Distraction of cyclists: how does it influence their risky behaviors and traffic crashes?

Different types of distractions, one single problem

Distractions on the road may have different origins, dynamics and consequences (Neyens & Boyle, 2007; Neyens & Boyle, 2008). However, most of them have been related to the same result: traffic crashes potentially cause preventable injuries among the involved road users (Kahn et al., 2015; Llerena et al., 2015). In the first place, cycling while being distracted by technology, especially for what concerns the use of cellphones, headphones and navigators, constitutes a relevant factor whose impact on the road safety of cyclists has been demonstrated by empirical studies such as the one performed by Ethan et al. (2016), and Wolfe et al. (2016), showing that headphones and visual/tactile devices could be the most prevalent distractors related to bicycle using. Nevertheless, there are many potential distracting sources other than technological devices, which may potentially influence the adverse outcomes of road safety (Stimpson, Wilson & Muelleman, 2013). Decker et al. (2015) highlighted that external distractions, such as billboards on the roadsides and elements inherently designed to attract the attention of drivers (Dukic et al., 2013) affect between 6% to 9% of collisions among motor vehicles caused by road users’ distractions. Other relevant external sources of distraction are related to the behavior of other road users, that, as in the case of distracted pedestrians or drivers, may influence the attention of cyclists during the riding task, causing potential impairments in their cycling performance and thus explaining many operational errors, subsequently deriving in potential traffic crashes. Finally, weather conditions and road features may play a significant role in catching the attention of road users; in fact, some environmental conditions modulated by weather, such as darkness and low visibility, are associated with a higher degree of perceptual errors such as distraction and lack of attention (Jägerbrand & Sjöbergh, 2016; Oikawa et al., 2016), and many infrastructural factors, such as the presence of obstacles in the way, may imply a substantial decrease in the cycling performance and higher probabilities of being involved in a traffic crash (Useche et al., 2018a).

Also, it is known that, although cyclists are highly prone to experience distractions on the road, there are some mitigating factors that differentiate them from motor vehicle drivers: bicycles are usually exempt from some typical distractors present in heavier vehicles (e.g., audio systems, integrated navigators and other on-board units), and the absence of noise-isolation makes cyclists more prone to detect sonorous stimuli potentially present in the road environment. Furthermore, the average low speed at which cyclists circulate, often related to the high density of traffic in urban areas, constitutes a factor which reduces the objective risk of crash and injury. Also, the growing availability of cycling lanes in various countries constitutes a relevant mitigating factor which not only prevents cyclists from suffering cycling crashes, but also reduces their severity (Kaplan, Vavatsoulas & Prato, 2014; Harris et al., 2013).

For what concerns the existing empirical studies on road distractions, most of them agree on the fact that distraction may explain a large part of traffic crashes suffered by road users, and their results have been useful for the improvement of policy and decision making regarding traffic accidents and their associated positive outcomes for traffic safety. For instance, following the results obtained by Dukic et al. (2013) when testing the potential impairment caused by billboards in the drivers’ attention, the Swedish authorities decided to remove a large set of electronic billboards under testing phase, considering their proven distracting effect on drivers, and the subsequent hazardous outcomes for their safety. Also, in Spain, billboards on the sides of inter-urban roads were prohibited in 1998 as a way of reducing the negative effects of road distractions on the performance of drivers. Also, Stutts et al. (2003) raised the urgent need for a better management of all forms of distraction, in order to ensure the road safety of users. Furthermore, road safety education of cyclists and drivers, together with a successful integration of cycling into normal traffic patterns, might help decrease the morbidity and mortality on the road (Kiburz et al., 1986). To sum up, these have been our main practical motivations for developing the present study, focusing on the specific case of bicycle users.

Objectives of the study

The main purposes of this study were: first, to explore the prevalence and trends of cycling distractions in an international sample of bike users, and second, to determine their influence on road crashes involving cyclists, simultaneously considering the explanatory role of risky behaviors (errors and traffic violations) as potentially mediating variables between cycling distractions and traffic crashes.

