This study examines the environmental predictors that classify street robbery hot spots and control street segments in Indianapolis. Empirical controls were generated by matching each hot spot to a corresponding set of zero-crime control and low-crime control units. Then, units were evaluated based on the presence of crime generators and attractors, which were downloaded from open data sources and spatially joined to the street segments, and disorder indicators obtained via systematic social observation using Google Street View. The findings provide information about the influence environmental predictors have on the presence of street robbery hot spots, and whether the composition of hot spots significantly differs from that of similar places that experienced no crime or low counts of crime.
hot spots, environmental criminology, systematic social observation, quasi-experiment, propensity score matching,
Connealy, N. (2020). Understanding the Predictors of Street Robbery Hot Spots: A Matched Pairs Analysis and Systematic Social Observation. Crime & Delinquency. https://doi.org/10.1177/0011128720926116
The author would like to thank Dr. Jean McGloin, Dr. Chris Sullivan, and the students at the graduate student summer research workshop for their feedback and contributions to the present manuscript.
Environmental criminology (EC), which situates crime in time and space (P. J. Brantingham & Brangintham, 1981), has become an increasingly prevalent framework for attempting to understand how the characteristics of the environment influence the likelihood of crime and the location of hot spots. One branch of EC emphasizes the importance of crime predictors like the types of businesses and establishments in the immediate vicinity. Another important branch of EC considers crime predictors related to the level of visible environmental disorder. Due to the pronounced nuances from unit to unit across even the most microenvironments, the different EC branches and sets of predictors used to understand the characteristics of hot spots are often tested separately. Resultantly, the insights and conclusions produced across the different EC domains and predictor types have largely been treated as distinct explanations of crime. If the field is to continue researching and implementing solutions related to hot spots, important questions remain about the specific environmental features that inform crime occurrence and hot spot composition.
The current study simultaneously incorporates different EC predictors to determine which features of the environment best characterize street robbery hot spot street segments in Indianapolis. Crime generators and attractors (CGAs) were incorporated through spatial data, and Google Street View was used to conduct a systematic social observation (SSO) to measure visible environmental disorder and decay. For the SSO, statistically similar control street segments were generated for comparison using propensity score matching. The presence, and co-location, of CGAs and environmental disorder was then tested to determine the environmental predictors that significantly distinguish hot spot street segments from empirical controls.
This study sets out to do three things. First, testing different sets of environmental predictors sheds insight on which features of the environment are significantly associated with street robbery hot spots. Such conclusions contribute to our understanding of how multiple domains of EC may be related to the presence of crime and hot spots. Second, the study uses a matching technique that rigorously tests the SSO framework in a case-controlled, quasi-experimental design. The matching approach in this study is the first to involve multiple covariates and geographic proximity requirements for treated and control units prior to conducting SSO. Third, the study leverages several open source data platforms to innovatively apply publicly accessible spatial datasets and uses a free, online virtual platform to conduct the SSO. The use of open data and SSO provides insights into the relevance of these data types and methods in micro-level crime and place research, especially for capturing phenomena like disorder and decay that are not regulated with official datasets.
Review of Relevant Literature
Hot Spots of Crime
Across the field, there has been a fundamental acceptance for the notion that crime is not evenly or randomly dispersed, and that crime tends to disproportionately concentrate at relatively few locations (Weisburd, 2015). The places where crime concentrates are called “hot spots” due to their elevated level of risk and crime relative to other places (Sherman & Weisburd, 1995). Crime and place research has started to prioritize the micro hot spot locations where crime concentrates the most, operationalizing the study of hot spots using units of analysis like block-groups (Boessen & Hipp, 2015), street segments (Weisburd, Groff & Yang, 2012) and street intersections (Braga, Papachristos & Hureau, 2010). Important conclusions have emerged demonstrating that crime disproportionately concentrates to few locations in a study area (Sherman, Gartin & Buerger, 1989; Weisburd & Amram, 2014), crime concentrations are reducible to micro spatial extents (Weisburd, 2015), and crime levels at hot spots tend to remain high and stable over time (Andresen, Linning & Malleson, 2017). Hot spot techniques have been widely used to identify and predict high crime locations (Chainey, Tompson & Uhlig, 2008), and large body of evidence exists highlighting place-based policing efforts and their effectiveness in reducing crime (Braga et al., 2019).
In addition to the growing repository of evidence on the efficacy of hot spots policing tactics and their ability to reduce crime, research has also attempted to determine the features that make a hot spot “hot.” Studies have examined the composition, contexts, and characteristics associated with hot spot environments, and evidence has emerged for perspectives related to the potential criminogenic influence of places like businesses and establishments (Bernasco & Block, 2011) as well as the presence of environmental disorder (Sampson & Raudenbush, 2001). Therefore, the impact of the environment on crime has been found to be multi-faceted, with several different predictor types exerting an influence on crime events and hot spot locations.
According to crime pattern theory (CPT), the most salient crime predictors are often classifiable as crime generators or crime attractors (Brantingham & Brantingham, 1995; 2008). Crime generators are features that are not inherently criminogenic but make a place more predisposed to crime due to the congregation of large amounts of people. Conversely, crime attractors increase the likelihood of crime by promoting and reinforcing well-known opportunities to engage in crime. Research employing CPT has underscored the influence of micro-level factors such as business types, shared outdoor spaces, and places of public interest and their relationship to crime through the concurrent creation of both activity and opportunity (Brantingham & Brantingham, 1993; 2008; Clutter, Henderson & Haberman, 2019; Tillyer, Wilcox & Walter, 2020). A multitude of techniques have been operationalized to assess the spatial risk and influence of such CGAs. Near repeat analyses (Knox, 1964; Morgan, 2000) focus on the places that experience multiple instances of crime by identifying and quantifying the characteristics that make an area more susceptible to crime. Similarly, recent scholarship has demonstrated that crime can be accurately forecasted based on the components of a location through techniques like machine learning (Wheeler & Steenbeek, 2020), predictive policing (Perry et al., 2013), and spatial analytics (McCord & Ratcliffe, 2007). The research base on CGAs is well-established, and widely accepted, and includes a wealth of studies specific to the risk for robbery (Barnum et al., 2017; Bernasco & Block, 2011; Bernasco, Block & Ruiter, 2013; Connealy & Piza, 2019; Haberman & Ratcliffe, 2015).
Another important set of crime predictors includes variables related to environmental disorder. Prior research has distinguished between two forms of environmentally-based disorder, labeling the forms of disorder that are short-term, often temporary, and easier to rectify as general physical disorder (graffiti, littering, etc...), and the long-term, more permanent, deleterious conditions associated with neglect as decay (dilapidation, deterioration, etc...) (Taylor, 2001; Taylor, Gottfredson & Brower, 1985; Wheeler, 2018). The characteristics, condition, and the level of visible disorder and decay present in an environment may vary from place to place, influencing the likelihood of crime in unique ways. Research has previously found that there is a positive relationship between disorder and crime (Skogan, 1990; Skogan & Steiner, 2004), and that crime hot spots are often characterized by the presence of disorder (Telep & Hibdon, 2019; Weisburd & Mazerolle, 2000). Capturing the level of disorder and decay within and across places may be an important consideration when determining the presence of hot spots, as these constructs are indicators of both crime (Skogan & Steiner, 2004) and perpetual disorder (Steenbeek & Hipp, 2011). However, there is also contrary evidence suggesting that disorder has little, if any, effect on crime occurrence when controlling for neighborhood level characteristics (Harcourt & Ludwig, 2006; Sampson & Raudenbush, 2004). Thus, it is important to include disorder measures, particularly at more micro-level units of analysis, to help untangle the debate around their influence on crime and hot spots. The potential presence of disorder and decay, and/or the co-location of these elements with CGAs, may help explain the environments influence on crime and the location of hot spots.
