Available online at www.sciencedirect.com ScienceDirect Available online at www.sciencedirect.com Transportation Research Procedia 00 (2017) 000–000 Transportation Research Procedia 00 (2017) 000–000 www.elsevier.com/locate/procedia ScienceDirect www.elsevier.com/locate/procedia Transportation Research Procedia 25C (2017) 3237–3256 www.elsevier.com/locate/procedia

World Conference on Transport Research - WCTR 2016 Shanghai. 10-15 July 2016 The Impact of Mass Transit on Public Security - A Study of in

a WANG Dia*

a a Shanghai Jiguang College, 2859 Shuichan Road, Shanghai, 201901

Abstract

In the United States, mass transit is considered to be related to crimes in some cases, on which scholars focusing on different cases have drawn different conclusions. In this research, spatial relation analysis between 12 types of major crimes and Bay Area Rapid Transit (BART) stations in San Francisco were conducted on different spatial levels from city, block groups to blocks and streets. It is demonstrated that mass transit and crimes are closely related in San Francisco. On the macro level, the extension direction of whole crimes and the BART service alignment are similar. On the meso-level, BART stations are significantly correlated with most types of crimes. No matter in the downtown area or not, the neighborhoods around stations are high criminal regions. On the micro level, crimes tend to cluster obviously around the stations. © 2017 The Authors. Published by Elsevier B.V. ©Peer 2017-review The Authors.under responsibili Publishedty by of ElsevierWORLD B.V. CONFERENCE ON TRANSPORT RESEARCH SOCIETY. Peer-reviewPeer-review under responsibilityresponsibility of WORLD CONFERENCE ON TRANSPORT RESEARCHRESEARCH SOCIETY.SOCIETY.

Keywords: Mass transit; Crime; San Francisco; Bay Area Rapid Transit

1. Introduction

The vehicle holding is 0.8 car per capita in the US in 2013 (U. S. Census Bureau, 2014), while each car’s average travelling distance is more than 20000km per year (Federal Highway Administration of U. S. Department of Transportation, 2014), which means each individual has a daily driving car averagely. Public transportation is therefore always seen as the last resort of travels (Lusk, 2001) especially outside downtown and is not popular in some neighborhoods for its attribute of welfare. According to the summary of researches (Dunphy et al., 2004), at least one third of the land values go down when transit stations open. As far as mass transit is concerned, it is

* Corresponding author. Tel.: +86-13611977976; fax: +86-21-59889644. E-mail address: [email protected]

2214-241X © 2017 The Authors. Published by Elsevier B.V. Peer-review under responsibility of WORLD CONFERENCE ON TRANSPORT RESEARCH SOCIETY.

2352-1465 © 2017 The Authors. Published by Elsevier B.V. Peer-review under responsibility of WORLD CONFERENCE ON TRANSPORT RESEARCH SOCIETY. 10.1016/j.trpro.2017.05.145

10.1016/j.trpro.2017.05.145 2352-1465 3238 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 2 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 3 generally acknowledged that this main type of public transportation brings large amounts of strangers, who will In addition, the change of crime is influenced by some important factors around transit rather than the transit break the peace of neighborhood, thus worsening public security and community environment. Therefore, there used itself. For example, the land use, infrastructures, population structures, forms of neighborhood space, time node, to be rejections and resistances to mass transit in history for diverse but similar reasons. For example, the norward critical events and so on. When these objective conditions change, the relationship between transit and crimes seems extension of BART in San Francisco was vetoed twice in 1990 and 1998. However, does mass transit really lead to to be changed. Thus some researches don’t interpret the relation between them and only describe the tendencies with high crime rate and worsen public security? This research is aimed to shed some light on the relationship between different conditions. crimes and mass transit. Bay Area Rapid Transit (BART) system of San Francisco is selected as the empirical case, for the city is isolated geographically with developed neighborhoods and mass transit. Spatial relation analysis 3. Research Object, Data and Methodology between 12 types of major crimes and BART stations will be conducted on multi-levels. On the macro level, the research analyzes overall distribution features of crimes of the whole city; on the meso-level, the research analyzes 3.1. Research Object and Scope attributes and criminal clusters of blocks; on the micro level, the research focuses on spatial details of criminal spots. Along with different levels of spatial analysis from city, block groups, blocks through streets, the impacts of other This research is focused on Bay Area Rapid Transit (BART) in the city of San Francisco (the same scope with factors such as population, land use intensity, neighborhood attribute and street scale are also partly taken into the county of San Francisco) to analyze the criminal impact of mass transit, taking following advantages: (1) In the account to make comprehensive analysis in order to reveal whether BART really has close relation to crimes. north of the peninsula, the city’s location effectively resist crime from exterior by natural barriers such as the sea to the north, west and east, five city parks and one state park along the south border. The data is independent in spatial 2. Literature Review analysis, free from externalities. (2) Population density of San Francisco is only second to New York in the U.S. Most districts have been developed, featuring stabilized social-spatial characteristics and criminal distribution Scholars in different cases have drawn different conclusions on this question. The first kind of viewpoints (often pattern. (3) BART has been under operation over 40 years, whose lines and stations are familiar to most people. based on common sense) is (mass) transit will increase crime rate. For example, the existing studies show that busy Passengers travel regularly, contributing stable ridership. places often generate higher numbers of criminal incidents. These busy and high crime areas are often served by mass transit (Yu, 2009). The crimes come from interactions between transit and its surroundings (Block and Block, 2000). These crimes seem to be related to overall crimes at the neighborhood (Loukaitou-Sideris et al., 2002). In addition, mass transit may increase convenience for offenders the same as normal passengers. It helps them get over distances (Liggett et al., 2003) and carry them to the destinations around mass transit (Brantingham et al., 1991). It can be seen from the cases that new transit station will worsen the security around it at least in short term or in some types of crimes (Poister, 1996; Liggett et al., 2001; Newton, 2004). Most of these crimes are in stations but not during travelling (Degeneste and Sullivan, 1994) and fear levels are higher while waiting (Crime Concern, 2004). The fear rather influences people not to choose transit (Hartgen et al., 1993; Lusk, 2001). This negative change of passengers’ attitude will have a negative impact on the regional environment which is associated with increased number of criminal incidents (Yu, 2009) and it will be a vicious spiral. The second kind of viewpoints is (mass) transit doesn’t have significant correlation with criminal increase. It is often the falsification of the first kind. For instance, the busy places such as stations will have more police equipped (Frischtak and Mandel, 2012) especially for preventing terrorist attacks. Most offenders prefer to select a familiar place (Brantingham and Brantingham, 1993; Willits et al., 2011) for ease of committing crime such as their residences rather than strange places by (mass) transit (Desroches, 1995; Tilley et al., 2004). It can be seen from the cases that the security around mass transit is not worse than even residence (Walker et al., 2006). New built mass transit will not increase number of criminal incidents significantly (Poister, 1996; LaVigne, 1996; Liggett et al., 2003; Sedelmaier, 2003) and the peacekeeping activities depend more on whether well-organized communities are established (Krivo and Peterson, 1996). The transit crime incidents are less serious (TCRP, 1997) and tend to be concentrating into small places instead of dispersing among the blocks (Tseloni and Pease, 2003). These two viewpoints are not independent. Some researchers appear in both two viewpoints. Furthermore, researches on same theories or hypothesis may draw tremendously different conclusions according to the empirical evidence because data’s integrity and reliability will impact the results. Data in different researches differ much in time, location, area, sample size and accuracy. The accumulated effects of these impacts may contribute more than the logic established in theories. Therefore, these empirical analyses only provide supplementary evidences for hypothesis or theories under certain circumstances. Both the viewpoints are based on same basic theories oriented from Chicago school. The most popularly employed theory to analyze transit crimes is routine activity theory (Cohen and Felson, 1979), which lists three crime premises: motivated offenders, suitable targets and the absence of capable guardians. Many researches study these premises around transit on micro level. Moreover, spatial analysis in crime mapping study, simulation technique and relevant technique in interdiscipline are methodologically implemented more frequently. Fig.1. Bay Area Rapid Transit Lines (Source: www.bart.com) Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3239 2 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 3 generally acknowledged that this main type of public transportation brings large amounts of strangers, who will In addition, the change of crime is influenced by some important factors around transit rather than the transit break the peace of neighborhood, thus worsening public security and community environment. Therefore, there used itself. For example, the land use, infrastructures, population structures, forms of neighborhood space, time node, to be rejections and resistances to mass transit in history for diverse but similar reasons. For example, the norward critical events and so on. When these objective conditions change, the relationship between transit and crimes seems extension of BART in San Francisco was vetoed twice in 1990 and 1998. However, does mass transit really lead to to be changed. Thus some researches don’t interpret the relation between them and only describe the tendencies with high crime rate and worsen public security? This research is aimed to shed some light on the relationship between different conditions. crimes and mass transit. Bay Area Rapid Transit (BART) system of San Francisco is selected as the empirical case, for the city is isolated geographically with developed neighborhoods and mass transit. Spatial relation analysis 3. Research Object, Data and Methodology between 12 types of major crimes and BART stations will be conducted on multi-levels. On the macro level, the research analyzes overall distribution features of crimes of the whole city; on the meso-level, the research analyzes 3.1. Research Object and Scope attributes and criminal clusters of blocks; on the micro level, the research focuses on spatial details of criminal spots. Along with different levels of spatial analysis from city, block groups, blocks through streets, the impacts of other This research is focused on Bay Area Rapid Transit (BART) in the city of San Francisco (the same scope with factors such as population, land use intensity, neighborhood attribute and street scale are also partly taken into the county of San Francisco) to analyze the criminal impact of mass transit, taking following advantages: (1) In the account to make comprehensive analysis in order to reveal whether BART really has close relation to crimes. north of the peninsula, the city’s location effectively resist crime from exterior by natural barriers such as the sea to the north, west and east, five city parks and one state park along the south border. The data is independent in spatial 2. Literature Review analysis, free from externalities. (2) Population density of San Francisco is only second to New York in the U.S. Most districts have been developed, featuring stabilized social-spatial characteristics and criminal distribution Scholars in different cases have drawn different conclusions on this question. The first kind of viewpoints (often pattern. (3) BART has been under operation over 40 years, whose lines and stations are familiar to most people. based on common sense) is (mass) transit will increase crime rate. For example, the existing studies show that busy Passengers travel regularly, contributing stable ridership. places often generate higher numbers of criminal incidents. These busy and high crime areas are often served by mass transit (Yu, 2009). The crimes come from interactions between transit and its surroundings (Block and Block, 2000). These crimes seem to be related to overall crimes at the neighborhood (Loukaitou-Sideris et al., 2002). In addition, mass transit may increase convenience for offenders the same as normal passengers. It helps them get over distances (Liggett et al., 2003) and carry them to the destinations around mass transit (Brantingham et al., 1991). It can be seen from the cases that new transit station will worsen the security around it at least in short term or in some types of crimes (Poister, 1996; Liggett et al., 2001; Newton, 2004). Most of these crimes are in stations but not during travelling (Degeneste and Sullivan, 1994) and fear levels are higher while waiting (Crime Concern, 2004). The fear rather influences people not to choose transit (Hartgen et al., 1993; Lusk, 2001). This negative change of passengers’ attitude will have a negative impact on the regional environment which is associated with increased number of criminal incidents (Yu, 2009) and it will be a vicious spiral. The second kind of viewpoints is (mass) transit doesn’t have significant correlation with criminal increase. It is often the falsification of the first kind. For instance, the busy places such as stations will have more police equipped (Frischtak and Mandel, 2012) especially for preventing terrorist attacks. Most offenders prefer to select a familiar place (Brantingham and Brantingham, 1993; Willits et al., 2011) for ease of committing crime such as their residences rather than strange places by (mass) transit (Desroches, 1995; Tilley et al., 2004). It can be seen from the cases that the security around mass transit is not worse than even residence (Walker et al., 2006). New built mass transit will not increase number of criminal incidents significantly (Poister, 1996; LaVigne, 1996; Liggett et al., 2003; Sedelmaier, 2003) and the peacekeeping activities depend more on whether well-organized communities are established (Krivo and Peterson, 1996). The transit crime incidents are less serious (TCRP, 1997) and tend to be concentrating into small places instead of dispersing among the blocks (Tseloni and Pease, 2003). These two viewpoints are not independent. Some researchers appear in both two viewpoints. Furthermore, researches on same theories or hypothesis may draw tremendously different conclusions according to the empirical evidence because data’s integrity and reliability will impact the results. Data in different researches differ much in time, location, area, sample size and accuracy. The accumulated effects of these impacts may contribute more than the logic established in theories. Therefore, these empirical analyses only provide supplementary evidences for hypothesis or theories under certain circumstances. Both the viewpoints are based on same basic theories oriented from Chicago school. The most popularly employed theory to analyze transit crimes is routine activity theory (Cohen and Felson, 1979), which lists three crime premises: motivated offenders, suitable targets and the absence of capable guardians. Many researches study these premises around transit on micro level. Moreover, spatial analysis in crime mapping study, simulation technique and relevant technique in interdiscipline are methodologically implemented more frequently. Fig.1. Bay Area Rapid Transit Lines (Source: www.bart.com) 3240 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 4 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 5

