Abstract
Objective:
Examine the distribution and sources of crime across freestanding businesses in San Antonio. We test hypotheses about the main and interactive effects of neighborhood and business characteristics on crime at the business, with a focus on busy contexts and busy businesses.
Method:
Police crime incident data are spatially joined to study area business parcels. Additional data sources include Infogroup USA Business Data, the American Community Survey, and an Environmental Protection Agency traffic activity indicator. Multilevel negative binomial regression models are estimated to observe the main and interactive effects of census block group and business variables on crime at the parcel.
Results:
Businesses located in block groups with more commercial property and high levels of vehicular traffic experience more crime. In addition, crime is higher at “busy” businesses, as indicated by employee size, sales volume, and square footage. Busy contexts and busy businesses do not appear to interact to increase crime at the parcel beyond their main effects.
Conclusions:
Crime is clustered at relatively few businesses, and this variation cannot be explained by business type alone. Both neighborhood and business characteristics are associated with crime at freestanding businesses, with busy businesses and those within busier block groups experiencing more crime.
In his 2014 Sutherland Address to the American Society of Criminology, Weisburd (2015) emphasized the need to focus on crime concentration at places and, among other things, encouraged the use of theoretically informed models to examine why crime concentrates at relatively few places. The study of crime and place in criminology has drawn from social disorganization and environmental criminology, with researchers relying on both the broader social and physical environment shaped by neighborhood characteristics, as well as micro features of places that create criminal opportunities to understand the spatial patterns of crime (Weisburd et al. 2016; Wilcox and Tillyer 2018). At the neighborhood level, the social disorganization tradition suggests that communities unable to realize common values and exert social control suffer higher crime rates (Bursik 1988; Bursik and Grasmick 1993; Shaw and McKay 1942). In recent decades, there has been a shift in emphasis to smaller spatial units of analysis in light of numerous empirical studies documenting the “tight coupling of crime at place” (Weisburd, Groff, and Yang 2012, p. 9). Environmental criminological theories generally focus on how the characteristics and routine activities of places and their users can create or reduce opportunities for crime, thus accounting for the concentration of crime at micro places (Brantingham and Brantingham 1995; Eck, Clarke, and Guerette 2007; Eck and Weisburd 1995; Weisburd, Morris, and Groff 2009).
While initially the influences of social disorganization and opportunity on the spatial distribution of crime were examined separately, more recent theoretical and empirical work acknowledges that they may operate simultaneously to affect crime at places (see, e.g., Weisburd 2012; Weisburd, Groff, and Yang 2014; Weisburd et al. 2017; Wilcox and Tillyer 2018; Wilcox, Gialopsos, and Land 2013). Related, a considerable amount of crime and place research has focused on specific forms of nearby places—such as schools, bars, and payday lenders (Kubrin et al. 2011; Roncek and Faggiani 1985; Roncek and Maier 1991)—with the assumption being that some places are inherently criminogenic because of the types of services they provide. Recently, however, Wilcox and Eck (2011) challenged the assumption that specific types of places—or what they call, “facilities”—are inherently criminogenic. Instead, they assert that “it is the traffic associated with the studied land uses, not the activities of the facilities per se, that likely lead to problems for the area” (p. 475). Wilcox and Eck (2011) predict there will be considerable variability within types of places, and that high-traffic places and contexts will experience more crime.
The validity of these predictions has important implications for businesses, which often use a variety of strategies to maximize traffic as a means to generate profits, including investing in real estate in high-traffic areas and implementing marketing activities to drive store traffic (Perdikaki, Kesavan, and Swaminathan 2012). Understanding the extent to which these strategies also have the unintended consequence of increasing crime risk can help businesses anticipate and address crime problems. Determining whether high-traffic places and contexts experience more crime—independent of the type of place or land use—also has important theoretical implications for an integrated multilevel approach aimed at explaining the distribution of crime across micro places (Wilcox et al. 2013).