Sample

The data was obtained from a full sample of 1,064 bike users from 20 different Latin American countries (831 participants, representing 78.1% of the sample; 38.6% females and 61.4% males), Europe (161 participants, representing 15.15% of the sample; 41% female cyclists and 59% males), and North America (72 participants, representing 6.75% of the sample; 37.5% females and 61.1% males), 413 (38.8%) females, and 651 (61.2%) males. A graphic contextualization of the geographical coverage of the project can be seen in the Fig. 1.

Regarding the educational level of participants, approximately half of the respondents (50.1%) had an undergraduate degree, and 29.9% a post-graduate degree. A total of 9.1% of them had received a technical training (more advanced than a high school diploma, but lower than a university degree); 10.4% only had a high school diploma, and the remaining 0.5% had a maximum educational level of primary studies. They were aged between 17 and 80, with a mean value of M = 32.83 (SD = 12.63) years.

Procedure

For this research, a convenience (non-probabilistic) sampling method was used, grounded on the availability and accessibility to the study population, and on their willingness to participate (or not) in the research. For this purpose, we designed an online (electronic) form in order to collect the data, and this was individually sent through an e-mail invitation to a wide sample of subjects contained in a pre-existent mailing list used for research purposes, composed of institutional members (staff, students and collaborators) and individuals who had previously participated in other studies.

The data collection was performed by the staff of the INTRAS-University of Valencia. During the presentation of the survey, potential participants accessing the online form were informed about the existing laws on data protection and about the fact that the collected information would only be used for statistical and research purposes. They were also asked, prior to the beginning of the electronic survey, to complete the questionnaire only if they were frequent bicycle users (it was suggested, as a guideline, “if you ride a bike with certain regularity, at least once per week, or once every few weeks”).

It is worth mentioning that, although this data collection method implies several advantages for the researchers, among which we find the increase of the efficiency, approachability and expediency of the data collection, the fact that subjects participating in this study belong to a pre-verified source, and that the questionnaire can be easily adapted to the time availability of participants (Tyrer & Heyman, 2016), it yet remains vulnerable to a reliability-related bias, given the impossibility to check, for instance, the veracity of the basic data (i.e., demographics) provided by the respondent (for further information, please see Limitations of the study).

Description of the questionnaire

The questionnaire (Appendix I) was administrated in Spanish, and consisted of four sections: The first part asked about individual and demographic variables, such as age, gender, region of provenance, educational level and occupation.

As for the second part, self-reported risky cycling behaviors were assessed using the raw item bank of the Cyclist Behavior Questionnaire (CBQ) (Useche et al., 2018c), a self-report measure of road behaviors specifically designed for measuring high-risk riding behaviors (errors and violations) performed by cyclists. This Likert scale is originally composed of 44 items distributed along three factors: Violations (V), consisting of 16 items; Errors (E), composed of 16 items; and Positive Behaviors (PB), consisting of 12 items. A global score of Risk Behaviors was built through the sum of Errors and Violations reported by respondents.

The entire questionnaire used a frequency-based response scale of 5 levels: 0 = never; 1 = hardly ever; 2 = sometimes; 3 = frequently; 4 = almost always. Regarding the study’s factors, the first two dimensions were grounded on the original factorial structure of the Driving Behavior Questionnaire (DBQ) (Reason et al., 1990). Additionally, this scale includes a supplementary factor (positive of protective behaviors) for measuring behaviors that, unlike the first two dimensions, may help to prevent the occurrence of traffic accidents. To measure the risk perception and the knowledge of traffic regulations among cyclists, the Cyclist Risk Perception and Regulation Scale (RPRS) (Useche et al., 2018a) was used; it is a Likert scale composed of 12 items (seven for risk perception and five for assessing general rules of bike use), in which the degree of risk perceived in objective risk factors and the knowledge of general road regulations are assessed in a scale from 0 (no knowledge/risk perceived) to 4 (highest knowledge/risk perceived).

Thirdly, for assessing cycling distractions, an eight-item scale was build using dichotomous questions (yes/no), aimed at presenting different potential distractors and determining their self-reported presence and influence on the participants’ common journeys.

