η = 0.16. The highest—but still low—correlations were shared by former accidents (η = 0.23), primary two-wheeler (η = 0.20), past car driving behavior (η = 0.21), and maximum speed (η = 0.25). Overall, the effect sizes were small to medium (from 0.025 to 0.065). Finally, there were no significant relationships with length of use and socio-demographic variables except gender (η = 17).

6 Discussion

6.1 Summary of results

For the purpose of facilitating a comparative approach extended to all types of two- wheelers, we developed and assessed the validity of the A-TRIBE inventory. This inventory addresses all two-wheelers: users of bicycles, bicycle and scooter type e-bikes, LPG and gasoline scooters, and motorcycles. This inventory is intended to assess the propensity of riders for risk-taking through the frequency of adoption of riding behaviors common to all these two-wheeler types. An internal validity test (reliability and factorial structure) was conducted and the criterion-related validity of the A-TRIBE inventory was examined in relation to external variables that have been generally shown in the literature to be associated with risk-taking and crash rates. Regarding intrinsic psychometric properties, the A-TRIBE turned out to be very satisfactory, with a final single-factor 12-item structure (52.81% of variance explained, α = 0.93), with very good fit indexes. Furthermore, based on previous research regarding risky driving behaviors and road accidents, the A-TRIBE inventory’s construct validity was investigated by assessing its relationships with accident history, two-wheeler types, different levels of riding/driving experience, and socio-demographic variables. Overall, the eta (η) correlation ratio tests run on the A-TRIBE sum-scores indicated significant correlations, as expected. More importantly, since one goal was to gather information about the risky nature of some riding behaviors, one of the highest correlations of the A-TRIBE inventory sum-scores was with past accidents. Assessing road risks in terms of crash rates is quite common in the literature (e.g. [17]. Besides these relationships with past accident history, inventory scores also significantly correlated with experience variables typically related to road risk. Only length of time primarily riding a specific vehicle type did not significantly correlate with A-TRIBE scores. One explanation could be the recent increase in urbanization and the abrupt transition to motorized modes of transportation dating back to the 2000s in China, after years of bicycle riding. The consequences are a lack of practice and poor knowledge of road safety and traffic rules, due to economic and historical factors [41]. Despite the lack of a significant relationship of inventory scores with occupation and age, it may be interesting to examine whether population displacement between districts is sometimes accompanied by social inequalities, which would lead to reliance on specific modes of transportation and modal shifts and/or to different traffic conditions (travel time, type of transport, traffic density, etc.). Another result was the high correlation between A-TRIBE inventory scores and the maximum speed of the vehicle. This result suggests a need to further compare two-wheeler riding behaviors based on this variable rather than based on their types. That may be preferable if there is serious suspicion of engine power tampering.

Overall, the low correlations and small to medium effect sizes were reasonable considering that riding behaviors might be explained by multiple factors at different levels.

6.2 Strengths, limitations and future research studies

This study was carried out in Shanghai where e-bikes have long been in frequent use. Beyond some practical and cultural differences, China can be viewed as a useful source of data to shed light on issues that are already emerging in Europe. With an application perspective, the data collected there could thus help in the early definition of safety measures based on scientific knowledge. Initially focusing on e-bike riders’ safety, the A-TRIBE facilitates the development of relevant road safety measures for all types of two-wheelers, while being adaptable to some of their specificities. This concerns both safety policies and measures such as, for example, media campaigns, enforcement, new regulations, in such a way as to be relevant to each type of two-wheelers users.

To further establish the A-TRIBE’s predictive validity, forthcoming research studies would examine the relationships between the items of risky riding behaviors and other variables, such as traffic violations, incidents or accidents. Another research direction will seek to identify among cognitive and psychosocial determinants of risky riding behaviors (such as those related to personality, conative, social or cultural dimensions) some levers for modifying riding in favor of greater safety. The effectiveness of future safety programs could be increased by investigating such determinants of risky riding that may for some be shared between several types of two-wheeler users or specific to one type of user.

The data collected here with this instrument are situational. However, this inventory might be easily transferred with some slight adaptations for application in the European context. Additionally, besides some local peculiarities, the variables along which risk was measured by this instrument will help to better understand the emerging risk factors in Europe. For example, it will be useful to consider the effect of experience with the current vehicle, prior experience with alternative two-wheeler types, and the possession of an automobile driver’s license as factors to be investigated in terms of their relationship to the frequency of risky behaviors.

7 Conclusion

Today, the growing use of e-bikes is noticeable in major cities around the world, and particularly in Europe, the mobility they afford has a number of individual and collective advantages. But for these benefits to not be offset by the negative consequences that can result from their increased use, namely the risky behavior of their users and the traffic accidents that can result from it, a number of measures must be implemented in terms of urban policies that encourage safer highway travel, aiming at zero death and serious injury. In fact, the evolution in types of road traffic implies a necessary adaptation of the road system. Riders of two-wheelers are considered vulnerable users because of the greater severity of the injuries they are likely to incur in the event of an accident compared to drivers of heavier vehicles. Despite the specificities attached to each type of two-wheelers, they share a certain number of physical and dynamic characteristics (e.g. maneuverability in traffic), as well as certain similar behaviors on the part of their users. These common features can lead to similar accident mechanisms and could suggest the implementation of similar safety measures. And as Hayworth and Debnath (2013) already mentioned with regard to cyclists and motorcyclists, road safety policies which have shown their effectiveness in preventing accidents for all types of users should prove beneficial to the safety of others. New joint actions integrating the current evolution of mobility should be developed.

To identify similar safety measures which are relevant for different types of two-wheeler riders, including e-bikes, there is a growing need to characterize and explain risky riding behaviors by e-bike users compared to users of other types of two-wheelers. The tool presented in this paper has shown its capacity to characterize the propensity toward risk-taking of e-bikers using a comparative approach with other types of two-wheelers. Regarding its design, this inventory might be easily transferred with some slight adaptations for application in the European context. It will be of significant interest when e-bikes, in addition to other traditional two-wheeler types, multiply on roads in large European cities and further complicate road interaction dynamics.

There are many implications for the prevention of accidents with injuries and/or fatalities concerning all road users, with a potential gain in investment in treatment. For example, faced with the same risky behaviors, educational content could be designed and implemented for education and training for users of different types of two-wheelers. Likewise, operational behavioral influence techniques could be developed (classical or technological persuasion, commitment, biding communication). Moreover, safety measures are needed for users of two-wheelers themselves but also for all road users who interact with them and face similar risky configurations regardless of the type of two-wheeler involved.

Road safety is now recognized as a shared responsibility between all stakeholders, integrating human fragility and the eventuality of human error. Data collection via A-TRIBE, adapted to the local context studied, should also be hel** in making the traffic environment safer for these new two-wheeled users. The measures to be implemented concern the road and street infrastructure, which must be designed, built, and maintained in such a way as to integrate e-bikes, in particular, into a more harmonious and protective traffic pattern for their users and those interacting with them (e.g., pedestrians). This involves the development of a layout that promotes traffic calming measures considering the physical and dynamic specificities of these vehicles and others, as well as the implications in terms of user behavior.