Figure 1 shows that BART lines lie in San Francisco city (Red region in the figure), peninsula area and east bay type intuitively. The long axe of oval reveals the spread direction of crime cases, while the short axe indicates the area. The lines connect cities in several directions except North Bay area, taking large amounts of passengers to agglomeration of crime cases. The deviation oval is expressed in the following formula: downtown San Francisco, featuring rush hour transportation. There are 8 BART stations in the city of San n n Francisco, among which Montgomery St, Powell St and Civic Center are situated in downtown area, Embarcadero (x X )2 (y Y )2 å i - å i - station is on the fringe of downtown. The Treasure Island is not within the research scope for its isolated location i 1 i 1 SDE = = ,SDE = = (2) with few crime events. x n y n The rotation angle: 3.2. Data Resource n n n n n 2 2 2 2 ( x~ - y~ ) + ( x~ - y~ )2 + 4( x~ y~ )2 The crime data from data.sfgov.org include all crime cases reported or detected in San Francisco from 2004 to å i å i å i å i å i i tan i =1 i =1 i =1 i =1 i =1 ( ) 2013, featuring type, time and location. 12 types of them were selected in this research on following principles: (1) q = n 3 specific motive and successful implementation; (2) harm to other people or society; (3) brought into force or stopped 2 x~ y~ due to external environment; (4) in plenty amount of samples. They take up 57.8 percent of all crime cases. å i i i =1 These cases can be divided into 3 categories: (1) violence crimes: assault, kidnapping, robbery and sex offenses; (2) property crimes: burglary, vehicle theft, fraud, larceny; (3) other crimes: arson, drug, prostitution, vandalism. where x~i and y~i stand for the deviation between mean center and x, y coordinates. θ has both possibilities of Violence crimes plus arson can be defined as serious crimes for much more harm to the victims physically and plus and minus. The main axe is rotated angle θ from north clockwise in case of plus and anti-clockwise in case of mentally. minus. The following formula calculates the standard deviation of x and y axes: n 2 n 2 3.3. Research Design and Methodology (x cos y sin ) (x cos y sin ) å ~i q - ~i q å ~i q + ~i q i 1 i 1 s = = ,s = = (4) The analysis of relationship between crimes and BART are studied on macro, meso and micro level. The overall x n y n city is analysed on macro level, whilst blocks on meso-level and crime spots on micro level. On one hand, this method will provide a complete logical framework. On the other hand, it is conducive to accept diverse factors to Weight is not referred to in the research. ( ) make a comprehensive analysis rather than single statistical technique. Since all the analyses are based on numerous 3 Nearest neighboring index. detailed crime cases which have individual interpretability respectively but lack abstract summarizing ability, the The index is the quotient of the average distance from each crime case to its nearest crime case and the average research will try to find a balance between detail and abstract. This means some extra studies such as the features of distance supposed the crime cases in random distribution. If the quotient is under 1 and meets the requirement of different criminal types which root in activities and land use will not be fully discussed in detail. The research wants test, the crimes are clustered. If it equals 1, the crimes are distributed in random. If over 1, the crimes have the to propose a basic judgment on the correlation between crimes and BART. Once an upper spatial level’s tendency of homogeneity distribution. The formula is as follows: 1 n phenomenon is put forward, it will depend on the interpretation of lower level’s analysis. Meanwhile, the deeper d reasons and development possibilities involving non-spatial factors will be analyzed partly and given an overview å i n i =1 for further studies on the basis of current judgment. R = (5) 1 3.3.1. Overall Description and Analysis n To begin with, the comparison is made both externally and internally to figure out the interrelationship and 2 tendency of general crime data. Secondly, overall spatial analysis of crime cases is made to obtain the layout and A change of the 12 types, in which the mean center method is implemented to compare the mean center of various where di stands for the distance from crime i to its nearest crime case. A stands for the area of San Francisco city. criminal types, and standard deviation oval is set up to contrast the spread direction and disperse scope, and the In the research, test standard is that critical value z (standard deviation) is under -2.58, the significance level p is nearest neighboring index is employed to inspect the cluster of crime cases as a whole. Finally, the mean centers and under 0.01and confidence interval is 99%. The formula of z is: ovals are compared with the distribution of BART stations. 1 n 1 d (1)Mean center. å i - n i =1 n The calculation of x and y mean coordinates of all crime cases of a certain type follows the formula: 2 n n A x y z = (6) å i å i 0.26136 X = i =1 ,Y = i =1 (1) n n n 2 where xi and yi stand for the coordinates of crime i, whilst n for the quantity of this criminal type. (2)Standard deviation oval. A Taking the mean center as the start point, the standard deviations of x and y coordinates of a crime case is 3.2.2. Meso-level Analysis calculated and defined as the axe of oval, which is called deviation oval. Both mean center and deviation oval are The research on meso-spatial level is focused on block groups and blocks, whose space feature can attract or implemented to analyze the spatial distribution of different crimes or to compare the tendency of certain criminal reject certain criminal types. The multiple regression analysis of social economic elements of block sheds light on the main elements influencing diverse criminal types and the relation between BART and criminal types. The space Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3241 4 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 5