The last part of the questionnaire consisted of a series of questions related to the use of bikes, such as the average use of the bicycle (including average distances traveled and length of trips) and reasons for using it as a mode of transportation. Finally, this section of the questionnaire also included two questions about the traffic crashes suffered by participants, as cyclists: first, if they had suffered (or not) cycling crashes during the previous five years—regardless of their severity, but specifying that they were non-fatal crashes—and, second, in case of an affirmative answer, the number of crashes suffered during this period while cycling. In this sense, the variable traffic crash rate can be understood in this study as “the total amount of traffic accidents or crashes suffered while cycling during the last 5 years”.

Ethics

In order to carry out this study, the Social Science in Health Research Ethics Committee of the University of Valencia was consulted. They certified that the research responded to the general ethical principles, was currently relevant for research in Social Sciences, and accorded with the Declaration of Helsinki, thus issuing a favorable opinion (IRB approval number H1517828884105). Furthermore, an informed consent statement containing ethical principles and data treatment details was used for all participants, explaining the objectives of the study, the average duration of the survey, the treatment of personal data and the voluntary participation, and it was always provided to the participants before they answered the questionnaire. Personal and/or confidential data were not used, and participation was anonymous, implying no potential risks for the integrity of our participants.

Statistical analysis

In addition to the descriptive analyses, conducted in order to obtain the frequencies of distractions experienced by cyclists, and to the average scores for the used scales, a correlation analysis was performed to establish potential relationships among the variables of the study. After testing basic parameters and comparability criteria, a one-way ANOVA and Welch’s mean difference tests were used for comparing the average scores of distractions between gender and age groups. Furthermore, the association between distractions, age and risk perception, traffic crashes suffered while cycling during the previous five years, and the potential mediation of this relationship with cycling errors and normative violations, were tested using structural equation modeling (SEM). Weighted Least Square Mean and Variance Corrected (WLSMV) estimation was used, given that some data were ordinal and that the assumption of multivariate normality was not met (Caycho-Rodríguez et al., 2018; Finney & DiStefano, 2013). The model fit was evaluated through several statistics and indices from different logics and families. In this particular case, all available types of indices for the method of estimation were used: the chi-square, CFI, and RMSEA. Fit was decided based on the cut-off criteria expanded in the literature. A CFI above .90 (better if above .95) and RMSEA below .08 (Marsh, Hau & Wen, 2004) were indicative of an adequate model fit. Significance of parameters was established at p < .05. SEM were estimated in Mplus 8.0 (Muthén & Muthén, 2007–2018).

Frequency analyses

Table 1 summarizes the first block of descriptive statistics (i.e., frequency analyses), in order to determine the prevalence of distracting sources in our sample of 1,064 participants. Overall, all the distractors had a relevant influence over cyclists, in a rage between 34.7% and 83.6%. Specifically, and based on the list contained in the table, the most frequently reported distracting sources which affected cyclists were the behaviors of other road users (83.6%), the presence of obstacles on the way (83.5%), and the current weather conditions (68.5%). On the other hand, the least reported distracting sources were: billboards (visual elements) with 34.7%, people that found them attractive (47.5%), and text messages or chats. It is worth mentioning that, although they are part of the same potentially distracting element (i.e., the cellphone), telephone calls (64.9%) represent a more reported distractor than text messages (46.4%).

As for the number of distracting sources reported by each participant, Table 1B shows that 60.1% (six of each 10) of cyclists commonly experience between four and six of the measured distracting factors during their cycling trips. Also, it shows that only 1.4% of them think they are not affected by any of these eight distracting sources.

Distractions, gender and age groups

For what concerns the second block of descriptive analysis, a continuous variable was built (i.e., distractions while riding); it summarized, based on the previous list, the number of distracting sources which commonly affected participants. In this regard, both cyclists who consider that they do not experience any of the distractions when using the bicycle and cyclists that report being affected by all of them were found.

Overall, the mean value obtained for the variable was M = 4.84 (SD = 1.79) distractions experienced by users, in a scale between 0 (minimum) and 8 (maximum). As for gender comparisons, the results showed that men are the users who get more distracted (M = 4.91; SD = 1.86), compared to women (M = 4.73; SD = 1.67), although the mean difference (ANOVA: F_(1,1062) = 2.333; p = 0.127; Welch: F_(1,944.907) = 2.447; p = 0.118) is not statistically significant at level p <0.05, and for that reason this data should be interpreted only as an observed trend.