Measurement, Methods and Matching
Both CGAs and environmental disorder are widely researched concepts across the field, but still are generally analyzed separately. With several potential predictor types influencing the level of crime and the location of hot spots, it is important for the field to consider all potential hot spot predictors and perspectives through integrated research approaches. Although only a few studies have jointly examined the potential linkages between CGAs and disorder (McCord & Ratcliffe, 2007; Rice & Smith, 2002; Weisburd, Groff & Yang, 2012), the theoretical grounds for the crossover between these predictors types has been previously advocated. Theorists have proposed that opportunity theory variables like CGAs can been combined with disorder-based constructs from social disorganization and broken windows theory to better understand the context of high crime places (Miethe & Meier, 1990; Weisburd, Groff & Yang, 2012; Braga & Clarke, 2014). However, despite this theoretically informed connection, there are several reasons different predictor types are not often simultaneously examined empirically, including researcher operationalization, data availability, and measurement limitations. Pertaining to researcher operationalization, studies on the influence of the environment tend to focus exclusively on the singular presence of CGAs or disorder as opposed to the potential interrelatedness of the variables. Since predictor sets are classified under different EC domains, the motivating angle for a research question or hypothesis often only draws from one explanation of why crime might occur as a result of the environment, therefore, limiting the number of predictor types included.
Data availability has also limited the breadth and scope of spatial research, as environmental studies testing EC predictors require robust datasets. Micro-level CGA research commonly necessitates datasets with highly specific locational information and often entails the inclusion of many unique variables. In a similar vein, because disorder data is not officially collected and maintained, disorder research tends to rely heavily on proxy measures of disorder like 311 calls that are related to damages or maintenance (O’Brien, Gordon & Baldwin, 2014; Wheeler, 2018). These proxy measures may introduce construct and content validity issues because they are not designed to effectively capture disorder, are often recorded at zip code or neighborhood-level spatial extents (Swatt et al., 2013; Wyant, 2008), and are subject to a high degree of subjectivity error (Gau & Pratt, 2008; Hipp, 2010). Measurement limitations have also constrained the ability to mutually consider the influence of different EC predictor types. For example, because disorder is not captured in an official, regulated dataset, new methods are required for measuring and evaluating the level of disorder and decay in micro-level research.
SSO is an example of an alternative way to capture place specific measures of disorder more objectively and reliably. SSO involves the visual evaluation of an environment through a rigorous, consistent, and replicable approach of observing and coding predetermined features of interest (Reiss, 1968; 1971). SSO has a long history of application to understanding policing functions, with research covering topics such as the use of force (Bayley, 1986; Terrill & Reisig, 2003), police stops and citizen interactions (Gould & Mastrofski, 2004), and officer allocation to hot spots (Sherman & Weisburd, 1995). The technique has also been previously applied to capture environmental disorder, but mostly at the neighborhood level to evaluate concepts like collective efficacy (Sampson & Raudenbush, 1999) and social control (Taylor, 1997). These early SSOs of disorder required onsite coding or the collection of video recordings that could be viewed and coded secondarily (Sampson & Raudenbush, 1999). Sampson and Raudenbush (1999) first used video recordings in a seminal SSO that evaluated the physical characteristics of disorder and their relationship to social cohesion and crime. One of the lasting implications of this research was the development of a physical disorder index, which was later expanded in a collaborative study with Odgers and colleagues (2012) to include an index for disorder constructs related to decay. However, as a result of data advances and technological capacities, SSO research has started to move beyond exclusively observing places and events in-person.
The concepts associated with environmental disorder are now more readily defined and can be studied through remote SSO with new technological interfaces like Google Earth and Street View (He, Paez & Liu, 2016; Hoeben, Steenbeek & Pauwels, 2018). Video and remote techniques are advantageous for SSO in that they provide permanently stored observations that can be revisited to help mitigate recall errors (Mastrofski, Parks & McClusky, 2010), they can be more readily tested for reliability across coding schema, coders, and time periods (Lindegaard & Bernasco, 2018), and they can be re-evaluated in light of important new developments (Sampson & Raudenbush, 1999). Subsequent testing has also found modern applications of remote SSO to be an effective way to approach measuring disorder. Clarke and colleagues (2010) studied the reliability and validity of remote SSO research and found the results to be consistent across site visits and virtual evaluations, and Google Street View (GSV) has demonstrated reliability in audit research evaluating structural conditions and environmental constructs (Ben-Joseph et al., 2013; Rundle et al., 2011). GSV has also emerged as a popular tool in criminal justice and social science research focused on characterizing small scale environments (He, Páez, & Liu, 2017; Hsu & Miller, 2017; Langton & Steenbeek, 2017; Vandeviver, 2014). Previous remote SSOs in criminology have used GSV to assess the sensory experiences of crime and the importance of area appearance in fear of crime and recounting victimization (Hur & Nasar, 2014; Toet & Van Schaik, 2012), and to evaluate the characteristics that influence drug transaction decision-making in Newark, NJ (Sytsma, Connealy & Piza, 2020). In public health and city planning, researchers have also used remote SSO to conduct environmental audits in order to designate different types and perceptions of built environments in urban settings (Salesses, Schechtner, & Hidalgo, 2013).
As SSO continues to gain traction for studying micro-environments, it is important to ground the results based on the methodological design of the study. When randomized control trials are not possible, as is often the case in criminal justice research, quasi-experimental designs are considered the optimal evaluation method (Eck, 2006; Eck, 2017; Ratcliffe, 2019). One of the first case-controlled applications of SSO involved comparing drug markets to controls to determine if there were significant differences in environmental characteristics (Eck, 1994). The case-controlled application of the design, however, only involved random matching of treated and control units in the same census block group without making other covariate considerations. Similarly, Langton and Steenbeek (2017) randomly matched units within a predefined spatial extent in their SSO of target selection for residential burglary. Most recently, Schnell, Grossman and Braga (2019) attempted to match high crime environments (their dependent variable) to low crime controls across several covariates. They were able to effectively reduce imbalance across their matched samples using two covariates but were unable to account for spatial proximity. Thus, they chose to use theoretically informed matching, but still matched on the dependent variable. The present study successfully used PSM to match on the characteristics of individual hot spot units while ensuring matched unit pairs were within the same neighborhood.
Scope of the Current Study
Micro-level hot spots research has gained momentum in criminology because the datasets needed to test these highly localized places are now more actively collected, maintained, and publicly available. Additionally, the evidence-base advocating for micro-level units of analysis is continuing to grow more robust (Weisburd, 2015). Despite the advances in data, methods, and evidence, though, a research gap still exists in that different EC predictor types are still treated as separate explanations for crime and hot spot presence. Due to the linkages both CGAs and disorder variables have demonstrated with crime, there is reason to believe both predictor types are significantly present and influence the location of street robbery hot spots. Such conclusions would inform our understanding of the most pertinent environmental characteristics of hot spots.
In addition to providing new evidence on hot spot composition, the present study’s use of SSO with a case-controlled design represents the next steps in this area of research. Using remote SSO to study disorder provides several strategic advantages over other disorder data sources including objective, consistent, and replicable results of specific areas of interest and the ability to record unique environmental characteristics and features that may not be collected by official records. This study also innovatively utilizes propensity score matching to compare hot spots to empirically derived controls. Empirical matching prior to SSO has not been done in the field before in a case-controlled design involving multiple covariates within fixed geographic parameters. This study provides a template for future SSO or other related studies that require “treated” units to be matched to real, individual controls as opposed to synthetic or mock units.
Study Setting and Data Sources
According to 2017 census population estimates, Indianapolis was the 16th most populous city in the United States and the 16th largest by land area. The Federal Bureau of Investigations (FBI) Uniform Crime Report (UCR) indicates that Indianapolis had a robbery rate of 99 per 100,000 people in 2017, which was essentially equivalent to the US rate of 98 per 100,000. As a larger metropolitan city mirroring the US average for robbery, Indianapolis provides an ideal backdrop to explore street robbery hot spot contexts and characteristics. Street robbery was selected as the crime of interest because it is more likely to be influenced by the structural and place features of the environment since it occurs outdoors and because it varies based on the level of activity in the immediate area (Haberman, Sorg, & Ratcliffe, 2017). Further, the street robbery counts in Indianapolis remained relatively stable over the five-year period of interest.
