- Email: [email protected]
A spatial and temporal analysis of child pedestrian crashes in Santiago, Chile
Accident Analysis and Prevention 50 (2013) 304–311
Contents lists available at SciVerse ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
A spatial and temporal analysis of child pedestrian crashes in Santiago, Chile Carola A. Blazquez ∗ , Marcela S. Celis Universidad Andres Bello, Department of Engineering Science, Sazie 2315, Piso 2, Santiago, Chile
a r t i c l e
i n f o
Article history: Received 20 January 2012 Received in revised form 19 April 2012 Accepted 1 May 2012 Keywords: Kernel Spatial autocorrelation Crashes Child pedestrian
a b s t r a c t This paper presents a spatial and temporal analysis of child pedestrian crash data in Santiago, Chile during the period 2000–2008. First, this study identified seven critical areas with high child pedestrian crash risk employing kernel density estimation, and subsequently, statistically significant clusters of the main attributes associated to these crashes in each critical area were determined in a geographic information systems environment. Moran’s I index test identified a positive spatial autocorrelation on crash contributing factors, time of day, straight road sections and intersections, and roads without traffic signs within the critical areas during the studied period, whereas a random spatial pattern was identified for crashes related to the age attribute. No statistical significance in the spatial relationship was obtained in child pedestrian crashes with respect to gender, weekday, and month of the year. The results from this research aid in determining the areas in which enhanced school-age child pedestrian safety is required by developing and implementing effective enforcement, educational, and engineering preventive measures. © 2012 Elsevier Ltd. All rights reserved.
1. Introduction The World Health Organization (WHO) and UNICEF published a report on child injury prevention in 2008, which statistically indicates that child death involved in road traffic crashes corresponds to 2% of total annual fatalities worldwide. In 2004, approximately 262,000 children were killed in pedestrian crashes representing 30% of the total road traffic crash casualties. Currently, crashes are ranked ninth globally among the leading causes of disability, and the ranking is projected to rise to third by 2020 (WHO, 2008). Furthermore, this report predicts that crashes will be the main cause of death in children under the age of 18 by 2030. Road traffic crashes among children are one of the most serious epidemic problems in developing countries due to rapid motorization and other factors (Nantulya and Reich, 2002). In Chile, traumatism derived from pedestrian vehicle crashes is the leading cause of death among children over one year of age. Although, statistics indicate that the total number of injured children has diminished in recent years, this value is still high when compared to developed countries producing significant social and economic consequences to the families, nation, and health organizations (Romero, 2007). On average, approximately 15% of the total number of crashes that occur in Chile each year are pedestrian crashes, and over 33% of these crashes involve children, and 40% of whom suffer serious and less serious injuries or death. Thus, there is a need for
∗ Corresponding author. Tel.: +56 2 661 8644; fax: +56 2 770 3076. E-mail address: [email protected] (C.A. Blazquez). 0001-4575/$ – see front matter © 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.aap.2012.05.001
a spatial and temporal analysis of child pedestrian crashes in Chile, and subsequent effective measures to prevent these crashes. Various studies have used Geographic Information Systems (GIS) as a tool for data management and spatial analysis (e.g., patterns and clusterization) of road traffic crashes, which some combine with statistical models to determine the relationship between crashes and causal factors (Levine et al., 1995; Ladron de Guevara et al., 2004; Ng et al., 2002; Flahaut et al., 2003; Meliker et al., 2004; Erdogan et al., 2008; Erdogan, 2009; Gundogdu, 2010; Anderson, 2009; Steenberghen et al., 2004). Additional studies have addressed the spatial problem of pedestrian crashes (Schneider et al., 2004; Kim and Yamashita, 2005; Truong and Somenahalli, 2011), and other researches have focused on pedestrian crashes that involve children. For example, some studies have employed GIS technology to locate zones with high child pedestrian severity and risk near schools (Banos and Huguenin-Richard, 2000; Austin et al., 1997; Miller, 2000; Clifton and Kreamer-Fults, 2007), or within high density residential areas (Lightstone et al., 2001; Dissanayake et al., 2009). The aforementioned studies focus on child pedestrian crashes that occurred in urban zones of Europe and North America. No previous research has analyzed pedestrian crashes (particularly with the involvement of children) both spatial and temporally, and that portrays the Chilean reality. Therefore, this paper presents, to the authors’ knowledge, the first results ever published for child pedestrian crashes in Chile using KDE and Moran’s I spatial autocorrelation statistics in a GIS environment over a period of time. The results of this study may be utilized as a supportive tool for decision making and appropriate allocation of resources
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
for safety enhancements in road infrastructure, and pedestrian education offered to children and parents to increase awareness and usage of warning signs and crossing signals. The first objective of this study is to identify critical areas with a high likelihood of child pedestrian crash occurrence using kernel density estimation (KDE) in a GIS environment. Many studies have employed the KDE technique to analyze road traffic crashes due to its simple implementation and easy understanding (Banos and Huguenin-Richard, 2000; Flahaut et al., 2003; Steenberghen et al., 2004; Schneider et al., 2004; Pulugurtha et al., 2007; Erdogan et al., 2008; Xie and Yan, 2008; Anderson, 2009; Kuo et al., 2011). The second objective of this study is to perform a spatial correlation test using Moran’s I index to assess whether the pattern of child pedestrian crashes had an average tendency to cluster in space and time within each of the identified critical areas, and whether a spatial dependence of these patterns exists relative to the main crash attributes (e.g., crash contributing factors, age, gender, time of day, month, road type, and traffic signs). The methodology presented in this paper was implemented with nine years of child pedestrian crash data for the city of Santiago, Chile. Santiago is the capital of Chile with over 900 km2 of extension and an estimated population of 6.5 million inhabitants in 2005, representing approximately 40% of the total population of Chile. 2. Methodology 2.1. Identification of critical areas KDE is a non-parametric method that involves introducing a symmetrical surface over each point feature, assessing the distance from the point to a reference location based on a mathematical function, and subsequently, adding the value of all the surfaces for that reference location (Levine, 2004). Eq. (1) defines KDE for a given set of observations with an unknown probability density function f: f (x) =
n 1 K(x − xi )
nh
i=1
h
(1)
where xi is the value of the variable X at the location i, n is the total number of locations, h is the bandwidth or smoothing parameter, and K is the kernel function. This research employed a normal distribution as a kernel function that weighs all points in the study area with near points having more weight than distant points. The outcome of the KDE depends significantly on the bandwidth and the cell size. According to Steenberghen et al. (2004), the cell size that provides the best concentration patterns of crashes in a dense urban network (such as Santiago) is smaller than the road segments between intersections and large enough to identify crash clustering. Therefore, after a number of tests, a bandwidth or search radius of 1000 m and cell size of 100 m by 100 m was selected for the crash density analysis for the studied period. 2.2. Spatial autocorrelation Subsequent to the identification of spatial clusters with high child pedestrian crashes over time, a spatial autocorrelation test was performed to determine spatial dependence between the main attributes of these crashes and to detect statistically significant clusters with respect to each of these attributes. If the level of correlation is higher than expected, then neighboring locations have similar values and the spatial autocorrelation is positive. When the level of correlation is lower than expected, high values of the variable are contiguous to low values and the autocorrelation is negative. Thus, the spatial patterns of crash data take into account