Figure 1 shows that BART lines lie in San Francisco city (Red region in the figure), peninsula area and east bay type intuitively. The long axe of oval reveals the spread direction of crime cases, while the short axe indicates the area. The lines connect cities in several directions except North Bay area, taking large amounts of passengers to agglomeration of crime cases. The deviation oval is expressed in the following formula: downtown San Francisco, featuring rush hour transportation. There are 8 BART stations in the city of San n n Francisco, among which Montgomery St, Powell St and Civic Center are situated in downtown area, Embarcadero (x X )2 (y Y )2 å i - å i - station is on the fringe of downtown. The Treasure Island is not within the research scope for its isolated location i 1 i 1 SDE = = ,SDE = = (2) with few crime events. x n y n The rotation angle: 3.2. Data Resource n n n n n 2 2 2 2 ( x~ - y~ ) + ( x~ - y~ )2 + 4( x~ y~ )2 The crime data from data.sfgov.org include all crime cases reported or detected in San Francisco from 2004 to å i å i å i å i å i i tan i =1 i =1 i =1 i =1 i =1 ( ) 2013, featuring type, time and location. 12 types of them were selected in this research on following principles: (1) q = n 3 specific motive and successful implementation; (2) harm to other people or society; (3) brought into force or stopped 2 x~ y~ due to external environment; (4) in plenty amount of samples. They take up 57.8 percent of all crime cases. å i i i =1 These cases can be divided into 3 categories: (1) violence crimes: assault, kidnapping, robbery and sex offenses; (2) property crimes: burglary, vehicle theft, fraud, larceny; (3) other crimes: arson, drug, prostitution, vandalism. where x~i and y~i stand for the deviation between mean center and x, y coordinates. θ has both possibilities of Violence crimes plus arson can be defined as serious crimes for much more harm to the victims physically and plus and minus. The main axe is rotated angle θ from north clockwise in case of plus and anti-clockwise in case of mentally. minus. The following formula calculates the standard deviation of x and y axes: n 2 n 2 3.3. Research Design and Methodology (x cos y sin ) (x cos y sin ) å ~i q - ~i q å ~i q + ~i q i 1 i 1 s = = ,s = = (4) The analysis of relationship between crimes and BART are studied on macro, meso and micro level. The overall x n y n city is analysed on macro level, whilst blocks on meso-level and crime spots on micro level. On one hand, this method will provide a complete logical framework. On the other hand, it is conducive to accept diverse factors to Weight is not referred to in the research. ( ) make a comprehensive analysis rather than single statistical technique. Since all the analyses are based on numerous 3 Nearest neighboring index. detailed crime cases which have individual interpretability respectively but lack abstract summarizing ability, the The index is the quotient of the average distance from each crime case to its nearest crime case and the average research will try to find a balance between detail and abstract. This means some extra studies such as the features of distance supposed the crime cases in random distribution. If the quotient is under 1 and meets the requirement of different criminal types which root in activities and land use will not be fully discussed in detail. The research wants test, the crimes are clustered. If it equals 1, the crimes are distributed in random. If over 1, the crimes have the to propose a basic judgment on the correlation between crimes and BART. Once an upper spatial level’s tendency of homogeneity distribution. The formula is as follows: 1 n phenomenon is put forward, it will depend on the interpretation of lower level’s analysis. Meanwhile, the deeper d reasons and development possibilities involving non-spatial factors will be analyzed partly and given an overview å i n i =1 for further studies on the basis of current judgment. R = (5) 1 3.3.1. Overall Description and Analysis n To begin with, the comparison is made both externally and internally to figure out the interrelationship and 2 tendency of general crime data. Secondly, overall spatial analysis of crime cases is made to obtain the layout and A change of the 12 types, in which the mean center method is implemented to compare the mean center of various where di stands for the distance from crime i to its nearest crime case. A stands for the area of San Francisco city. criminal types, and standard deviation oval is set up to contrast the spread direction and disperse scope, and the In the research, test standard is that critical value z (standard deviation) is under -2.58, the significance level p is nearest neighboring index is employed to inspect the cluster of crime cases as a whole. Finally, the mean centers and under 0.01and confidence interval is 99%. The formula of z is: ovals are compared with the distribution of BART stations. 1 n 1 d (1)Mean center. å i - n i =1 n The calculation of x and y mean coordinates of all crime cases of a certain type follows the formula: 2 n n A x y z = (6) å i å i 0.26136 X = i =1 ,Y = i =1 (1) n n n 2 where xi and yi stand for the coordinates of crime i, whilst n for the quantity of this criminal type. (2)Standard deviation oval. A Taking the mean center as the start point, the standard deviations of x and y coordinates of a crime case is 3.2.2. Meso-level Analysis calculated and defined as the axe of oval, which is called deviation oval. Both mean center and deviation oval are The research on meso-spatial level is focused on block groups and blocks, whose space feature can attract or implemented to analyze the spatial distribution of different crimes or to compare the tendency of certain criminal reject certain criminal types. The multiple regression analysis of social economic elements of block sheds light on the main elements influencing diverse criminal types and the relation between BART and criminal types. The space 3242 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 6 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 7 auto correlation analysis gives verdict to whether there exists crime cluster dominated by BART in blocks which are station entrance-exits on crimes. near or containing BART stations. Crimes in San Francisco around all stations’ all entrance-exits are calculated according to the formula. On one (1)Social economic element analysis. hand, it is sufficient to make a total statistic of every entrance-exit in each station. On the other hand, the result will The social, economic and demographic features of blocks are collected to make factor analysis and multiple not only be influenced by other city in case of large statistic scope, but also will make repeated calculation of crimes regression analysis. around other stations. Therefore, the catchment area of road network should be determined at the beginning, (2)Space auto correlation analysis. calculating general K function of each station’s every entrance-exit with the valid range of 600 meters. The formula Since there are spatial correlations between crimes, Moran’s I index is implemented to analyze auto correlation, is: divided into global auto correlation and local auto correlation. nA t Global auto correlation describes the average correlation between crimes in San Francisco, expressed as å n( ) following formula: t 1 i =1 p A i K ( ) = (10) n n n n AB p r n w (x x )(x x ) w z z A i B A å å ij i - j - å å ij i j n i =1 j =1 n i =1 j =1 where nA stands for the entrance-exit collection of the station. In order to secure the validity and independence, I = = ( ) n n 7 crimes within other station’s hinterland are not considered. S S 2 0 (x x )2 0 z å i - å i i =1 i =1 4. Macro level Analysis n n S w where wij stands for the spatial weight between crime i and j , 0 = å å ij is the total of all weights. 4.1. Crime Review i =1 j =1 4.1.1. External Contrast z = (x - x ), z = (x - x ) . Moran’s I is within [-1,1]. On certain significance level, crimes i i j j San Francisco city covers 121 km2 (U. S. Census Bureau, 2011) with a population of 837 thousands in 2013 (U. are positively correlated in case of Moran’s I’ plus, negatively correlated in case of minus and random distribution S. Census Bureau, 2012), ranking 14th among the 279 cities over 100 thousands population in the U.S. In 2012, total in case of around 0. crimes in San Francisco ranks 14th and the crime rate is 75th (FBI, 2012), higher than national average crime rate, in Local auto correlation serves to analyze the relationship among crimes near BART stations and regional which sex offenses and assault obviously lower than the average, while robbery, larceny and vehicle theft difference. Both local indicators of spatial association and Moran scatter diagram are employed here. The former is significantly higher than the average. Provided the motive of robbery is property, the property crime comes in large expressed by Moran’s Ii, standing for local Moran’s I, in the following formula: amounts, while the violence crime is relatively fewer in this city. z w z ( ) I i= i å ij j 8 j 4.1.2. Internal Contrast The observed value has been transformed to standardization, whilst row-standardized form is employed in spatial The quantity of crimes is listed in table 1, which indicates that larceny, assault, drug, vandalism, vehicle theft and weights. The results and Moran scatter diagram are integrated to make intuitive recognition through Moran burglary take up a bigger proportion. significance level diagram. In the software of Geoda, local spatial relationships are generated in 5 classifications, Table 1. Quantity of Crimes in the Study which are high criminal region encircled by high criminal region, high criminal region encircled by low criminal region, low criminal region encircled by high criminal region, low criminal region encircle by low criminal region Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Total and not significant space relationship. Assault 13162 11864 12443 12518 12667 12257 10851 10262 9801 10369 116194 Kidnapping 290 298 318 345 343 478 347 406 267 513 3605 3.2.3. Micro Level Analysis Robbery 3415 3615 4131 4027 4229 3578 3304 3348 3904 4156 37707 The study on micro level will go deep into the streets in blocks to analyze the criminal accessibility of specific Sex offenses 775 722 588 627 670 661 566 576 573 587 6345 spots. This helps to find the law of criminal clusters around BART stations. The cluster of different crimes around BART stations entrance-exit could be calculated through bivariate network K-function (Okabe et al., 2002) on the Burglary 6777 7096 7004 5453 5678 5377 4790 4783 5962 5844 58764 standard accessibility of road network with following formula: Fraud 3076 2766 2591 2344 2554 2496 2465 2541 2469 2577 25879 t 1 t Larceny 24822 25623 27347 25765 25796 25543 23880 25123 29979 35646 269524 ( ) K AB( ) = n( ) 9 Vehicle Theft 8639 8696 7291 6459 6052 5182 4350 4750 6096 6207 63722 p Ai rB p Ai Arson 280 236 240 246 248 222 206 194 228 228 2328 t where n( ) stands for the amount of criminal type B in the shortest distance t from the ith BART station Drug 10014 8527 8909 10351 11462 11744 9034 6724 6230 6682 89677 p A i Prostitution 1551 1112 1287 1862 1663 1460 637 445 269 269 10555 Vandalism 6563 7068 7688 7566 7342 7603 7604 6910 7503 6653 72500 entrance-exit(pA), r B for the density of station entrance-exits in road network. It is supposed that the crimes and Total 79364 77623 79837 77563 78704 76601 68034 66062 73281 79731 756800 station entrance-exits stand alone. The random distribution functions are calculated to form envelope lines (significance level 95%). Then the empirical calculation results are compared to the envelope lines, in order to judge the distribution (over the envelope lines, cluster; inside, random; below, dispersion) and indicate the influence of Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3243 6 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 7 auto correlation analysis gives verdict to whether there exists crime cluster dominated by BART in blocks which are station entrance-exits on crimes. near or containing BART stations. Crimes in San Francisco around all stations’ all entrance-exits are calculated according to the formula. On one (1)Social economic element analysis. hand, it is sufficient to make a total statistic of every entrance-exit in each station. On the other hand, the result will The social, economic and demographic features of blocks are collected to make factor analysis and multiple not only be influenced by other city in case of large statistic scope, but also will make repeated calculation of crimes regression analysis. around other stations. Therefore, the catchment area of road network should be determined at the beginning, (2)Space auto correlation analysis. calculating general K function of each station’s every entrance-exit with the valid range of 600 meters. The formula Since there are spatial correlations between crimes, Moran’s I index is implemented to analyze auto correlation, is: divided into global auto correlation and local auto correlation. nA t Global auto correlation describes the average correlation between crimes in San Francisco, expressed as å n( ) following formula: t 1 i =1 p A i K ( ) = (10) n n n n AB p r n w (x x )(x x ) w z z A i B A å å ij i - j - å å ij i j n i =1 j =1 n i =1 j =1 where nA stands for the entrance-exit collection of the station. In order to secure the validity and independence, I = = ( ) n n 7 crimes within other station’s hinterland are not considered. S S 2 0 (x x )2 0 z å i - å i i =1 i =1 4. Macro level Analysis n n S w where wij stands for the spatial weight between crime i and j , 0 = å å ij is the total of all weights. 4.1. Crime Review i =1 j =1 4.1.1. External Contrast z = (x - x ), z = (x - x ) . Moran’s I is within [-1,1]. On certain significance level, crimes i i j j San Francisco city covers 121 km2 (U. S. Census Bureau, 2011) with a population of 837 thousands in 2013 (U. are positively correlated in case of Moran’s I’ plus, negatively correlated in case of minus and random distribution S. Census Bureau, 2012), ranking 14th among the 279 cities over 100 thousands population in the U.S. In 2012, total in case of around 0. crimes in San Francisco ranks 14th and the crime rate is 75th (FBI, 2012), higher than national average crime rate, in Local auto correlation serves to analyze the relationship among crimes near BART stations and regional which sex offenses and assault obviously lower than the average, while robbery, larceny and vehicle theft difference. Both local indicators of spatial association and Moran scatter diagram are employed here. The former is significantly higher than the average. Provided the motive of robbery is property, the property crime comes in large expressed by Moran’s Ii, standing for local Moran’s I, in the following formula: amounts, while the violence crime is relatively fewer in this city. z w z ( ) I i= i å ij j 8 j 4.1.2. Internal Contrast The observed value has been transformed to standardization, whilst row-standardized form is employed in spatial The quantity of crimes is listed in table 1, which indicates that larceny, assault, drug, vandalism, vehicle theft and weights. The results and Moran scatter diagram are integrated to make intuitive recognition through Moran burglary take up a bigger proportion. significance level diagram. In the software of Geoda, local spatial relationships are generated in 5 classifications, Table 1. Quantity of Crimes in the Study which are high criminal region encircled by high criminal region, high criminal region encircled by low criminal region, low criminal region encircled by high criminal region, low criminal region encircle by low criminal region Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Total and not significant space relationship. Assault 13162 11864 12443 12518 12667 12257 10851 10262 9801 10369 116194 Kidnapping 290 298 318 345 343 478 347 406 267 513 3605 3.2.3. Micro Level Analysis Robbery 3415 3615 4131 4027 4229 3578 3304 3348 3904 4156 37707 The study on micro level will go deep into the streets in blocks to analyze the criminal accessibility of specific Sex offenses 775 722 588 627 670 661 566 576 573 587 6345 spots. This helps to find the law of criminal clusters around BART stations. The cluster of different crimes around BART stations entrance-exit could be calculated through bivariate network K-function (Okabe et al., 2002) on the Burglary 6777 7096 7004 5453 5678 5377 4790 4783 5962 5844 58764 standard accessibility of road network with following formula: Fraud 3076 2766 2591 2344 2554 2496 2465 2541 2469 2577 25879 t 1 t Larceny 24822 25623 27347 25765 25796 25543 23880 25123 29979 35646 269524 ( ) K AB( ) = n( ) 9 Vehicle Theft 8639 8696 7291 6459 6052 5182 4350 4750 6096 6207 63722 p Ai rB p Ai Arson 280 236 240 246 248 222 206 194 228 228 2328 t where n( ) stands for the amount of criminal type B in the shortest distance t from the ith BART station Drug 10014 8527 8909 10351 11462 11744 9034 6724 6230 6682 89677 p A i Prostitution 1551 1112 1287 1862 1663 1460 637 445 269 269 10555 Vandalism 6563 7068 7688 7566 7342 7603 7604 6910 7503 6653 72500 entrance-exit(pA), r B for the density of station entrance-exits in road network. It is supposed that the crimes and Total 79364 77623 79837 77563 78704 76601 68034 66062 73281 79731 756800 station entrance-exits stand alone. The random distribution functions are calculated to form envelope lines (significance level 95%). Then the empirical calculation results are compared to the envelope lines, in order to judge the distribution (over the envelope lines, cluster; inside, random; below, dispersion) and indicate the influence of 3244 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 8 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 9