The crime data were downloaded from the Indianapolis Open Data Portal and included all robbery incidents from 2013-20171. The robbery data was parsed out by an existing column that indicated if the incident was a street robbery, and five years of data were incorporated to identify the segments that were persisting hot spots. The micro-level CGA datasets used in this study came from several different sources. In total, the study accounted for 25 micro-level CGAs.2 The variables used are common to the EC body of work on CGAs and were more specifically drawn from prior micro-level spatial studies conducted in Indianapolis (Connealy, 2019; Piza & Carter, 2018). The first data source used to glean data, the Indianapolis Open Data Portal, is a publicly accessible platform that includes a multitude of city-maintained datasets in different vector formats.3 The business data provider InfoGroup was also used in the study to gather business-related, crime generator datasets.4 InfoGroup has been previously used in spatial research (Connealy & Piza, 2019; Miller, Caplan, & Ostermann, 2016; Piza, Caplan, & Kennedy, 2014) and rigorously maintains their business datasets to ensure active licensure. The Environmental Systems Research Institute (ESRI) repository of spatial data, ArcGIS online5, which is generally maintained by geographic information systems (GIS) professionals, was also used to download spatial data. Lastly, Google MyMaps,6 which allows users to search for establishment types and download the search results into a format amenable in a GIS, was used to collect data. In addition to the sources previously noted, data for the covariate matching and statistical analyses also came from the US Census Bureau American Community Survey.
Unit of Analysis
The unit of analysis was individual street segments in Indianapolis. There are 62,667 segments in Indianapolis, with individual segments comprising both block faces of a street between two intersections. Street segments have demonstrated utility as a unit of analysis because they minimize geocoding errors by providing fixed unit boundaries, they capture micro-level social activity (Braga, Papachristos & Hureau, 2010), and have become a focal, semi-standardized unit size for crime and place research (Weisburd, Bernasco & Bruinsma, 2008; Weisburd, Groff, & Yang, 2012). All street robbery incidents were joined to the street segment they fell on and a total count field of all street robbery incidents at an individual segment from 2013-2017 was calculated. Of the 62,667 segments in Indianapolis, 5,811 experienced at least one count of street robbery during the study period and were considered for the subsequent hot spot sample. These 5,811 segments were then broken down by standard deviations, with segments greater than or equal to three standard deviations from the mean level of robberies considered hot spots. Thus, the 99th percentile consisted of 144 segments (0.23% of the study area) that experienced between 6-28 robberies during the five-year period. This approach ensures that only the segments experiencing the greatest frequencies of street robbery were designated as hot spots (Haberman, 2017). The low number of selected hot spot segments also makes the coding efforts feasible, and far exceeds the number of hot spots generally identified in studies predicated on generating actionable results at specific places (Haberman, 2017).
In order to determine the EC predictors present in hot spots, empirical counterfactuals were generated. Segments that did not experience any instances of street robbery during the five-year study period comprised the first control pool, deemed the “zero-crime controls.” The street segments that experienced street robbery but were not classified as hot spots were removed from this pool of controls, leaving 56,856 segments that experienced no instances of street robbery over the five-year time period as potential control units. This control approach allows for the identified hot spots to be compared to characteristically similar street segments that did not experience any street robbery. Comparing the hot spots to units that did not experience crime better isolates the relationship between crime and the environment and helps determine if there are features that are more significantly present in hot spots then in places without crime.
A second control pool was also utilized to produce more robust findings. This control pool only drew from the 5,667 segments that experienced at least one instance of street robbery during the five-year study period but were not classified as hot spots.7 The hot spot segments were also independently compared to these “low crime control” segments. If a feature is found to be significantly present in hot spots relative to the low crime units, it suggests that the feature is salient to hot spot presence. Alternatively, criminogenic features that do not inform hot spot presence can be separated by identifying the features that distinguish crime environments (hot spots and low crime controls) from places without crime, but do not demonstrate a significant presence in hot spots compared to the low crime units. This distinction allows for environmental CGAs and disorder to be better understood in their influence on crime and hot spots.
The operationalization of two control pools allows for the diagnosis of the significant environmental factors for both crime occurrence and for hot spots, and the identification of the factors that have no effect on crime. This helps move the literature forward on the specific features that distinguish crime environments from other locations. An additional benefit of multiple control groups is the more effective comparison of treated units to empirically similar units. Most studies utilize a singular control pool that often comprises the entirety of the “non-treated” study area. Wheeler and Steenbeek (2020) suggest that the presence and influence of environmental features at the locations where crime concentrates may be washed out when compared against the entirety of a study areas non-treated landscape.
Prior to the SSO, matching was carried out to pair each identified hot spot unit to a zero-crime control unit and a low crime control unit. Additional data was collected for the matching phase and several empirically derived covariates were operationalized in the matching sequence. This study utilized covariates pertaining to unit size, activity setting, and proximate crime level to ensure similar control segments were selected. Other commonly used community-level matching considerations were excluded as those measures were generally captured and held constant through the exact match specification at the neighborhood-level.
Street Segment Length: this covariate ensured that longer segments, which may contain more risky features or greater levels of activity, were matched accordingly. Initially measured in feet, continuous segments lengths were categorically recoded into quartiles.
Street Type Classification: this covariate designates each segment by its assigned road type, which may influence traffic and activity levels. The types include expressway, freeway, primary arterial, secondary arterial, two-lane arterial, and local streets.
Land Usage: this covariate included residential, commercial and mixed land use classifications based on zoning data. Classifying segments by their primary land use helps facilitate better contextual settings for the singular unit matches.
Ambient Population: the ambient population was captured using the Oak Ridge National Laboratory’s LandScan database, which provides 24-hour satellite imagery estimates of the on-street population within 1km2 grids.8 This measure controls for the approximate average number of people actively moving about within each 1km grid (Andresen, 2011, p. 195). The ambient population measure was calculated as a five-year index for each grid, which included the summation of average yearly values for each year in the study period after z-score transformation for standardization purposes. Segments within each grid were assigned the value of the grid they resided within. The five-year period more adequately captures the estimated on-street population across the study period.
Spatial Lag: the spatial lag variable used a queen’s contiguity matrix to identify all units adjacent to a hot spot segment (Wheeler, 2018). The concentration of hot spots, which may have an additive or displaced effect on surrounding units, was scored differently than isolated units by assessing first order adjacencies to a hot spot street segment.
Neighborhood: an exact match was specified to ensure that each matched pair was in the same neighborhood so that the level of disadvantage, demographics, and other community measures were held constant. Indianapolis has 99 unique neighborhoods.9
The PSM “MatchIt” package in R was used to generate the matched pairs for the zero-crime and low crime control groups (Ho, Imai, King, & Stuart, 2011). PSM was employed to match on the characteristics of the dependent variable (hot spots), as opposed to its more common application of balancing covariates and independent variables (Schnell, Grossman, & Braga, 2019).10 The present study’s application of PSM invoked nearest neighbor matching, which has proved to be a valid PSM technique for generating one-to-one matches (Austin, 2011; Gu & Rosenbaum, 1993). The matching sequence was conducted without replacement and used a caliper width of 0.2 standard deviations for each individual case to reduce the probability of mismatching cases (Austin, 2010). The “discard” option was also specified, which removed treated and control units that fell outside the area of common support in the preprocessing phase before matching (Iacus, King, & Porro, 2012; Stuart, 2010).11 The matching results for the zero-crime controls indicated that 129 of the 144 (90%) treated cases produced a one-to-one match. Similarly, the matching results for the low crime control group indicate that 134 of the 144 (93%) cases successfully matched. Across the matched sets, 126 hot spots (87.5%) successfully matched to both control units. The subsequent SSO and regression analyses were completed on only the 126 matched hot spots and their associated control units.12
Table 1: Propensity Score Matching Paired Match Results
Low Crime Controls
All Units Matched
Zero Crime Controls
Low Crime Controls
Post-matching, the level of imbalance was rigorously examined across both matched datasets using the “cobalt” package in R (Greifer, 2019). In both control groups, all continuous covariates had balanced standardized mean difference statistics and the adjusted variance ratios were also balanced for all the categorical covariate variables. The variance ratios for the covariates achieved balance at the field threshold of two (Zhang et al. 2019). The full model results also indicated high degrees of balance across the samples, indicating successful unit matching across both groups of controls.14 One caveat of the matching process, though, was that the highly variant ambient population variable did not achieve full balance in the final matching models for both control group datasets.15 However, a decision was made to include it in the analysis because the variable helps ensure that any relationships between the EC predictors of interest and street robbery presence is attributable to the predictor itself and not the corresponding level of foot traffic. The level of foot traffic has been previously found to be independently associated with crime outcomes and has been shown to mediate and/or exacerbate the impact of CGAs (Gerell, 2018; Malleson & Andresen, 2016; Perry et al., 2013).