305
simultaneously crash locations and their attribute values by measuring for attribute similarity and location proximity into one single index named Moran’s I (Truong and Somenahalli, 2011). This index is formally expressed by Eq. (2): I=
N
N N i=1
w (x − x¯ )(xj / i ij i j=1,j = N S0 i=1 (xi − x¯ )2
− x¯ )
∀i = 1, ..., n ∧ ∀ j = 1, ..., n (2)
where wij are the elements of a spatial binary contiguity matrix with weights representing proximity relationships between location i and neighboring location j, S0 is the summation of all elements wij , xi is the variable value at a particular location i, xj is the variable value at a another location (i = / j), x¯ is the mean of the variable, and n is the total number of locations. The values of Moran’s I index range from −1 to 1, where the former represents a strong negative autocorrelation (i.e., perfect dispersion or clusterization of dissimilar values) and the latter represents a strong positive autocorrelation (i.e., perfect concentration or clusterization of similar values). A value of Moran’s I near zero indicates a spatially random pattern. The results from the spatial autocorrelation are always interpreted within the context of its null hypothesis, which states that the attribute being analyzed is randomly distributed among the features in the study area. The statistical significance of the Moran’s I index is computed with the Z score method assuming a standard normal distribution with mean equal to zero and a variance of one. A positive Z score for a point indicates that the neighboring features have similar values, whereas a negative Z score denotes that the feature is surrounded by dissimilar values (Getis and Ord, 1994). 3. Data description This study employed the data provided by the Chilean National Road Safety Commission (CONASET), which maintains a database with all road traffic crashes that occur in the country. Of 6078 child pedestrian crashes that arose in Santiago between 2000 and 2008, 5152 (84.8%) were successfully geocoded in a GIS environment, as shown in Fig. 1. This research focuses on pedestrian crashes that involve children between the ages of 5 and 18 and that normally attend school. These school-age children are from now on referred to as “children”. Interactive manual intervention was required to match the crash address information with the spatial database. The remaining percentage of crash locations was not geocoded due to non-existent or incomplete addresses. Currently, police officers fill out a paper form at the scene to report crashes with their main characteristics (e.g., time, date, address, weather, road condition, etc.). Subsequently, this information is manually entered to a computer. Hence, the reported crash data is prone to yield discrepancies in both manual data entry processes. CONASET classifies road crashes into 42 main contributing factors. This study employed 14 of these factors, which account for 96% of all pedestrian crashes. Table 1 presents the percentages according to injury type for each of the 14 factors, of which 67.2% was the responsibility of the pedestrian, 26.3% was the driver’s responsibility, and 6.6% were undetermined causes. Approximately 75% of all child pedestrian crashes were produced by the following factors: Pedestrian crosses road surprisingly or carelessly; imprudence of pedestrian; and driver violates crosswalk. In terms of injury type, over 60% of all children involved in a pedestrian crash during the studied period suffered slight injuries, while only 0.73% of the children were unharmed. In other words, 99.27% of all child pedestrian crashes yielded some degree of physical injury. In addition, 55.8% of child pedestrian crashes that occurred in Santiago during the studied period were boys and 58.1% of all child pedestrian crashes
306
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
Fig. 1. Child pedestrian crashes that occurred in Santiago between 2000 and 2008.
Table 1 Percentage of injuries for each contributing factor to pedestrian–vehicle crashes. Pedestrian crash contributing factor
% Unharmed
% Slightly injured
% Less seriously injured
% Seriously injured
% Fatalities
Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Total
0.54 1.40 0.96 0.00 0.00 0.00
57.16 57.87 53.99 58.27 53.93 60.47
15.00 11.66 13.74 14.39 13.48 16.28
26.66 28.09 27.80 25.18 26.97 16.28
0.64 0.98 3.51 2.16 5.62 6.98
41.01 14.36 6.31 2.80 1.80 0.87 67.16
Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Total
0.72 1.35 0.00 1.69 0.00 0.00 2.44
66.50 62.16 75.41 66.10 64.58 73.91 48.78
12.23 13.51 4.92 15.25 16.67 6.52 17.07
18.40 20.27 18.03 15.25 18.75 19.57 26.83
2.16 2.70 1.64 1.69 0.00 0.00 4.88
19.63 1.49 1.23 1.19 0.97 0.93 0.83 26.27
0.61 0.73
60.12 59.65
18.10 13.88
19.33 24.23
1.84 1.51
6.58 100.00
Other Undetermined causes Total
% Total
307
600 500 400 300 200 100
11:00 PM
9:00 PM
10:00 PM
7:00 PM
8:00 PM
5:00 PM
6:00 PM
3:00 PM
4:00 PM
2:00 PM
1:00 PM
12:00 PM
11:00 AM
9:00 AM
10:00 AM
7:00 AM
8:00 AM
6:00 AM
5:00 AM
4:00 AM
3:00 AM
2:00 AM
1:00 AM
0 12:00 AM
Average Number of Child Pedestrian Crashes
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
Time of Day
Fig. 2. Average number of child pedestrian crashes per hour between 2000 and 2008.
Table 2 Percentage of child pedestrian crashes for each contributing factor by road type.
concentrated children between 12 and 18 years of age with the highest number of fatalities at the average age of 12 years old. With respect to the time of the day, the average number of pedestrian crashes increased drastically after 7 am coinciding with the beginning of the school day (see Fig. 2). Approximately 3000 pedestrian crashes occurred between 1 p.m. and 8 p.m. corresponding to 57% of the total daily child pedestrian crashes. This time period includes the end of the school day for both morning and afternoon school shifts when most children return home. Fig. 3 illustrates that most of the child pedestrian crashes occurred on weekdays, particularly Wednesdays and Fridays, which accounted for 32.8% of the crashes during the studied period. The academic year (i.e., March through December) was the worst for child pedestrian crashes, while a lower number of these crashes were perceived during the summer time (i.e., January and February), as presented in Fig. 4. The largest quantity of crashes (51.8%) occurred on straight road sections (i.e., between intersections) during the period 2000–2008, as presented in Table 2, where 24.7% were caused by the pedestrian crossing the road in a surprisingly or carelessly manner, and 7% by the driver violating crosswalks. Conversely, this table shows that approximately 48% of the pedestrian crashes occurred at intersections with the pedestrian crossing roads surprisingly or carelessly being the major contributing factor to pedestrian crashes. Table 3 indicates that approximately 67% of child pedestrian crashes occurred on road segments with traffic signs. The pedestrian’s responsibility was the main contributing factor of these crashes by disobeying traffic sign laws or violating crosswalk locations. The driver was responsible for nearly 10% of the total number
Contributing factor Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Other Undetermined causes Total
Average Number of Child Pedestrian Crashes
900 800 700 600 500 400 300 200 100 0 Tuesday
Intersections
24.7
17.2
7.1 4.8 1.9 0.1 0.5
7.5 1.7 1.0 1.8 0.4
7.0 1.1 0.7 0.9 0.6 0.1 0.5
13.0 0.5 0.6 0.3 0.4 0.8 0.3
1.8
2.7
51.8
48.2
of pedestrian. These results suggest that pedestrians and/or drivers traveling in a distracted or careless manner are more likely to cause crashes at locations with traffic signs. Whereas, the occurrence of pedestrian crashes on road sections without traffic signs is less frequent since pedestrians cross roads with more care knowing the absence of traffic signs.
1000
Monday
Straight road sections
Wednesday
Thursday
Friday
Saturday
Sunday
Day of the Week
Fig. 3. Average number of child pedestrian crashes per day of the week between 2000 and 2008.