To this end, the present study builds on and extends the existing crime concentration literature by examining the distribution and sources of crime at freestanding businesses in San Antonio, TX. 1 Using an integrated theoretical approach informed by both social disorganization and opportunity perspectives, we embrace the importance of micro place characteristics but also acknowledge that such places are embedded within structural contexts that may directly shape opportunities for crime, as well as moderate the effects of place-level features (Wilcox et al. 2013). To date, crime and place studies emphasizing smaller microgeographic units have focused primarily on street segments, with analyses often accounting for the proximity of specific types of land uses. Our study differs in that we focus on crime at the business (i.e., the parcel level) rather than the effect of the business on crime in the surrounding area (i.e., at the street segment or larger aggregate). One benefit of studying crime at small spatial units such as freestanding businesses, as opposed to street segments or shopping malls comprised of multiple entities and business owners, is that freestanding businesses have distinct place management with the authority to respond to place-based risk factors for crime by implementing strategies for prevention.
Building on previous research on the concentration of crime at place, we examine the distribution of crime across all study businesses, as well as the distribution within types of business, to determine whether crime is nonrandomly distributed across businesses, including businesses that provide similar services. We then test hypotheses related to the effects of business characteristics and the surrounding neighborhood variables on crime at the business. Specifically, we focus on indicators of “busy” businesses including sales volume, employee size, and square footage of the business building. We also examine the main and moderating effects of the surrounding neighborhood, approximated using census block groups (CBGs), on crime at the businesses, with a focus on indicators of “busy” contexts—such as traffic activity and nonresidential land use—that may help shape opportunities for crime in the area.
Crime Concentration and Freestanding Businesses
Research on crime and place has consistently demonstrated that crime is not randomly distributed across microgeographic places. For example, in their seminal study of crime in Minneapolis, Sherman, Gartin, and Buerger (1989) found that 3.3 percent of all addresses and intersections generated over half of all calls for service to the police, suggesting that there is considerable variation in predatory crime within communities, regardless of variation in crime across communities (see also Andresen and Malleson 2011; O’Brien and Winship 2017). Similar distributions have been observed in other studies focused on crime along street segments. Weisburd et al. (2004), for example, studied crime incidents along street segments in Seattle from 1989 to 2002, noting that 4.5 percent of segments account for approximately half of all crimes. Officially recorded juvenile crime in Seattle is also highly concentrated among street segments, and these patterns are relatively stable across the 14-year study period (Weisburd et al. 2009). Braga and colleagues have also noted similar patterns across street segments and intersections in Boston over a 29-year period, with less than 3 percent of places accounting for more than half of all gun assault incidents (Braga, Papachristos, and Hureau 2010) and 5 percent of places accounting for approximately two-thirds of all robbery incidents (Braga, Hureau, and Papachristos 2011). Curman, Andresen, and Brantingham (2015) examined crimes along street blocks in Vancouver, BC, from 1991 to 2006 and report that 5.25 percent of blocks account for 50 percent of all crimes. Finally, Gill, Wooditch, and Weisburd (2017) examine crime incidents along street segments in Brooklyn Park, a suburban city outside of Minneapolis, from 2000 to 2014. They report that just 2 percent of all street segments account for half of all crimes. Collectively, these findings have led Weisburd (2015) to call for more research on the law of crime concentration, which states that “for a defined measure of crime at a specific microgeographic unit, the concentration of crime will fall within a narrow bandwidth of percentage for a defined cumulative proportion of crime” (p. 138).
Although much of the research in this area has focused on crime distributions across street segments, there are a few micro place studies that examine crime using different units of analysis. In a related body of research that is perhaps more closely aligned with the present study, some scholars have examined the distribution of crime at specific types of places. Eck et al. (2007) report distributions of crime within various types of places, or what they call facilities, noting a “J curve” that illustrates how risky facilities account for much of the overall total crime count that occurs within-facility type (see also Blair, Wilcox, and Eck 2017). These distributions illustrate what Wilcox and Eck (2011) refer to as “The Iron Law of Trouble Places.” That is, just as a relatively small proportion of street segments can account for a large proportion of crimes observed in a city, there is considerable within-type variation in crime across places, with a small proportion of places accounting for a majority of crimes at those types of places.