Furthermore, age comparisons were performed in order to determine the hypothesized differences in distractions among different age groups. For this reason, five different age intervals were created for conducting the comparative analysis, as summarized in Table 2. In this regard, the highest mean value of cycling distractions was reported by bike users belonging to the interval between 46–55 years of age (M = 5.09; SD = 1.98), and the lowest one was found in the case of cyclists under 26.

Statistical analyses allowed us to find out that age is one variable differentiating distraction rates among bike users (F_(4,1058) = 7.998; p < 0.001), and we also observed a growing tendency in the scores they reported among the different age groups, as shown in Fig. 2.

Correlation Analysis

The bivariate (Pearson) correlation analysis (see Table 3) for σ coefficients and significance levels permitted the identification of significant associations among cycling distractions, other individual factors, and traffic crash rates of the participants. Specifically, distractions were significantly related to cycling intensity (i.e., the average hours riding per week) [−], age [+], cycling errors [+], protective behaviors [−], and psychological distress of cyclists [+]. On the other hand, traffic crashes suffered during the last five years were significantly correlated to age [−], cycling intensity [+], cycling errors [+], traffic violations [+], level of knowledge of traffic rules [−], and psychological distress of users [+].

A specific correlation analysis based on the region of provenance of participants allowed us to confirm the high similarity in the directions of the association measures (Pearson’ coefficients) between study variables, when considering three subsamples (i.e., Latin America, Europe and North America). It is interesting to note how traffic crash rates of Latin American participants show differential values of significance if compared to cyclists from Europe and North America in the case of demographic (age) and cycling-related factors (cycling intensity). Despite the initial disproportionality of sub-samples used for this complementary analysis, most of the significance levels and magnitudes keep similar values, as shown in Table 3B.

Structural equation modelling

With the aim of testing the hypothesis of the effect of cycling distractions, age and risk perception in traffic crashes involving cyclists, and the potential mediating role of errors and traffic violations within an explicative dynamic, a structural equation model (SEM) was built according to the empirical directions reported in the introduction. Age, distractions (built up through the sum of an eight-item scale, as shown in Fig. 3) and risk perception were independent variables, violations and errors were mediators and the final outcome was the traffic crash rates. This initial completely a priori model did not fit the data well: χ²(55) = 746.46, p < .001; CFI = .810; RMSEA = .109, 90% CI [.102–.116]. Therefore, several modifications were made. Firstly, two non-significant and very low paths from distractions to traffic crash rates and to violations were set to zero. Secondly, a very large modification index that pointed out a relevant relationship between the two first indicators of distractions was included. With these three modifications, that made the model even more parsimonious, the model fit resulted adequate: χ²(56) = 408.83, p < .001; CFI = .903; RMSEA = .077, 90% CI [.070–.084]. All standardized parameter estimates are presented in Fig. 3, in which the unidirectional arrows indicate the direction of the predictive relationship, and bidirectional ones the correlational association between study variables.

In this regard, errors while riding (β = .094; p < . 05) and traffic violations (β = .308; p < .05) have a direct positive effect on the rate of traffic crashes suffered by cyclists (dependent variable). On the other hand, the explicative association of age (β =  − .173; p < .05) occurs in a negative direction, as observed in the unidirectional arrow linking age of cyclists and cycling crashes suffered. In other words, the higher the age of cyclists, the less prone to experience cycling accidents they are. As for risk perception (β = .065, p > .05), this variable showed a statistically non-significant explanatory effect on traffic crashes reported by cyclists, suggesting that it has not a direct influence on crashes.

Regarding the latter, and as for the hypothesized mediating role of errors and violations in the relationships among distractions, age, risk perception (independent variables) and traffic crashes (dependent variable), tested through this procedure, our results showed that errors while riding have significant relationships with age (β =  − .173; p < .05), risk perception (β =  − .157; p < .05), and distractions (β = .24; p < .05), as they are predictors of the dependent variable, i.e., the number of accidents suffered in the last five years, but essentially when they are linked to the commission of cycling errors.