The spatial joins of all CGAs were completed using ArcMap 10.7. All street segments in Indianapolis were used in order to place the features on the closest street segment regardless of hot spot or control classification. The environmental disorder data coded in the SSO were derived from two indices used in the social science field, which collectively encompass elements of both physical disorder (Sampson & Raudenbush, 1999) and decay (Odgers et al. 2012). The SSO was completed using a GSV plugin16 in ArcMap that allowed the author to view the on-map GSV location for each segment in the GIS while simultaneously coding the disorder data. This technique ensures the CGA spatial joins and the disorder SSO had consistent boundaries.
Following the matching procedure, the SSO included coding for 12 individual disorder variables that were operationalized as dichotomous measures of presence or absence at any point on the street segment.17 GSV was used to complete the coding and provides 360-degree, static images that allow the user to zoom, pan and scroll within the virtual environment. Starting at the intersection, each segment was coded on both block faces until the next intersection was reached (Weisburd et al., 2004), and then the direction was reversed to code back across the segment in the opposite direction. GSV also allows the user to select the year the imagery was recorded in if multiple years of observations are archived. The researcher ensured that the imagery evaluated fell within the years designated as the study-period, which coincides with practices observed in other recent SSO studies (Langton & Steenbeek, 2017).18 To reduce bias, the coding of hot spot and control segments was done after removing identifiers and randomizing the segments to ensure that the status of the unit as a hot spot or a control was unknown to the coder. The distribution of the hot spot and control cases across Indianapolis is depicted in Figure 1.
Figure 1: Distribution of Street Robbery Hot Spots and Controls 19
The author individually coded each segment in the study to ensure consistent construct interpretation was maintained across all variables and units. Single coders have been previously utilized in prior research (Sytsma, Connealy & Piza, 2020) and have been deemed appropriate when the coding schema requires less subjective interpretation (Potter & Levine-Donnerstein, 1999). The nature of the present study involves the coding of variables that are binary, operationalized as visibly present or absent, and are not highly subjective or discretionary. However, to ensure the intra-rater reliability of the single coder, test-retest analyses were completed and measured through percent agreement scores, which have been previously used in test-retest analyses in lieu of Kappa coefficients when all variables are binary (McHugh, 2012). The test-retest process tests the ability of the coder to make consistent decisions, while simultaneously testing the coders ability to navigate GSV effectively. Of all the segments coded, 10% (38 segments) were randomly selected and recoded.20 The original coding was completed in June 2019 with the test-retest coding completed in February 2020. This substantial lapse in time minimizes the possibility of memory effects influencing the results (Porter et al., 2018). Results of the test-retest analyses indicate there was an average agreement of 95% across the 12 disorder indicators, suggestive of a high level of intra-rater reliability and GSV navigability.
As a result of the low frequencies across individual CGAs due to the small number of selected hot spots and controls, dichotomous typologies were created to meet the necessary conditions of regression. Seven unique typologies were constructed by merging like CGAs into a singular measure based on the general business type, establishment usage or intention, or other theoretically or empirically informed criteria that provided a definable link.21 The typologies were constructed by the author thematically, with groups formed around a central service (ex: money issuers), a common product (ex: alcohol), or a definable purpose (short-term stays). Similar groupings of CGAs have been previously operationalized through land use designations (Kinney et al., 2008) or through the classification of places as crime radiators or crime absorbers based on their level of internal and external crime risk (Bowers, 2014). Table 2 details the frequencies for the individual predictors and typologies for both the hot spots and controls.
Table 2: Environmental Criminology Crime Generators and Attractors
Body Art Shops
Points of Interest
*Numbers reflect the total count of the risk factor in the study sample.
Generator and Attractor Typologies
Bars, nightclubs/lounges, liquor stores
Bus stops, parks, points of interest, schools
Shopping centers, supercenters, supermarkets, dollar stores
ATMs, banks, credit unions, check cashing/payday loans
Single Service and Stay
Body art shops, laundromats, restaurants
Small Retail / Variety
Convenience stores, small grocery stores, pharmacies, pawnshops
*Numbers reflect the count of segments where the risk factor was dichotomously present, not the count of the risk factor.
Like the micro-level CGAs, the environmental disorder characteristics were also transformed into two typologies: physical disorder and decay. Physical disorder constructs such as litter and graffiti may be more transient and actionable than the structural forms of decay like dilapidated buildings or poor infrastructure, thus, the variables in each construct are different in both the way they develop and persist. Table 3 below outlines the disorder variables included in the present study and their classification as physical disorder or decay. In addition, a dichotomous variable denoting the presence of an apartment complex in the coding area was also recorded during the SSO. Apartments are a unique residential environment, as they foster high population densities in non-commercial settings and are a more permanent form of a residence than the living spaces that characterize the “short-term residence” typology. This distinct environment type provides a special setting where the risk for street robbery may be more likely and independent from other known criminogenic considerations.
Table 3: Environmental Criminology Disorder Variables
Environmental Disorder Typologies
*Numbers reflect the count of segments where each environmental characteristic was dichotomously present.
A series of regressions were used to test the presence of EC predictor variables in hot spots relative to the controls. The regressions included the covariates from the matching algorithm to ensure doubly robust estimates (Ho et al. 2007; Na, 2016), and clustered standard errors were calculated at the block-group level as matching considerations were aggregated to the neighborhood.25 Prior research has also suggested that the relationship between disorder and crime is mediated by sociodemographic characteristics (Harcourt & Ludwig, 2006; Sampson & Raudenbush, 2004). Thus, block-group measures26 including a concentrated disadvantage index,27 the percent of foreign born residents, and the percent of residents who moved homes within the last year were added to the regressions to account for community-level conditions within neighborhoods that may influence the relationship between predictors and crime levels.
The first analysis is a multinomial logistic regression model testing the hot spot and low crime control segments against the zero-crime control segments. This analysis helps determine if the places that experienced street robbery (hot spots and low crime controls) are different in their environmental composition compared to places that recorded no instances of street robbery. Significant results in this model indicate which environmental predictors distinguish hot spot and low crime units from zero-crime units separately. Then, an additional logistic regression model will be executed comparing the hot spot and low crime control segments. The results of this model will depict which predictors are significantly present in hot spots relative to low crime places, thereby identifying the environmental features that are responsible for distinguishing hot spots from just occasionally crime prone places. Also, this analysis will uncover the variables that are potentially criminogenic, but not hot spot related, by identifying the variables that were significantly more present in the crime environments when compared against the zero-crime places but were not significant in the comparison of hot spots and low crime places.