The present study extends this line of work by examining the distribution and sources of crime at freestanding businesses, with a specific focus on the criminogenic effects of busy businesses and contexts. While Eck et al.’s (2007) work cited above demonstrates the concentration of crime within homogeneous sets of facilities, their univariate analyses did not empirically explore predictors explaining this concentration. More recently, Wilcox and Eck (2011) posited that traffic at and around places may increase crime risk, an assertion worth empirical scrutiny given that “busyness” is often viewed favorably by businesses as a means to generate profit. We explore this possibility using data on freestanding businesses, a decision informed by conceptual, methodological, and practical considerations. First, a freestanding business is consistent with common criminological conceptualizations of a micro place, including Sherman et al.’s (1989, p. 31) definition of “a fixed physical environment that can be seen completely and simultaneously, at least on its surface, by one’s naked eyes.” Focusing on freestanding businesses, rather than strip malls or larger shopping centers, minimizes the modifiable area unit problem (Openshaw 1984), or the aggregation bias that results from averaging variability at smaller geographic units.
Moreover, the retail literature documents important differences in business activity across categories of locations (Guy 1998; Litz and Rajaguru 2008; Tokatli and Boyaci 1998), thus highlighting that freestanding businesses are conceptually distinct from individual businesses located in strip malls or shopping centers. Malls and shopping centers, for example, have set rules and guidelines for their multiple tenants. They produce a traffic flow that can lead to comparison shopping or impulse buying, with one or more anchor stores often generating most of the traffic. Because these businesses often share parking and security and are bound by lease agreement restrictions governing the property as a whole, it is difficult to attribute a crime occurring in the parking lot, for example, to a particular business and its practices. In short, some individual businesses in strip malls and shopping centers may not be conceptually distinct “places.”
In contrast, freestanding locations often have fewer restrictions, more parking, and are more likely to be the primary destination point for their customers (Burnaz and Topcu 2006; Reimers and Clulow 2004). By limiting the analyses to parcels containing a single business, the effects of business characteristics on crime at the parcel can be estimated, while also accounting for the broader physical and social context that may also facilitate or restrict criminal opportunities. Finally, and perhaps most important, unlike street segments or shopping malls that are comprised of multiple entities and business owners, a freestanding business has distinct place management that can implement prevention strategies to address specific risk factors for crime at the location, a point we return to in the Discussion section.
Explaining Crime at Place
Researchers have largely drawn on social disorganization and environmental criminological theories, or “opportunity theories,” to understand the spatial distribution of crime. Historically, the social disorganization tradition has focused on the importance of the neighborhood context for understanding patterns of crime and delinquency. Social disorganization theory asserts that disorganized communities—that is, those often marked by poverty, population heterogeneity, and residential mobility—have higher crime rates because they are unable to exert the effective informal social control necessary to address crime problems (Bursik 1988; Bursik and Grasmick 1993; Shaw and McKay 1942). Similarly, Sampson, Raudenbush, and Earls (1997) have argued that residential instability and concentrated disadvantage within neighborhoods can increase crime rates, as these communities lack mutual trust among residents and the willingness to intervene for the common good (Sampson et al. 1997). Numerous studies have documented the importance of social structural variables in explaining crime and victimization across various spatial units of analysis (Miethe and McDowall 1993; Osgood and Chambers 2000; Pratt and Cullen 2005; Sampson and Groves 1989; Sampson et al. 1997; Wilcox et al. 2004; Wilcox Rountree, Land, and Miethe 1994).
Conversely, researchers focused on understanding crime at smaller units of analysis, or micro places, have largely embraced opportunity theories of crime that focus on how the characteristics and routine activities of places and their users can create or reduce opportunities for crime (Brantingham and Brantingham 1995; Cohen and Felson 1979; Eck and Weisburd 1995; Eck et al. 2007; Weisburd et al. 2009). Opportunity theories assume offender decision-making is more or less rational, with offenders assessing the risks, efforts, and rewards associated with crime based on perceived environmental cues (Brantingham and Brantingham 1982; Clarke and Cornish 1985). Offenders are likely to search for and encounter targets at locations of major activities or the routes traveled between the locations. As a result, crime events will be clustered along these nodes and paths of activities (Brantingham and Brantingham 1993). Some locations may serve as crime generators because they draw in large concentrations of potential targets, while other places are crime attractors because they are known by offenders to be rich in criminal opportunities (Brantingham and Brantingham 1995). Therefore, places with more high-quality targets, those that are in close proximity to or attract offender populations, and those with weak place management are expected to have more crime (Eck et al. 2007).
