Journal of Criminal Justice 44 (2016) 21–29
Contents lists available at ScienceDirect
Journal of Criminal Justice
Bringing the physical environment back into neighborhood research: The utility of RTM for developing an aggregate neighborhood risk of crime measure Grant Drawve a,⁎, Shaun A. Thomas b, Jeffery T. Walker c a b c
Department of Psychology and School of Criminal Justice, Rutgers University, 101 Warren St., Smith Hall 3rd Floor, Newark, NJ 07102, United States Department of Sociology & Criminal Justice, University of Arkansas, 211 Old Main, Fayetteville, AR 72701, United States Department of Justice Sciences, University of Alabama Birmingham, 1201 University Blvd., Birmingham, AL 35294, United States
a r t i c l e
i n f o
Article history: Received 7 September 2015 Received in revised form 9 December 2015 Accepted 10 December 2015 Available online xxxx Keywords: Risk terrain modeling Environmental criminology Neighborhoods and crime
a b s t r a c t Purpose: The current research examines the utility of risk terrain modeling (RTM) in developing an aggregate neighborhood risk of crime (ANROC) measure. RTM is often employed at the micro-place, forecasting future crime by street segment from attributes of the physical environment. Controlling for concentrated socioeconomic disadvantage and residential stability, we examine the ability of RTM to forecast neighborhood-level violent crime rates in Little Rock, Arkansas. Methods: Grounded in the extant literature and our knowledge of the area, we identified 14 risk factors expected to influence violent crimes. Once a RTM was constructed on 2013 violent crimes, the risk of crime per cell was averaged by neighborhood (Census tract), developing an aggregate neighborhood risk of crime measure. The ANROC measure was used to predict 2014 neighborhood violent crime rates. Results: This measure significantly increases the understanding of variation in neighborhood violent crime rates. The regression analyses indicated all three measures were significant predictors of neighborhood violent crime rates in Little Rock. Conclusions: The overall pattern of results supported our contention that the development of a macro- or neighborhood-level measure reflecting risk for criminal opportunities contributes substantively to the neighborhoods and crime literature. © 2015 Elsevier Ltd. All rights reserved.
Introduction Risk terrain modeling (RTM) is a spatial diagnostic technique capable of identifying subtle variation in victimization risk across environments based on features of the landscape (Caplan, Kennedy, & Miller, 2011). Essentially, RTM is an analytical tool that allows for the creation of an intuitive and detailed analysis of the underlying risk factors present in a landscape, assisting in our understanding of why crime is occurring in specific locales. RTM has thus far been utilized, almost exclusively, to forecast criminal events in micro-units (i.e. street segments) (Caplan et al., 2011; Drawve, 2014; Drawve, Moak, & Berthelot, 2014; Dugato, 2013; Kennedy, Caplan, & Piza, 2011; Moreto, Piza, & Caplan, 2014). What is insufficiently understood is the ability of RTM to forecast criminal events in diverse units of analysis (tract, neighborhood, etc.). The neighborhood effect on crime is well established (Morenoff & Sampson, 1997; Sampson, 2012). We propose RTM may be uniquely ⁎ Corresponding author. E-mail addresses:
[email protected] (G. Drawve),
[email protected] (S.A. Thomas),
[email protected] (J.T. Walker).
http://dx.doi.org/10.1016/j.jcrimjus.2015.12.002 0047-2352/© 2015 Elsevier Ltd. All rights reserved.
capable of advancing the extant neighborhoods and crime literature due to the techniques ability to provide a quantifiable estimate of risk for neighborhoods based on risk levels across subordinate or microunits. In utilizing RTM to construct an estimate of risk of crime or victimization at the neighborhood or other macro-social unit of analysis, it may become possible to more accurately predict neighborhood crime rates by simultaneously accounting for both characteristics of the built environment and social factors known to be robust correlates of levels of crime and violence. RTM draws on concepts derived from environmental criminological perspectives, prior literature, and a familiarity with an area to forecast crime in reference to the environmental backcloth (Brantingham & Brantingham, 1995). RTM identifies significant risk factors that delineate criminogenic places within the built environment or physical landscape in relation to an outcome event (i.e. robbery, residential burglary, gun crimes, etc.). Criminogenic places, also known as crime generators and attractors (CGAs) (Brantingham & Brantingham, 1995), are, in short, places that influence crime in the surrounding area. Examples of CGAs include bus stops, bars, liquor stores, pawn shops, and other locations where criminal opportunities arise due to the converge of offenders, targets, and a lack of guardianship in time and space. The
22
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
estimated risk of crime produced by RTM can be interpreted as higher values indicate an increased likelihood of crime at a specific place. While detailed characteristics of the built environment or physical landscape are utilized by RTM in the creation of risk scores, social structural characteristics describing the population at-large have largely been overlooked and are not incorporated when identifying risk factors (see Drawve et al., 2014a for exception). This is, in part, due to a focus on micro-places in prior RTM based research and the requirement of xy coordinates in RTMDx (discussed below). Data concerning social characteristics are publically available, but at higher order or aggregate levels of analysis, such as the block group and tract.1 The current study contributes to the established literature on neighborhoods and crime and the emerging literature on RTM by developing an aggregate neighborhood risk of crime (ANROC) measure and testing its ability to predict 2014 violent crime rates in Little Rock, Arkansas. It is expected that, beyond traditional social structural correlates of crime, an ANROC measure capturing risk associated with the physical landscape will play an influential role in predicting 2014 neighborhood violent crime rates. In simultaneously exploring the influence of social and physical characteristics of neighborhoods on levels of crime, our analyses provide a more comprehensive assessment of the association between levels of crime and the environmental backcloth (Brantingham & Brantingham, 1981). In outlining how such analyses contribute to the extant literature and the development of ecological theories of crime, we draw on the disorganization theoretical framework as well as the theory of risky places (Kennedy & Caplan, 2013), routine activities theory (Cohen & Felson, 1979) and concepts from crime pattern theory (Brantingham & Brantingham, 1993). Review of literature The neighborhood context of crime is a well-researched social problem. Ecological perspectives highlighting the importance of place, space, and crime appeared in the work of Park, Burgess, and McKenzie (1925) and continue to be relevant (Sampson, 2012). Social disorganization is a prominent theoretical framework, which underscores the importance of neighborhood level social correlates of crime, particularly residential instability, racial and cultural heterogeneity, and poverty (Shaw & McKay, 1942). The concentration of over-lapping forms of structural deficits or concentrated socioeconomic disadvantage fosters normative confusion and isolation from mainstream society. As a result, structurally disadvantaged communities experience attenuation in the ability to transmit mainstream culture and the proliferation of cultural adaptations that undermine social organization (Anderson, 1999; Krivo & Peterson, 2000; Warner, 2003). Through such mechanisms, communities become socially disorganized which entails a generalized inability to maintain effective formal and informal social control mechanisms thereby undermining the ability to regulate behavior (Bursik, 1988). The development of the systemic model of community attachment refined the social disorganization perspective in conceptualizing the community as a “complex system of friendship and kinship networks and formal and informal associational ties rooted in family life and the ongoing socialization processes” (Kasarda & Janowitz, 1974:329). The systemic model emphasizes the role of residential instability in understanding the ecological distribution of crime. Population instability often endemic in “disorganized” communities undermines social cohesion, family stability, normative consensus, organizational participation, and institutional attachment which leaves residents with fewer opportunities to build friendship ties, associational networks, or participate in organizational activities (Burchfield, 2009; Bursik, 1988; Kasarda & Janowitz, 1974; Sampson, 1988; Sampson & Groves, 1989; Warner, 2014). Under such circumstances, the effectiveness of formal and informal social control mechanisms are attenuated, which fosters an increase in crime. An important theoretical connection between the disorganization and systemic frameworks concerns the link between socioeconomic