Criminal types have different tendencies in recent 5 years. Assault, drug and prostitution declined obviously. It can be found that most criminal types change little in recent years in either mean center or standard deviation Robbery, larceny, vehicle theft and burglary fluctuated. Sex offenses decreased stably. Vandalism and fraud oval. The more amount crimes have, the more stably they are distributed. Therefore the mean center and standard changed little. deviation oval of 10 years’ sum of each criminal type and that of BART stations are contrasted in figure 3. It The correlation analysis of annual change of crimes by the software of SPSS shows that most criminal types are indicates that violence crimes have similar mean centers and oval shapes. The mean centers are close to that of significantly correlated to sex offenses, as many as 6 types. 5 criminal types are correlated to arson and assault. BART stations and also is the spread direction. As far as property crimes are concerned, person-targeted crimes are Robbery, kidnapping, larceny are independent, with little correlation to other criminal types. Vandalism and fraud concentrated in downtown area, while the others have universe distribution, both of whose extension direction have negative correlation with sex offenses, which indicates the possibility of criminal transfer. similar to BART stations. As to other crimes, drug and prostitution have smallest scope with almost coincide mean center, showing the same result as the above correlation analysis of annual change. 4.2. Mean Center and Standard Deviation Oval

4.2.1. Mean Center and Standard Deviation Oval of All Crimes The mean center and standard deviation oval of BART stations, crimes in the study and all crimes in San Francisco are showed in figure 2. It is obvious that there is little distinction between crimes chosen in this research and all crimes in San Francisco, for mean center and standard deviation oval almost coincide, confirming the accordance and stability of their distribution. Thus, the crime chosen in the study is typical. Mean center of crimes is located in north-east city region, close to mean center of BART stations, around 16th and Mission Station. The long axis of oval extends along northeast-southwest, resembling to the oval long axis direction of BART stations, which shows the closely similar extension direction of crimes and BART stations.

Fig.3. Mean Centers and Standard Directional Ellipses of Various Criminal Types in the Study and BART Stations

4.3. The Nearest Neighborhood Index

As the road network in San Francisco mainly features grids, Manhattan Distance is employed in the nearest dots calculating. According to the calculation, the cluster of crimes has changed little in recent 10 years. Therefore, only the index of 2012 is listed in table 2 (prostitution data listed in 2011 for great change in 2012 and 2013), divided to not significant, weak, general, significant cluster.

Table 2. Nearest Neighborhood Index of Crimes in the Study in 2012 Fig.2. Mean Centers and Standard Directional Ellipses of BART Stations, Crimes in the Study and All Crimes in San Francisco Criminal Types R z p Cluster Levels Assault 0.448532 -104.444617 0.000000 General 4.2.2. Mean Center and Standard Deviation Oval of Certain Criminal Type Kidnapping 0.954827 -1.412087 0.157924 Not Significant Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3245 8 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 9

Criminal types have different tendencies in recent 5 years. Assault, drug and prostitution declined obviously. It can be found that most criminal types change little in recent years in either mean center or standard deviation Robbery, larceny, vehicle theft and burglary fluctuated. Sex offenses decreased stably. Vandalism and fraud oval. The more amount crimes have, the more stably they are distributed. Therefore the mean center and standard changed little. deviation oval of 10 years’ sum of each criminal type and that of BART stations are contrasted in figure 3. It The correlation analysis of annual change of crimes by the software of SPSS shows that most criminal types are indicates that violence crimes have similar mean centers and oval shapes. The mean centers are close to that of significantly correlated to sex offenses, as many as 6 types. 5 criminal types are correlated to arson and assault. BART stations and also is the spread direction. As far as property crimes are concerned, person-targeted crimes are Robbery, kidnapping, larceny are independent, with little correlation to other criminal types. Vandalism and fraud concentrated in downtown area, while the others have universe distribution, both of whose extension direction have negative correlation with sex offenses, which indicates the possibility of criminal transfer. similar to BART stations. As to other crimes, drug and prostitution have smallest scope with almost coincide mean center, showing the same result as the above correlation analysis of annual change. 4.2. Mean Center and Standard Deviation Oval

4.2.1. Mean Center and Standard Deviation Oval of All Crimes The mean center and standard deviation oval of BART stations, crimes in the study and all crimes in San Francisco are showed in figure 2. It is obvious that there is little distinction between crimes chosen in this research and all crimes in San Francisco, for mean center and standard deviation oval almost coincide, confirming the accordance and stability of their distribution. Thus, the crime chosen in the study is typical. Mean center of crimes is located in north-east city region, close to mean center of BART stations, around 16th and Mission Station. The long axis of oval extends along northeast-southwest, resembling to the oval long axis direction of BART stations, which shows the closely similar extension direction of crimes and BART stations.

Fig.3. Mean Centers and Standard Directional Ellipses of Various Criminal Types in the Study and BART Stations

4.3. The Nearest Neighborhood Index

As the road network in San Francisco mainly features grids, Manhattan Distance is employed in the nearest dots calculating. According to the calculation, the cluster of crimes has changed little in recent 10 years. Therefore, only the index of 2012 is listed in table 2 (prostitution data listed in 2011 for great change in 2012 and 2013), divided to not significant, weak, general, significant cluster.

Table 2. Nearest Neighborhood Index of Crimes in the Study in 2012 Fig.2. Mean Centers and Standard Directional Ellipses of BART Stations, Crimes in the Study and All Crimes in San Francisco Criminal Types R z p Cluster Levels Assault 0.448532 -104.444617 0.000000 General 4.2.2. Mean Center and Standard Deviation Oval of Certain Criminal Type Kidnapping 0.954827 -1.412087 0.157924 Not Significant 3246 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 10 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 11

Robbery 0.574817 -50.823182 0.000000 General neighborhood attributes (table 3). Metric of Kaiser-Meyer-Olkin is 0.757, supporting the appropriate factor analysis. Sex offenses 0.780318 -10.060095 0.000000 Weak The account accumulation of factor 1 is 46.58%, while that of factor 2 is 21.34%. The sum is 67.92%. Burglary 0.695653 -44.956844 0.000000 Weak Factor 1 featuring poverty, congestion, lack of self-occupation attract, represents the disadvantageous social- economic situation. Factor 2 standing for the proportion of the higher education and employment, represents the Fraud 0.707738 -27.782053 0.000000 Weak population quality and the favorable humanity situation. Larceny 0.416992 -193.114021 0.000000 General Vehicle Theft 0.73106 -40.170581 0.000000 Weak 5.1.2. Multiple regression analysis Arson 0.930738 -2.000746 0.04542 Not Significant The total of 10 years crimes is taken as dependent variable. BART stations (including each block group if Drug 0.25542 -112.431076 0.000000 Significant entrance-exits belong to different block groups since the stations are always at road intersections), 2 principal Prostitution 0.16735 -33.602621 0.000000 Significant factors, other low mutual-correlated factors are taken as independent variable. The method of enter is implemented in the multiple regression analysis. The value of F is significant in the regression equation and the collinearity Vandalism 0.67426 -53.978381 0.000000 Weak statistic VIF value is less than 2, manifesting the collinearity within permissible range.

The long axis angles of standard deviation ovals of significant and general types are similar to BART stations Table 4. Multiple Linear Regression of Crimes in the Study except prostitution, indicating the possible influence of the BART stations and necessity of further analysis on the Coefficient t Sig lower level. Constant 7.674 .583 .560

5. Meso-level Analysis BART Station 258.801 5.651 .000 Factor1 21.886 2.432 .015 5.1. Social and Economic Factors Analysis Factor2 -16.643 -1.941 .053 Recreation and Community Centers 27.851 2.950 .003 5.1.1. Factor Analysis Schools (High School or Below) -6.225 -.703 .482 Block group level data are implemented in the analysis. Most social-economic data censuses are on block group Health Facilities 51.615 3.182 .002 level in the U.S., which are detailed enough to analyze the main factors of crimes and universally employed in most city studies. Data such as population density, poverty rate, employment rate, renter rate, median contract rent, no Banks and Financial Facilities -26.899 -6.162 .000 vehicle owned rate, percentage of owners moved in recent 5 years, percentage of education below high school, Food Vendors 11.277 9.298 .000 number of recreation and community centers, number of schools (high school or below), number of health facilities, Alcohol Outlets 7.358 1.012 .312 number of banks and financial facilities, number of alcohol outlets and number of food vendors are all obtained from U.S. Census Bureau (year of 2009), official website of San Francisco Municipal Government and libraries of The result in table 4 is in favor of the positive correlation between BART stations and crimes. Factor 1, factor 2, UCB. recreation and community centers, health facilities, food vendors have significantly positive correlation with crimes. Banks and financial facilities appear to be negatively correlated due to safeguard measures and monitoring Table 3. Matrix of Factor Components equipment. It is a little surprising that there is not significant correlation between alcohol outlets and crimes, the possible reason of which is no drinking inside transferring drunken issues elsewhere. The contribution to crimes of Factor Components BART stations exceeds other independent variables remarkably according to the coefficient. Each BART station 1 2 entrance-exit’s influence on crime is equivalent to 5 health facilities, or 9 recreation centers, or 23 food vendors. Population Density .660 .274 Provided each criminal type is independent, multiple regression analysis of 10 year’s sum of each criminal type is Poverty Rate .793 -.058 made through the above method. The result shows that independent variables have different degrees of influence on Renter Rate .648 .649 different criminal types. BART stations have outstanding positive correlation with almost all criminal types except fraud, arson and prostitution and make greatest contribution to assault, kidnapping, robbery, larceny and drug, Median Contract Rent -.777 .248 indicating probable to occur around stations. The result is consistent with the hypothesis of cluster reasons according No Vehicle Owned Rate .828 .364 to nearest neighborhood index in preceding macro level analysis. Food vendor is another remarkable correlative Percentage of Owners Moved in Recent 5 Years -.642 -.035 factor to many types of crimes but much lower correlative coefficient than BART. Other conclusions include: poor Employment Rate -.377 .798 community environment tends to foster violent crimes and drug; communities of high quality inhabitants restrain Percentage of Education below High School .630 -.613 violent crimes, person-targeted property crime and vandalism; banks and financial facilities bring down most crimes; recreation and community centers tend to produce violent crimes, prostitution and vandalism, but less

property crimes; there is few property around crimes health facilities but more around alcohol outlets. In general, the These social and economic factors could be divided into two parts. One part is the attribute of neighborhood accessibility of BART will undoubtedly attract diverse facilities, leading to mixed layout of land use and causing all based on residential characteristics; the other part is the attribute of activities based on the number of facilities. As sorts of complex activities. The crimes around BART stations will arise due to these activities and result from a far as downtown area is concerned, activities of employment and relaxation dominate rather than resident. Kernel series of criminal interactions such as promotion, elimination and transfer. The result will further exert an influence principal component extraction is employed in SPSS to combine similar factors into one type which are mainly upon the attribute of neighborhood and bring changes of crimes. Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3247 10 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 11

Robbery 0.574817 -50.823182 0.000000 General neighborhood attributes (table 3). Metric of Kaiser-Meyer-Olkin is 0.757, supporting the appropriate factor analysis. Sex offenses 0.780318 -10.060095 0.000000 Weak The account accumulation of factor 1 is 46.58%, while that of factor 2 is 21.34%. The sum is 67.92%. Burglary 0.695653 -44.956844 0.000000 Weak Factor 1 featuring poverty, congestion, lack of self-occupation attract, represents the disadvantageous social- economic situation. Factor 2 standing for the proportion of the higher education and employment, represents the Fraud 0.707738 -27.782053 0.000000 Weak population quality and the favorable humanity situation. Larceny 0.416992 -193.114021 0.000000 General

Vehicle Theft 0.73106 -40.170581 0.000000 Weak 5.1.2. Multiple regression analysis Arson 0.930738 -2.000746 0.04542 Not Significant The total of 10 years crimes is taken as dependent variable. BART stations (including each block group if Drug 0.25542 -112.431076 0.000000 Significant entrance-exits belong to different block groups since the stations are always at road intersections), 2 principal Prostitution 0.16735 -33.602621 0.000000 Significant factors, other low mutual-correlated factors are taken as independent variable. The method of enter is implemented in the multiple regression analysis. The value of F is significant in the regression equation and the collinearity Vandalism 0.67426 -53.978381 0.000000 Weak statistic VIF value is less than 2, manifesting the collinearity within permissible range.