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311 600 500 400 300 200 100
r
r
be ec em D
em
be
er ov N
r be
O ct ob
t em Se
pt
ly
us Au g
Ju
ne Ju
ay M
ril Ap
M
ua r br
ar Fe
nu Ja
ar ch
y
0
y
Average Number of Child Pedestrian Crashes
308
Month
Fig. 4. Average number of child pedestrian crashes per month between 2000 and 2008.
4. Results and discussion Varying number of critical areas (ranging from 1 to 14) with high child pedestrian crash risk was identified using KDE for each year of the period 2000–2008. As a result from a yearly comparison analysis, critical areas 1, 2, 4, 5, 6, 7, and 9 (depicted as ellipses) were selected, as shown in Fig. 5, since they maintained high crash concentrations and frequencies for five or more years. These critical areas have density levels between 8.7 and 26.3 child pedestrian crashes per square kilometer. The selected critical areas cover large segments of major roads with high traffic flow that traverse more than one district such as critical areas 1, 4, and 5 (see Table 4). In addition, all critical areas are densely populated with mixed residential/commercial land use. Particularly, critical area 1 has perceived an increased residential population in recent years, and concentrates major educational, political, administrative, and financial activities of the city. Table 4 shows that critical areas 6 and 7 present a large number of child pedestrian crashes between 2000 and 2008, which correspond to high child population zones located in the districts of Puente Alto and Maipu, respectively. Children from both of these districts with lower middle socioeconomic level commute alone to
Table 3 Percentage of child pedestrian crashes on road segments with and without traffic signs for each contributing factor. Contributing factor Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Other Undetermined causes Total
With traffic signs
Without traffic signs
18.8
16.9
10.4 2.8 1.5 3.7 0.8
6.3 0.7 0.6 0.0 0.1
23.0 0.5 0.9 0.5 0.4 1.7 0.3
4.2 0.5 0.3 0.2 0.3 0.0 0.3
1.9
2.4
67.2
32.8
and from school using public transportation, and therefore, they are more prone to be exposed to pedestrian crashes. When contrasting the identified critical areas with the school density throughout the city, critical areas 1 and 7 situated in the districts of Santiago, Estacion Central, and Maipu overlap the highest school concentration with more than 11 schools per square kilometer. Further analysis in each of these areas is needed, in order to determine effective and strategic measures at school entrances and adjacent roads to prevent child pedestrian crashes. These measures may include improving intersection crossing times, increasing child pedestrian visibility, and enhancing police enforcement of speed laws and obedience to traffic signs (Braddock et al., 1994). Moran’s I spatial autocorrelation statistics was utilized to determine statistically significant clustering patterns of child pedestrian crashes within selected critical areas based on location proximity and their attribute similarities. In other words, a spatial and temporal analysis determined whether the main crash variables present a cluster, dispersion, or random distribution within high risk zones for child pedestrians. Moran’s I index was computed multiple times with different increasing distance thresholds until a distance of 80 m was obtained that maximized the Z score with a value of 60. Therefore, this distance is appropriate for maximizing the spatial autocorrelation (i.e., child pedestrian crashes are clustered up to a threshold distance of 80 m with a statistical significance of 0.01). In addition, Moran’s index was calculated using Euclidian distances and a 95% confidence coefficient under the assumption of a normal distribution. The first results of this analysis indicated that the attributes of child pedestrian crashes related to particular cohorts gender, weekday, and month of the year were not statistically significant within the identified critical areas. The spatial autocorrelation analysis in the selected critical areas showed an average positive autocorrelation for the crash contributing factors listed in Table 5. The four crash contributing factors presented in this table have a spatial dependence and tend to cluster in these areas. These factors Table 4 Districts within each identified critical area. Critical area
Districts
Total number of crashes 2000–2008
1 2 4 5 6 7 9
Santiago, Estacion Central Lo Prado San Ramon, La Cisterna El Bosque, San Bernardo Puente Alto Maipu ˜ Penalolen
589 162 238 389 414 431 221
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
309
Fig. 5. Selected critical areas with high child pedestrian crash risk in Santiago.
represent 81.31% of all the crash contributing factors and coincide with the major causes of child pedestrian crashes that occurred during the studied period. The remaining crash contributing factors presented randomness within the critical areas. Moran’s I coefficients were computed for children involved in crashes with an age range between 5 and 18. These coefficients yielded an average Moran’s I value of 0.009 and standard deviation of 0.047 with a Z score of 1.131, suggesting that no spatial autocorrelation exists between this variable and child pedestrian crashes within the critical areas. Therefore, there is no evidence of
dispersion or clustering of pedestrian crashes with respect to the children’s age. In terms of time of the day, there is a clear spatial pattern of pedestrian crashes reflected by the fact that the spatial autocorrelation tests resulted to be significant, as presented in Table 6. This table shows coefficient values solely for the years that the critical areas were identified as a result from the KDE analysis. The Table 6 Average yearly Moran’s I coefficients for each critical area with respect to the time of the day. Year
Table 5 Spatial autocorrelation analysis results per crash contributing factor. Factor
Average Moran’s I index
Standard deviation
Z-score
Pedestrian crosses surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Driver violates crosswalk
0.262
0.080
2.662*
0.253 0.233 0.248
0.022 0.027 0.047
2.651* 2.608* 2.600*
*
p < 0.005.
2000 2001 2002 2003 2004 2005 2006 2007 2008
Critical areas 1
2
4
5
6
7
9
0.275 0.287 0.278 0.274 0.270 0.266 0.307 0.258 0.254
0.290 0.365
0.268 0.306 0.244
0.315 0.267 0.284 0.266
0.321 0.290
0.248 0.274 0.321 0.287
0.307 0.262
0.254 0.266 0.248
0.310 0.274 0.338 0.277 0.266
0.287 0.278 0.274 0.287 0.288 0.241
0.306
0.278
0.295 0.286
0.304
310
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
average standardized Z score value for these coefficient values is 2.6 (p < 0.004). Therefore, there is a strong dependence between the time of day and the child pedestrian crashes in these areas over the study period. The spatial autocorrelation results for the road type attribute show a positive autocorrelation with an average Moran’s I value of 0.282 and standard deviation of 0.062 (Z = 2.762 and p < 0.003) on straight road sections in critical areas 4, 5 and 6, whereas intersections present a random spatial pattern in these areas. These critical areas comprise long segments of road between crosswalks with high traffic flow traveling at high speeds. Hence, pedestrians have a tendency to cross at any location, which exposes them to a higher pedestrian crash risk. An average positive spatial concentration of pedestrian crashes was obtained at intersections in critical areas 1, 2, 7 and 9, while crashes on straight sections for these areas were randomly distributed during the nine years of the study. In contrast to critical areas 4, 5, and 6, these areas are situated in urban areas with a large density of intersections. Thus, pedestrians are more likely to cross roads at intersections, and crashes may be caused due to the driver’s imprudence and the carelessness of pedestrians. Further analysis is required to examine in more detail the spatial relationship between road type and child pedestrian crashes in these critical areas. Finally, a positive autocorrelation is present for roads without traffic signs in each critical area with an average Moran’s I index value of 0.247 and standard deviation of 0.102 (Z = 2.610 and p < 0.004) for the studied period. Hence, the pattern of crashes that occurred on these road segments has an average tendency to cluster in space and time. On the other hand, road segments with traffic signs in all seven critical areas contained yielded an average index of 0.019 and standard deviation of 0.039 (Z = 1.183 and p < 0.005), which implies an absence of spatial autocorrelation between this variable and child pedestrian crash locations. Hence, the generation of pedestrian crash on road segments with traffic signs is randomly distributed. Additional investigation is needed to identify the sources of child pedestrian crashes that occurred on road sections with no traffic signs.