disadvantage, residential instability, and crime. It is proffered that such factors undermine the creation and maintenance of neighborhood network ties and attachment to place. When community attachment is undermined it becomes increasingly difficult to maintain effective formal and informal social control mechanisms, opening the door for social dislocations such as crime and violence. Prior studies support these linked perspectives with certain community social characteristics, particularly residential instability and concentrated socioeconomic disadvantage emerging as robust correlates of variation in crime across neighborhoods and larger macro-social units of analysis (Land, McCall, & Cohen, 1990; Lee, 2000; McCall, Land, & Parker, 2010; Peterson & Krivo, 2010; Sampson, 2012; Sampson & Groves, 1989; Thomas & Shihadeh, 2013; Wilson, 1987, 1996, 2009). The communities and crime literature has played a large role in shaping our understanding of variation in crime across diverse macrosocial units of analysis. The focus on larger units of analysis within the empirical literature has however been largely a result of data availability as opposed to theoretical relevance. To assist in a more comprehensive development of our understanding of variations in crime across contextual environments there has been a push within the extant literature for a focus on micro-units (Weisburd, Groff, & Yang, 2012). The conceptualization of what are often referred to as environmental perspectives and concepts, which are also ecological theories of crime, provide a basis expecting a more variations in crime across neighborhoods. A number of researchers have proposed routine activities theory (RAT) is a micro-level theory with macro-level implications (Eck, 1995; Groff, 2007). In fact, Weisburd et al. (2012) suggested crime patterns present at the micro-place (i.e. street segment) may resemble variation in criminal opportunities within a larger aggregate unit of analysis. Cohen and Felson (1979) developed RAT to explain the relation between crime rate trends and the type of activities people encounter throughout days/weeks/months. A substantive insight drawn from RAT underscoring the current study is the premise that greater movement of individuals across space increases criminal opportunities and thus levels of crime (Cohen & Felson, 1979). Cohen and Felson (1979) proffered that individuals have broad areas in which routine activities occur: home, work, and other activities away from their home, creating varying opportunities for crime. As individuals increase time spent on outside activities, capable guardianship naturally fluctuates, leading to variation in their suitability as a criminal target. For example, Kennedy and Forde (1990) found young males whose activity space routinely involved drinking establishments, restaurants, and sporting events experienced higher victimization rates. Such findings suggest certain locations may lack capable guardianship while also attracting both victims and offenders, which generates criminal opportunities and, in turn, an increase in victimization. Crime generators and crime attractors (CGAs) (Brantingham & Brantingham, 1993) were originally proposed to be distinct yet related concepts referencing how nonresidential land use influences crime. Victims may be attracted to a business for the type of products or services offered by the establishment, and offenders may be attracted to the location for the same reason or because of the type of patrons who frequent that area whom are identified as potentially attractive and suitable victims. The current study is not concerned with distinctions between these concepts as both are expected to influence crime at and around their locations. Generally, CGAs attract both victims and offenders for their respective reasons. The convergence of victims and offenders at these locations is relevant to RAT in that an increase in the relative prevalence of individuals motivated to offend in conjunction with an increase in potential victims who lack capable guardianship is likely to result in more criminal opportunities that are acted upon. As an example, businesses such as check-cashing stores and pawn shops offer quick cash transactions for patrons but become attractive locations for robberies, especially if their physical location may be viewed as conducive to a lack of guardianship or facilitative of a quick exit. Offenders target such businesses and their customers because there are
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
more potential targets and they are perceived to be more lucrative (i.e. carrying cash or displaying jewelry) (Wright & Decker, 1996). Prior studies suggest bars are another CGA that influence the level of crime in the surrounding area (Groff, 2011; Ratcliffe, 2012). Such findings suggest alcohol consumption influences the prevalence of suitable targets and motivated offenders while the nature of the environment facilitates decreased guardianship due to individuals' unwillingness to supervise or intervene. Drawve, Thomas, & Walker (2014) developed a complimentary argument in an examination of arrests suggesting intoxicated patrons may be suitable targets while intoxicated offenders may act impulsively rather than rationally assessing a criminal opportunity. RAT and the environmental concepts discussed above impact how criminal opportunities vary across a landscape. Expanding upon this work, Kennedy and Caplan (2013) developed the theory of risky places that delves into how approaching crime from a vulnerability-exposure framework enhances our understanding of crime in space. The theory of risk places offers three primary propositions. The first asserts that, while risk is omnipresent, there is considerable variation in risk across places as a result of the spatial influence of criminogenic features. For example, being in close proximity to a high school may result in a contextual environment in which risk of robbery is exacerbated compared to distant places. That is, the criminogenic influence of high schools is spatially based and exhibits a greater influence on risk in the immediate proximity. Kennedy and Caplan (2013) also proposed that vulnerability is greatest where multiple criminogenic features have overlapping spatial influence. This overlap or co-location in space results in overlapping spatial influences and an exacerbated vulnerability for risk of criminal victimization. Kennedy and Caplan's (2013) final proposition connects their previous assertions in suggesting the effect risky places have on crime is a function of variation in both vulnerability and exposure across the physical landscape. Exposure refers to the observation that crime is not random but rather tends to be concentrated in certain places. As such, exposure to risk and criminal opportunities varies from place to place. Moreover, not all places with increased levels of vulnerability exhibit the same degree of exposure, however, being vulnerable does identify an exacerbated potential for offending and victimization. In developing the theory of risky places Kennedy and Caplan (2013) provided a more refined framework for understanding why and to what extent we can expect risk and thus offending and victimization to vary across the physical landscape. Kennedy, Caplan, Piza, and Buccine-Schraeder (2015) tested the applicability of this vulnerability-exposure framework in an examination of assaults utilizing a geospatial technique known as risk terrain modeling (RTM). Their analysis provided consistent and robust support for each of the propositions outlined in the theory of risky places. Risk terrain modeling (RTM) RTM is a technique that examines criminogenic features of the physical landscape, most often from the built environment, to develop a risk of crime score for micro-units. The advent of such techniques has been influential to the development of the extant literature examining CGAs. Drawve (2014) highlighted the ability of RTM to reliably forecast crime across micro-units when compared to multiple retroactive techniques. While such examinations have significantly advanced our understanding of criminal events, prior literature has but scratched the surface of the potential of RTM. For example, RTM has the ability to identify environments conducive to future crime based on unique behavioral settings in which CGAs co-locate. Such substantive advancements are precisely why further research is warranted on the testing and exploring of the analytical technique. Much of the existing RTM research has focused on crime forecasting, and has primarily been grounded in a policing framework. This is, in part, the result of RTM being conceptualized as, “… a major step forward in predictive policing”