The long axis angles of standard deviation ovals of significant and general types are similar to BART stations Table 4. Multiple Linear Regression of Crimes in the Study except prostitution, indicating the possible influence of the BART stations and necessity of further analysis on the Coefficient t Sig lower level. Constant 7.674 .583 .560

5. Meso-level Analysis BART Station 258.801 5.651 .000 Factor1 21.886 2.432 .015 5.1. Social and Economic Factors Analysis Factor2 -16.643 -1.941 .053 Recreation and Community Centers 27.851 2.950 .003 5.1.1. Factor Analysis Schools (High School or Below) -6.225 -.703 .482 Block group level data are implemented in the analysis. Most social-economic data censuses are on block group Health Facilities 51.615 3.182 .002 level in the U.S., which are detailed enough to analyze the main factors of crimes and universally employed in most city studies. Data such as population density, poverty rate, employment rate, renter rate, median contract rent, no Banks and Financial Facilities -26.899 -6.162 .000 vehicle owned rate, percentage of owners moved in recent 5 years, percentage of education below high school, Food Vendors 11.277 9.298 .000 number of recreation and community centers, number of schools (high school or below), number of health facilities, Alcohol Outlets 7.358 1.012 .312 number of banks and financial facilities, number of alcohol outlets and number of food vendors are all obtained from U.S. Census Bureau (year of 2009), official website of San Francisco Municipal Government and libraries of The result in table 4 is in favor of the positive correlation between BART stations and crimes. Factor 1, factor 2, UCB. recreation and community centers, health facilities, food vendors have significantly positive correlation with crimes. Banks and financial facilities appear to be negatively correlated due to safeguard measures and monitoring Table 3. Matrix of Factor Components equipment. It is a little surprising that there is not significant correlation between alcohol outlets and crimes, the possible reason of which is no drinking inside transferring drunken issues elsewhere. The contribution to crimes of Factor Components BART stations exceeds other independent variables remarkably according to the coefficient. Each BART station 1 2 entrance-exit’s influence on crime is equivalent to 5 health facilities, or 9 recreation centers, or 23 food vendors. Population Density .660 .274 Provided each criminal type is independent, multiple regression analysis of 10 year’s sum of each criminal type is Poverty Rate .793 -.058 made through the above method. The result shows that independent variables have different degrees of influence on Renter Rate .648 .649 different criminal types. BART stations have outstanding positive correlation with almost all criminal types except fraud, arson and prostitution and make greatest contribution to assault, kidnapping, robbery, larceny and drug, Median Contract Rent -.777 .248 indicating probable to occur around stations. The result is consistent with the hypothesis of cluster reasons according No Vehicle Owned Rate .828 .364 to nearest neighborhood index in preceding macro level analysis. Food vendor is another remarkable correlative Percentage of Owners Moved in Recent 5 Years -.642 -.035 factor to many types of crimes but much lower correlative coefficient than BART. Other conclusions include: poor Employment Rate -.377 .798 community environment tends to foster violent crimes and drug; communities of high quality inhabitants restrain Percentage of Education below High School .630 -.613 violent crimes, person-targeted property crime and vandalism; banks and financial facilities bring down most crimes; recreation and community centers tend to produce violent crimes, prostitution and vandalism, but less

property crimes; there is few property around crimes health facilities but more around alcohol outlets. In general, the These social and economic factors could be divided into two parts. One part is the attribute of neighborhood accessibility of BART will undoubtedly attract diverse facilities, leading to mixed layout of land use and causing all based on residential characteristics; the other part is the attribute of activities based on the number of facilities. As sorts of complex activities. The crimes around BART stations will arise due to these activities and result from a far as downtown area is concerned, activities of employment and relaxation dominate rather than resident. Kernel series of criminal interactions such as promotion, elimination and transfer. The result will further exert an influence principal component extraction is employed in SPSS to combine similar factors into one type which are mainly upon the attribute of neighborhood and bring changes of crimes. 3248 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 12 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 13

5.2. Spatial Auto Correlation Analysis Table 5. Types of Blocks around BART Stations High-High High-Low Low-High Low-Low Not Significant Unit is refined to block level to make spatial auto correlation analysis of criminal density, revealing the cluster Blocks Containing BART entrance-exits 17 4 0 0 4 tendency of crimes. Firstly, global auto correlation is made in Geoda software. The neighborhood relations between blocks are handled by distance-based weight matrix. The weight is defined by distance threshold value. If the First Level Blocks (Queen Contiguity) 27 2 11 13 12 straight-line distance between the polygon geometric centers of the two blocks is in the range of threshold value, the Second Level Blocks (Queen Contiguity) 54 1 21 25 23 weight is 1. Otherwise, 0. Due to relatively longer station distance of BART lines outside downtown area, the Third Level Blocks (Queen Contiguity) 55 1 53 43 53 distance is defined as 600 meters, a little longer than common subway station walking distance - 500 meters. Global auto correlation result is Moran’I= 0.264, revealing that the cluster exists, the same as the result of nearest The blocks containing BART entrance-exits have a very high proportion of high-high and high-low relationships, neighborhood index on macro level analysis. surpassing the average proportion of city remarkably. In detail, with the increasing distance from the station in Secondly, local auto correlation is made with 999 permutations and significance level p=0.05. Local spatial downtown, local spatial relationships change from high-high to low-high to no significant gradually. As far as areas relationship of criminal density is generated (figure 4). There are 5 classifications: high-high, high-low, low-high, outside downtown are considered, they change from high-low to low-low to no significant rapidly, manifesting the low-low and no significant. The high-high blocks distribute near downtown and apparently extend along BART high criminal density in BART station blocks and the decline along with the distance from the stations, whether or lines to the south. not in downtown area. Therefore, it can be inferred that BART stations have notably positive influence on block’s criminal clusters. It should be pointed that this spatial relationship of criminal density is quite different from criminal density, and also has nothing to do with land use intensity. The maximum criminal density is in the blocks to the north of civic center (not exactly the blocks with maximum land use density) and much higher than most of the station blocks. But these station blocks are still higher than nearby blocks, forming diminishing circles. This shows that whether or not the station block has the highest criminal density in the city, it must have the highest criminal density among its service scope. Hence, it can be seen from figure 4 there is an obvious turn of density relationship along with the BART route, from Market Street to Mission Street, which is consistent with the overall distribution in preceding macro level analysis.

6. Micro Level Analysis

Influence region of road network is confirmed by means of plug-in program named SANET in GIS to begin the micro level analysis. The crimes within 600 meters from BART stations along the road are calculated in bivariate network K (cross K in the figure) function by SANET. There is a similarity in different years among clusters. The criminal types in large numbers tend to have stronger similarity because of less accidental choices for crime sites. The farther crimes are away from stations, the weaker similarity appears. Figure 5 shows the result of fraud around Powell Street Station in 2011, 2012 and 2013. There is an obviously similar trend of crime clusters in these years, which shows a random distribution around entrance-exit and a first sudden rise over the envelope line at 20 meters (feet unit in the figure), and simultaneously other sudden rises at 70 meters, 135 meters and 290 meters. Thus, data in 2013 are selected to make analysis, defining the minimum distance unit as 5 meters. Since small quantity of crimes doesn’t contribute to effective statistic, 10 years’ data of certain criminal type are implemented, shown with “*” in the table. Table 6 to 13 shows the results.

Fig.4. Local Spatial Relationship of Crime Density

Types of blocks around BART stations are classified. The neighborhood level is made clear by means of Queen Contiguity, namely, blocks with adjacent dots and lines are regarded as the neighborhood blocks, for the move of crime is along the roads, exerting influence on blocks sharing crossroads. Given the blocks in San Francisco are rectangles with different width and length, the periphery is valued 600 meters, avoiding the distortion of large blocks such as parks. The result is showed in table 5. Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3249 12 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 13

5.2. Spatial Auto Correlation Analysis Table 5. Types of Blocks around BART Stations High-High High-Low Low-High Low-Low Not Significant Unit is refined to block level to make spatial auto correlation analysis of criminal density, revealing the cluster Blocks Containing BART entrance-exits 17 4 0 0 4 tendency of crimes. Firstly, global auto correlation is made in Geoda software. The neighborhood relations between blocks are handled by distance-based weight matrix. The weight is defined by distance threshold value. If the First Level Blocks (Queen Contiguity) 27 2 11 13 12 straight-line distance between the polygon geometric centers of the two blocks is in the range of threshold value, the Second Level Blocks (Queen Contiguity) 54 1 21 25 23 weight is 1. Otherwise, 0. Due to relatively longer station distance of BART lines outside downtown area, the Third Level Blocks (Queen Contiguity) 55 1 53 43 53 distance is defined as 600 meters, a little longer than common subway station walking distance - 500 meters. Global auto correlation result is Moran’I= 0.264, revealing that the cluster exists, the same as the result of nearest The blocks containing BART entrance-exits have a very high proportion of high-high and high-low relationships, neighborhood index on macro level analysis. surpassing the average proportion of city remarkably. In detail, with the increasing distance from the station in Secondly, local auto correlation is made with 999 permutations and significance level p=0.05. Local spatial downtown, local spatial relationships change from high-high to low-high to no significant gradually. As far as areas relationship of criminal density is generated (figure 4). There are 5 classifications: high-high, high-low, low-high, outside downtown are considered, they change from high-low to low-low to no significant rapidly, manifesting the low-low and no significant. The high-high blocks distribute near downtown and apparently extend along BART high criminal density in BART station blocks and the decline along with the distance from the stations, whether or lines to the south. not in downtown area. Therefore, it can be inferred that BART stations have notably positive influence on block’s criminal clusters. It should be pointed that this spatial relationship of criminal density is quite different from criminal density, and also has nothing to do with land use intensity. The maximum criminal density is in the blocks to the north of civic center (not exactly the blocks with maximum land use density) and much higher than most of the station blocks. But these station blocks are still higher than nearby blocks, forming diminishing circles. This shows that whether or not the station block has the highest criminal density in the city, it must have the highest criminal density among its service scope. Hence, it can be seen from figure 4 there is an obvious turn of density relationship along with the BART route, from Market Street to Mission Street, which is consistent with the overall distribution in preceding macro level analysis.

