You are on page 1of 7

Matthew Benson

GIS for Health Data Analysis


Henry
12/14/17
Philadelphia, PA Bike Crash Incidence from 2011-2014
Introduction
The goal of this report is to identify statistically significant clusters of bike crashes (BC)
in the city of Philadelphia, Pennsylvania primarily in order to understand where BC are
happening most often and secondarily to find out if there are any areas where BC are
unexpectedly occurring with significant frequency and what the implications of such results may
be. Hot spot analysis was employed for the main goals stated above, and central feature functions
were used to aid the analysis. Street class by BC incidence will also be analyzed in the form of a
table as further supplementary analysis. Clusters in and around Center City are expected because
of high traffic volume and little room for bikes and cars to share the road.

Data and Methods


All of the data used in this report was gathered from OpenDataPhilly. The layers and
shapefiles used were a 2010 census tracts shapefile, a neighborhoods shapefile, vehicular crashes
point data from 2011-2014, and street centerlines data for Philadelphia.
I was unable to find bike lane data for Philadelphia, which would certainly aid the
analysis, as bike lanes are intended to make the road safer for cyclists. A question I would have
liked to have asked would have been, are bike lanes indeed making Philly streets safer for
bikers? And if they are, how successful are bike lanes in doing so? This is an important city
planning question for Philadelphia as there are probably thousands of bikers on the road
everyday in the city, and as may be gleaned from the results detailed below, many vehicular
crashes involving bikes do occur. According to the data, on average 1.8 BC happen in the city
everyday.
Hotspot analysis was used to identify statistically significant clusters of BC by tract (see
figure 1), which I felt was a more useful aggregation than neighborhood for this analysis because
tracts are larger aggregates and contain more crash points. The neighborhoods shapefile was then
used to identify neighborhoods that fell mostly within hotspots by manual selection. A central
feature function was used to reveal the spatial pattern of BC incidents and streets where BC
incidents occurred versus the spatial pattern of all vehicular crashes and all streets (see figure 2).
Lastly, a select by location function was used with a buffer of several feet (crash point data was
several feet askew from the centerline data) to select all segments that intersect with BC point
features (BC segments) and was displayed as a map (see figure 3).
Street class by segment was also analyzed in the form of a table (see table 1) which
displays: the number of BC segments there are in each class, the total number of BC segments,
the number of segments there are in each class in total in the city, the total number of segments
defined by the dataset in the city, the percent of total segments each class represents, and the
percent of BC segments each BC segment class represents. Differences for these last two fields
were then calculated and displayed.