The multinomial logistic regression (MLR) independently compared the hot spots and low crime units to the zero-crime units. There are several EC typologies that distinguish hot spots from places without crime, including a hot spot having significantly more decay (2.2), small retail stores (17.3), money institutions (7.9) and alcohol establishments (5.1). Although significant, the effects of the generators were substantially larger than the finding for the presence of decay. Decay characteristics distinguish hot spots from units without crime, but the effect is much more pronounced for the CGAs. The presence of apartments (10.4) and the number of people changing residences (1.05) were also significantly higher in the hot spots. The model also compared low crime control units to the units without crime. Interestingly, the presence of apartments (4.0) and the spatial lag variable (.78) were the only significant variables. This suggests there is not a significant, observable difference in feature presence between places without crime and places with low crime as it pertains to CGAs and environmental disorder.
Table 4: Multinomial Logistic Regression Results - Zero Crime Control Baseline
Hot Spot Units
Low Crime Control Units
.852 – 3.369
.975 – 3.706
1.09 – 4.372
.627 – 2.492
Risk Factor Typologies
.437 – 79.03
.167 – 23.32
Small Retail Stores
3.29 – 90.87
.657 – 21.57
Single Service Establishments
.995 – 8.158
.522 – 5.308
Generators / Population Concentrators
.889 – 4.567
.629 – 3.403
Large Retail Establishments
.826 – 47.37
.367 – 17.18
1.45 – 42.86
.292 – 12.10
1.35 – 19.06
.331 – 6.887
Presence of Apartment Complex
4.17 – 25.71
1.56 – 10.45
.645 - .9197
.637 - .9527
Census Block Group Variables
1.00 – 1.067
Measures of Model Fit
686.06 / 882.81
p< 0.05*, < 0.01**, < 0.001***
The second model assessed the classification of hot spots against the pool of low crime controls through logistic regression. The model demonstrated that neither physical disorder nor decay is significantly more present on hot spot segments than in places that experienced low levels of crime. This finding suggests that although environmental disorder (particularly decay) may distinguish hot spots from places without crime, and even though it may generally influence the likelihood of crime, environmental disorder is not uniquely present at hot spots. However, several generator typologies were significantly more present in hot spots than the low crime controls. Small retail stores (5.3) and money institutions (6.0) were both significantly more present in hot spots. This finding suggests that these generator types yield a strong influence over the environment, and they may be catalysts for manipulating crime prone locations into hot spots. Moreover, the effects of some particularly salient crime generators may not be fully realized as a result of low frequencies and model volatility. For example, hot spots are responsible for 12 of 13 liquor stores in the study. This underscores the importance and influence of specific CGAs on the environment, a finding that is especially magnified in the present study through the differences in observed CGA frequencies across empirically similar hot spots and control locations (see Table 2). The results in Table 5 show that the hot spots were also found to be distinguishable from low crime units by the significant presence of apartments (2.5), on-street populations (1.3), and disadvantage (2.2).
Table 5: Logistic Regression Results - Hot Spot and Low Crime Control Units
.319 – 1.859
.660 – 3.659
Risk Factor Typologies
.622 – 455.2
Small Retail Stores
1.49 – 19.41
Single Service Establishments
.345 – 4.586
Generators / Population Concentrators
.409 – 3.372
Large Retail Establishments
.870 – 21.30
1.38 – 25.97
.806 – 19.63
Presence of Apartment Complex
1.02 – 5.920
1.12 – 1.416
Census Block Group Variables
Concentrated Disadvantage Index
1.54 – 3.080
Measures of Model Fit
Area Under ROC
Note: SSO = systematic social observation; ROC = receiver operating characteristic
p< 0.05*, < 0.01**, < 0.001***, < 0.09^ (approaching significance)
Discussion and Conclusions
The first contribution of this study is the identification of the EC variables that were significantly more present in hot spots relative to the controls. The present study also began to tease apart which features of the environment influence hot spot presence as opposed to general crime occurrence through the operationalization of multiple control groups. Several different EC predictor variables across each domain were found to be significantly present in hot spots compared to places that did not experience crime, which suggests that both CGAs and environmental disorder influence crime occurrence and potentially the formation of hot spots. It is time the field moves toward more regularly considering perspectives from both EC domains in theoretical and research endeavors, and in practice such as situational crime prevention efforts.
Specifically examining the significant typologies, several generator-based typologies demonstrated a more significant presence in hot spots relative to the controls. Constructs like small retail stores, money institutions and the presence of apartments demonstrated a significantly higher presence in hot spots compared to either type of control and may be some of the most salient predictors of hot spots. These features are distinguishable from other known generators that may influence the likelihood of crime occurrence, as these constructs demonstrated that they are significantly more present in hot spots even when compared to control units that experienced crime. As it pertains to micro-level situational crime prevention and more nuanced environmental risk factor identification, these generator constructs should be given priority in our attempts to understand the relationship between the environment and crime.
In the multinomial logistic regression model, the results indicated that hot spots have significantly more decay when compared to the zero-crime controls. This suggests that decay significantly influences the potential occurrence of crime in the immediate environment, although physical disorder does not. Physical disorder may possess less influence on hot spot presence because of its adaptive nature. Deterioration, dilapidation, and undeveloped vacant spaces may demonstrate a greater impact on crime and hot spots because they may better capture the lack of care, concern, and territorial reinforcement across community members and government stakeholders over time. Areas do not decay overnight in the same way that an area experiencing graffiti can experience physical disorder overnight, for example. Decay was not found to be significantly present in hot spots compared to the low crime controls, though. This finding indicates that physical disorder and decay may be important considerations in determining the likelihood of crime occurrence, but on a higher level, they may have less influence on the composition and location of a hot spot.
The present study provides evidence that environmental disorder is pertinent to understanding crime occurrence, however, the effect may be capped there. Several generator types were found to significantly influence hot spots relative to the low crime controls, whereas the disorder indicators tested were not distinguishably and significantly different between hot spots and low crime controls. The present study suggests disorder may be an important environmental consideration, but their effect on hot spots may be lesser than that of CGAs. Prior research has found disorder is linked to crime (Boggess & Maskaly, 2014; O’Shea, 2006), but the link may be through their association with CGAs (Tower & Groff, 2016).The relationships between CGAs, disorder, and crime need to be further teased out in mediation and moderation analyses that detail the potential for interactions across constructs to be the driving force behind crime occurrence and hot spot presence. Future studies should examine the influence, and implications, of the interaction between CGAs and disorder at individual locations.
The second contribution of the study is the formation of an empirical, case-controlled, quasi-experimental matching design that allowed for matched units to be paired for the SSO. The rigorous matching sequence employed in this study facilitates stronger conclusions than past comparative studies using SSO by matching on more covariate conditions and successfully matching within a fixed geography (Schnell, Grossman, & Braga, 2019). The field standard requires conclusions to be made relative to a control group, and a framework now exists for conducting quasi-experimental research in the context of SSO. The matching approach allows for the examination of unit-to-unit matches while still meticulously ensuring balance across covariates in the treated and control samples. The conclusions produced in the present study are more robust than past research using SSO absent an empirical control group.
The third contribution of this study is the effective and novel use of several sources of publicly available data and methods. Many environmental variables that were previously unrecorded are now publicly accessible, updated, and maintained, allowing for the study of new research questions and the revisiting of old conclusions. In the present study, the use of remote SSO to examine environmental disorder provides a new way to more objectively incorporate relevant disorder data at the micro-level compared to other data sources that may be plagued by subjectivity or spatial extent. SSO has been previously utilized to predict crime likelihood (Langton & Steenbeek, 2017), and to conduct environmental audits in related sciences (Clarke et al., 2010), but this study represents a new area of research by applying SSO to assess the qualitative characteristics that make up identified high-crime places (Braga & Clarke, 2014). The use of remote SSO also provided a quick, cost-effective and reliable way to measure disorder, which was determined to have some observable impact on crime and street robbery hot spots.
Despite its contributions, this study suffers from several limitations. There are many techniques to generate hot spots and a limitation of the selected approach was the use of five years of crime data. Although this ensures that persisting hot spots are evaluated, the temporal order of crime and disorder is unanswered. Questions of temporality have previously been raised, with research suggesting that disorder is both a cause and consequence of crime (Steenbeek & Hipp, 2011). However, the central aim of the study was to test for the presence of disorder in hot spots compared to controls, irrespective of whether it was the cause or consequence.