6. Micro Level Analysis

Influence region of road network is confirmed by means of plug-in program named SANET in GIS to begin the micro level analysis. The crimes within 600 meters from BART stations along the road are calculated in bivariate network K (cross K in the figure) function by SANET. There is a similarity in different years among clusters. The criminal types in large numbers tend to have stronger similarity because of less accidental choices for crime sites. The farther crimes are away from stations, the weaker similarity appears. Figure 5 shows the result of fraud around Powell Street Station in 2011, 2012 and 2013. There is an obviously similar trend of crime clusters in these years, which shows a random distribution around entrance-exit and a first sudden rise over the envelope line at 20 meters (feet unit in the figure), and simultaneously other sudden rises at 70 meters, 135 meters and 290 meters. Thus, data in 2013 are selected to make analysis, defining the minimum distance unit as 5 meters. Since small quantity of crimes doesn’t contribute to effective statistic, 10 years’ data of certain criminal type are implemented, shown with “*” in the table. Table 6 to 13 shows the results.

Fig.4. Local Spatial Relationship of Crime Density

Types of blocks around BART stations are classified. The neighborhood level is made clear by means of Queen Contiguity, namely, blocks with adjacent dots and lines are regarded as the neighborhood blocks, for the move of crime is along the roads, exerting influence on blocks sharing crossroads. Given the blocks in San Francisco are rectangles with different width and length, the periphery is valued 600 meters, avoiding the distortion of large blocks such as parks. The result is showed in table 5. 3250 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 14 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 15

Vandalism 50,130 - 0,90 Sudden rise at 130 meters

Embarcadero station is located on the fringe of downtown, featuring random distributions of vehicle theft and prostitution, and clusters of other criminal types around entrance-exits. The peaks of crimes appear at about 50 and 130 meters away from entrance-exits. Many criminal types increase dramatically at about 130 meters.

Table 7. Clusters of Various Criminal Types around Montgomery St Station Cluster Dispersion Random Remarks Assault 0 - 185 Sudden rise at 90 meters Kidnapping* - - 0 Robbery 0 - - Sudden rise at 80 meters Sex offenses* 90 0,165 Sudden rise at 90 meters Burglary 410,495 0,450 Sudden rise at 495 meters Fraud 90 - 0 Larceny 0 - 170 Sudden rise at 90 meters Vehicle Theft - 245 0 Arson 25 - 0,30 Drug 0 - 130 Weak cluster Prostitution* 85,270 0,145,300 Vandalism 0 - 115 Weak cluster

Montgomery St station is in downtown. Almost all crimes cluster near station entrance-exits except robbery and vehicle theft. Many clusters form at the entrance-exits, with a peak around 80 meters.

Table 8. Clusters of Various Criminal Types around Powell St Station Cluster Dispersion Random Remarks Assault 15 - 0 Kidnapping - - 0 Robbery 20 - 0 Sudden rises at 20 and 135 meters Sex offenses 140 0,450 Fig.5. Observed and Expected Cross K Function Curves of Frauds around Powell St Station (2011-2013) Burglary 20 - 0 Sudden rises at 30 and 300 meters Fraud 20 - 0 Sudden rises at 20 and 70 meters Table 6. Clusters of Various Criminal Types around Larceny 20 - 0 Sudden rises at 20 and 135 meters Cluster Dispersion Random Remarks Vehicle Theft - 105 0 Assault 60 - 0 Sudden rise at 135 meters Arson - - 0 Kidnapping* 130 - 0,195 Weak cluster Drug 80,440 - 0,160 Sudden rise at 440 meters Robbery 50 - 0 Sudden rise at 90 meters Prostitution* - - 0 Sex offenses* 130 - 0,240 Sudden rise at 130 meters Vandalism 30 - 0,195 Burglary 125 - 0,145 Weak cluster

Fraud 130 - 0 Sudden rises at 125 and 180 meters Powell St station is in downtown. Clusters appear at the entrance-exits except robbery, vehicle theft, arson and Larceny 50,135,355 - 0,65,235 Sudden rises at 125 and 355 meters prostitution. Two respective criminal peaks are around 20 and 135 meters. Vehicle Theft - - 0 Table 9. Clusters of Various Criminal Types around Civic Center Station Arson* 10 - 0,75 Weak cluster Cluster Dispersion Random Remarks Drug 10 - 0,375 Sudden rise at 125 meters , , , Prostitution* - - 0 Assault 25 145 - 0 90 360 Sudden rise at 145meters Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3251 14 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 15

Vandalism 50,130 - 0,90 Sudden rise at 130 meters

Embarcadero station is located on the fringe of downtown, featuring random distributions of vehicle theft and prostitution, and clusters of other criminal types around entrance-exits. The peaks of crimes appear at about 50 and 130 meters away from entrance-exits. Many criminal types increase dramatically at about 130 meters.

Table 7. Clusters of Various Criminal Types around Montgomery St Station Cluster Dispersion Random Remarks Assault 0 - 185 Sudden rise at 90 meters Kidnapping* - - 0 Robbery 0 - - Sudden rise at 80 meters Sex offenses* 90 0,165 Sudden rise at 90 meters Burglary 410,495 0,450 Sudden rise at 495 meters Fraud 90 - 0 Larceny 0 - 170 Sudden rise at 90 meters Vehicle Theft - 245 0 Arson 25 - 0,30 Drug 0 - 130 Weak cluster Prostitution* 85,270 0,145,300 Vandalism 0 - 115 Weak cluster

Montgomery St station is in downtown. Almost all crimes cluster near station entrance-exits except robbery and vehicle theft. Many clusters form at the entrance-exits, with a peak around 80 meters.

Table 8. Clusters of Various Criminal Types around Powell St Station Cluster Dispersion Random Remarks Assault 15 - 0 Kidnapping - - 0 Robbery 20 - 0 Sudden rises at 20 and 135 meters Sex offenses 140 0,450 Fig.5. Observed and Expected Cross K Function Curves of Frauds around Powell St Station (2011-2013) Burglary 20 - 0 Sudden rises at 30 and 300 meters Fraud 20 - 0 Sudden rises at 20 and 70 meters Table 6. Clusters of Various Criminal Types around Embarcadero Station Larceny 20 - 0 Sudden rises at 20 and 135 meters Cluster Dispersion Random Remarks Vehicle Theft - 105 0 Assault 60 - 0 Sudden rise at 135 meters Arson - - 0 Kidnapping* 130 - 0,195 Weak cluster Drug 80,440 - 0,160 Sudden rise at 440 meters Robbery 50 - 0 Sudden rise at 90 meters Prostitution* - - 0 Sex offenses* 130 - 0,240 Sudden rise at 130 meters Vandalism 30 - 0,195 Burglary 125 - 0,145 Weak cluster

Fraud 130 - 0 Sudden rises at 125 and 180 meters Powell St station is in downtown. Clusters appear at the entrance-exits except robbery, vehicle theft, arson and Larceny 50,135,355 - 0,65,235 Sudden rises at 125 and 355 meters prostitution. Two respective criminal peaks are around 20 and 135 meters. Vehicle Theft - - 0 Table 9. Clusters of Various Criminal Types around Civic Center Station Arson* 10 - 0,75 Weak cluster Cluster Dispersion Random Remarks Drug 10 - 0,375 Sudden rise at 125 meters , , , Prostitution* - - 0 Assault 25 145 - 0 90 360 Sudden rise at 145meters 3252 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 16 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 17

Kidnapping - - 0 Arson* 240 - 0,350 Robbery 140 - 0 Sudden rise at 150 meters Drug 20 - 0 Sudden rise at 35 meters Sex offenses - - 0 Prostitution* - 160 0 Burglary 150 - 0,300 Weak cluster Vandalism 15 - 0 Fraud 45,150 - 0,80,500 Larceny 150 - 0 Sudden rise at 150 meters All crimes cluster around entrance-exits except vehicle theft and prostitution, generating a peak at 15 meters. Vehicle Theft - 115 0,510 Table 12. Clusters of Various Criminal Types around Arson - - 0 Cluster Dispersion Random Remarks Drug 270 45 0 Sudden rise at 270 meters Assault 65 340 0,155 Prostitution* 225 - 0,390 Sudden rise at 225 meters Kidnapping* - 365 0 Vandalism - - 0 Robbery 0 - 0,390 Sudden rises at 65 and 270 meters

Sex offenses* 65 495 0,180 Sudden rise at 65 meters Civic Center station is in downtown. It is to the west of city hall and the municipal square equipped with some , monitoring devices and two blocks away from the northern regional police station. The statistical result is Burglary 0 - 0 240 influenced, putting up fewer crimes in the northwest. However, some criminal types still cluster around entrance- Fraud 70 - 0,370 Sudden rise at 70 meters exits, forming a peak at 145 meters. Larceny 0 555 0,330 Sudden rise at 65 meters Vehicle Theft 65 375 0,210 Table 10. Clusters of Various Criminal Types around 16th & Mission Station Arson* 60 - 0,150 Cluster Dispersion Random Remarks Drug - 450 0 Assault 20 - 0 Sudden rise at 20 meters Prostitution - - - Insufficient sample Kidnapping 90 - 0 Sudden rise at 90 meters Vandalism 60 - 0,330 Sudden rise at 105 meters Robbery 15 - 0 Sudden rises at 20 and 120 meters

Sex offenses 90 - 0 Sudden rises at 90 and 150 meters Most crimes cluster around Glen Park Station in certain scope and randomly distribute or disperse out of the , Burglary 135 - 0 445 Sudden rise at 135 meters scope, proving that the concentration area is around station. Crimes cluster except robbery, drug and prostitution, Fraud 20 - 0 Sudden rise at 20 meters forming a peak at 65 meters. Larceny 20 - 0 Sudden rise at 20 meters Table 13. Clusters of Various Criminal Types around Vehicle Theft - - 0 Cluster Dispersion Random Remarks Arson* 20 - 0 Sudden rise at 120 meters Assault - 355 0 Drug 15 - 0 Sudden rises at 20 and 130 meters Kidnapping* 35 - 0,40 Prostitution* 495 0,530 510 Sudden rise at 495 meters Robbery 15 - 0,355 Sudden rises at 15 and 175 meters Vandalism 20 - 0 Sex offenses* 50 265 0,55

, 16th & Mission station is outside downtown. All the crimes except vehicle theft cluster close to the entrance- Burglary - 465 0 545 exits, forming two obvious peaks at 20 and 120 meters. Fraud - - 0 Larceny 20 185 0,75 Table 11. Clusters of Various Criminal Types around 24th & Mission Station Vehicle Theft - - 0 Cluster Dispersion Random Remarks Arson* - 285 0 Assault 0 - - Sudden rise at 15 meters Drug - - 0 Kidnapping 100,370 - 0,260,480 Prostitution - - - Insufficient sample Robbery 0 - - Sudden rise at 15 meters Vandalism 50,175 435 0,105,265 Sudden rise at 175 meters Sex offenses 80 - 0

, Burglary 20 - 0 30 Weak cluster Balboa Park station has a specific location with a freeway on the west, a rail car station yard on the east. The Fraud 35 - 0 Sudden rise at 35meters north of the station enjoys a wide view of sports park lawn, at the corner of which locates the regional police station. Larceny 15 - 0 Sudden rise at 15 meters The entrance-exits of this station are on the south and north. Most crimes are distributed in the southern blocks as Vehicle Theft - - 0 expected. Only five types cluster. Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3253 16 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 17