In recent years, theoretical and empirical integration of social disorganization theory and opportunity theories have demonstrated the utility of taking a more integrated approach to understanding crime and place. Smith, Frazee, and Davison (2000), for example, report that both social disorganization and routine activity variables are predictive of street robberies across face blocks in a midsized southeastern U.S. city. More recently, Weisburd et al. (2014) found that both social characteristics (e.g., property value, housing assistance, and physical disorder) and indicators of criminal opportunity (e.g., bus stops, arterial roads, and public facilities) can distinguish low-crime street segments from hot spots in Seattle.
Wilcox and colleagues have argued for a multilevel approach to understanding crime and victimization (Wilcox and Tillyer 2018; Wilcox et al. 2013; Wilcox, Land, and Hunt 2003), recognizing that potential places and victims of crime are embedded within broader neighborhood crime markets. At the neighborhood level, variables tapping social and physical structure have come to be seen as indicators of aggregate criminal opportunity that can serve to shape offender decision-making. In addition to its direct influence, aggregate criminal opportunity can interact with micro-level characteristics to influence crime and victimization. Wilcox et al. (2013, p. 586) have posited that the effects of “place-level crime-event risk factors are greater in high-opportunity contexts than low-opportunity contexts” because there is a greater market exposure to crime in such contexts. A lack of security features at a store, for example, may be particularly criminogenic when the store is embedded in a high-risk neighborhood. Indeed, there is now an established body of research that demonstrates how neighborhood context interacts with individual routine activities to influence victimization risk (e.g., Miethe and McDowall 1993; Wilcox, Madensen, and Tillyer 2007; Wilcox et al. 1994). The current study uses this integrated multilevel theoretical approach to examine the main and interactive effects of busy contexts and busy businesses on crime.
Busy Places and Contexts
Scholars have also considered the influence of nearby land uses on crime, with these studies continuing to draw from social disorganization and opportunity frameworks to understand the spatial distribution of crime. This line of inquiry suggests that physical structure may influence social and physical processes related to crime including increasing criminal opportunity. For example, nonresidential land uses that bring strangers into a community serve to undermine neighbor networks required for effective social control, as well as produce physical disorder or deterioration (Kurtz, Koons, and Taylor 1998; Steenbeek et al. 2012; Wilcox et al. 2004). Kurtz et al. (1998) examined land use, physical deterioration, resident-based control, and calls for police service along Philadelphia street blocks. They found that nonresidential land use was related to decreased perceptions of control among residents, as well as increased physical deterioration, both of which were related to an increase of calls for service to the police. Wilcox et al. (2004) extended this work, using data aggregated from residents in Seattle’s 100 census tracts and controlling for neighborhood social structural variables. Although they found little evidence that land use influenced “neighboring” (i.e., various forms of interaction among neighbors), business-oriented public land use did increase violence and burglary at the census tract level, an effect that was partially mediated by physical disorder.
Other research has focused on specific types of nearby nonresidential land uses assumed to be particularly criminogenic because of the services they provide and/or the type and volume of users they attract (e.g., Kinney et al. 2008; Sohn 2016). Using data on Seattle census tracts, Kubrin et al. (2011) report that the concentration of payday lenders is associated with higher violent and property crime rates. Groff and Lockwood (2014) examined how exposure of street segments to various facility types influences violent, property, and disorder crimes using three threshold distances (400, 800, and 1,200 feet). Exposure to bars and subway stations was positively related to all forms of crime using all distance thresholds, while the influence of halfway houses and drug treatment centers differed depending on distance and crime type. Research by Roncek and colleagues has documented higher crime on street blocks with bars and high schools (Roncek and Bell 1981; Roncek and Faggiani 1985; Roncek and LoBosco 1983; Roncek and Maier 1991). Additional studies have documented higher crime near malls (LaGrange 1999), neighborhood parks (Groff and McCord 2012), and motels (Smith et al. 2000).