23
(Brantingham, 2011, p. 202). As Koss (2015) discussed, RTM offers a quantifiable methodology for police to target high-crime areas instead of relying on officers' beliefs. However, Koss (2015) also recognized high-crime areas are typically characterized as lower income, minority neighborhoods. The current study focuses on Koss (2015) qualifying statement. Specifically, we examine the development of an RTM based neighborhood level measure of criminal opportunities and hence victimization risk. We then explore this measures ability to enhance our understanding of variation in violent crime across neighborhoods. That is, we develop an Aggregated Neighborhood Risk Of Crime (ANROC) measure and explore its relevance to predicting victimization risk across neighborhoods. We examine the effect of this ANROC measure on neighborhood crime levels above and beyond the impact of structural demographic and socioeconomic factors known to be strongly associated with levels of crime (Land et al., 1990; Sampson, 2013). The current study advances therefore advances the extant literature by exploring the potential diversity and robustness of RTM in terms of applying the technique to assess variation in risk of crime at the neighborhood-level. Data & methodology The current analyses examine Little Rock, Arkansas, which has about 200,000 residents across 121 mile2 (quickfacts.census.gov/qfd/states/ 05/0541000.html). According to Uniform Crime Reports, Little Rock had the 14th (2012) and the 7th (2013) highest violent crime rate of cities with more than 100,000 residents. As such, Little Rock offers a unique environment in which to conduct the current research considering it is a medium size yet relatively violent place. That said, like all cities, crime is not evenly or randomly dispersed across Little Rock neighborhoods. Our goal is to simultaneously explore the ability of neighborhood physical and social characteristics to assist in the prediction of variation in crime across neighborhoods. Neighborhoods were operationalized via Census tracts and analyses are based on data for 50 tracts.2 Geocoding was accomplished using a 20 ft. offset rather than to the centerline file because tract boundaries often coincide with streets. If the data were geocoded to the street centerline, crimes could be counted twice, once for each tract. It is important to acknowledge that tracts may not be representative of actual neighborhoods but do “offer the best compromise with respect to size, homogeneity, data availability, and comparability” (White, 1987:19). In order to examine how the ANROC measure influences violent crime rates, tracts were used, in part, to facilitate merging sociodemographic data for analyses. Allowing data availability to drive the research is not ideal; but, as we explore in this research, it is possible findings are robust across units of analysis (Land et al., 1990). In addition, census tracts are appropriate for this analysis because we are not making a direct argument of neighborhood change. Rather, we are extending the ability of RTM to be applied across diverse units. Dependent variable The dependent variable in this study was 2014 neighborhood violent crime rates per 1000 persons based on data obtained from the Little Rock Police Department (LRPD). Violent crimes in our analyses include: homicide, aggravated assault, and business/street/individual robberies.3 Multiple crime types are combined and examined simultaneously because our focus is on violent crime in general as opposed to a single form of violence. LRPD data included 1965 violent crimes for 2014, 1860 of which were geocoded (95%). Constructing a measure of risk of crime The current study includes fourteen risk factors based on extant literature and our personal knowledge of the study area4 (Bernasco &
24
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
Block, 2011; Bichler, Schmerler, & Enrinquez, 2013; Brantingham & Brantingham, 1995; Drawve, 2014; Groff, 2011; McCord & Ratcliffe, 2007; LeBeau, 1997, 2011; Levine, Wachs, & Shirazi, 1986; McCord & Ratcliffe, 2007; McCord, Ratcliffe, Garcia, & Taylor, 2007; Pridemore & Grubesic, 2011; Spelman, 1995; Ratcliffe, 2012; Roncek & Bell, 1981; Roncek & Pravatiner, 1989; Roncek & Maier, 1991; Wright & Decker, 1996). The risk factors include: banks, big box retail, bus stops, checkcashing, convenience stores, fast-food restaurants, grocery stores, hotel/motels, on-site alcohol establishments, liquor stores, lottery retailers, pawn shops, high schools, and tattoo/piercing shops. Some of the less tested risk factors such as, lottery retailers and tattoo/piercing parlors do not necessarily generate crime but could rather attract crime in similar fashions as pawn shops and check-cashing stores. For example, with lottery retailers, there are often quick exchanges of currency meaning patrons are more likely to carry cash/payment forms on them to purchase lottery tickets. The inclusion of lesser known potential risk factors was to determine what the unique factors were in creating conducive environments for violence. The data on these risk factors were drawn from five sources: Little Rock Treasury Department, Little Rock School District, Arkansas Scholarship Lottery, Arkansas Beverage Control, and MetroPlan. Data on all businesses with an active license in 2013 were obtained from the Little Rock Treasury Department. These data provided the business name, address, and a general category (i.e. liquor store, convenient store, and pawn shop). These data were used as a starting point in determining what types of businesses could be examined as risk factors. Ten businesses that served as potential risk factors were extracted from these data: banks, big box retail stores, check-cashing stores, convenient marts, fast food restaurants, grocery store/supermarkets, hotel/motels, liquor stores, pawn shops, and tattoo/piercing parlor. In our attempt to comprehensively account for risk factors associated with violent crime, we sought out additional data sources that complimented the data on businesses. Four additional risk factors were drawn from unique sources. First, public high school locations were obtained from Little Rock School District (http://www.lrsd. org/). Second, data obtained from the Arkansas Scholarship Lottery (myarkansaslottery.com) allowed for the identification of the location of lottery retailers in 2014. 5 Third, data identifying establishments with a liquor license were obtained from the Arkansas Beverage Control (ABC). 6 These data identify permit type (beer only, mixed drink) and if an establishment had an on-site consumption or off-site permit. Only on-site data, bars and restaurants serving alcohol, were utilized. Finally, bus stop location identifiers were obtained from MetroPlan (metroplan.org) for 2014, the only year available. MetroPlan is an organization aiding municipalities in transportation and planning investments. While the dependent variable is the 2014 violent crime rate, constructing a RTM requires an outcome event, in this case, 2013 violent crime data. That is, the RTM used to develop a measure of the relative prevalence of criminal opportunities or victimization risk is based on prior reported offenses. This is not to say that prior crime were used to predict future crime levels but rather that 2013 crime data were utilized to identify the spatial operationalization and influence risk factors have on crime so as to determine associated risk. Data on violent crimes in 2013 were provided by LRPD and utilized as the outcome event to determine the appropriate operationalization and significant risk factors associated with violence across Little Rock neighborhoods. In 2013, LRPD recorded 2173 violent crimes of which 1999 were geocoded (92%). The RTMDx software was used to determine the significant risk factors necessary for developing a measure of risk (Caplan & Kennedy, 2013). The RTM analysis is a multistage process. First, RTMDx was used to operationalize and quantify the influence of risk factors on 2013 violent crime for micro-units (i.e. risk of crime by street). Next, using the RTMDx output, an ANROC measure was constructed to predict 2014 neighborhood violent crime rates.
