4. Neighborhood effects
With the grouping analysis we looked at total overall accidents and total fatal accidents from 2007 to 2017 and grouped them based on the variables of road location (at intersections=1; at mid-blocks=0), if speeding was involved (yes=1; no=0), and whether the accidents happened during morning (yes=1; no=0) or evening rush hours (yes=1; no=0). We first joined each of the points of accidents by neighborhood with by table to better identify if there were areas that had specific characteristics of crashes and fatalities. Since each of the points of crashes were attributed to a neighborhood from the dataset, the joining process was applicable as this eliminated the issue of spatially joining the points to the neighborhoods where the road would be on boundary lines and points would be randomly joined to neighborhoods. Since joining the neighborhoods only shows one point. As a result, this eliminates instances where multiple accidents would happen at the same location. Therefore, we took the sum of the accidents as well as the sum of our binary variables and normalized them against the sum of the accidents to get a proportion of the input of these variables on a spectrum of high to low. For example, since all our variables were binary 1 was representative of the affirmative and 0 was representative of the negative, where taking the sum of the 1s would tell us what proportion of accidents in that location included the variable. The grouping would then take these proportions of the variables where the results would tell us on a spectrum of high to low the occurrence of the variables at locations, where a top result on the boxplot results would mean such variable is very apparent in accidents at those locations and a low result on the boxplot meant it was not apparent or of the opposite such as with road locations.
In the results of our grouping, we decided to categorize each of the groups based on the results indicating high levels from their boxplot results of each variable were apparent in the grouping to distinguish them (Figure 1, 2). These groups were as a result divided between neighbourhoods where accidents happened involving speeding, during morning rush hours, during evening rush hours, and neutral where none of the variables showed high or low results. The grouping colours of the summary of results in the table for both total overall accidents and total fatal accidents have been changed to reflect any similarities in grouping between the two so comparing colours between the boxplot results then no longer match up (Table 1, 2, Figure 1, 2). These different groupings were also then compared to the land use zoning of Toronto to see if there were any relationships between land use and type of group.
Table 1. Summary results of the grouping analysis and land use reference for total overall accidents in Toronto, Canada between 2007 and 2017
Table 2. Summary results of the grouping analysis and land use reference for total fatal accidents in Toronto, Canada between 2007 and 2017
Figure 1. Boxplot results of the grouping analysis of total overall accidents
Comparing within total overall accidents we see high incidents of speeding grouped with high incidents at mid-blocks, where the time of day are both medium or neutral for rush hours for accidents grouped in green (Table 1, Figure 3). These neighborhoods show clustering around the East and West peripheries and the center of the study area and form small groups of approximately three neighborhoods (Figure 3). In the yellow group, there is a high rate of incidents at intersections and also involving a medium-high level of speeding during morning rush hours at a high level (Table 1, Figure 3). Grouped in blue, it shows high levels of accidents happening during evening rush hours that also happen at intersections on a medium-high level (Table 1, Figure 3). These areas are spatially clustered as many neighborhoods are adjacent to others and their separate groups are also not that far from another (Figure 3). Most of these neighborhoods also cluster towards the middle areas of Toronto and are not shown along the Eastern and Western perimeters of Toronto (Figure 3). Grouped in red, these show areas that do not show any high levels of one variable but show a low level of speeding overall. Therefore, the red grouping show accidents that equally are variable between both road locations and both morning and evening rush hours, but also do not involve speeding (Table 1, Figure 3). These neighborhoods are grouped mostly in the middle to middle-West of Toronto and are more clustered and adjacent to each other (Figure 3).
Comparing within total fatal accidents grouped in green, there are high levels of speeding coinciding with high rates at mid-blocks at mostly equal variance between rush hour times but skewing towards evening rush hours (Table 2, Figure 8). These neighborhoods are predominantly on the East side and are more clustered in the East, gradually getting less clustered moving towards the West (Figure 8). Also in yellow, the grouping shows fatal accidents having high rates of incidents during morning rush hour, but with an equal medium variance between road locations (Table 2, Figure 8). For high rates of evening rush hours grouped in blue, there are also high rates of speeding involved and high occurrences at intersections (Table 2, Figure 8). These groupings are spatially distributed evenly across the study area where none of them are adjacent except one (Figure 8). In comparison to the number of groups, there are a lot less than shown for overall accidents. In the red group, also showing generally neutral and even rates between all variables but skewing to more incidents of speeding involvement and towards evening rush hours over morning rush hours (Table 3, Figure 8). These neighborhoods were quite spatially clustered as each neighborhood was adjacent to at least one other neighborhood, which spans across the study area (Figure 8). More so these neighborhoods were towards to West side and the Western periphery of Toronto, with gaps towards the Eastern periphery and the Northern Periphery.