5. Conclusions This study identified seven critical areas with high child pedestrian crash risk in the city of Santiago, Chile using kernel density estimation (KDE) in a GIS environment. These areas are located in districts with lower middle socioeconomic status population. Therefore, they have low investment capabilities on traffic safety (e.g., traffic signs, speed bumps, crosswalks, etc.) exposing children that travel alone to and from schools to a higher pedestrian crash risk. The results of this study determined the main attributes of child pedestrian crashes (e.g., crash contributing factors, age, time of day, road type, and traffic signs) that present a statistical significance in the identified critical areas. Moran’s I spatial autocorrelation analysis indicates that the responsibility of pedestrians is the major contributing factor for the generation of child pedestrian crashes with a tendency to cluster in space and time. Similarly, there is a spatial clustering distribution of crashes in terms of time of the day. This analysis also suggests an absence of spatial autocorrelation between the age of the victims and child pedestrian crashes in the selected critical areas. Thus, there is no dependence of crashes due to the age attribute. A positive spatial autocorrelation of pedestrian crash on roads without traffic signs imply the occurrence of these crashes on neighboring roads without traffic signs. Further analysis is needed to explore the spatial and temporal relationship of child pedestrian crashes with respect to the road type variable.
The results of this research contributed by analyzing the crash reported data and determining critical areas where enhanced pedestrian safety is needed by developing and implementing effective enforcement, educational, and engineering strategies. Children must be taught to be alert and have greater caution where pedestrian crashes are more likely to occur, particularly in areas with high volume traffic and large quantity of parked vehicles. In addition, children that travel unsupervised to and from school should be taught how to cross roads and safely use public transportation. Future research includes the spatial and temporal analysis of child pedestrian crashes taking into account other variables such as land use (in particular, proximity to schools), socioeconomic level, population distribution, and trip purpose. Acknowledgements Financial support from the Universidad Andres Bello Project No. DI 97 -12/R is acknowledged. Anonymous review of this paper is also gratefully thanked. References Anderson, T., 2009. Kernel density estimation and K-means clustering to profile road accident hotspots. Accident Analysis and Prevention 41, 359–364. Austin, K., Tight, M., Kirby, H., 1997. The use of geographical information systems to enhance road safety analysis. Transportation Planning and Technology 20 (3), 249–266. Banos, A., Huguenin-Richard, F., 2000. Spatial distribution of road accidents in the vicinity of point sources application to child pedestrian accidents. In: Geography and Medicine. Elsevier, pp. 54–64. Braddock, M., Lapidus, G., Cromley, E., Cromley, R., Burke, G., Banco, L., 1994. Using a geographic information system to understand child pedestrian injury. American Journal of Public Health 84 (7), 1158–1161. Clifton, K., Kreamer-Fults, K., 2007. An examination of the environmental attributes associated with pedestrian–vehicular crashes near public schools. Accident Analysis and Prevention 39, 708–715. Dissanayake, D., Aryaija, J., Wedagama, D.M., 2009. Modelling the effects of land use and temporal factors on child pedestrian casualties. Accident Analysis and Prevention 41, 1016–1024. Erdogan, S., 2009. Explorative spatial analysis of traffic accident statistics and road mortality among the provinces of Turkey. Journal of Safety Research 40, 341–351. Erdogan, S., Yilmaz, I., Baybura, T., Gullu, M., 2008. Geographical information systems aided traffic accident analysis system case study: city of Afyonkarahisar. Accident Analysis and Prevention 40, 174–181. Flahaut, B., Mouchart, M., San Martin, E., Thomas, I., 2003. The local spatial autocorrelation and the kernel method for identifying black zones: a comparative approach. Accident Analysis and Prevention 35, 991–1004. Getis, A., Ord, J., 1994. The analysis of spatial association by use of distance statistics. Geographical Analysis 24 (3), 189–206. Gundogdu, I., 2010. Applying linear analysis methods to GIS-supported procedures for preventing traffic accidents: case study of Konya. Safety Science 48, 763–769. Kim, K., Yamashita, E., 2005. Using K-means clustering algorithm to examine patterns of pedestrian involved crashes in Honolulu, Hawaii. Journal of Advanced Transportation 41 (1), 69–89. Kuo, P.-F., Lord, D., Walden, T., 2011. Using geographical information systems to effectively organize police patrol routes by grouping hot spots of crash and crime data. In: Proceedings of the 3rd International Conference on Road Safety and Simulation, Indianapolis. Ladron de Guevara, F., Washington, S., Oh, J., 2004. Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arisona. Transportation Research Board 1897, 191–199. Levine, N., Kim, K., Nitz, L., 1995. Spatial analysis of Honolulu motor vehicle crashes. I. Spatial patterns. Accident Analysis and Prevention 27 (5), 663–674. Levine, N., 2004. CrimeStat III: A Spatial Statistics Program for the Analysis of Crime Incident Locations. Chapter 8: Kernel Density Interpolation. Ned Levine & Associates/The National Institute of Justice, Houston, TX/Washington, DC. Lightstone, A., Dhillon, P., Peek-Asa, C., Kraus, J., 2001. A geographic analysis of motor vehicle collisions with child pedestrian in Long Beach, California: comparing intersection and midblock incident locations. Injury Prevention 7, 155–160. Meliker, J., Maio, R., Zimmerman, M., Kim, H., Smith, S., Wilson, M., 2004. Spatial analysis of alcohol-related motor vehicle crash injuries in southeastern Michigan. Accident Analysis and Prevention 36, 1129–1135. Miller, J., 2000. Geographic information systems: unique analytic capabilities for the traffic safety community. Transportation Research Record 1734, 21–28. Nantulya, V., Reich, M., 2002. The neglected epidemic: road traffic injuries in developing countries. British Medical Journal 324 (7346), 1139–1141. Ng, K.-S., Hung, W.-T., Wong, W.-G., 2002. An algorithm for assessing the risk of traffic accident. Journal of Safety Research 33, 387–410.
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311 Pulugurtha, S., Krishnakumar, V., Nambisan, S., 2007. New methods to identify and rank high pedestrian crash zones: an illustration. Accident Analysis and Prevention 39, 800–811. Romero, P., 2007. Accidentes en la infancia: Su prevención tarea prioritaria en este milenio. Revista Chilena de Pediatría 78, 57–73. Schneider, R., Ryznar, R., Khattak, A., 2004. An accident waiting to happen: a spatial approach to proactive pedestrian planning. Accident Analysis and Prevention 36, 193–211. Steenberghen, T., Dufays, T., Thomas, I., Flahaut, B., 2004. Intra-urban location and clustering of road accidents using GIS: a Belgian example. International Journal Geographical Information Science 18 (2), 169–181.
311
Truong, L., Somenahalli, S., 2011. Using GIS to identify pedestrian–vehicle crash hot spots and unsafe bus stops. Journal of Public Transportation 14 (1), 99–114. Xie, Z., Yan, J., 2008. Kernel density estimation of traffic accidents in a network space. Computers, Environment and Urban Systems 32, 396–406. World Health Organization, 2008. Accidents and injuries. Children’s environmental health, http://www.who.int.