RTMDx RTMDx requires both the selection of parameters and input of data when determining a crime model, including: study area, block length, raster cell size, type of model, outcome event, risk factors, and the operationalized strategies of the spatial influence (see Caplan, Kennedy, & Piza, 2013).7 The Little Rock boundary file was input for the model study area. Based on guidelines recommended by Caplan et al. (2013) we determined block length (432 ft) in Little Rock by calculating the average of street segments. Next, raster cell size must be determined. Raster cells are akin to draping a net over Little Rock that divides the city into thousands of cells with unique characteristics, similar to a grid. The characteristics of a cell or cell attributes were determined based on RTMDx results and assigned accordingly. Once the spatial influence of each risk factor is determined, it is visually represented by the surrounding cells. For example, if lottery retailers were found to have a spatial influence of one-block on violent crime, cells within one-block of the retailer would be assigned a value of 1 (risky). Based on suggestions from Caplan et al. (2013), a cell size of 216 ft was chosen to reflect about half the average street segment length. By having the cell size equal to half a block, the output from RTMDx can be expressed in possible half block increments of spatial influence (discussed below). There were a total of 74,004 cells within Little Rock. Next, we set the model type parameter based on the expectation that risk factors are positively correlated with 2013 violent crime locations. As such, an ‘aggravating’ model was selected within RTMDx to test for this positive spatial relation between risk factor locations and the xy coordinates identifying where violent crimes occurred. The next step in the RTMDx analysis was utilizing the program to operationalize risk factors. Table 1 provides a synopsis of risk factors and their appropriate operationalizations within RTMDx. We set the parameters to the max spatial influence of 4 blocks that are tested at halfblock increments. For risk factors with fewer than 20 locations, proximity was used, based on the results of average nearest neighbor analyses, as a spatial operationalization. Once all parameters were set in RTMDx, the program determines the best model based on the risk factors input and operationalization decisions. Based on our operationalization of risk factors, a total of 192 variables were created (see Table 1). For example, because banks were operationalized as both, proximity and density, there were 8 proximity and 8 density measures at half-block increments up to 4 blocks. As Heffner (2013) discussed, spurious relations may emerge when a large number of variables are included in the model. To compensate, RTMDx utilizes cross-validation to build a penalized Poisson regression model in which any variable with a non-zero coefficient is considered for potential inclusion in the model. When constructing the best fitting model, RTMDx implements a stepwise regression process with variable
Table 1 Operationalization of CGAs in RTMDx. CGAs
n
OP
SI
Increments
Total
Banks Big box retail Bus stop Check cashing Conv. mart Fast food Grocery store Hotel/motel Liquor-on Liquor-off Lottery Pawn Public high school Tattoo/piercing Total
23 51 1155 6 88 90 61 57 250 133 51 16 5 10
Both Both Both Proximity Both Both Both Both Both Both Both Proximity Proximity Proximity
4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks 4-Blocks
Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-block Half-Block
16 16 16 8 16 16 16 16 16 16 16 8 8 8 192
OP = Operationalization; SI = Spatial Influence.
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
selection based on p ≤ .05. The model is then justified through a Bayesian Information Criterion (BIC) score. Lower BIC scores are indicative of better overall model fit. Variables are added and removed from the model to determine the best fit, with the goal of minimizing the BIC score. Considering criminal events may be related to one another, their spatial distribution violates the Poisson distribution assumption. To compensate, two models are estimated: one assuming a Poisson distribution and a second assuming a negative binomial distribution. The model with the lower BIC score is reported. The end result is a list of significant risk factors and appropriate operationalizations for the best fitting model in relation to the criminal event outcome. In the model output, significant risk factors are reported with the type of operationalization (density or proximity) and the spatial influence increments (i.e. half- or whole-block). Two values are reported, the coefficient and the relative risk value. This output reflects the best fitting model based on the risk factors included in the analysis and the outcome measure. For measures operationalized as proximity, cells within the determined distance are assigned a value of 1 (risky) and all other cells are assigned 0. For density measures, the Kernel Density Estimation (KDE) tool in ArcGIS is utilized for the operationalization function based on the RTMDx output. Once the KDE tool is used with the specific risk factors, the cells with values greater than 2 standard deviations from the mean are assigned a value of 1 (risky). Based on the proximity or density operationalization, cells identified as “risky” were assigned values of 1 for each significant risk factor. Because each layer is comprised of 1's and 0's, only risky cells in each layer are assigned a risk value in relation to the overall model (Caplan et al., 2013). Once a final risk map is created, the overall risk in Little Rock is determined and discussed in terms of odds. The lowest odds of crime occurring would be a score of one (1) and any cell with a value greater than 1 would be interpreted in the following manner: a cell with a value of 11.18 has an expected rate of crime 11.18 times greater than cells with the value of 1 (Caplan et al., 2013). The aggregate neighborhood (i.e. tract) risk of crime (ANROC) value was computed as the average risk score for the raster cells located within the neighborhood. The centroids of each individual cell were used to determine which cells would be averaged to form the aggregate neighborhood measure. That is, all cells within each tract were used when determining the average risk value for a tract. The neighborhood-level risk value was then utilized in the regression model. Neighborhood-level social measures This study examines whether variation in neighborhood violent crime rates can be more comprehensively understood and predicted by simultaneously considering elements of the physical environment and structural level social and demographic correlates of crime. Prior research indicates certain characteristics of the social environment are consistent and robust correlates of levels of crime and violence (Land et al., 1990). However, one limitation in the extant literature concerns whether these measures operate in a similar fashion when employed jointly with a risk of crime measure based on the physical environment. Characteristics of the social environment examined in the current study are: concentrated socioeconomic disadvantage and residential stability (Morenoff, Sampson, & Raudenbush, 2001; Sampson, Raudenbush, & Earls, 1997), which are established structural level crime correlates. These measures were obtained from American Community Survey (ACS) 2009–2013 5-year estimates. Our measure of concentrated socioeconomic disadvantage is a summary index comprised of the average of six standardized items: median income (inverse), percent unemployed, percent of households in poverty, percent of households receiving public assistance (food stamps/ SNAP), percent of residents that are African American, and percent of households headed by a single female with children. A principle components analysis indicated these measures converged on a single
25
dimension with all factor loadings in excess of .806 with a Cronbach's α of .942. These findings supported the reliability of this summary index. We expected neighborhoods with greater levels of concentrated socioeconomic disadvantage to exhibit exacerbated violent crime rates. A summary measure capturing neighborhood levels of residential stability was also examined. The stability index was constructed through a principal component analysis that included percent of households that are owner-occupied and those in which residents have been in the community for at least a year. In support of the reliability of this summary index, results indicated the measures converged on a single dimension with all factor loadings in excess of .698 with a Cronbach's α of .696. As discussed in prior literature, residential stability is a neighborhood-level protective factor (Boggess & Hipp, 2010; Bursik & Grasmick, 1993; Kasarda & Janowitz, 1974; Sampson et al., 1997) whereas concentrated disadvantage is a well-established aggravating factor (Kubrin & Weitzer, 2003; Sampson et al., 1997). As such, the stability index was expected to be negatively associated with violent crime rates such that greater levels of residential stability will be associated with attenuated violent crime rates. Analytical strategy Our goal was to conduct a neighborhood level assessment of the influence of a measure estimating risk for criminal opportunities based on the physical environment. Once the RTM was estimated and spatially operationalized using 2013 violent crime, the average risk value per tract was determined. When spatially operationalizing the RTMDx, cells in Little Rock were limited to those within 20 ft. of streets. This excluded areas where streets were not located and where crime data could not be geocoded. This resulted in a final n of 33,449 cells that were capable of being aggregated to their respective census tracts based on cell centroids. This provided an average risk estimate of crime at the neighborhood-level (Census tract). This mismatch in units of analysis (i.e. street segments analyzed in RTM versus census defined areas) creates an issue known as the modifiable aerial unit problem (MAUP) (Bailey & Gatrell, 1995) because of the administrative boundaries at higher aggregate levels. The current study acknowledges this issue and advances the literature by examining the ability of RTM to construct an aggregate neighborhood risk of crime (ANROC) measure. Next, ordinary least squares (OLS) regression models predicting 2014 neighborhood violent crime rates across Little Rock neighborhoods were estimated while controlling for the social constructs described above. Separate regression models were employed to determine the unique influence the ANROC measure had on crime when the social factors were and were not included in the analysis. Results Prior to conducting regression analyses, the RTM analysis was performed to develop the ANROC measure. Fourteen risk factors were included in the RTMDx model examining 2013 violent crimes Table 2 RTMDx output for 2013 violent crimes. Risk factor
OP
SI (ft)
Coefficient
RRV
Public high schools Hotel/motels Lottery retailers Big box retail Bus stops Fast-food Conv. marts Bars/restaurants Grocery store Liquor stores Intercept
Proximity Density Density Density Proximity Density Proximity Proximity Proximity Proximity –
216 216 216 216 1512 216 432 432 1728 1728 –
2.656 2.380 2.104 2.048 1.563 1.237 0.774 0.599 0.5929 0.392 −4.991
14.240 10.801 8.202 7.756 4.775 3.445 2.168 1.821 1.809 1.480 –
Model: Negative binomial; BIC: 12,881.