Although the above studies highlight the criminogenic effects of particular types of places, Wilcox and Eck (2011) argue that such studies may simply be observing a general effect of most places on crime: busy places, rather than specific types of facilities, offer criminal opportunity. They argue that “areas with high-traffic facilities have relatively more crime than comparable areas without high-traffic facilities (or with a lower density of such places)” (Wilcox and Eck 2011, pp. 474-75, italics in original). To illustrate their point, Wilcox and Eck (2011) suggest that a low-traffic bar might be less problematic for a neighborhood than a high-traffic church. Furthermore, they note that these busy places tend to cluster together spatially, creating high-traffic contexts rich in criminal opportunities.
Indeed, there is research to support the idea that busy contexts and places are associated with more crime. For example, Boivin and Felson (2017) examined the impact of visitor inflows to a census tract on crime and found that increased visitor inflow is significantly associated with both visitor and resident crime. Lockwood and Stillings (1998) report that a traffic calming project in West Palm Beach, FL, designed to address resident complaints of speeding, cut-through traffic, and unwanted driver behavior in the neighborhood resulted in reductions in prostitution, drug crimes, and overall street crime. Hanaoka (2018), using hourly population estimates based on cell phone user location data, found that ambient populations increase “snatch-and-run” offenses in Japan.
Additional studies support the idea that crime is elevated at locations near “busy” places. Yu and Maxfield (2014) examined burglary in Newark and found that areas with businesses that provide on-site services, are frequently patronized by customers, and serve neighborhood residents (i.e., food stores, eating and drinking places, educational services, and bus stops) are at elevated risk of burglary. Recently, Askey et al. (2018) argued that revenue data from fast-food restaurants and convenience stores can be used as an indicator of human activity. They found that sales volume is positively and significantly associated with crime counts across street blocks in Seattle.
Beyond these main effects, places and the environments in which they are embedded may interact to affect crime at place (Wilcox et al. 2013). With respect to busy places and contexts, Wilcox and Eck (2011) argue that it is the clustering of such places that create criminal opportunity, suggesting an interaction between an individual facility and the aggregate features of surrounding places that define environmental opportunity. A recent study by Deryol, Wilcox, Logan, and Wooldredge (2016) found support for the idea that proximity to crime generators is particularly criminogenic for locations embedded in environments assumed to be rich in criminal opportunities. Specifically, their results show that the interactive effects of carryout liquor establishments, on-premise drinking establishments, and bus routes on crime at nearby addresses were stronger in CBGs with a higher density of commercial land use.
The current study extends the research on crime and place to focus on crime at the parcel a business is located on rather than how the characteristics of the business influence crime at the larger aggregate. This approach allows us to isolate the effects of micro place characteristics and is consistent with arguments for using small spatial units, such as addresses or parcels, to avoid the aggregation bias that often accompanies the agglomeration process (Brantingham et al. 2009). As noted above, understanding crime at the parcel level carries more practical relevance for prevention than examining larger aggregates with less defined place management responsibility. We predict that crime is higher at “busy” businesses—as indicated by sales volume, employee size, and square footage—and those businesses located in “busy” contexts, that is, CBGs with more commercial land use and more vehicular traffic. In addition, we examine whether the effects of busy businesses on crime are enhanced in busy contexts. Our focus differs from Deryol et al. (2016) in that we estimate crime at the business parcel rather than the effects of the business on crime at nearby addresses. Furthermore, and consistent with Wilcox and Eck’s (2011) assertion that it is the traffic associated with facilities and contexts that creates criminal opportunity, we estimate the main and interactive effects of business activity and area vehicular traffic activity.