Contents lists available at SciVerse ScienceDirect
Accident Analysis and Prevention journal homepage: www.elsevier.com/locate/aap
A spatial and temporal analysis of child pedestrian crashes in Santiago, Chile Carola A. Blazquez ∗ , Marcela S. Celis Universidad Andres Bello, Department of Engineering Science, Sazie 2315, Piso 2, Santiago, Chile
a r t i c l e
i n f o
Article history: Received 20 January 2012 Received in revised form 19 April 2012 Accepted 1 May 2012 Keywords: Kernel Spatial autocorrelation Crashes Child pedestrian
a b s t r a c t This paper presents a spatial and temporal analysis of child pedestrian crash data in Santiago, Chile during the period 2000–2008. First, this study identified seven critical areas with high child pedestrian crash risk employing kernel density estimation, and subsequently, statistically significant clusters of the main attributes associated to these crashes in each critical area were determined in a geographic information systems environment. Moran’s I index test identified a positive spatial autocorrelation on crash contributing factors, time of day, straight road sections and intersections, and roads without traffic signs within the critical areas during the studied period, whereas a random spatial pattern was identified for crashes related to the age attribute. No statistical significance in the spatial relationship was obtained in child pedestrian crashes with respect to gender, weekday, and month of the year. The results from this research aid in determining the areas in which enhanced school-age child pedestrian safety is required by developing and implementing effective enforcement, educational, and engineering preventive measures. © 2012 Elsevier Ltd. All rights reserved.
1. Introduction The World Health Organization (WHO) and UNICEF published a report on child injury prevention in 2008, which statistically indicates that child death involved in road traffic crashes corresponds to 2% of total annual fatalities worldwide. In 2004, approximately 262,000 children were killed in pedestrian crashes representing 30% of the total road traffic crash casualties. Currently, crashes are ranked ninth globally among the leading causes of disability, and the ranking is projected to rise to third by 2020 (WHO, 2008). Furthermore, this report predicts that crashes will be the main cause of death in children under the age of 18 by 2030. Road traffic crashes among children are one of the most serious epidemic problems in developing countries due to rapid motorization and other factors (Nantulya and Reich, 2002). In Chile, traumatism derived from pedestrian vehicle crashes is the leading cause of death among children over one year of age. Although, statistics indicate that the total number of injured children has diminished in recent years, this value is still high when compared to developed countries producing significant social and economic consequences to the families, nation, and health organizations (Romero, 2007). On average, approximately 15% of the total number of crashes that occur in Chile each year are pedestrian crashes, and over 33% of these crashes involve children, and 40% of whom suffer serious and less serious injuries or death. Thus, there is a need for
∗ Corresponding author. Tel.: +56 2 661 8644; fax: +56 2 770 3076. E-mail address: [email protected] (C.A. Blazquez). 0001-4575/$ – see front matter © 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.aap.2012.05.001
a spatial and temporal analysis of child pedestrian crashes in Chile, and subsequent effective measures to prevent these crashes. Various studies have used Geographic Information Systems (GIS) as a tool for data management and spatial analysis (e.g., patterns and clusterization) of road traffic crashes, which some combine with statistical models to determine the relationship between crashes and causal factors (Levine et al., 1995; Ladron de Guevara et al., 2004; Ng et al., 2002; Flahaut et al., 2003; Meliker et al., 2004; Erdogan et al., 2008; Erdogan, 2009; Gundogdu, 2010; Anderson, 2009; Steenberghen et al., 2004). Additional studies have addressed the spatial problem of pedestrian crashes (Schneider et al., 2004; Kim and Yamashita, 2005; Truong and Somenahalli, 2011), and other researches have focused on pedestrian crashes that involve children. For example, some studies have employed GIS technology to locate zones with high child pedestrian severity and risk near schools (Banos and Huguenin-Richard, 2000; Austin et al., 1997; Miller, 2000; Clifton and Kreamer-Fults, 2007), or within high density residential areas (Lightstone et al., 2001; Dissanayake et al., 2009). The aforementioned studies focus on child pedestrian crashes that occurred in urban zones of Europe and North America. No previous research has analyzed pedestrian crashes (particularly with the involvement of children) both spatial and temporally, and that portrays the Chilean reality. Therefore, this paper presents, to the authors’ knowledge, the first results ever published for child pedestrian crashes in Chile using KDE and Moran’s I spatial autocorrelation statistics in a GIS environment over a period of time. The results of this study may be utilized as a supportive tool for decision making and appropriate allocation of resources
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
for safety enhancements in road infrastructure, and pedestrian education offered to children and parents to increase awareness and usage of warning signs and crossing signals. The first objective of this study is to identify critical areas with a high likelihood of child pedestrian crash occurrence using kernel density estimation (KDE) in a GIS environment. Many studies have employed the KDE technique to analyze road traffic crashes due to its simple implementation and easy understanding (Banos and Huguenin-Richard, 2000; Flahaut et al., 2003; Steenberghen et al., 2004; Schneider et al., 2004; Pulugurtha et al., 2007; Erdogan et al., 2008; Xie and Yan, 2008; Anderson, 2009; Kuo et al., 2011). The second objective of this study is to perform a spatial correlation test using Moran’s I index to assess whether the pattern of child pedestrian crashes had an average tendency to cluster in space and time within each of the identified critical areas, and whether a spatial dependence of these patterns exists relative to the main crash attributes (e.g., crash contributing factors, age, gender, time of day, month, road type, and traffic signs). The methodology presented in this paper was implemented with nine years of child pedestrian crash data for the city of Santiago, Chile. Santiago is the capital of Chile with over 900 km2 of extension and an estimated population of 6.5 million inhabitants in 2005, representing approximately 40% of the total population of Chile. 2. Methodology 2.1. Identification of critical areas KDE is a non-parametric method that involves introducing a symmetrical surface over each point feature, assessing the distance from the point to a reference location based on a mathematical function, and subsequently, adding the value of all the surfaces for that reference location (Levine, 2004). Eq. (1) defines KDE for a given set of observations with an unknown probability density function f: f (x) =
n 1 K(x − xi )
nh
i=1
h
(1)
where xi is the value of the variable X at the location i, n is the total number of locations, h is the bandwidth or smoothing parameter, and K is the kernel function. This research employed a normal distribution as a kernel function that weighs all points in the study area with near points having more weight than distant points. The outcome of the KDE depends significantly on the bandwidth and the cell size. According to Steenberghen et al. (2004), the cell size that provides the best concentration patterns of crashes in a dense urban network (such as Santiago) is smaller than the road segments between intersections and large enough to identify crash clustering. Therefore, after a number of tests, a bandwidth or search radius of 1000 m and cell size of 100 m by 100 m was selected for the crash density analysis for the studied period. 2.2. Spatial autocorrelation Subsequent to the identification of spatial clusters with high child pedestrian crashes over time, a spatial autocorrelation test was performed to determine spatial dependence between the main attributes of these crashes and to detect statistically significant clusters with respect to each of these attributes. If the level of correlation is higher than expected, then neighboring locations have similar values and the spatial autocorrelation is positive. When the level of correlation is lower than expected, high values of the variable are contiguous to low values and the autocorrelation is negative. Thus, the spatial patterns of crash data take into account