As for the second mediating variable (traffic violations while cycling), this relationship follows the predictive model when considering age, i.e., more violations explain more traffic crashes (β =  − .242; p < .05), and risk perception (β =  − .151; p < .05). When indirect (mediation) effects were calculated together with traffic crash rates, the indirect effect of age was statistically significant in traffic crash rates (β =  − .088; p <  .05), as well as in risk perception (β =  − .061; p < .05), but not in distractions (β = .023; p > .05).

In other words, the observed relationships between study variables suggest that both errors and violations need to be kept in mind when considering the causal chain, in order to establish a predictive link between distractions and risk perception (independent variables), and the number of crashes suffered by cyclists during the last five years: a younger age and risk perception, and a higher score in cycling distractions predict more traffic crashes, through the commission of both non-deliberate (errors) and deliberate (violations) risky behaviors while riding.

Limitations of the study and further research

Finally, some potential sources of bias and two essential facts related to the collection of data and to the analyses performed in this study should be mentioned. First, while this research employed a considerably large sample of cyclists, the survey was conducted through an electronic questionnaire and based on self-reported information.

Regarding this, and although research on individual differences in traffic safety—including most of the studies dealing with risky behaviors—has been developed mainly on the basis of self-reported data (af Wåhlberg & Dorn, 2015), it is worth mentioning the potential biasing effect of this data collection method on the reliability of our results. Specifically, some bias related to the use of self-report and to the convenience sampling, potentially affecting the results of the study, are: response bias, that may range from the social desirability (even when the questionnaire is anonymous, the tendency to show one’s “best look” in behaviors and attitudes may still be present, in this case related to road safety habits) to responses provided with a poor understanding of the questions or the survey dynamics (Rosenman, Tennekoon & Hill, 2011); these potential biases can be present even when instruments are carefully revised several times according to different ages and educational levels. Another bias may be the potential under/overrepresentation of the study’s population, that could make it difficult to perform further comparisons among subjects with (in this case) other habits for transportation; and, finally, it is worth mentioning the selection bias, i.e., the possibility that some population members were more likely to be asked to participate than others. For these reasons, it is important to remark the need for researchers to be careful when making generalizations based on the available data, since it may exceed the actual characteristics of the study’s sample.

And, also, we wish to remark that this is an international study, a commonly positive aspect which can nevertheless imply a negative side: the heterogeneity of traffic dynamics. In this regard, we should bear in mind that the specific traffic patterns and infrastructural advances in different countries may vary largely (Lahrmann et al., in press; Reynolds et al., 2009), factor which could be important when generalizing the results obtained for groups of cyclists that, although having similar individual characteristics, may be exposed to differential road risk factors modulating their road behavior. Specifically, and it is also important, the lack of data on other complementary factors (e.g., distractions from other road users, traffic volumes, average speed) could represent a major limitation for explaining a higher variance rate in the prediction of traffic crashes involving cyclists. For this reason, it is important to remark the need of considering multi-level designs and accident/crash related information (i.e., objective records) in order to enhance the predictive ability and external validity of the findings. Furthermore, it may be very useful to consider factors such as average speeds, the severity of reported cycling crashes and the traffic volumes through the Annual Average Daily Traffic (AADT) data—an important couple of missing element in this study—as categories to contrast the potentially differential impact of distractions on traffic crash outcomes of cyclists.

Regarding the SEM model fit, although the coefficients of the final model are good overall (CFI > .90; RMSEA < .80), the cutoff point for these coefficients may vary depending on the theoretical approach. Also, and although it was fulfilled in this research, the root mean square error of approximation (RMSEA) should not be pursued as the single way for determining the model fit (Chen et al., 2008). This is the reason why it is suggestible to consider the use of larger samples and the test of the model fit.

Finally, and speaking about further researches which may be conducted in this field, it is also suggestible to use samples with similar proportionalities in order to make potential comparisons between cyclist of different regions/countries, task that may contribute to obtain relevant information on the impact of cycling distractions within different contexts.