The study was conducted in Toronto, Canada. Two related analyses are presented, using collision data extracted from routinely collected Toronto Police Service pedestrian-motor vehicle collision reports from 2000–2011 (Fig. 1). Police-reported pedestrian-motor vehicle collisions include those resulting in no injury, minimal injury, minor injury (seen in the emergency department), major injury (admitted to hospital), and fatal injury. The location of the incident was provided as longitudinal and latitudinal coordinates. Each collision record included the date and time of the collision, and the age of the pedestrian(s) involved in the collisions.
The first analysis used a quasi-experimental design to determine the effects of newly implemented school crossing guards. The Toronto Police Service supplied details of all newly implemented crossing guards during the period 2003 – 2011. The information included the implementation date, the intersections at which all crossing guards were located and the school crossing guard duty time. When time was not specifically indicated, duty times were assumed to be the standard guard contracted times across the city; 08:00 – 09:00, 11:30 – 1:00 and 3:00 – 4:15 on weekdays only. Major school holidays across Toronto for the study period were determined from the Toronto District School Board website, and these were identified as non-duty times.
The analysis focused only on collisions from 2000–2011 which occurred at locations where new guards were implemented from 2003–2011 (Fig. 1). The outcome of interest was the collision count for all ages within 50 m of the newly implemented guard location. The independent variable was a dichotomous variable, designated as either pre or post guard implementation. Covariates included a dichotomous winter variable (October– March), and several built environment features measured at the school crossing guard location. Road type at the crossing location was provided by the City of Toronto, and was included as a categorical variable (Midblock location: collector road, local road, major road; Intersection location: local/local, major/major, major/local, collector/collector and collector/local). The age of the neighborhood was calculated based on the era of development from the 2006 Census; this binary variable was based on >50 % of neighborhood constructed before or after 1960 (0 = majority of neighborhood constructed after 1960, 1 = majority of neighborhood constructed prior to 1960). The demarcation between pre and post 1960 was used as it reflected the different characteristics of school neighbourhoods where crossing guards are located in Toronto. In the Canadian context, typical modern suburbs developed during the period of urban renewal that occurred in many cities in the 1960s. Most of the neighbourhoods constructed before 1960 are within the older central city core of Toronto. A land use mix variable was also constructed based on parcel level data from the Municipal Property Assessment Corporation (MPAC). Land use mix was calculated based on a 1000 m straight line buffer from the guards location (measured using an entropy index where scores of 0 = single land use, 1 = equal distribution of all land use classifications, i.e. residential, commercial, institutional, industrial, recreational/open space) [24, 25].
The second analysis involved a retrospective cohort formed from all child PMVCs, ages 4–12 from 2000–2011 throughout Toronto (Fig. 1). The analysis was conducted to assess where all school crossing guards were located (i.e. not just the ones where implementation dates were available) in relation to child PMVCs and the proportions of these collisions which occurred during school travel times. The Toronto Police Service supplied the location of all crossing guards in Toronto for the years 2010–11, which was assumed to be a consistent location over the collision study period. A bivariate variable was created indicating whether the collision occurred during school travel times as defined in the first analysis (08:00 – 09:00, 11:30 – 1:00 and 3:00 – 16:15 on weekdays only) and excluded major school holidays. Implementation dates were not available for all of these guards and therefore, could not be included in the quasi-experimental analysis However, these data provided some descriptive information regarding the location of child PMVCs specifically (i.e. near or far from locations identified as needing school crossing guards) and school travel times throughout the city.
Statistical analysis
The unit of analysis was the crossing guard location, with the location of guards and PMVCs mapped using ArcGIS (ESRI: ArcGIS, version 10). A straight line (Euclidean) distance buffer of 50 m was created around crossing guards to assess the proportion of PMVCs occurring within the vicinity of a crossing guard. The 50 m buffer was selected to locate the collision as close as possible to the crossing guard, while still maintaining a meaningful number of collisions and allowing for some minor collision location error. The recorded PMVC locations were superimposed and the collision counts, with reference to the 50 m buffers, were determined. Statistical analysis was conducted using SAS (SAS, version 9.3).
Analysis 1: Newly implemented school crossing guards (2003–2011)
The period at risk, i.e. number of months of exposure pre and post crossing guard implementation was determined, and the number of all PMVCs per month/school crossing guard with reference to the 50 m buffer, was calculated. Descriptive statistics were conducted of the PMVCs occurring within the 50 m buffer of the crossing guards. Repeated measures Poisson regression was used to model counts of PMVCs per school crossing guard month (subsequently referred to as SCG month) pre and post intervention, which controlled for winter and environmental features and accounted for repeated measures of PMVCs around each crossing guard location. Over-dispersion of the data was tested and was not found to be present in these data. The offset used in the regression analysis was calculated as time per school crossing guard-month, as pedestrian volume data was unavailable. Model-based standard error estimates were used, due to the small sample size [26]. Incident rate ratios ((IRR with 95 % confidence intervals (CI)) were calculated by exponentiating the beta coefficients from the regression models. The model was rerun for just collisions that occurred during times that school crossing guard duty time.
Analysis 2: All City of Toronto school crossing guards (2010–2011)
The proportions of PMVCs which occurred at locations where school crossing guards were implemented and during school travel times hours were calculated for children (4–12 years).