305
simultaneously crash locations and their attribute values by measuring for attribute similarity and location proximity into one single index named Moran’s I (Truong and Somenahalli, 2011). This index is formally expressed by Eq. (2): I=
N
N N i=1
w (x − x¯ )(xj / i ij i j=1,j = N S0 i=1 (xi − x¯ )2
− x¯ )
∀i = 1, ..., n ∧ ∀ j = 1, ..., n (2)
where wij are the elements of a spatial binary contiguity matrix with weights representing proximity relationships between location i and neighboring location j, S0 is the summation of all elements wij , xi is the variable value at a particular location i, xj is the variable value at a another location (i = / j), x¯ is the mean of the variable, and n is the total number of locations. The values of Moran’s I index range from −1 to 1, where the former represents a strong negative autocorrelation (i.e., perfect dispersion or clusterization of dissimilar values) and the latter represents a strong positive autocorrelation (i.e., perfect concentration or clusterization of similar values). A value of Moran’s I near zero indicates a spatially random pattern. The results from the spatial autocorrelation are always interpreted within the context of its null hypothesis, which states that the attribute being analyzed is randomly distributed among the features in the study area. The statistical significance of the Moran’s I index is computed with the Z score method assuming a standard normal distribution with mean equal to zero and a variance of one. A positive Z score for a point indicates that the neighboring features have similar values, whereas a negative Z score denotes that the feature is surrounded by dissimilar values (Getis and Ord, 1994). 3. Data description This study employed the data provided by the Chilean National Road Safety Commission (CONASET), which maintains a database with all road traffic crashes that occur in the country. Of 6078 child pedestrian crashes that arose in Santiago between 2000 and 2008, 5152 (84.8%) were successfully geocoded in a GIS environment, as shown in Fig. 1. This research focuses on pedestrian crashes that involve children between the ages of 5 and 18 and that normally attend school. These school-age children are from now on referred to as “children”. Interactive manual intervention was required to match the crash address information with the spatial database. The remaining percentage of crash locations was not geocoded due to non-existent or incomplete addresses. Currently, police officers fill out a paper form at the scene to report crashes with their main characteristics (e.g., time, date, address, weather, road condition, etc.). Subsequently, this information is manually entered to a computer. Hence, the reported crash data is prone to yield discrepancies in both manual data entry processes. CONASET classifies road crashes into 42 main contributing factors. This study employed 14 of these factors, which account for 96% of all pedestrian crashes. Table 1 presents the percentages according to injury type for each of the 14 factors, of which 67.2% was the responsibility of the pedestrian, 26.3% was the driver’s responsibility, and 6.6% were undetermined causes. Approximately 75% of all child pedestrian crashes were produced by the following factors: Pedestrian crosses road surprisingly or carelessly; imprudence of pedestrian; and driver violates crosswalk. In terms of injury type, over 60% of all children involved in a pedestrian crash during the studied period suffered slight injuries, while only 0.73% of the children were unharmed. In other words, 99.27% of all child pedestrian crashes yielded some degree of physical injury. In addition, 55.8% of child pedestrian crashes that occurred in Santiago during the studied period were boys and 58.1% of all child pedestrian crashes
306
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
Fig. 1. Child pedestrian crashes that occurred in Santiago between 2000 and 2008.
Table 1 Percentage of injuries for each contributing factor to pedestrian–vehicle crashes. Pedestrian crash contributing factor
% Unharmed
% Slightly injured
% Less seriously injured
% Seriously injured
% Fatalities
Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Total
0.54 1.40 0.96 0.00 0.00 0.00
57.16 57.87 53.99 58.27 53.93 60.47
15.00 11.66 13.74 14.39 13.48 16.28
26.66 28.09 27.80 25.18 26.97 16.28
0.64 0.98 3.51 2.16 5.62 6.98
41.01 14.36 6.31 2.80 1.80 0.87 67.16
Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Total
0.72 1.35 0.00 1.69 0.00 0.00 2.44
66.50 62.16 75.41 66.10 64.58 73.91 48.78
12.23 13.51 4.92 15.25 16.67 6.52 17.07
18.40 20.27 18.03 15.25 18.75 19.57 26.83
2.16 2.70 1.64 1.69 0.00 0.00 4.88
19.63 1.49 1.23 1.19 0.97 0.93 0.83 26.27
0.61 0.73
60.12 59.65
18.10 13.88
19.33 24.23
1.84 1.51
6.58 100.00
Other Undetermined causes Total
% Total
307
600 500 400 300 200 100
11:00 PM
9:00 PM
10:00 PM
7:00 PM
8:00 PM
5:00 PM
6:00 PM
3:00 PM
4:00 PM
2:00 PM
1:00 PM
12:00 PM
11:00 AM
9:00 AM
10:00 AM
7:00 AM
8:00 AM
6:00 AM
5:00 AM
4:00 AM
3:00 AM
2:00 AM
1:00 AM
0 12:00 AM
Average Number of Child Pedestrian Crashes
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
Time of Day
Fig. 2. Average number of child pedestrian crashes per hour between 2000 and 2008.
Table 2 Percentage of child pedestrian crashes for each contributing factor by road type.
concentrated children between 12 and 18 years of age with the highest number of fatalities at the average age of 12 years old. With respect to the time of the day, the average number of pedestrian crashes increased drastically after 7 am coinciding with the beginning of the school day (see Fig. 2). Approximately 3000 pedestrian crashes occurred between 1 p.m. and 8 p.m. corresponding to 57% of the total daily child pedestrian crashes. This time period includes the end of the school day for both morning and afternoon school shifts when most children return home. Fig. 3 illustrates that most of the child pedestrian crashes occurred on weekdays, particularly Wednesdays and Fridays, which accounted for 32.8% of the crashes during the studied period. The academic year (i.e., March through December) was the worst for child pedestrian crashes, while a lower number of these crashes were perceived during the summer time (i.e., January and February), as presented in Fig. 4. The largest quantity of crashes (51.8%) occurred on straight road sections (i.e., between intersections) during the period 2000–2008, as presented in Table 2, where 24.7% were caused by the pedestrian crossing the road in a surprisingly or carelessly manner, and 7% by the driver violating crosswalks. Conversely, this table shows that approximately 48% of the pedestrian crashes occurred at intersections with the pedestrian crossing roads surprisingly or carelessly being the major contributing factor to pedestrian crashes. Table 3 indicates that approximately 67% of child pedestrian crashes occurred on road segments with traffic signs. The pedestrian’s responsibility was the main contributing factor of these crashes by disobeying traffic sign laws or violating crosswalk locations. The driver was responsible for nearly 10% of the total number
Contributing factor Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Other Undetermined causes Total
Average Number of Child Pedestrian Crashes
900 800 700 600 500 400 300 200 100 0 Tuesday
Intersections
24.7
17.2
7.1 4.8 1.9 0.1 0.5
7.5 1.7 1.0 1.8 0.4
7.0 1.1 0.7 0.9 0.6 0.1 0.5
13.0 0.5 0.6 0.3 0.4 0.8 0.3
1.8
2.7
51.8
48.2
of pedestrian. These results suggest that pedestrians and/or drivers traveling in a distracted or careless manner are more likely to cause crashes at locations with traffic signs. Whereas, the occurrence of pedestrian crashes on road sections without traffic signs is less frequent since pedestrians cross roads with more care knowing the absence of traffic signs.
1000
Monday
Straight road sections
Wednesday
Thursday
Friday
Saturday
Sunday
Day of the Week
Fig. 3. Average number of child pedestrian crashes per day of the week between 2000 and 2008.
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311 600 500 400 300 200 100
r
r
be ec em D
em
be
er ov N
r be
O ct ob
t em Se
pt
ly
us Au g
Ju
ne Ju
ay M
ril Ap
M
ua r br
ar Fe
nu Ja
ar ch
y
0
y
Average Number of Child Pedestrian Crashes
308
Month
Fig. 4. Average number of child pedestrian crashes per month between 2000 and 2008.