A limitation of the matching sequence was the use of a larger fixed geography than is commonly seen in micro-level research. Similarly to what plagued Schnell, Grossman and Braga’s (2019) application of dependent variable matching, cases were not able to successfully match at the census block-group or tract level because there were often too few units within those geographic extents for an adequate pool of matches. The use of neighborhoods still represents a widely applied proxy extent for community-based studies that is similar in size to census boundaries. Another limitation within the matching sequence was that 18 hot spots were dropped for not matching to both a zero-crime and a low crime control unit. Although a small percentage of the sample, and not necessarily comprised of the highest crime segments, these cases may represent unique environments that are not accounted for in the conclusions.
With a small sample and low frequencies, the predictor variables were transformed into dichotomous presence/absence typologies instead of treated as individual measures. This resulted in some high standard errors, some high but insignificant odds ratios, and the inability to fully determine the factor that was most salient in a typology. To speculate on which specific CGA or disorder element may have driven hot spot classification within a typology, the breakdowns of each feature across the sample groups is included in Table 3. For example, of the 13 liquor stores in the sample 12 were in hot spots (one at a zero-crime unit). Similarly, the typology “short-term stays” has low frequencies and is categorized towards hot spots, which produced highly volatile odds ratios and standard errors across the models. This underscores the strong correlations between some features and street robbery that may dictate hot spot classification for an entire typology. Relatedly, this finding also demonstrates the importance of understanding hot spot compositions, as certain environmental features disproportionately concentrate in hot spot environments even when compared to empirically similar controls that also experienced crime.
The use of remote SSO has been previously proved to be reliable (Clarke et al., 2010; Edwards et al., 2013), but GSV does have some inherent limitations. The cross-sectional images require that a construct be present at that individual point in time in order to be recorded. Further, remote SSO’s do not allow for the coder to become fully immersed in the environment of interest, taking in the sights, smells, and general feeling of the setting (Sampson, 2012). These concerns may question the validity of some of the more transient forms of physical disorder, and moreover, there may be some forms of disorder and decay that are not adequately captured by imagery. This is especially the case for important measures of social disorder and social interactions, which are predicated on the presence of people or other dynamic constructs. Social disorder may exert a direct influence on crime, physical disorder and decay indicators, and the presence of hot spots that is not measured in the present study. This inherent limitation should be fleshed out through future research examining the potential differences between remote methods and site observations as it pertains to studying the indicators of disorder (particularly social).
The typologies in the study were also collapsed to dichotomous values as opposed to counts, which may have better reflected the severity of the measure. Thus, the study can discern between differences in the presence of CGAs and environmental disorder across units, but not differences in magnitude. The coder is also limited to the years archived by GSV. Although all the hot spots and controls were coded within the study timeframe, there is an unaccounted-for degree of variance in the year each specific site was coded. Additionally, there are instances when certain aspects of the environment are less observable due to obstructions that inhibit the range of visibility that could be achieved by site visits. GSV also only collects images during the day, preventing the viewing of environments at night when crime may be more likely to occur. However, the measures coded for in the present study tend to be more fixed, less transient, and not contingent on day or night hours. Moreover, the money and time saved by remote SSO, in addition to its demonstrated credibility, may override rare instances of viewshed impairment and time stamp limitations in comparison to conducting hundreds of site visits (Edwards et al., 2013).
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
The author received no financial support for the research, authorship, and/or publication of this article.
Andresen, M.A. (2011). The ambient population and crime analysis. Professional Geographer, 63(2), 193-212.
Andresen, M. A., Jenion, G. W., & Reid, A. (2012). An evaluation of ambient population estimates for use in crime analysis. Crime Mapping: A Journal of Research and Practice, 4, 8–31.
Andresen, M. & Linning, S. (2012). The (in)appropriateness of aggregating across crime types. Applied Geography, 35(1-2), 275-282.
Andresen, M., Linning, A., & Malleson, S. (2017). Crime at Places and Spatial Concentrations: Exploring the Spatial Stability of Property Crime in Vancouver BC, 2003–2013. Journal of Quantitative Criminology, 33(2), 255-275.
Austin, P. C. (2010). Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharmaceutical Statistics, 10, 150–161.
Austin, P.C. (2011). An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behavioral Research, 46(3), 399-424.
Barnum, J., Caplan, J., Kennedy, L., & Piza, E. (2017). The crime kaleidoscope: A cross-jurisdictional analysis of place features and crime in three urban environments. Applied Geography, 79(C), 203-211.
Bayley, D. H. (1986). The tactical choices of police patrol officers. Journal of Criminal
Justice, 14(4), 329-348.
Ben-Joseph, E., Lee, J., Cromley, E., Laden, F., & Troped, P. (2013). Virtual and actual: Relative accuracy of on-site and web-based instruments in auditing the environment for physical activity. Health & Place, 19, 138-150.
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.
Bernasco, W., Ruiter, S., & Block, R. (2017). Do street robbery location choices vary over time of day or day of week? a test in Chicago. Journal of Research in Crime and Delinquency, 54(2), 244-275.
Boessen, A., & Hipp, J. (2015). Close‐ups and the scale of ecology: Land uses and the geography of social context and crime. Criminology, 53(3), 399-426.
Boggess, L., & Maskaly, J. (2014). The spatial context of the disorder-crime relationship in a study of Reno neighborhoods. Social Science Research, 43, 168-183.
Bowers, K. (2014). Risky facilities: Crime radiators or crime absorbers? A comparison of internal and external levels of theft. Journal of Quantitative Criminology, 30(3), 389-414.
Braga, A., & Clarke, R. (2014). Explaining high-risk concentrations of crime in the city: social disorganization, crime opportunities, and important next steps. Journal of Research in Crime and Delinquency, 51(4), 480-498.
Braga, A., Papachristos, A., & Hureau, A. (2010). The concentration and stability of gun violence at micro places in Boston, 1980–2008. Journal of Quantitative Criminology, 26(1), 33-53.
Braga, A.A., Papachristos, A.V., Hureau, D.M. (2014). The effects of hot spots policing on crime: An updated systematic review and meta-analysis. Justice Quarterly, 31(4), 633-663.
Braga, A., Turchan, B., Papachristos, A. & Hureau, D. (2019). Hot spots policing and crime reduction: An update of an ongoing systematic review and meta-analysis. Journal of Experimental Criminology, 15(3), 289-311.
Brantingham, P.J., & Brangintham, P.L. (1981). Environmental Criminology. Beverly Hills, CA: Sage Publications.
Brantingham, P.L. & Brantingham, P.J. (1993). Nodes, paths, and edges: Considerations on the complexity of crime and the physical environment. Journal of Environmental Psychology, 13(1), 3-28.
Brantingham, P. J., & Brantingham, P. L. (1995). Criminality of place: Crime generators and crime attractors. European Journal of Criminal Policy, 3(3), 5–26.
Brantingham, P.J. & Brantingham, P.L. (2008). Crime pattern theory pg. 78-93 in Environmental Criminology and Crime Analysis, edited by R. Wortley & L. Mazerolle. Cullompton, UK: Wilson.
Chainey, S., Tompson, L. & Uhlig, S. (2008). The utility of hotspot mapping for predicting spatial patterns of crime. Security Journal, 21(2), 4-28.
Clarke, P., Ailshire, J., Melendez, R., Bader, M., & Morenoff, J. (2010). Using Google Earth to conduct a neighborhood audit: Reliability of a virtual audit instrument. Health & Place, 16, 1224–1229.
Clutter, J., Henderson, S., & Haberman, C. (2019). The impact of business improvement district proximity on street block robbery counts. Crime & Delinquency, 65(8), 1050-1075.
Connealy, N. & Piza, E. (2019). Risk factor and high-risk place variations across different robbery targets in Denver, Colorado. Journal of Criminal Justice, 60, 47-56.