26
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
employing negative binomial regression. The RTMDx output, presented in Table 2, represents the risk of crime (ROC) at a micro-unit based on a RAT approach. The results identified 10 significant risk factors, 6 operationalized as proximity risk factors and 4 as density. For example, being within three and a half blocks (1512 ft) of a bus stop is a risky environment, putting individuals at an elevated risk for violent victimization. The greatest risk factor associated with violent crime was the presence of a public high school. The relative risk values (RRVs) reported in Table 2 allow for comparison between risk factors. The RRV indicated being within close proximity (OP), 216 ft. (SI), to a public high school increased the likelihood of a violent crime by more than 14 times compared to cells with a risk value of 1. Big box retail stores were about half the risk of public high schools (7.756) but about 5 times as risky as liquor stores. The range of value for the ANROC measure was 1.00 to 46.06, with a mean of 9.36. Because the measure was positively skewed (2.50) and leptokurtic (8.70) we used a square-root transformation to induce normality.8 The transformed ANROC measure ranged from 1.00 to 5.79 with a mean of 2.83 and standard deviation of 1.17. In addition to conducting the RTM analysis to develop the ANROC measure, we examined descriptive statistics before conducting regression analyses. Table 3 summarizes the descriptive statistics and Table 4 presents correlations between each of the measures included in the regression analyses. The pattern of correlations presented in Table 4, specifically the association between 2014 crime rates and the predictive measures, as well as the associations between the predictor measures themselves, support our expectations. That is, our ANROC measure exhibits a significant and robust positive correlation with 2014 crime rates. In addition, the ANROC measure is positively associated with concentrated socioeconomic disadvantage (an aggravating factor) and negatively associated with residential stability (a protective factor). Substantively, these associations support an argument that the structure of routine activities and hence criminal opportunities and risk of victimization are exacerbated in structurally disadvantaged contextual environments yet attenuated in neighborhoods characterized by greater levels of residential stability. Next, we turn our attention to testing the utility of the ANROC measure in predicting 2014 crime rates. The results of our neighborhood (tract) level OLS regression analyses are presented in Table 5. Model 1 of Table 5 examines the effect of the ANROC measure on 2014 violent crime rates before the social demographic measures were included in the prediction equation. As expected, the ANROC measure was significantly and positively associated with violent crime rates, and accounted for a substantial portion (36%) of the variation in violence across neighborhoods. This finding supports the extant literature on the importance of routine activities for understanding and predicting levels of crime by providing an initial layer of evidence suggesting the prevalence of criminal opportunities available in neighborhoods assists in explaining variation in the relative prevalence of violent crime. To more comprehensively gauge the extent to which the ANROC measure may contribute to future analyses of the neighborhood context of crime and violence, additional measures were included in the analysis. In model 2 of Table 5, we added the concentrated disadvantage index to the prediction equation. As expected, both the ANROC measure and the concentrated disadvantage index were significantly and positively associated with variation in violent crime across neighborhoods. Table 3 Descriptive statistics.
Independent variables ANROC Con. dis. Stability
2014 crime rate ANROC Con. dis. Stability
Maximum
Mean
S.D.
0.000
43.870
10.657
9.645
1.000 −1.780 −2.420
6.790 1.830 1.780
2.834 0.000 0.000
1.166 0.880 0.876
2014 crime rate
ANROC
Con. dis.
Stability
–
0.612⁎⁎ –
0.567⁎⁎ 0.361⁎ –
−0.568⁎⁎ −0.509⁎⁎ −0.275 –
⁎ p b 0.05. ⁎⁎ p b 0.01.
These measures together explain 49% of the variation in violence across neighborhoods.9 It is important to note these effects were in the expected direction and highlight the relevance of both the physical and social environment in understating variation in violent crime across neighborhoods. That is the results suggest the ANROC and concentrated disadvantage measures are empirically distinct indicators of the social (disadvantage) and physical (ANROC) correlates of crime and violence in a neighborhood. Each factor offers a unique yet complimentary contribution to developing our understanding of neighborhood-level variation in violent crime. In model 3 of Table 5, we remove the concentrated disadvantage index but add the residential stability index to the prediction equation. Once again, the results support our expectations, with both the ANROC measure and the residential stability index exhibiting significant effects on violent crime rates but in divergent directions. The results support prior research indicating neighborhoods with a more established or tenured residential population and less population turnover experience attenuated levels of crime. These results contribute to the extant literature in indicating such findings on the social correlates of violent crime are robust to the inclusion of our ANROC measure, which taps the influence of routine activities, the built environment, and the prevalence of criminal opportunities on violent crime. Moreover, these results provide another layer of support for our contention that the social (residential stability) and physical (ANROC) correlates of crime and violence in a neighborhood are conceptually and empirically distinct. The residential stability and ANROC measures combine to account for 44% of the variance in 2014 neighborhood violent crime rates. Our final model, model 4, includes all three measures to offer a more comprehensive examination into whether the ANROC measure based on routine activities and criminal opportunities emanating from the built or physical environment is empirically distinct from multiple established social correlates of neighborhood levels of crime. When all three measures were simultaneously included in the analysis, they combined to explain 55% of variation in neighborhood violent crime rates.10 While each measure remained significant
Table 5 OLS regression results. Model 1
Model 2
Model 3
Model 4
ANROC
0.612⁎⁎⁎ 5.059 (0.944)
0.435⁎⁎⁎ 3.600 (0.944)
Con. dis.
–
Stability
–
0.468⁎⁎⁎ 3.870 (0.903) 0.398⁎⁎⁎ 4.366 (1.197) –
0.326⁎⁎ 2.694 (0.953) 0.367⁎⁎⁎ 4.019 (1.131) −0.302⁎⁎
Constant Minimum
Dependent variable 2014 crime rates
Table 4 Correlation matrix.