The Present Study
In response to Weisburd’s (2015) call for theoretically informed studies examining crime concentration at micro places, the present study embraces an integrated approach to understanding crime at places. Our perspective is informed by the social disorganization tradition as well as environmental criminology, with an explicit focus on the main and interactive effects of busy businesses and contexts. We test the following hypotheses related to the distribution and correlates of crime at freestanding businesses in San Antonio, TX:
Data and Methods
San Antonio is located in South Central Texas and makes up the southwestern corner of the Texas Triangle megaregion. With nearly 300 years of history, it is the seventh most populated city in America, with approximately 1.5 million people residing in the 460 square miles of the city limits. San Antonio is a majority-minority city, with over 60 percent of the population identified as Hispanic or Latino. The city’s median household income is lower than the United States (approximately US$48,000 compared to US$55,000), and the poverty rate is higher (19.5 percent compared to 12.7 percent; American Community Survey 5-Year Estimates 2012-2016). San Antonio has positioned itself as one of the fastest growing cities with a flourishing business community. With a low cost of living enticing economic activity and development, the city has focused on streamlining the expansion of business operations, preparing the workforce to maximize capacity, providing incentives to attract new businesses to the region, and offering capital, working spaces, and educational programs for start-ups (San Antonio Economic Development Foundation N.d.).
Several sources of data were combined to test the present study’s hypotheses. Data on businesses within the city limits of San Antonio come from Infogroup USA Business Data. The present study focuses on those businesses that fall within 41 business type categories (see Appendix). 2 These categories were based on the North American Industry Classification System descriptions and company names. A city boundary shapefile provided by the City of San Antonio was used to select all freestanding businesses in the city boundaries. As discussed above, this study is limited to properties that contain one business, so that the effects of business characteristics on crime at the parcel can be estimated. Therefore, the study does not include businesses in strip malls or parcels that house multiple businesses.
A parcel shapefile was obtained from Bexar County Property Appraisal District. All parcels that contained a single business were selected and used as the unit of analysis for this study, resulting in 9,545 parcels with freestanding businesses. 3 Parcels were used because crime counts were generated for each place using the parcel boundary, which would include the area maintained by the business and their parking lot. This small unit of analysis was selected because each place has specific attributes that would be lost if summarized at an aggregate scale. A spatial join was conducted between the businesses and parcels. This attached a property ID to each business. Listwise deletion based on missing data resulted in 9,028 parcels with freestanding businesses for analysis that fall within 904 CBGs. 4 Figure 1 displays the spatial distribution of cases across the city.
Study area freestanding business parcels.
Crime Data
The dependent variable is crime counts at the parcel (see Table 1 for descriptive statistics for all study variables). The San Antonio Police Department (SAPD) provided x and y coordinates for crime incident data for 2014-2016. 5 These data include both Part I and Part II Uniform Crime Report offenses. The x and y coordinates were first displayed in ArcGIS as an event series and then exported as a shapefile. The crime data shapefile was spatially joined to the study area business parcels, so that all crimes analyzed in the current study fell within the parcel boundaries of the businesses, including crimes that occurred both indoors and outdoors. There were 17,536 crimes within the study area business parcels during the study period.
Study Variables.
Note. n = 9,028 businesses. VIFs ≤ 1.50.
Business Variables
We created a busy business index based on business characteristics captured in the Infogroup USA Business Data. 6 Specifically, the busy business index is the mean of the z-scores of dollar amount in sales generated annually (logged), employee size (logged), and total square footage of the building (α = .89), with higher values indicating busier businesses. Given the present study’s focus on place-level features, we also control for code violations, which may indicate weak place management. Code enforcement is viewed by local governments as a proactive approach to mitigate the potential impact of nuisance properties that may become havens for criminal activity. Typical code violations include junked vehicles, garbage, and weeds. The code violation data for 2015 with x and y coordinates were obtained from the City of San Antonio’s Building Department. The code violations at each parcel were summed to derive a total count of code violations. Finally, the multivariate analyses also control for parcel size in acres (logged) as a measure of exposure, as well as business type (see Appendix for categories).
Neighborhood Variables
The present study uses CBGs to approximate neighborhoods. Land cover of commercial properties in acres was calculated by using the State Code in the Bexar County Parcel Shapefile. 7 All properties with a commercial code were selected, and a new shapefile was created out of the commercial parcels. A spatial join between the CBG shapefile and the commercial parcel shapefile was conducted. Total acreage for each commercial parcel was summed for each CBG.