4. Results and discussion Varying number of critical areas (ranging from 1 to 14) with high child pedestrian crash risk was identified using KDE for each year of the period 2000–2008. As a result from a yearly comparison analysis, critical areas 1, 2, 4, 5, 6, 7, and 9 (depicted as ellipses) were selected, as shown in Fig. 5, since they maintained high crash concentrations and frequencies for five or more years. These critical areas have density levels between 8.7 and 26.3 child pedestrian crashes per square kilometer. The selected critical areas cover large segments of major roads with high traffic flow that traverse more than one district such as critical areas 1, 4, and 5 (see Table 4). In addition, all critical areas are densely populated with mixed residential/commercial land use. Particularly, critical area 1 has perceived an increased residential population in recent years, and concentrates major educational, political, administrative, and financial activities of the city. Table 4 shows that critical areas 6 and 7 present a large number of child pedestrian crashes between 2000 and 2008, which correspond to high child population zones located in the districts of Puente Alto and Maipu, respectively. Children from both of these districts with lower middle socioeconomic level commute alone to
Table 3 Percentage of child pedestrian crashes on road segments with and without traffic signs for each contributing factor. Contributing factor Responsibility of pedestrian Pedestrian crosses road surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Pedestrian remains on the road Pedestrian disobeys red light signal Pedestrian under influence of alcohol Responsibility of driver Driver violates crosswalk Vehicle reversing Driving on left side of the road Loss of control of vehicle Unreasonable or imprudent speed Disobeying yield sign Driving under influence of alcohol Other Undetermined causes Total
With traffic signs
Without traffic signs
18.8
16.9
10.4 2.8 1.5 3.7 0.8
6.3 0.7 0.6 0.0 0.1
23.0 0.5 0.9 0.5 0.4 1.7 0.3
4.2 0.5 0.3 0.2 0.3 0.0 0.3
1.9
2.4
67.2
32.8
and from school using public transportation, and therefore, they are more prone to be exposed to pedestrian crashes. When contrasting the identified critical areas with the school density throughout the city, critical areas 1 and 7 situated in the districts of Santiago, Estacion Central, and Maipu overlap the highest school concentration with more than 11 schools per square kilometer. Further analysis in each of these areas is needed, in order to determine effective and strategic measures at school entrances and adjacent roads to prevent child pedestrian crashes. These measures may include improving intersection crossing times, increasing child pedestrian visibility, and enhancing police enforcement of speed laws and obedience to traffic signs (Braddock et al., 1994). Moran’s I spatial autocorrelation statistics was utilized to determine statistically significant clustering patterns of child pedestrian crashes within selected critical areas based on location proximity and their attribute similarities. In other words, a spatial and temporal analysis determined whether the main crash variables present a cluster, dispersion, or random distribution within high risk zones for child pedestrians. Moran’s I index was computed multiple times with different increasing distance thresholds until a distance of 80 m was obtained that maximized the Z score with a value of 60. Therefore, this distance is appropriate for maximizing the spatial autocorrelation (i.e., child pedestrian crashes are clustered up to a threshold distance of 80 m with a statistical significance of 0.01). In addition, Moran’s index was calculated using Euclidian distances and a 95% confidence coefficient under the assumption of a normal distribution. The first results of this analysis indicated that the attributes of child pedestrian crashes related to particular cohorts gender, weekday, and month of the year were not statistically significant within the identified critical areas. The spatial autocorrelation analysis in the selected critical areas showed an average positive autocorrelation for the crash contributing factors listed in Table 5. The four crash contributing factors presented in this table have a spatial dependence and tend to cluster in these areas. These factors Table 4 Districts within each identified critical area. Critical area
Districts
Total number of crashes 2000–2008
1 2 4 5 6 7 9
Santiago, Estacion Central Lo Prado San Ramon, La Cisterna El Bosque, San Bernardo Puente Alto Maipu ˜ Penalolen
589 162 238 389 414 431 221
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
309
Fig. 5. Selected critical areas with high child pedestrian crash risk in Santiago.
represent 81.31% of all the crash contributing factors and coincide with the major causes of child pedestrian crashes that occurred during the studied period. The remaining crash contributing factors presented randomness within the critical areas. Moran’s I coefficients were computed for children involved in crashes with an age range between 5 and 18. These coefficients yielded an average Moran’s I value of 0.009 and standard deviation of 0.047 with a Z score of 1.131, suggesting that no spatial autocorrelation exists between this variable and child pedestrian crashes within the critical areas. Therefore, there is no evidence of
dispersion or clustering of pedestrian crashes with respect to the children’s age. In terms of time of the day, there is a clear spatial pattern of pedestrian crashes reflected by the fact that the spatial autocorrelation tests resulted to be significant, as presented in Table 6. This table shows coefficient values solely for the years that the critical areas were identified as a result from the KDE analysis. The Table 6 Average yearly Moran’s I coefficients for each critical area with respect to the time of the day. Year
Table 5 Spatial autocorrelation analysis results per crash contributing factor. Factor
Average Moran’s I index
Standard deviation
Z-score
Pedestrian crosses surprisingly or carelessly Imprudence of pedestrian Pedestrian violates crosswalks Driver violates crosswalk
0.262
0.080
2.662*
0.253 0.233 0.248
0.022 0.027 0.047
2.651* 2.608* 2.600*
*
p < 0.005.
2000 2001 2002 2003 2004 2005 2006 2007 2008
Critical areas 1
2
4
5
6
7
9
0.275 0.287 0.278 0.274 0.270 0.266 0.307 0.258 0.254
0.290 0.365
0.268 0.306 0.244
0.315 0.267 0.284 0.266
0.321 0.290
0.248 0.274 0.321 0.287
0.307 0.262
0.254 0.266 0.248
0.310 0.274 0.338 0.277 0.266
0.287 0.278 0.274 0.287 0.288 0.241
0.306
0.278
0.295 0.286
0.304
310
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311
average standardized Z score value for these coefficient values is 2.6 (p < 0.004). Therefore, there is a strong dependence between the time of day and the child pedestrian crashes in these areas over the study period. The spatial autocorrelation results for the road type attribute show a positive autocorrelation with an average Moran’s I value of 0.282 and standard deviation of 0.062 (Z = 2.762 and p < 0.003) on straight road sections in critical areas 4, 5 and 6, whereas intersections present a random spatial pattern in these areas. These critical areas comprise long segments of road between crosswalks with high traffic flow traveling at high speeds. Hence, pedestrians have a tendency to cross at any location, which exposes them to a higher pedestrian crash risk. An average positive spatial concentration of pedestrian crashes was obtained at intersections in critical areas 1, 2, 7 and 9, while crashes on straight sections for these areas were randomly distributed during the nine years of the study. In contrast to critical areas 4, 5, and 6, these areas are situated in urban areas with a large density of intersections. Thus, pedestrians are more likely to cross roads at intersections, and crashes may be caused due to the driver’s imprudence and the carelessness of pedestrians. Further analysis is required to examine in more detail the spatial relationship between road type and child pedestrian crashes in these critical areas. Finally, a positive autocorrelation is present for roads without traffic signs in each critical area with an average Moran’s I index value of 0.247 and standard deviation of 0.102 (Z = 2.610 and p < 0.004) for the studied period. Hence, the pattern of crashes that occurred on these road segments has an average tendency to cluster in space and time. On the other hand, road segments with traffic signs in all seven critical areas contained yielded an average index of 0.019 and standard deviation of 0.039 (Z = 1.183 and p < 0.005), which implies an absence of spatial autocorrelation between this variable and child pedestrian crash locations. Hence, the generation of pedestrian crash on road segments with traffic signs is randomly distributed. Additional investigation is needed to identify the sources of child pedestrian crashes that occurred on road sections with no traffic signs.