Kidnapping - - 0 Arson* 240 - 0,350 Robbery 140 - 0 Sudden rise at 150 meters Drug 20 - 0 Sudden rise at 35 meters Sex offenses - - 0 Prostitution* - 160 0 Burglary 150 - 0,300 Weak cluster Vandalism 15 - 0 Fraud 45,150 - 0,80,500 Larceny 150 - 0 Sudden rise at 150 meters All crimes cluster around entrance-exits except vehicle theft and prostitution, generating a peak at 15 meters. Vehicle Theft - 115 0,510 Table 12. Clusters of Various Criminal Types around Glen Park Station Arson - - 0 Cluster Dispersion Random Remarks Drug 270 45 0 Sudden rise at 270 meters Assault 65 340 0,155 Prostitution* 225 - 0,390 Sudden rise at 225 meters Kidnapping* - 365 0 Vandalism - - 0 Robbery 0 - 0,390 Sudden rises at 65 and 270 meters

Sex offenses* 65 495 0,180 Sudden rise at 65 meters Civic Center station is in downtown. It is to the west of city hall and the municipal square equipped with some , monitoring devices and two blocks away from the northern regional police station. The statistical result is Burglary 0 - 0 240 influenced, putting up fewer crimes in the northwest. However, some criminal types still cluster around entrance- Fraud 70 - 0,370 Sudden rise at 70 meters exits, forming a peak at 145 meters. Larceny 0 555 0,330 Sudden rise at 65 meters Vehicle Theft 65 375 0,210 Table 10. Clusters of Various Criminal Types around 16th & Mission Station Arson* 60 - 0,150 Cluster Dispersion Random Remarks Drug - 450 0 Assault 20 - 0 Sudden rise at 20 meters Prostitution - - - Insufficient sample Kidnapping 90 - 0 Sudden rise at 90 meters Vandalism 60 - 0,330 Sudden rise at 105 meters Robbery 15 - 0 Sudden rises at 20 and 120 meters

Sex offenses 90 - 0 Sudden rises at 90 and 150 meters Most crimes cluster around Glen Park Station in certain scope and randomly distribute or disperse out of the , Burglary 135 - 0 445 Sudden rise at 135 meters scope, proving that the concentration area is around station. Crimes cluster except robbery, drug and prostitution, Fraud 20 - 0 Sudden rise at 20 meters forming a peak at 65 meters. Larceny 20 - 0 Sudden rise at 20 meters Table 13. Clusters of Various Criminal Types around Balboa Park Station Vehicle Theft - - 0 Cluster Dispersion Random Remarks Arson* 20 - 0 Sudden rise at 120 meters Assault - 355 0 Drug 15 - 0 Sudden rises at 20 and 130 meters Kidnapping* 35 - 0,40 Prostitution* 495 0,530 510 Sudden rise at 495 meters Robbery 15 - 0,355 Sudden rises at 15 and 175 meters Vandalism 20 - 0 Sex offenses* 50 265 0,55

, 16th & Mission station is outside downtown. All the crimes except vehicle theft cluster close to the entrance- Burglary - 465 0 545 exits, forming two obvious peaks at 20 and 120 meters. Fraud - - 0 Larceny 20 185 0,75 Table 11. Clusters of Various Criminal Types around 24th & Mission Station Vehicle Theft - - 0 Cluster Dispersion Random Remarks Arson* - 285 0 Assault 0 - - Sudden rise at 15 meters Drug - - 0 Kidnapping 100,370 - 0,260,480 Prostitution - - - Insufficient sample Robbery 0 - - Sudden rise at 15 meters Vandalism 50,175 435 0,105,265 Sudden rise at 175 meters Sex offenses 80 - 0

, Burglary 20 - 0 30 Weak cluster Balboa Park station has a specific location with a freeway on the west, a rail car station yard on the east. The Fraud 35 - 0 Sudden rise at 35meters north of the station enjoys a wide view of sports park lawn, at the corner of which locates the regional police station. Larceny 15 - 0 Sudden rise at 15 meters The entrance-exits of this station are on the south and north. Most crimes are distributed in the southern blocks as Vehicle Theft - - 0 expected. Only five types cluster. 3254 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 18 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 19

Micro level analysis confirms the obvious cluster of crimes around BART stations both in and outside downtown, explore the relation between peak distances and spatial characteristics of blocks based on the widespread double including non-significant correlated-with-BART criminal types on the meso-level analysis. The reason of peaks of clusters; to predict and prevent crimes in view of the above results. uncorrelation lies in that the cluster only exists around certain stations according to the different attributes of Related studies can be extended to medium capacity transit systems such as MUNI in San Francisco; to larger neighborhood and activities. As to the stations outside downtown area, the influence of worsening public security is geographic scope such as the whole bay area spatially; to more vertical researches such as the change of public not limited to the nearby neighborhoods. Robbery and larceny cluster at all the 8 stations and all have peaks except security before and after the construction of a mass transit; to more cities based on various culture and social Balboa Park Station. Assault, sex offenses, fraud and vandalism cluster at 7 stations, featuring peaks except background to discover whether the similar criminal clusters exist and give insight into the detailed nature and vandalism. Burglary and drug cluster at 6 stations. Burglary clusters in a short distance from most stations, giving causes of crime. comfort to residents far away from stations. Arson and prostitution have more different standard deviation ovals with BART than other crimes and they only cluster around few stations, explaining the phenomenon on macro level Acknowledgement analysis. Arson has the maximum nearest neighboring index and mainly clusters around stations outside downtown, while prostitution has the minimum index and mainly in downtown. Kidnapping is relatively few and only clusters The author would like to express appreciation to Jennifer Wolch, the Dean of College of Environment Design in around stations outside downtown. Vehicle theft rarely clusters around stations. Generally speaking, the larger University of California at Berkeley, who helped to provide the pretty good study environment. amount of crimes occurs (except vehicle theft), the more obvious tendency of cluster around stations appears. Defined as severe crimes, violence crimes have more clusters around BART stations, thus exerting great negative References influence on people mentally. It is worth noting that the distance of criminal peaks in a way has something to do with the scale of blocks and Block, R., Block, C. R., 2000. Charter11: The Bronx and Chicago: Street Robbery in the Environs of Rapid Transit Stations, in “Analyzing Crime the width and length of the streets. A large amount of crimes seem to occur at the intersection not merely by chance. Patterns: Frontiers to Practice”. In: Goldsmith, V., McGuire, P. G., Mollenkopf, J. H., Ross, T. A. (Ed.). SAGE Publications,Thousand Oaks, CA. It is much easier for offenders to hide, observe and escape at the corner. Therefore, the spatial characteristic and Brantingham, P. L., Brantingham, P. J., 1993. Paths, Nodes and Edges: Considerations on the Complexity and Crime and the Physical urban design around stations remains to be deeply studied. Environment. Journal of Environmental Psychology 13, 3-28. Interestingly, the “dangerous” BART stations seem to be friendly to vehicles. But it doesn’t mean the advocation Brantingham, P. J., Brantingham, P. L., Wong, P. S., 1991. How Public Transit Feeds Private Crime: Notes on the Vancouver “Skytrain” of “P+R” because both macro level crime extension and crime contribution on meso-level, and even the space auto Experience. Security Journal 2, 91-95. correlation analysis of vehicle theft all demonstrate the obvious correlations with BART stations. The construction Cohen, L. E., Felson, M., 1979. Social Change and Crime Rate Trends: a Routine Activity Approach. American Sociological Review 44, 588– 608. of parking lots is likely to give a dramatic rise in vehicle crimes. Crime transfer will not happen either, considering Desroches, F. J., 1995. Force and Fear: Robbery in Canada. Canadian Scholars' Press, Toronto. the correlation of criminal types. Degeneste, H., Sullivan, J. P., 1994. Policing Transportation Facilities. Charles Thomas Publisher, Springfield. Dunphy. R., Cervero, R., Dock, F., McAvey, M., Porter, D. R., 2004. Developing around Transit - Strategies and Solutions That Work. Urban 7. Conclusions and Future Studies Land Institute, Washington, DC. FBI. Preliminary Annual Uniform Crime Report, January-December, 2012. Federal Highway Administration of U. S. Department of Transportation. http://www.fhwa.dot.gov/ohim/onh00.html. 7.1. Conclusions Frischtak, C., Mandel, B. R., 2012. Crime, House Prices, and Inequality: the Effect of UPPs in Rio. Federal Reserve Bank of New York Staff Reports, (no. 542). Federal Reserve Bank of New York, New York, NY. On the macro level, the mean center and long axis of standard deviation oval of crimes are very close to that of Krivo, L., Peterson, R., 1996. Extremely Disadvantaged Neighborhoods and Urban Crime. Social Forces 75, 619-648. BART stations, revealing the close relationship between the extension of crimes and the only mass transit line in LaVigne, N. G., 1996. Chapter6: Safe Transport: Security by Design on the Washington Metro, in “Preventing Mass Transit Crime”. In: Clarke, San Francisco. R. V. (Ed.). Criminal Justice Press, Monsey, NY. Liggett, R., Loukaitou-Sideris, A., Iseki, H., 2003. Journey to Crime: Assessing the Effects of a Light Rail Line on Crime in the Neighborhoods. On the meso-level, BART stations in block groups are significantly correlated to 9 types of crimes, making great Journal of Public Transportation 6, 85-115. contribution to assault, robbery, kidnapping, larceny and drug. The high-high and high-low local spatial Liggett, R., Loukaitou-Sideris, A., Iseki, H., 2001. Bus Stop-environment Connection: Do Characteristics of the Built Environment Correlate relationships of criminal density occupy larger proportion in blocks with BART stations, which appear to be high with Bus Stop Crime? Transportation Research Record. (No. 1760), 20 - 27. criminal density regions whether or not in downtown area, and public security tends to be better with the increased Loukaitou-Sideris, A., Liggett, R., Iseki, H., 2002. The Geography of Transit Crime: Documentation and Evaluation of Crime Incidence on and distance from the stations. around the Green Line Stations in Los Angeles. Journal of Planning Education and Research 22, 131-151. Lusk, A., 2001. Bus and Bus Stop Designs Related to Perceptions of Crime. Federal Transit Administration of U. S. Department of On the micro level, crimes cluster significantly around BART stations, featuring many sudden rises of clusters Transportation, Washington, DC. and second clusters within a short distance (double peaks). The larger amount crimes have, the more obvious Newton, A. D., 2004. Crime and Disorder on Buses: Toward an Evidence Base for Effective Crime Prevention. Ph.D. Thesis, University of tendency of cluster around stations appears. The microcosmic spatial characteristics seem to lead to different Liverpool, Liverpool. criminal clusters. The fear of regional crimes rises brought by stations is not a fike. Furthermore, the violence crimes Newton, A. D., 2004. Crime on Public Transport: “Static” and “Non-static” (Moving) Crime Events. Western Criminology Review 5, 25-42. aggravate the fears mentally. Okabe, A., Okunuki, K. I., Shiode, S., 2002. SANET: A Toolbox For Spatial Analysis on A Network. Centre for Spatial Information Science, University of Tokyo, Tokyo. In a word, mass transit has close relation to crimes in San Francisco. The crimes result from a series of criminal Poister, T. H., 1996. Transit-related Crime in Suburban Areas. Journal of Urban Affairs 18, 63-75. interactions rooted in the diverse city functions attracted by the mass transit. Whether being the real reason of crimes Sampson, R. J., Raudenbush, S. W., Earls, F., 1997. Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy. Science 277, or just the centralized reflection, the mass transit seems to seriously worsen public security psychologically. 918–924. San Francisco Municipal Government. http://data.sfgov.org. 7.2 Future Studies Sedelmaier, C. M., 2003. Railroaded: The Effects of a New Public Transportation System upon Local Crime Patterns. Ph.D. Thesis, Rutgers University, Rutgers NJ. Stafford, J., Pettersson, G., 2004. People's Perceptions of Personal Security and Their Concerns about Crime on Public Transport: Literature There are several paths in future studies to distinguish between causes and effects and analyze in the social- Review. Crime-Concern Report for DfT. Department for Transport, London, England. economic level rather than statistical work: to discover the criminal causes and the relation between the various Spooner, P., Lunt, I. D., Okabe, A., Shiode, S., 2004. Spatial Analysis of Roadside Acacia Populations on a Road Network Using the Network K- criminal types, especially depending on non-spatial factors; to analyze the time distribution of criminal clusters; to function. Landscape Ecology 19, 491-499. Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 3255 18 Author name / Transportation Research Procedia 00 (2017) 000–000 Author name / Transportation Research Procedia 00 (2017) 000–000 19