Connealy, N. (2019). Can we trust crime predictors and crime categories? Expansions on the potential problem of generalization. Applied Spatial Analysis and Policy, In-press.
Eck, J. (1994). Drug markets and drug places: A case-control study of the spatial structure of illicit drug dealing. PhD Dissertation, Department of Criminology and Criminal Justice. University of Maryland, College Park, MD.
Eck, J. (2006). When is a bologna sandwich better than sex? A defense of small-n case study evaluations. Journal of Experimental Criminology, 2(3), 345-362.
Eck, J. (2017). Some solutions to the evidence-based crime prevention problem: In Knutsson, J. & Tompson, L. Advances in Evidence-Based Policing. New York, NY: Routledge.
Edwards, N., Hooper, P., Trapp, G. S., Bull, F., Boruff, B., & Giles-Corti, B. (2013). Development of a Public Open Space Desktop Auditing Tool (POSDAT): A remote sensing approach. Applied Geography, 38, 22–30
Gau, J., & Pratt, T. (2008). Broken windows or window dressing? Citizens’ (in)ability to tell the difference between disorder and crime. Criminology & Public Policy, 7, 163-194.
Gerell, M. (2018). Bus stops and violence, are risky places really risky? European Journal on Criminal Policy and Research, 24, 351-371.
Gould, J. B., & Mastrofski, S. D. (2004). Suspect searches: Assessing police behavior under the
U.S. constitution. Criminology & Public Policy, 3(3), 315–362.
Greifer, N. (2019). Cobalt: Covariate balance tables and plots. R package version 3.7.0.
Gu, X.S. & Rosenbaum, P. (1993). Comparison of multivariate matching methods: Structures, distances, and algorithms. Journal of Computational and Graphical Statistics, 2(4), 405-420.
Haberman, C. (2017). Overlapping hot spots? Examination of the spatial heterogeneity of hot spots of different crime types. Criminology & Public Policy, 16(2), 633-660.
Haberman, C., & Ratcliffe, J. (2015). Testing for temporally differentiated relationships among potentially criminogenic places and census block street robbery counts. Criminology, 53(3), 457-483.
Haberman, C.P., Sorg, E.T. & Ratcliffe, J.H. (2017). The seasons they are a changin’: Testing for seasonal effects of potentially criminogenic places on street robbery. Journal of Research in Crime and Delinquency, 55(3), 425-459.
Harcourt, B., & Ludwig, J. (2006). Broken windows: New evidence from New York City and a five-city social experiment. The University of Chicago Law Review, 73, 271-320
He, L., Paez, A. & Liu, D. (2016). Built environment and violent crime: An environmental audit approach using Google Street View. Computers, Environment and Urban Systems, 66, 83-95.
Hipp, J. (2010). Resident perceptions of crime and disorder: How much is “bias,” and how much is social environment differences? Criminology, 48, 475-508.
Ho, D., Imai, K,, King, G. & Stuart, E. (2007). Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Analysis, 15(3), 199–236.
Ho, D., Imai, K., King, G. & Stuart, E. (2011). MatchIt: Nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 42(8), 1-28.
Hoeben, E., Steenbeek, W. & Pauwels, L. (2018). Measuring disorder: Observer bias in systematic social observations at streets and neighborhoods. Journal of Quantitative Criminology, 34(1), 2221-249.
Hur, M. & Nasar, J. (2014). Physical upkeep, perceived upkeep, fear of crime and neighborhood satisfaction. Journal of Environmental Psychology, 38, 186-194.
Hsu, K.H. & Miller, J. (2017). Assessing the situational predictors of drug markets across street segments and intersections. Journal of Research in Crime & Delinquency, 54(4), 902-929.
Iacus, S., King, G. & Porro, G. (2012). Causal inference without balance checking: Coarsened exact matching. Political Analysis, 20(1), 1-24.
Kinney, B., Brantingham, P., Wuschke, K., Kirk, M., & Brantingham, P. (2008). Crime attractors, generators and detractors: Land use and urban crime opportunities. Built Environment, 34(1), 62-74.
Knox, G. (1964). Epidemiology of childhood leukemia in Northumberland and Durham. British Journal of Preventative Social Medicine, 18, 17-24.
Kurland, J., Johnson, S. D., & Tilley, N. (2014). Offenses around stadiums: A natural experiment on crime attraction and generation. Journal of Research in Crime and Delinquency, 51, 5–28
Langton, S. & Steenbeek, W. (2017). Residential burglary target selection: An analysis at the property-level using Google Street View. Applied Geography, 86, 292-299.
Lindegaard, M. R., & Bernasco, W. (2018). Lessons learned from crime caught on camera. Journal of Research in Crime and Delinquency, 55(1), 155–186.
Malleson, N., & Andresen, M. (2016). Exploring the impact of ambient population measures on London crime hotspots. Journal of Criminal Justice, 46(C), 52-63.
Mastrofski, S., Parks, R., and McClusky, J. (2010). Systematic social observation in criminology. In Piquero A. and Weisburd, D. (eds.) Handbook of Quantitative Criminology, 225-247.
Springer: New York.
Mccord, E., & Ratcliffe, J. (2007). A micro-spatial analysis of the demographic and criminogenic environment of drug markets in Philadelphia. Australian & New Zealand Journal of Criminology, 40(1), 43-63.
McHugh, M. (2012). Interrater reliability: The kappa statistic. Biochemia Medica, 22(3), 276-282.
Miethe, T., & Meier, R. (1990). Opportunity, choice, and criminal victimization: a test of a theoretical model. Journal of Research in Crime and Delinquency, 27(3), 243-266.
Miller, J., Caplan, J., & Ostermann, M. (2016). Home nodes, criminogenic places, and parolee failure: Testing an environmental model of offender risk. Crime & Delinquency, 62(2), 169-199.
Morenoff, J.D., Sampson, R.J. & Raudenbush, S.W. (2001). Neighborhood inequality, collective efficacy, and the spatial dynamics of urban violence. Criminology. 39, 517-560.
Morgan, F. (2000). Repeat burglary in a Perth suburb: Indicator of short-term or long-term risk. In G. Farrell & K. Pease (eds.) Repeat Victimization: Crime Prevention Studies (Vol. 12). Monsey, NJ: Criminal Justice Press.
Na, C. (2016). The consequences of fatherhood transition among disadvantaged male offenders: Does timing matter? Journal of Developmental Life Course Criminology, 2: 182-208.
O’Brien, D.T., Gordon, E. & Baldwin, J. (2014). Caring about the community, counteracting disorder: 311 reports of public issues as expressions of territoriality. Journal of Environmental Psychology, 40(11), 320-331.
Odgers, C., Caspi, A., Bates, C., Sampson, R. & Moffitt, T. (2012). Systematic social observation of children’s neighborhoods using Google Street View: A reliable and cost-effective method. The Journal of Child Psychology and Psychiatry, 53(10), 1009-1017.
O’shea, T. (2006). Physical deterioration, disorder, and crime. Criminal Justice Policy Review, 17(2), 173-187.
Perry, W., McInnis, B., Price, C., Smith, S., & Hollywood, J. (2013). Predictive policing: The role of crime forecasting in law enforcement operations. RAND Corporation.
Piza, E., Caplan, J. M., & Kennedy, L. W. (2014). Analyzing the influence of micro-level factors on CCTV camera effect. Journal of Quantitative Criminology, 30(2), 237-264.
Piza, E. & Carter, J. (2018). Predicting initiator and near repeat events in spatiotemporal crime patterns: An analysis of residential burglary and motor vehicle theft. Justice Quarterly, 35(5), 842-870.
Porter, A.K., Wen, F., Herring, A.H., Rodríguez, D.A., Messer, L.C., Laraia, B.A., & Evenson, K.R. (2018). Reliability of one-year stability of the PIN3 neighborhood environmental audit in urban and rural neighborhoods. Journal of Urban Health, 95, 431-439.