5. Conclusions This study identified seven critical areas with high child pedestrian crash risk in the city of Santiago, Chile using kernel density estimation (KDE) in a GIS environment. These areas are located in districts with lower middle socioeconomic status population. Therefore, they have low investment capabilities on traffic safety (e.g., traffic signs, speed bumps, crosswalks, etc.) exposing children that travel alone to and from schools to a higher pedestrian crash risk. The results of this study determined the main attributes of child pedestrian crashes (e.g., crash contributing factors, age, time of day, road type, and traffic signs) that present a statistical significance in the identified critical areas. Moran’s I spatial autocorrelation analysis indicates that the responsibility of pedestrians is the major contributing factor for the generation of child pedestrian crashes with a tendency to cluster in space and time. Similarly, there is a spatial clustering distribution of crashes in terms of time of the day. This analysis also suggests an absence of spatial autocorrelation between the age of the victims and child pedestrian crashes in the selected critical areas. Thus, there is no dependence of crashes due to the age attribute. A positive spatial autocorrelation of pedestrian crash on roads without traffic signs imply the occurrence of these crashes on neighboring roads without traffic signs. Further analysis is needed to explore the spatial and temporal relationship of child pedestrian crashes with respect to the road type variable.
The results of this research contributed by analyzing the crash reported data and determining critical areas where enhanced pedestrian safety is needed by developing and implementing effective enforcement, educational, and engineering strategies. Children must be taught to be alert and have greater caution where pedestrian crashes are more likely to occur, particularly in areas with high volume traffic and large quantity of parked vehicles. In addition, children that travel unsupervised to and from school should be taught how to cross roads and safely use public transportation. Future research includes the spatial and temporal analysis of child pedestrian crashes taking into account other variables such as land use (in particular, proximity to schools), socioeconomic level, population distribution, and trip purpose. Acknowledgements Financial support from the Universidad Andres Bello Project No. DI 97 -12/R is acknowledged. Anonymous review of this paper is also gratefully thanked. References Anderson, T., 2009. Kernel density estimation and K-means clustering to profile road accident hotspots. Accident Analysis and Prevention 41, 359–364. Austin, K., Tight, M., Kirby, H., 1997. The use of geographical information systems to enhance road safety analysis. Transportation Planning and Technology 20 (3), 249–266. Banos, A., Huguenin-Richard, F., 2000. Spatial distribution of road accidents in the vicinity of point sources application to child pedestrian accidents. In: Geography and Medicine. Elsevier, pp. 54–64. Braddock, M., Lapidus, G., Cromley, E., Cromley, R., Burke, G., Banco, L., 1994. Using a geographic information system to understand child pedestrian injury. American Journal of Public Health 84 (7), 1158–1161. Clifton, K., Kreamer-Fults, K., 2007. An examination of the environmental attributes associated with pedestrian–vehicular crashes near public schools. Accident Analysis and Prevention 39, 708–715. Dissanayake, D., Aryaija, J., Wedagama, D.M., 2009. Modelling the effects of land use and temporal factors on child pedestrian casualties. Accident Analysis and Prevention 41, 1016–1024. Erdogan, S., 2009. Explorative spatial analysis of traffic accident statistics and road mortality among the provinces of Turkey. Journal of Safety Research 40, 341–351. Erdogan, S., Yilmaz, I., Baybura, T., Gullu, M., 2008. Geographical information systems aided traffic accident analysis system case study: city of Afyonkarahisar. Accident Analysis and Prevention 40, 174–181. Flahaut, B., Mouchart, M., San Martin, E., Thomas, I., 2003. The local spatial autocorrelation and the kernel method for identifying black zones: a comparative approach. Accident Analysis and Prevention 35, 991–1004. Getis, A., Ord, J., 1994. The analysis of spatial association by use of distance statistics. Geographical Analysis 24 (3), 189–206. Gundogdu, I., 2010. Applying linear analysis methods to GIS-supported procedures for preventing traffic accidents: case study of Konya. Safety Science 48, 763–769. Kim, K., Yamashita, E., 2005. Using K-means clustering algorithm to examine patterns of pedestrian involved crashes in Honolulu, Hawaii. Journal of Advanced Transportation 41 (1), 69–89. Kuo, P.-F., Lord, D., Walden, T., 2011. Using geographical information systems to effectively organize police patrol routes by grouping hot spots of crash and crime data. In: Proceedings of the 3rd International Conference on Road Safety and Simulation, Indianapolis. Ladron de Guevara, F., Washington, S., Oh, J., 2004. Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arisona. Transportation Research Board 1897, 191–199. Levine, N., Kim, K., Nitz, L., 1995. Spatial analysis of Honolulu motor vehicle crashes. I. Spatial patterns. Accident Analysis and Prevention 27 (5), 663–674. Levine, N., 2004. CrimeStat III: A Spatial Statistics Program for the Analysis of Crime Incident Locations. Chapter 8: Kernel Density Interpolation. Ned Levine & Associates/The National Institute of Justice, Houston, TX/Washington, DC. Lightstone, A., Dhillon, P., Peek-Asa, C., Kraus, J., 2001. A geographic analysis of motor vehicle collisions with child pedestrian in Long Beach, California: comparing intersection and midblock incident locations. Injury Prevention 7, 155–160. Meliker, J., Maio, R., Zimmerman, M., Kim, H., Smith, S., Wilson, M., 2004. Spatial analysis of alcohol-related motor vehicle crash injuries in southeastern Michigan. Accident Analysis and Prevention 36, 1129–1135. Miller, J., 2000. Geographic information systems: unique analytic capabilities for the traffic safety community. Transportation Research Record 1734, 21–28. Nantulya, V., Reich, M., 2002. The neglected epidemic: road traffic injuries in developing countries. British Medical Journal 324 (7346), 1139–1141. Ng, K.-S., Hung, W.-T., Wong, W.-G., 2002. An algorithm for assessing the risk of traffic accident. Journal of Safety Research 33, 387–410.
C.A. Blazquez, M.S. Celis / Accident Analysis and Prevention 50 (2013) 304–311 Pulugurtha, S., Krishnakumar, V., Nambisan, S., 2007. New methods to identify and rank high pedestrian crash zones: an illustration. Accident Analysis and Prevention 39, 800–811. Romero, P., 2007. Accidentes en la infancia: Su prevención tarea prioritaria en este milenio. Revista Chilena de Pediatría 78, 57–73. Schneider, R., Ryznar, R., Khattak, A., 2004. An accident waiting to happen: a spatial approach to proactive pedestrian planning. Accident Analysis and Prevention 36, 193–211. Steenberghen, T., Dufays, T., Thomas, I., Flahaut, B., 2004. Intra-urban location and clustering of road accidents using GIS: a Belgian example. International Journal Geographical Information Science 18 (2), 169–181.
311
Truong, L., Somenahalli, S., 2011. Using GIS to identify pedestrian–vehicle crash hot spots and unsafe bus stops. Journal of Public Transportation 14 (1), 99–114. Xie, Z., Yan, J., 2008. Kernel density estimation of traffic accidents in a network space. Computers, Environment and Urban Systems 32, 396–406. World Health Organization, 2008. Accidents and injuries. Children’s environmental health, http://www.who.int.