We conducted a cross-sectional survey using a two-stage stratified cluster sampling design. The strata were geographical (16 mandals, equivalent to boroughs) and administrative (types of school management).
There are three main types of schools in Hyderabad: government, semi-private and private schools. ‘Government’ schools are run by the Central or State Government; ‘semi-private’ schools are government-aided schools which are managed privately but receive regular maintenance grant from the government, local body or any other public authority; and ‘private’ schools which are run by a Society or a Trust without government aid . There are 802 government schools, 342 semi-private schools, and 1,899 private schools in Hyderabad. We considered type of school to be a marker of socio-economic status and parental influence: generally, government schools cater to lower income families, semi-private schools cater to middle income families and children from higher income families attend private schools.
We obtained lists of all schools in each mandal in Hyderabad district with grades 6–9 (typically children aged 11–14 years) from the District Education Office. We selected one school of each type from each mandal at random, using random numbers generated using the software R. In each school selected the principal randomly selected two sections (i.e. classrooms which normally have 30–40 children) in grades 6–9. Where schools had only one section in grades 6–9, it was selected. All children in grades 6–9 who were present on the day of the survey were included in the study. Assuming that the true prevalence of walking to school was 50 % , we estimated that a sample of 6,000 children would be required to be 95 % confident that the sample estimate would be within 5 % of the true prevalence.
We prepared a self-completion questionnaire with 21 questions about distance and mode of travel to school and conducted extensive piloting of the questionnaire . The questionnaire collected information on the usual mode of travel to school, mode of travel during wet or dry weather conditions, parental permissions for independent travel, children’s perception of safety, and physical activity after school. We used an English version of the questionnaire in private schools, and a Telugu version (which was the language of instruction) in government and semi-private schools. The questionnaire was administered during a regular class period and could be completed in 15–20 min.
The outcome variable was children’s usual mode of travel to school. The exposure variable was distance to school. Potential confounding variables were grade, gender, school type, physical activity, and parental permissions for independent mobility. We estimated distance from home to school using Google EarthTM based on the school location and self-reported nearest landmark to home. The estimated distance has been shown to be accurate to within 65m (-30m to 159m) for walking and cycling and to within 325m (-664m to 1314m) for motorised transport .
Modes of transport were categorised as walking, cycling, auto-rickshaw and cycle rickshaw (commercial three-wheeled passenger vehicles), school bus (private), RTC bus (public road transport corporation bus), motorised two-wheeler (motorbike), car and train. We assessed independent mobility by asking whether children were allowed to cycle and to cross main roads on their own. Distance to school was categorised as: 0.25 to 0.5km; 0.5 to 0.75km; 0.75 to 1km; 1.0 to 1.25 km; 1.25 to 1.5km; 1.5 to 2km; 2 to 2.5km; 2.5 to 3km; 3 to 5km and >5 km. These distance categories were chosen to ensure similar sample sizes in each group. Grades were categorised as grade 6, 7, 8 or 9. Physical activity was categorised as the number of days and hours exercised after school during the past week.
Research assistants with survey and interview experience conducted the survey in the schools, in the presence of the class teachers. The survey was conducted from November 2013 to February 2014. Each question was read out aloud by a study investigator, allowing plenty of time for the children to give their responses. Only after all children in a class had answered one question did the study investigator read out the next question, until all questions had been answered. This ensured that any questions, or doubts, that children had were attended to immediately, so no child would feel left out. The study investigator made monitoring visits to schools to ensure that each question was read out and explained to the children.
For each stratum, we estimated the probability of each school being selected (first stage of sampling), followed by the probability of each section being selected (second stage). The probability of selection at the first stage was the reciprocal of the number of schools in each stratum. The probability of selection at the second stage was the number of sections of each grade selected by principals, divided by the number of sections of each grade in each school (which was recorded when principals selected the sections). We checked the probability weights by comparing the population size estimated when applying the weights, with the numbers of children in grades 6–9 in each mandal recorded in state education department reports [18, 19].
We examined associations between travel mode and distance to school, stratified by school type. We used logistic regression to estimate odds ratios with 95 % confidence intervals for the association between walking and cycling and distance to school, adjusting for potential confounding factors (e.g. grade, gender, school type, independent mobility, physical activity). We used the ‘survey’ commands in Stata to account for stratification, clustering and unequal probability of selection, and the ‘test’ command to test the associations in the logistic regression models. We retained variables that remained statistically significant at the 5 % level in the ‘best fit’ model. We analysed data using STATA/SE V.12.0 (Stata Corporation, Texas, USA).