Potter, W., & Levine-Donnerstein, D. (1999). Rethinking validity and reliability in content analysis. Journal of Applied Communication Research, 27(3), 258-284.
Ratcliffe, J. (2019). Reducing Crime: A Companion for Police Leaders. Routledge.
Reiss, A. (1968). Stuff and nonsense about social surveys and observation: In H. Becker , et al (Edn), Institutions and the Person. Chicago, IL: Aldine.
Reiss, A. (1971). Systematic Observation of Natural Social Phenomena. Sociological Methodology, 3, 3-33.
Rice, K., & Smith, W. (2002). Socioecological models of automotive theft: integrating routine activity and social disorganization approaches. Journal of Research in Crime and Delinquency, 39(3), 304-336.
Rundle, A., Bader, M., Richards, C., Neckerman, K., & Teitler, J. (2011). Using Google Street View to audit neighborhood environments. American Journal of Preventive Medicine, 40(1), 94-100.
Salesses, P., Schechtner, K., & Hidalgo, C. A. (2013). The collaborative image of the city: Mapping the inequality of urban perception. PloS One, 8(7), e68400.
Sampson, R. & Raudenbush, S. (1999). Systematic social observation of public spaces: A new look at disorder in urban neighborhoods. American Journal of Sociology, 105, 603-651.
Sampson, R., Raudenbush, S. & National Institute of Justice. (2001). Disorder in urban neighborhoods: Does it lead to crime (Research in brief). Washington, D.C.: U.S. Dept. of Justice, Office of Justice Programs, National Institute of Justice.
Sampson, R., & Raudenbush, S. (2004). Seeing disorder: Neighborhood stigma and the social construction of “Broken windows.” Social Psychology Quarterly, 67, 319-342
Sampson, R., Raudenbush, S. & Earls, F. (1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918-924.
Schnell, C., Grossman, L. & Braga, A. (2019). The routine activities of violent crime places: A retrospective case-control study of crime opportunities on street segments. Journal of Criminal Justice, 60, 140-153.
Sherman, L. W., Gartin, P. R., & Buerger, M. E. (1989). Hot spots of predatory crime: Routine activities and the criminology of place. Criminology, 27(1), 27–55.
Sherman, L., & Weisburd, D. (1995). General deterrent effects of police patrol in crime "hot spots": A randomized, controlled trial. Justice Quarterly, 12(4), 625-648.
Skogan, W. (1990). Disorder and Decline: Crime and the Spiral of Decay in American Cities. New York: The Free Press
Skogan, W., & Steiner, L. (2004). Crime, disorder and decay in Chicago’s Latino community. Journal of Ethnicity in Criminal Justice, 2(1-2), 7-26.
Steenbeek, W. & Hipp, J. (2011). A longitudinal test of social disorganization theory: Feedback effects among cohesion, social control, and disorder. Criminology, 49(3), 833-871.
Stuart, E. (2010). Matching methods for causal inference: A review and a look forward. Statistical Science, 25(1), 1-21.
Swatt, M.L., Varano, S.P., Uchida, C.D., & Solomon, S.E. (2013). Fear of crime, incivilities, and collective efficacy in four Miami neighborhoods. Journal of Criminal Justice, 41(1), 1-11.
Sytsma, V., Connealy, N. & Piza, E. (2020). Environmental predictors of a drug offender crime script: a systematic social observation of Google Street View images and CCTV footage. Crime & Delinquency. In-press.
Szkola, J., Piza, E. & Drawve, G. (2019). Risk terrain modeling: Seasonality and predictive validity. Justice Quarterly, DOI: https://doi.org/10.1080/07418825.2019.1630472.
Taylor, R. (2001). Breaking away from broken windows: Baltimore neighborhoods and the nationwide fight against crime, grime, fear, and decline. Boulder, CO: Westview Press.
Taylor, R. (1997). Social order and disorder of street blocks and neighborhoods: Ecology, microecology, and the systemic model of social disorganization. Journal of Research in Crime and Delinquency, 34(1), 113-155.
Taylor, R., Gottfredson, S. & Brower, S. (1985). Attachment to place: Discriminant validity, and impacts of disorder and diversity. American Journal of Community Psychology, 13(5).
Terrill, W., & Reisig, M. D. (2003). Neighborhood context and police use of force. Journal of
Research in Crime and Delinquency, 40(3), 291-321.
Tillyer, M.S., Wilcox, P. & Walter, R.J. (2020). Crime generators in context: examining ‘place in neighborhood’ propositions. Journal of Quantitative Criminology. DOI: https://doi.org/10.1007/s10940-019-09446-5
Toet, A. & Van Schaik, M. (2012). Effects of signals of disorder on fear of crime in real and virtual environments. Journal of Environmental Psychology, 32(3), 260-276.
Tower, S. & Groff, E. (2016). Examining the disorder-crime connection in Philadelphia parks. Security Journal, 9(3), 520-537.
Vandeviver, C. (2014). Applying google maps and google street view in environmental criminological research. Crime Science, 3(1), 1-16.
Weisburd, D. (2015). The law of crime concentration and the criminology of place. Criminology, 53(2), 133–157.
Weisburd, D., & Amram, S. (2014). The law of concentrations of crime at place: The case of Tel Aviv-Jaffa. Police Practice and Research: Trends in Israeli Policing: Terrorism, Community, Victimization and Crime Control, 15(2), 101-114.
Weisburd, D., Bernasco, W. & Bruinsma, G. (2008). Putting Crime in its Place: Units of Analysis in Geographic Criminology. New York, NY: Springer.
Weisburd, D. L., Bushway, S., Lum, C., & Yang, S. M. (2004). Trajectories of crime at places: A longitudinal study of street segments in the city of Seattle. Criminology, 42, 283–322
Weisburd, D., Groff, E. & Yang, S.M. (2012). The Criminology of Place; Street Segments and Our Understanding of the Crime Problem. New York: Oxford University Press.
Weisburd, D., & Mazerolle, L. (2000). Crime and disorder in drug hot spots: Implications for theory and practice in policing. Police Quarterly, 3(3), 331-349.
Wheeler, A. (2018). The effect of 311 calls for service on crime in D.C. at microplaces. Crime & Delinquency, 64(14), 1882-1903.
Wheeler, A. & Steenbeek, W. (2020). Mapping the risk terrain for crime using machine learning. Journal of Quantitative Criminology. Pre-print: https://osf.io/preprints/socarxiv/xc538/.
Wyant, B.R. (2008). Multilevel impacts of perceived incivilities and perceptions of crime risk on fear of crime: Isolating endogenous impacts. Journal of Research in Crime and Delinquency, 45(1), 39-64.
Zhang, Z., Jung-Kim, H., Lonjon, G., & Zhu, Y. (2019). Balance diagnostics after propensity score matching. Annals of Translational Medicine, 7(1), 1-11.
Appendix A: Hot Spot and Zero Crime Control Matching Fit
Figure A1: Absolute Mean Differences Pre/Post Matching
Figure A2: Jitter Plot and Propensity Scores Distribution across Hot Spots and Zero Crime Controls
Figure A3: Distributional Balance for Segment Length
Figure A4: Distributional Balance for Ambient Population
Figure A5: Distributional Balance for Street Type
Figure A6: Distributional Balance for Spatial Lag
Figure A7: Distributional Balance for Land Use
Appendix B: Hot Spot and Low Crime Control Matching Fit
Figure B1: Absolute Mean Differences Pre/Post Matching
Figure B2: Jitter Plot and Propensity Scores Distribution across Hot Spots and Zero Crime Controls
Figure B3: Distributional Balance for Segment Length
Figure B4: Distributional Balance for Ambient Population
Figure B5: Distributional Balance for Street Type
Figure B6: Distributional Balance for Spatial Lag
Figure B7: Distributional Balance for Land Use
Appendix C: Logistic Regression Fit – Hot Spots and Low Crime Controls
C1: Sensitivity and Specificity Against Probability Cutoffs
C2: Area Under the Curve and Receiver Operating Characteristic