Model summary R2 Adjusted R2 Total # tracts ⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001.
–
−3.680 (2.889)
−0.289 (2.760)
−0.347⁎⁎ −3.822 (1.367) 0.456 (−3.822)
.374 .361 50
.512 .492 50
.464 .441 50
−3.326 (1.232) 3.021 (2.852)
.579 .552 50
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
and in the expected direction, there was a degree of attenuation in the effect of the ANROC measure on violent crime. However, this is to be expected due to the correlation between the predictor measures. Moreover, this interrelationship between the social and physical correlates of violent crime may be viewed as suggesting certain social correlates of crime and violence are, at least partially, the result of the impact of socioeconomic disadvantage and residential instability on routine activities and the convergence of motivated offenders, suitable targets, and capable guardians in time and space. Regardless, the overall pattern of results support our contention that the development of a macro- or neighborhood-level measure reflecting the risk for criminal opportunities contribute substantively to the future development of the neighborhoods and crime literature. While RTM is often employed as a micro-unit spatial analytical technique, these findings suggest RTM results for microunits can inform criminological research at larger aggregate units of analysis. Discussion Risk terrain modeling (RTM) is almost exclusively utilized in analyses of micro-units in the extant criminal justice research. Typically, RTM analyses are used within a policing framework to forecast future crime locations and diagnose risk factors in the physical environment. A singular focus on micro-units, often street segment or raster cells, offers a restricted perspective of the potential utility of RTM to diverse strands of criminological literature. The current study addressed this limitation by utilizing RTM to develop an indicator of neighborhood level victimization risk or criminal opportunity. RTM quantifies the risk of crime for micro-units, which allows for the calculation of the average risk across micro-units within Census tracts (i.e. neighborhood) and the creation of an Aggregated Neighborhood Risk Of Crime (ANROC) measure. Based on the premise that opportunities, and hence crime, are not random or uniformly dispersed across neighborhoods, our operationalization of an aggregate measure of risk was guided by insights gained from ecological criminological perspectives such as routine activities theory and the theory of risky places. Examining the correlates of variation in levels of crime across communities has long been a question of interest among sociologist and criminologists. In fact, the relationship between levels of crime and characteristics of communities was a concern for early Chicago School researchers (Shaw & McKay, 1942). The current study sought to bridge two largely distinct lines of research, RTM and communities and crime, to examine whether advances in these literatures potentially dovetail to provide a more comprehensive understanding of and ability to predict future levels of crime and violence. Specifically, we examined whether empirical advancements in RTM could increase our understanding of variation in neighborhood crime rates. Toward this end, we developed the ANROC measure using characteristics of the built or physical environment and tested whether the effect of this measure on crime rates was empirically distinct from characteristics of the social environment known to be robust predictors of crime. Results indicated that, controlling for established and theoretically relevant structural level social correlates of crime (i.e. concentrated disadvantage and residential stability), the ANROC measure significantly increased our ability to explain variation in neighborhood violent crime rates. The ANROC measure accounted for approximately 36% of the variation in violent crime rates (total variance explained about 55%). The convergence of criminal opportunities in time and space with motivated offenders is not a random phenomenon as crime clusters in specific places and often only during specific time periods (Sherman, 1995; Sherman & Weisburd, 1995). This has become a consistent finding in previous spatial analyses of crime, which overlaps considerably with an abundance of aggregate level criminological research completed over the prior 90 years. Drawing on routine
27
activities theory, a primary explanation for this variation in crime and violence across space and time is largely due to the potential for convergence between offenders, targets, and the lack of guardians. As shown in prior literature and the current study, RTM is an analytical tool capable of quantifying this variation in criminal opportunities and hence crime. Social or human ecology based criminological theories are rooted in the intersection of people and place. Both the physical environment and social structure are proffered to play a causal role in the occurrence, frequency, and distribution of crime by differentially exposing individuals to conditions conducive for offending and victimization. Social disorganization is an ecological theory emphasizing the role of the physical and social landscape in examining how community structure generates circumstances conducive to crime and hence exacerbated levels of crime. However, the emphasis on intervening mechanisms such as mainstream norms, network ties, and informal social control has led many to focus instead on the cultural deviance elements of social disorganization theory (Kornhauser, 1978). Indeed, the extant literature rooted in the disorganization framework has predominantly focused on analyses of concentrated socioeconomic disadvantage, residential stability, and informal social control. RTM is uniquely situated to allow researchers to substantively advance the literature concerning ecologically based theories of crime. RTM will contribute to advancing the extant literature by serving as a novel method that allows for the development of more refined measures of structural processes highlighted throughout classic and contemporary ecological theories (Park et al., 1925; Peterson & Krivo, 2010; Sampson, 2012; Shaw & McKay, 1942). Prior research has extensively investigated the relation between crime and elements of social structure yet there are limited empirical explorations of the physical environment or the simultaneous influence of the physical and social environments on crime. The results of the current study exemplify the utility of RTM in developing an indicator of victimization risk associated with the built or physical environment. Moreover, results indicate the impact of the physical and social environment are empirically distinct, opening the door for more comprehensive analyses and hence advancements to ecologically based theories of crime. The current research examined a novel application of quickly advancing technological abilities in developing measures of variation in criminal opportunities. This represents a substantive contribution to the emerging RTM and traditional communities and crime literatures by supporting the application of RTM methods toward advancing our understanding of neighborhood level criminological analyses. We believe the development of an ANROC measure to be both an innovative use of the most current and sophisticated analytical methods as well as a substantive addition to our understanding of crime patterns. We encourage future researchers to continue in this vein by expanding our methodological tool kit by drawing on RTM to construct indicators of risk for criminal opportunities and exploring the capabilities of such methods in addressing key questions and dilemmas concerning the ecological nature of crime. Additional rigorous tests of the capabilities of RTM will allow for a better understanding of the capabilities and limitations of the diagnostic tool in enhancing criminological research. One issue that will be critical to the development of future researchers is the availability of the requisite data that will allow for the accurate quantification of risk. RTM reflects the risk for crime based on the physical landscape. The current study collected data from a variety of sources to develop indicators of potential risk factors. However, there are a number of complications and hurdles researchers are likely to encounter when seeking to obtain similar data in different locales. In fact, gathering the requisite data may not be possible in every city, limiting our ability to fully quantify the riskiness of a study area. A limitation of the current study that should be explored in future research concerns our reliance on Census tracts as a proxy for neighborhoods. Tracts represent a compromise between data availability, unit size, comparability, and homogeneity (White, 1987), however,
28
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29