Figure 3. Grouping analysis map of total fatal accidents of Toronto
Figure 8. Grouping analysis map of total fatal accidents of Toronto
Then we also compared between total overall accidents and total fatal accidents of these results to see if there were any changes in accidents in general and fatal accidents. With the group showing high levels of speeding and high incidents at mid-blocks, there was the least change between overall accidents and fatal accidents; except that for fatal accidents there was a slight skew towards accidents happening during evening rush hours at a medium-high level but rates of morning rush hours were still medium level (Table 1, 2). Where the fatal accidents are spatially located means that more on the East side there are fatal accidents involving speeding at mid-blocks than overall accidents which are distributed all over (Figure 5, 10). For areas grouped as mostly happening during morning rush hours, there was a shift where fatal accidents involved speeding less from a medium-high level to a medium-low level and the variance between road locations were both even at medium levels compared to high levels for intersections at an overall accident level (Table 1, 2). This shows that fatal accidents during morning rush hours are spatially different than overall accidents and speeding involvement is not as important (Table 2). This could be a result of how these groupings are spatially different since overall accidents were grouped more spatially distributed across the study area except for the East side but for fatal accidents, they are grouped more around the middle third of Toronto if divided vertically and are spatially more clustered around each other (Figure 6, 11). However, both still are located around the downtown areas and the type of land use these neighborhoods occupy are relatively similar of residential and commercial areas and open and industrial areas moving away from the downtown area (Figure 6, 11). For areas of high evening rush hour fatal accidents, the variability between road locations skewed more to the intersection from medium-high levels to high levels and additionally in fatal accidents coincided with high levels of speeding involvement where overall accidents show only medium-low levels (Table 1, 2). In cases of fatalities during evening rush hours, they are then more so related to intersections and drastically higher levels of speeding involvement. The similarities in grouping variables could be due to their similarities in spatial distribution as most of the neighborhoods for fatal accidents are represented in overall accidents (Figure 7, 12). The changes in the grouping of neutral or equally as apparent variables in the fatal accidents showed a large increase in speeding involvement from low to medium-high while the road locations became more evenly variable as both became medium and rush hour times became a bit more variable as morning rush hours decreased to medium-low and evening rush hours increased to medium-high (Table 1, 2). For this grouping, this was the only group that had more neighborhoods in looking at fatal accidents than overall accidents, meaning that when all variables are equally present there are specifically more fatal accidents than when there are just some of the variables. The inclusion of all the variables could be a reason that more neighborhoods were grouped into this category. Spatially, there is also a difference between neighborhoods in the less clustered and sparse areas between the fatal and overall accidents, which could be a reason why there were some changes in neighborhoods (Figure 8, 13). Overall, looking at the changes fatal accidents produced there was an increase in speeding involvement in both groupings of evening rush hours and the neutral grouping (Table 1, 2). Although there was a decrease of speeding in the morning rush hour group this had a lesser overall input as in comparison more neighborhoods were grouped to include higher involvement of speeding in fatal accidents.
In relation to land uses, there was little difference between the land use of overall accidents and fatal accidents. Most of the changes between the two were a reduction around institutional land uses in fatal accidents during evening rush hours, but this could be due to changes spatially in which neighborhoods were grouped (Table 2, Figure 12). The only biggest difference was with the speeding group as industrial areas were much less apparent in fatal accidents, but again this could be due to the fewer number of neighborhoods with such land uses (Figure5, 10). Also, the distribution of land uses was very much mixed and distributed within neighborhoods all around the study area. There were only a few neighborhoods that really represented predominant land uses and these were only located more around the Eastern and Western peripheries of the study area (Figure 5). Aside from these areas, it does not seem that the land uses have much influence on the severity of accidents or the variables of road location, speeding involvement, and rush hours on accidents either.