Micro level analysis confirms the obvious cluster of crimes around BART stations both in and outside downtown, explore the relation between peak distances and spatial characteristics of blocks based on the widespread double including non-significant correlated-with-BART criminal types on the meso-level analysis. The reason of peaks of clusters; to predict and prevent crimes in view of the above results. uncorrelation lies in that the cluster only exists around certain stations according to the different attributes of Related studies can be extended to medium capacity transit systems such as MUNI in San Francisco; to larger neighborhood and activities. As to the stations outside downtown area, the influence of worsening public security is geographic scope such as the whole bay area spatially; to more vertical researches such as the change of public not limited to the nearby neighborhoods. Robbery and larceny cluster at all the 8 stations and all have peaks except security before and after the construction of a mass transit; to more cities based on various culture and social Balboa Park Station. Assault, sex offenses, fraud and vandalism cluster at 7 stations, featuring peaks except background to discover whether the similar criminal clusters exist and give insight into the detailed nature and vandalism. Burglary and drug cluster at 6 stations. Burglary clusters in a short distance from most stations, giving causes of crime. comfort to residents far away from stations. Arson and prostitution have more different standard deviation ovals with BART than other crimes and they only cluster around few stations, explaining the phenomenon on macro level Acknowledgement analysis. Arson has the maximum nearest neighboring index and mainly clusters around stations outside downtown, while prostitution has the minimum index and mainly in downtown. Kidnapping is relatively few and only clusters The author would like to express appreciation to Jennifer Wolch, the Dean of College of Environment Design in around stations outside downtown. Vehicle theft rarely clusters around stations. Generally speaking, the larger University of California at Berkeley, who helped to provide the pretty good study environment. amount of crimes occurs (except vehicle theft), the more obvious tendency of cluster around stations appears. Defined as severe crimes, violence crimes have more clusters around BART stations, thus exerting great negative References influence on people mentally. It is worth noting that the distance of criminal peaks in a way has something to do with the scale of blocks and Block, R., Block, C. R., 2000. Charter11: The Bronx and Chicago: Street Robbery in the Environs of Rapid Transit Stations, in “Analyzing Crime the width and length of the streets. A large amount of crimes seem to occur at the intersection not merely by chance. Patterns: Frontiers to Practice”. In: Goldsmith, V., McGuire, P. G., Mollenkopf, J. H., Ross, T. A. (Ed.). SAGE Publications,Thousand Oaks, CA. It is much easier for offenders to hide, observe and escape at the corner. Therefore, the spatial characteristic and Brantingham, P. L., Brantingham, P. J., 1993. Paths, Nodes and Edges: Considerations on the Complexity and Crime and the Physical urban design around stations remains to be deeply studied. Environment. Journal of Environmental Psychology 13, 3-28. Interestingly, the “dangerous” BART stations seem to be friendly to vehicles. But it doesn’t mean the advocation Brantingham, P. J., Brantingham, P. L., Wong, P. S., 1991. How Public Transit Feeds Private Crime: Notes on the Vancouver “Skytrain” of “P+R” because both macro level crime extension and crime contribution on meso-level, and even the space auto Experience. Security Journal 2, 91-95. correlation analysis of vehicle theft all demonstrate the obvious correlations with BART stations. The construction Cohen, L. E., Felson, M., 1979. Social Change and Crime Rate Trends: a Routine Activity Approach. American Sociological Review 44, 588– 608. of parking lots is likely to give a dramatic rise in vehicle crimes. Crime transfer will not happen either, considering Desroches, F. J., 1995. Force and Fear: Robbery in Canada. Canadian Scholars' Press, Toronto. the correlation of criminal types. Degeneste, H., Sullivan, J. P., 1994. Policing Transportation Facilities. Charles Thomas Publisher, Springfield. Dunphy. R., Cervero, R., Dock, F., McAvey, M., Porter, D. R., 2004. Developing around Transit - Strategies and Solutions That Work. Urban 7. Conclusions and Future Studies Land Institute, Washington, DC. FBI. Preliminary Annual Uniform Crime Report, January-December, 2012. Federal Highway Administration of U. S. Department of Transportation. http://www.fhwa.dot.gov/ohim/onh00.html. 7.1. Conclusions Frischtak, C., Mandel, B. R., 2012. Crime, House Prices, and Inequality: the Effect of UPPs in Rio. Federal Reserve Bank of New York Staff Reports, (no. 542). Federal Reserve Bank of New York, New York, NY. On the macro level, the mean center and long axis of standard deviation oval of crimes are very close to that of Krivo, L., Peterson, R., 1996. Extremely Disadvantaged Neighborhoods and Urban Crime. Social Forces 75, 619-648. BART stations, revealing the close relationship between the extension of crimes and the only mass transit line in LaVigne, N. G., 1996. Chapter6: Safe Transport: Security by Design on the Washington Metro, in “Preventing Mass Transit Crime”. In: Clarke, San Francisco. R. V. (Ed.). Criminal Justice Press, Monsey, NY. Liggett, R., Loukaitou-Sideris, A., Iseki, H., 2003. Journey to Crime: Assessing the Effects of a Light Rail Line on Crime in the Neighborhoods. On the meso-level, BART stations in block groups are significantly correlated to 9 types of crimes, making great Journal of Public Transportation 6, 85-115. contribution to assault, robbery, kidnapping, larceny and drug. The high-high and high-low local spatial Liggett, R., Loukaitou-Sideris, A., Iseki, H., 2001. Bus Stop-environment Connection: Do Characteristics of the Built Environment Correlate relationships of criminal density occupy larger proportion in blocks with BART stations, which appear to be high with Bus Stop Crime? Transportation Research Record. (No. 1760), 20 - 27. criminal density regions whether or not in downtown area, and public security tends to be better with the increased Loukaitou-Sideris, A., Liggett, R., Iseki, H., 2002. The Geography of Transit Crime: Documentation and Evaluation of Crime Incidence on and distance from the stations. around the Green Line Stations in Los Angeles. Journal of Planning Education and Research 22, 131-151. Lusk, A., 2001. Bus and Bus Stop Designs Related to Perceptions of Crime. Federal Transit Administration of U. S. Department of On the micro level, crimes cluster significantly around BART stations, featuring many sudden rises of clusters Transportation, Washington, DC. and second clusters within a short distance (double peaks). The larger amount crimes have, the more obvious Newton, A. D., 2004. Crime and Disorder on Buses: Toward an Evidence Base for Effective Crime Prevention. Ph.D. Thesis, University of tendency of cluster around stations appears. The microcosmic spatial characteristics seem to lead to different Liverpool, Liverpool. criminal clusters. The fear of regional crimes rises brought by stations is not a fike. Furthermore, the violence crimes Newton, A. D., 2004. Crime on Public Transport: “Static” and “Non-static” (Moving) Crime Events. Western Criminology Review 5, 25-42. aggravate the fears mentally. Okabe, A., Okunuki, K. I., Shiode, S., 2002. SANET: A Toolbox For Spatial Analysis on A Network. Centre for Spatial Information Science, University of Tokyo, Tokyo. In a word, mass transit has close relation to crimes in San Francisco. The crimes result from a series of criminal Poister, T. H., 1996. Transit-related Crime in Suburban Areas. Journal of Urban Affairs 18, 63-75. interactions rooted in the diverse city functions attracted by the mass transit. Whether being the real reason of crimes Sampson, R. J., Raudenbush, S. W., Earls, F., 1997. Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy. Science 277, or just the centralized reflection, the mass transit seems to seriously worsen public security psychologically. 918–924. San Francisco Municipal Government. http://data.sfgov.org. 7.2 Future Studies Sedelmaier, C. M., 2003. Railroaded: The Effects of a New Public Transportation System upon Local Crime Patterns. Ph.D. Thesis, Rutgers University, Rutgers NJ. Stafford, J., Pettersson, G., 2004. People's Perceptions of Personal Security and Their Concerns about Crime on Public Transport: Literature There are several paths in future studies to distinguish between causes and effects and analyze in the social- Review. Crime-Concern Report for DfT. Department for Transport, London, England. economic level rather than statistical work: to discover the criminal causes and the relation between the various Spooner, P., Lunt, I. D., Okabe, A., Shiode, S., 2004. Spatial Analysis of Roadside Acacia Populations on a Road Network Using the Network K- criminal types, especially depending on non-spatial factors; to analyze the time distribution of criminal clusters; to function. Landscape Ecology 19, 491-499. 3256 Di Wang et al. / Transportation Research Procedia 25C (2017) 3237–3256 20 Author name / Transportation Research Procedia 00 (2017) 000–000

Tilley, N., Smith, J., Finer, S., Erol, R., Charles, C., Dobby, J., 2004. Problem-solving Street crime: Practical Lessons from the Street Crime Initiative. Great Britain Home Office Research Development and Statistics Directorate, London. Transit Cooperative Research Program(TCRP), 1997. The Role of Transit in Creating Livable Metropolitan Communities. Report 22. National Academy Press, Washington, DC. Tseloni, A., Pease, K., 2003. Repeat Personal Victimization: “Boost” or “Flag”? British Journal of Criminology. 43, 196-212. U. S. Census Bureau. http://www.census.gov/main/www/access.html. U. S. Census Bureau, 2011. State & County QuickFacts, San Francisco (city), California. U. S. Census Bureau, 2013. Annual Estimates of the Resident Population for Counties. Walker, A., Kershaw, C., Nicholas, S., 2006. Home Office Statistical Bulletin: Crime in England and Wales 2005/06. National Statistics, London, UK. Willits, D., Broidy, L., Gonzales, A., Denman, K., 2011. Place and Neighborhood Crime: Examining the Relationship between Schools, Churches, and Alcohol Related Establishments and Crime. Report to the Justice Research Statistics Association. New Mexico Statistical Analysis Center, Institute for Social Research, University of New Mexico, Albuquerque, NM. Yu, S. V., 2009. Bus Stops and Crime: Do Bus Stops Increase Crime Opportunities in Local Neighborhoods? Ph.D. Thesis. The State University of New Jersey, Newark, NJ.