Data source and participants
Analyses were based on the sixth (2009/10) Canadian cycle of the Health Behaviour in School-Aged Children (HBSC) Study . HBSC is a World Health Organization affiliated, cross-national study that involves study of health behaviours, their potential determinants and their consequences among 11–15 year olds. It employs an international mandatory questionnaire with additional modules that are used by sub-samples of participating countries . In the sixth cycle in Canada, in addition to data obtained from individual students, geographic measures were obtained to describe characteristics of school neighborhoods, as well as each student’s travel patterns and distances from home to school.
The Canadian version of the 2009–10 survey was completed by 26,078 students (77% of eligible participants) from 8 provinces and the three northern territories (11 of 13 eligible jurisdictions) . Schools were sampled with replacement; when a school was unable or unwilling to participate a neighbouring school with similar demographic characteristics was selected for study. The current analysis was restricted to urban students (n = 3,997) for which there was clear evidence that they lived proximally (within 1.6 km) to their school as assessed by postal code centroids or commuting times. This inclusion criterion limited the study population to students who were most likely ineligible for transportation in school buses. Consent for participation in the study (active or passive as dictated by local school board customs) was obtained at three levels – school boards, school principals, and parents/guardians. The HBSC study protocol was approved by the General Research Ethics Board of Queen’s University and the Health Canada and Public Health Agency of Canada Research Ethics Board. The current analysis received subsequent approval from the Queen’s University Health Sciences Research Ethics Board.
Active transportation (exposure variable)
Participants who indicated “walking” or “bicycling” as their main method of transportation to school were classified as engaged in active transportation. Those reporting that their main method of transportation to school was by “bus”, “train”, “streetcar”, “subway”, “boat/ferry”, “car”, “motorcycle”, “moped”, or “other” forms of transportation were classified as not being engaged. The items used to measure active transportation (type and duration) had been tested previously [17,18]. The first item (identifying the type of transportation used to arrive at school) had a high reported level of agreement between participants’ reports (Cronbach’s alpha ≥0.80) . The second item (identifying the duration of transportation to school) has also been tested psychometrically (percentage agreement range 74%-96%) .
Bullying (outcome variables)
Students were classified as being perpetrators or victims of bullying, categorized in a dichotomous fashion (yes or no) based upon a threshold frequency of the behaviour at least “2-3 times per month”. This classification was applied to an overall measure of bullying and for each of four specific subtypes (verbal, relational, physical, and cyber). Verbal bullying was defined as hurtful teasing delivered verbally. Relational bullying has the intended effect of ostracizing an individual from a group, possibly achieved through rumour spreading or social exclusion. Physical bullying referred to the use of physical means to assert dominance over an individual, (i.e., hitting, kicking, shoving). Lastly, cyber bullying involved Internet use in order to control an individual, (i.e., a composite question that addressed: someone sent mean instant messages, wall postings, emails and text messages, or created a Web site that made fun of me). These bullying items have been tested for face validity and reliability; the questions identifying involvement in most specific types of bullying were developed and evaluated originally by Olweus [1,19]. Questions used to measure the frequency of bullying involvement represented variations of the original questionnaires and were found to produce results that were congruent with the Olweus Bullying Victimization Questionnaire .
In addition to the above bullying module, we also used an additional item, collected for descriptive purposes, to determine whether “worrying about being bullied or attacked” was an impediment to active transportation to school” (yes or no). This was one of several questions used in the HBSC that documented facilitators and potential barriers to engagement in active transportation.
Possible confounders of the relationship between exposure to active transportation and victimization by, and separately, perpetration of, adolescent bullying were documented based upon the extant literature. Variables that had been identified as risk factors for bullying and also met classic statistical criteria for confounding  were retained in our statistical models. Confounders available for study therefore included: gender (male or female) [2,22] age (in years), [2,21] adiposity (body mass index categorized using age and gender-specific cut-points for normal weight, overweight and obese), [23,24] engagement in arguments with parents, (5 response options of “strongly agree” through “strongly disagree”)  communication with fathers and mothers (how easy is it for you to talk to the following persons about things that really bother you?; 5 response options of “very easy” through “don’t have or see this person”),  parental trust (my parents trust me; 5 response options of “strongly agree” through “strongly disagree”),  neighbourhood trust (you can trust people around here; 5 response options of “strongly agree” through “strongly disagree”),  sense of belonging at school (I feel I belong at this school; 5 response options of “strongly agree” through “strongly disagree”),  and support from teachers (I feel a lot of trust in my teachers; 5 response options of “strongly agree” through “strongly disagree”) .
We performed analyses using SPSS version 21.0 (IBM Corp., Armonk, NY). We first characterized the sample demographically. Patterns of bullying and its specific types (victimization then perpetration) were then described by gender and school grade. Tests for statistical significance of observed group differences were performed. Similar analyses were conducted for reports of active transportation, then also potential barriers to active transportation. We then used multiple logistic regression to examine relations between engagement in active transportation and reports of bullying, both as a victim and then as a perpetrator. We examined overall (any bullying) then specific bullying outcomes in these analyses. These analyses were viewed as exploratory. Confounder retention during the modeling process was informed by past literature, backwards elimination (a liberal p-value of 0.15 for inclusion), and a change in odds ratio of 10% or greater between unadjusted and adjusted models.