neighborhoods are socially defined as opposed to being encapsulated by arbitrary administrative boundaries (Chainey & Ratcliffe, 2005). That said, tracts have greater face validity in terms of representing neighborhoods or communities compared to a vast literature that draws on substantially larger and diverse units such as cities, counties, or metropolitan statistical areas. In addition, the use of tracts as proxies for neighborhoods is supported by prior research indicating the correlates of crime and violence are stable across various units of analysis (Land et al., 1990). Further, Sampson (2013, p. 9) suggests, “the empirical search for the correct operational definition of neighborhood or place is misplaced…and the concept of place ranges over units big and small. The phenomenon of crime does not privilege one type of place or ecological unit either. Crime varies within societies, states, counties, cities and certainly within neighborhood.” Future analyses of within neighborhood variation in risk are likely to be particular important to advancing our understanding of place and crime. The micro-units upon which our RTM analysis was constructed exhibit considerable variation, both between and within Census tracts. That is, similar to the social correlates of crime and crime itself, there is considerable heterogeneity in our measure of risk within neighborhoods. While our goal was to examine the utility of a measure of risk aggregated to the neighborhood level, heterogeneity in this measure of risk across micro-units is an indicator that future research should explore the potential impact of aggregation bias. The ANROC measure creates a global measure of risk for each census tract, in a fashion similar to the creation of the disadvantage and residential stability indices. At issue, is that such aggregation bias may obfuscate the true relation between risk and crime. This aggregate measure obscures the very precise distinction in the variation of risk and criminal opportunities across the geographic landscape. As a result, some element of the association between the ANROC measure and crime may be somewhat of a statistical artifact or “noise” resulting from the true relation between risk and crime operating at the micro-unit level. Such concerns, which are not unique to the current study, coupled with the nature of the data from which measures of risk must be developed provide both a rationale and the means by which to conduct future hierarchical analyses. Multilevel analyses would allow future research to take into account the fact that micro-units are nested within larger units of analysis and risk is not a global or homogenous characteristic of census tracts. Such analyses would clarify the level or levels at which the association between risk and crime is most evident, allowing for the simultaneous exploration of micro- and macro-unit predictors of crime, and could lead to the development of more refined policy implications centered on examining the nature of micro-risky places in relation to the larger physical and social landscape. Notes 1 Publically available individual block level data do exist but many measures are not available because of the potential that individuals may be identifiable. Future research should examine RTM in relation to individual-block-level data to determine the applicability. 2 Two tracts were excluded from the analyses based on the lack of cells that were located within them during the RTM analysis. The two tracts were only partially in Little Rock and had fewer than 100 cells, meaning that there were few streets that intersected those census tracts. 3 Aggravated assaults pertaining to law enforcement officers were excluded. 4 The distinction between crime generators and attractors was not examined. It is possible, depending on time of day, that certain types of establishments could be crime generators and, at other times, crime attractors. 5 Data for lottery retailers in 2013 were not available. 6 Data for establishments with a liquor license in 2013 or 2014 were not available. 7 Our discussion of RTMDx centers on the analytical capabilities of RTM as it relates to our research design. Additional resources are available online through Rutgers Center on Public Security (rutgerscps.weebly.com). 8 This transformation provides a more accurate estimate but supplemental analyses indicated this transformation did not substantively alter the results presented below. 9 Variance Inflation Factor (VIF) values for all models presented were below 1.5.
10 Supplemental analyses of residuals from the OLS regression were undertaken to examine if spatial autocorrelation was problematic. Spatially lagged regression was utilized to examine neighborhood violent crime rates and found the measures remained significant predictors.
References Anderson, E. (1999). Code of the street: Decency, violence, and the moral life of the inner city. New York: W.W. Norton & Co. Bailey, T.C., & Gatrell, A.C. (1995). Interactive spatial data analysis. Essex, UK: Addison Wesley Longman. Bernasco, W., & Block, R. (2011). Robberies in Chicago: A block-level analysis of the influence of crime generators, crime attractors, and offender anchor points. Journal of Research in Crime and Delinquency, 48(1), 33–57. Bichler, G., Schmerler, K., & Enrinquez, J. (2013). Curbing nuisance motels: An evaluation of police as place regulators. Policing: An International Journal of Police Strategies & Management, 36(2), 437–462. Boggess, L.N., & Hipp, J.R. (2010). Violent crime, residential instability and mobility: Does the relationship differ in minority neighborhoods? Journal of Quantitative Criminology, 26(3), 351–370. Brantingham, P. (2011). Crime and place: Rapidly evolving research methods in the 21st century. Cityscape, 13(3), 199–203. Brantingham, P.J., & Brantingham, P.L. (1981). Notes on the geometry of crime. In P.J. Brantingham, & P.L. Brantingham (Eds.), Environmental criminology (pp. 27–54). Thousand Oaks, CA: SAGE. Brantingham, P.L., & Brantingham, P.J. (1993). Environment, routine and situation: Toward a pattern theory of crime. In R.V. Clarke, & M. Felson (Eds.), Advances in criminological literature (pp. 259–294). New Brunswick, NJ: Transaction Books. Brantingham, P., & Brantingham, P. (1995). Criminality of place. European Journal on Criminal Policy and Research, 3(3), 5–26. Burchfield, K.B. (2009). Attachment as a source of informal social control in urban neighborhoods. Journal of Criminal Justice, 37, 45–54. Bursik, R.J. (1988). Social disorganization and theories of crime and delinquency: Problems and prospects. Criminology, 26, 519–552. Bursik, R.J., & Grasmick, H.G. (1993). Neighborhoods and crime: The dimensions of effective community control. New York, NY: Lexington Books. Caplan, J.M., & Kennedy, L.W. (2013). Risk terrain modeling diagnostics utility (version 1.0). Newark, NJ: Rutgers Center on Public Safety. Caplan, J.M., Kennedy, L.W., & Miller, J. (2011). Risk terrain modeling: Brokering criminological theory and GIS methods for crime forecasting. Justice Quarterly, 28(2), 360–381. Caplan, J.M., Kennedy, L.W., & Piza, E.L. (2013). Risk terrain modeling diagnostics utility user manual (version 1.0). Newark, NJ: Rutgers Center on Public Security. Chainey, S., & Ratcliffe, J.H. (2005). GIS and crime mapping. London: John Wiley and Sons. Cohen, L.E., & Felson, M. (1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588–608. Drawve, G. (2014). A metric comparison of predictive hot spot techniques and RTM. Justice Quarterly. http://dx.doi.org/10.1080/07418825.2014.904393 [Available online]. Drawve, G., Moak, S.C., & Berthelot, E.R. (2014a). Predictability of gun crimes: A comparison of hot spot and risk terrain modeling techniques. Policing and Society. http://dx. doi.org/10.1080/10439463.2014.942851 [Available online]. Drawve, G., Thomas, S.A., & Walker, J.T. (2014b). The likelihood of arrest: A routine activity theory approach. American Journal of Criminal Justice, 39(3), 450–470. Dugato, M. (2013). Assessing the validity of risk terrain modeling in a European city: Preventing robberies in the city of Milan. Crime Mapping, 5(1), 63–89. Eck, J. (1995). A general model of illicit retail marketplaces. In J. Eck, & D. Weisburd (Eds.), Crime and place (pp. 67–93). New York: Criminal Justice Press. Groff, E. (2007). Simulation for theory testing and experimentation: An example using routine activity theory and street robbery. Journal of Quantitative Criminology, 23(2), 75–103. Groff, E. (2011). Exploring ‘near’: Characterizing the spatial extent of drinking place influence on crime. Australian and New Zealand Journal of Criminology, 44(2), 156–179. Heffner, J. (2013). Statistics of the RTMDx utility. In J. Caplan, L. Kennedy, and E. Piza, Risk terrain modeling diagnostics utility user manual (Version 1.0). Newark, NJ: Rutgers Center on Public Security. Kasarda, J.D., & Janowitz, M. (1974). Community attachment in mass society. American Sociological Review, 39, 328–339. Kennedy, L.W., & Caplan, J.M. (2013). A theory of risk places. Research brief. Newark: Rutgers Center on Public Security. Kennedy, L.W., & Forde, D.R. (1990). Routine activities and crime: An analysis of victimization in Canada. Criminology, 28(1), 137–152. Kennedy, L.W., Caplan, J.M., & Piza, E. (2011). Risk clusters, hotspots, and spatial intelligence: Risk terrain modeling as an algorithm for police resource allocation strategies. Journal of Quantitative Criminology, 27(3), 339–362. Kennedy, L.W., Caplan, J.M., Piza, E.L., & Buccine-Schraeder, H. (2015). Vulnerability and exposure to crime: Applying risk terrain modeling to the study of assault in Chicago. Applied Spatial Analysis and Policy [Online]. Kornhauser, R.R. (1978). Social sources of delinquency: An appraisal of analytic models. Chicago, IL: University of Chicago Press. Koss, K.K. (2015). Leveraging predictive policing algorithms to restore fourth amendment protections in high-crime areas in a post-Wardlow world. Chicago-Kent Law Review, 90(1), 301–334.