Results
Statistically significant clusters of BC were observed in North Philadelphia
neighborhoods, South Philadelphia neighborhoods, West Philadelphia neighborhoods, the Kelly
Drive/ Martin Luther King Drive Fairmount Park area, and Center City and some of its
immediately surrounding areas. The Kelly Drive/ Martin Luther King Drive Fairmount Park
area, University City area, and the Mantua, Powelton, Passyunk Square, Queen Village,
Hawthorne, Bella Vista, Center City, Callowhill, and Logan Square neighborhood areas were
identified as 99% confidence level hot spots. The West and East Passyunk, Lower
Moyamensing, Center City, Hunting Park and Feltonville neighborhood areas were identified as
95% confidence level hotspots. And the University City, Rittenhouse, Northern Liberties, Old
Kensington, North Central, and Fairhill/Mcguire neighborhood areas were identified as 90%
confidence level hotspots.
The central feature for all vehicular crashes and all street centerline segments was
determined to be in almost exactly the same spot in central North Philadelphia; and for BC and
BC segments was determined to be in the Spring Garden/ Callowhill neighborhood, right on top
of Community College of Philadelphia/ JR Masterman High School, south and west of the
central features for all vehicular crashes and all centerline segments and just north of Center
City.
The results of table 1 report that all Philadelphia street segments total to 40,986. BC
segments total to 1,978. BC occured on 4.8% of total street segments, with a small margin of
error as some BC segments were unaccounted for an unknown reason. The percentage of total
segments for each class and the percentage of BC segment classes out of total BC segments are
distributed differently.
Classes that have high differences (given the null hypothesis - that the two columns
displaying the ratios detailed above should be the same) are: major arterial and minor arterial.
Classes that have low differences are: local and collector. I will not discuss the other 7 classes, as
the number of BC that occurred there and the incidence of these class streets in the first place in
the city are very small to none.
Classes displaying differences from highest to lowest as (sign-neutral magnitudes here)
are: local, major arterial, minor arterial, and lastly collector, which displayed only a relatively
small differences of 2.9. Local roads make up about 37% of segments in Philadelphia according
to the centerline data, but only 11.9% of BC segments, and have the highest differences of 25.1.
Major and minor arterial roads are similarly deviant, with differences of 14.8 and 14.7
respectively.
Conclusion
First, I will interpret the results of table 1. Differences are high enough for me to
comfortably state that it appears that major and minor arterial roads have particularly high rates
of BC incidence, and local roads seem to have the lowest rates proportional to the total numbers
of these streets, as well as collector roads, even though this class’ difference was low. This
makes sense, as arterial roads tend to have more traffic and higher speed limits than local roads.
Most clusters were observed in and around areas close to Center City, areas with high
traffic and street segments that are close together indicating high population density, which is not
a surprise. However this study proves that Center City and surrounding areas are indeed more
dangerous for cyclists than other areas of the city. The primary goal of this study has thus been
reached. The following will discuss the more interesting clusters observed.
The 95% confidence level Hunting Park/ Feltonville cluster is unusual because the street
segments there are more widely spaced apart, when clusters in most other areas occur where the
street segments are close together, as was previously stated indicating high population density.
An analysis of street class in this cluster among other factors may be necessary to understand
why this cluster appears here. It is also unusual geographically as it is the farthest cluster from
Center City.
The other interesting hotspot I will discuss is the Kelly Drive/ Martin Luther King Drive
Fairmount Park area 99% confidence level cluster. The drives, especially the Kelly Drive are
widely used for recreational and transportational cycling, as well as for many other types of
recreation. They serve as parks for the city, and are even closed off to traffic every Sunday for a
few hours for the purpose of bike riding and recreation. As an official bike riding hub for the
city, it would again make sense that there is a cluster of BC here, however the fact that the
cluster is so strong as to be 99% confidently a cluster may be evidence that this park area needs
to be safer for its users.
This report makes no claims as to whether the Kelly and Martin Luther King Drives need
to be safer, but the presence of a such a cluster in this area may be cause for further investigation.
Variables that would be useful in analyzing this cluster are: motor vehicle of all types-bike
collisions per ridership inside the cluster compared to outside the cluster, bike-pedestrian
collisions per ridership inside the cluster compared to outside the cluster, bike-bike collisions per
ridership inside the cluster compared to outside the cluster, total BC per ridership inside the
cluster compared to outside the cluster, and perhaps something like average age of the ridership
inside the cluster versus outside the cluster, among others; such as analysis of the state of repair
of the path, how much room there is from the path to the road and a safety index of this, and the
safety of car-bike crossings inside the cluster.
This study has looked into the complexities of bike safety analysis in Philadelphia and
serves as a window into potential research methods for further analysis, as the safety of
commuters and people enjoying recreation using bicycles is very important and sometimes a
matter of life and death for thousands of Philadelphians, and many millions of people around the
world.
Tables and Figures

Table 1
Street Class BC All Phila. Segment BC Segment Differences
By Segment Segments Segments Class/ Class/
2011-2014 Total Phila. Total BC
Segments Segments

Navy Yard 0 1 ~0% 0% ~0

Expressways 20 417 1% 1% 0

Major Arterial 448 3,193 7.8% 22.6% 14.8

Minor Arterial 572 5,816 14.2% 28.9% 14.7

Collector 690 15,486 37.8% 34.9% -2.9

Local 236 15,161 37% 11.9% -25.1

Driveway 0 78 0.02% 0% ~0

Low Speed Ramps 5 121 0.03% 0.003% ~0

High Speed 3 157 0.04% 0.002% ~0


Ramps

Non Travable 2 388 1% 0.001% ~1

City Boundary 2 156 0.04% 0.001% ~0

Walking 0 12 ~0% 0% ~0
Connector
Total 1,978 40,986

Figure 1
Figure 2
Figure 3

You might also like