G. Drawve et al. / Journal of Criminal Justice 44 (2016) 21–29 Krivo, L.J., & Peterson, R.D. (2000). The structural context of homicide: Accounting for racial differences in process. American Sociological Review, 65, 547–559. Kubrin, C.E., & Weitzer, R. (2003). Retaliatory homicide: Concentrated disadvantage and neighborhood crime. Social Problems, 50, 157–180. Land, K.C., McCall, P.L., & Cohen, L.E. (1990). Structural covariates of homicide rates: Are there any invariances across time and social space? American Journal of Sociology, 95(4), 922–963. LeBeau, J.L. (1997). Demonstrating the analytical utility of GIS for police operations. Washington, DC: National Institute of Justice. LeBeau, J.L. (2011). Sleeping with strangers: Hotels and motels as crime attractors and crime generators. In M.A. Andresen, & J.B. Kinney (Eds.), Patterns, prevention, and geometry of crime (pp. 77–102). New York: Routledge. Lee, M.R. (2000). Concentrated poverty, race, and homicide. The Sociological Quarterly, 41 (2), 189–206. Levine, N., Wachs, M., & Shirazi, E. (1986). Crime at bus stops: A study of environmental factors. Journal of Architectural and Planning Research, 3, 339–361. McCall, P.L., Land, K.C., & Parker, K.F. (2010). What do we know about the structural covariates of homicide rates? A return to a classic twenty years later. Homicide Studies, 14(3), 219–243. McCord, E.S., & Ratcliffe, J.H. (2007). A micro-spatial analysis of the demographic and criminogenic environment of drug markets in Philadelphia. The Australian and New Zealand Journal of Criminology, 40(1), 43–63. McCord, E.S., Ratcliffe, J.H., Garcia, R.M., & Taylor, R.B. (2007). Nonresidential crime attractors and generators elevate perceived neighborhood crime and incivilities. Journal of Research in Crime and Delinquency, 44(3), 295–320. Morenoff, J.D., & Sampson, R.J. (1997). Violent crime and the spatial dynamics of neighborhood transition: Chicago, 1970–1990. Social Forces, 76(1), 31–64. Morenoff, J.D., Sampson, R.J., & Raudenbush, S.W. (2001). Neighborhood inequality, collective efficacy, and the spatial dynamics of urban violence. Criminology, 39(3), 517–560. Moreto, W.D., Piza, E.L., & Caplan, J.M. (2014). “A plague on both your houses?”: Risks, repeats and reconsiderations of urban residential burglary. Justice Quarterly, 31(6), 1102–1126. Park, R.E., Burgess, E.W., & McKenzie, R.D. (1925). The city: Suggestions for investigation of human behavior in the urban environment. Chicago, IL: University of Chicago Press. Peterson, R.D., & Krivo, L.J. (2010). Divergent social worlds: Neighborhood crime and the racial-spatial divide. New York: Russell Sage. Pridemore, W. A., & Grubesic, T. H. (2011). Alcohol outlets and community levels of interpersonal violence: Spatial density, outlet type, and seriousness of assault. Journal of Research in Crime and Delinquency, 50(1), 132–159. Ratcliffe, J.H. (2012). The spatial extent of criminogenic places: A changepoint regression of violence around bars. Geographical Analysis, 44(4), 302–320. Roncek, D.W., & Bell, R. (1981). Bars, blocks, and crimes. Journal of Environmental Systems, 11(1), 35–47.
29
Roncek, D., & Maier, P. (1991). Bars, blocks, and crime revisited: Linking the theory of routine activities to the empiricism of ‘hot spots’. Criminology, 29, 725–754. Roncek, D.W., & Pravatiner, M.A. (1989). Additional evidence that taverns enhance nearby crime. Sociology and Social Research, 73(4), 185–188. Sampson, R.J. (1988). Local friendship ties and community attachment in mass society: A multilevel systemic model. American Sociological Review, 53, 766–779. Sampson, R.J. (2012). Great American city: Chicago and the enduring neighborhood effect. Chicago, IL: The University of Chicago Press. Sampson, R.J. (2013). The place of context: A theory and strategy for criminology's hard problems. Criminology, 51(1), 1–31. Sampson, R.J., & Groves, W.B. (1989). Community structure and crime: Testing social disorganization theory. American Journal of Sociology, 94, 774–802. Sampson, R.J., Raudenbush, S.W., & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918–924. Shaw, C.R., & McKay, H.D. (1942). Juvenile delinquency in urban areas. Chicago: University of Chicago Press. Sherman, L. (1995). Hot spots of crime and criminal careers of places. In J. Eck, & D. Weisburd (Eds.), Crime and place: Crime prevention studies 4. Monsey, NY: Willow Tree Press. Sherman, L., & Weisburd, D. (1995). General deterrent effects of police patrol in crime “hot spots”: A randomized controlled trial. Justice Quarterly, 12, 626–648. Spelman, W. (1995). Criminal careers of public places. Crime and Place, 4, 115–144. Thomas, S.A., & Shihadeh, E.S. (2013). Institutional isolation and crime: The mediating effect of disengaged youth on levels of crime. Social Science Research, 42, 1167–1179. Warner, B.D. (2003). The role of attenuated culture in social disorganization theory. Criminology, 41, 73–97. Warner, B.D. (2014). Neighborhood factors related to the likelihood of successful informal social control efforts. Journal of Criminal Justice, 42(5), 421–430. Weisburd, D., Groff, E.R., & Yang, S. (2012). The criminology of place: Street segments and our understanding of the crime problem. New York, NY: Oxford University Press. White M.J., American neighborhoods and residential differentiation (1987). Russell Sage. New York: NY. Wilson, W.J. (1987). The truly disadvantaged: The inner city, the underclass, and public policy. Chicago: The University of Chicago Press. Wilson, W.J. (1996). When work disappears: The world of the New urban poor. New York: Alfred A. Knopf. Wilson, W.J. (2009). More than just race: Being black and poor in the inner city. New York: W.W. Norton. Wright, R.T., & Decker, S.H. (1996). Burglars on the job: Streetlife and residential break-ins. NH: Northeastern University Press.