Given that commercial properties are not uniformly busy, we also measure vehicular traffic in the CBG to capture “busy” contexts that may offer more criminal opportunities. Traffic activity was measured using data from Environmental Justice Screening and Mapping Tool. Specifically, we used the traffic proximity and volume indicator to capture the average annual daily traffic, measured as the count of vehicles per day within 500 meters of a block centroid, divided by distance in meters, presented as the population-weighted average of blocks in each block group (U.S. Environmental Protection Agency [EPA] 2017). 8 This variable was logged in the analysis due to skewness.
Socioeconomic and demographic characteristics were collected for CBGs in San Antonio using the 2015 American Community Survey 5-Year Estimates. Concentrated disadvantage was comprised of percent living in poverty, percent unemployed, percent of residents receiving food stamps, percent of residents without a high school diploma, GED, or a higher educational degree, and percent non-White (α = .87). 9 Residential stability measures the proportion of residents who have remained in their homes since 2009 or longer. The current study also controls for the proportion of residents who are foreign born. Population density measures the number of residents per square mile in the CBG.
Analyses and Findings
We began by examining the distribution of crime across all freestanding businesses in the study. Consistent with prior research examining crime at other types of microgeographic places (see Braga et al. 2010, 2011; Eck, Gersh, and Taylor 2000; Gill et al. 2017; Sherman et al. 1989; Weisburd et al. 2009), we found that crime was concentrated among freestanding businesses, with the distribution deviating significantly from the simple Poisson model of chance (p < .001) using the one-sample Kolmogorov-Smirnov test. Most businesses experienced no crime during the three-year study period, while few businesses accounted for a majority of all crimes observed (see Table 2). Specifically, there were no crimes at 64.16 percent of businesses during the study period, while 3.97 percent of those businesses (i.e., those that experienced 10 or more crimes during the study period) accounted for 61.55 percent of all crimes.
Distribution of Crimes across Freestanding Businesses in San Antonio, 2014-2016.
Given Eck et al.’s (2007) research on risky facilities, in which they found substantial variability in crime within heterogeneous types of facilities, we also examined the distribution of crime for each of the business types represented in the study. Of the 41 business types, 34 had crime distributions that deviated significantly from the simple Poisson model of chance (see Appendix), demonstrating substantial crime concentration within business types; an additional three business types lacked sufficient power due to small sample size (n ≤ 10). Furthermore, business type does not appear to insulate a place from crime problems. Of the 41 business types examined in the current study, 29 are represented in the top 3 percent of high-crime businesses and 34 business types are represented in the top 5 percent.
We draw on the theories and research reviewed above to identify relevant business and neighborhood variables to explain this variation in crime across freestanding businesses. To test the hypotheses predicting counts of crime at the business parcel, we estimated a series of multilevel mixed effects negative binomial regression models using Stata 13.1 (StataCorp, College Station, TX). Negative binomial regression was selected due to the count nature of the dependent variable and its skewed distribution (Cameron and Trivedi 2013). This decision was further supported by the significance of the likelihood ratio test of the overdispersion parameter (p < .001). Variance inflation factors were all less than or equal to 1.52. 10
Table 3 presents the results of the multivariate regression analysis. 11 We began by examining the effects of business characteristics that might shape opportunities for crime at the business. Specifically, we hypothesized that busier businesses—as indicated by employee size, sales volume, and square footage of the building—would experience more crime, net of other characteristics, as these high-traffic places would offer more criminal opportunities (Wilcox and Eck 2011). As predicted in Hypothesis 1, the busy business index was positively and significantly related to crime at the business. In addition, both business-level control variables—code violations and parcel size—were positively and significantly related to crime. There were also significant differences across business type. Controlling for the other variables in the model, convenience stories experienced the most crime during the study period (see Appendix for all business type effects).
Multilevel Negative Binomial Regression Predicting Crime Counts at Businesses.
Note. n = 9,028. The multilevel negative binomial regression model also controls for the 41 business types listed in the Appendix (significant incident rate ratios available in the Appendix). For those concerned about multiple testing resulting from the large number of business types controlled for in the model, a Bonferroni correction would require a p value less than .001 for rejection.
*p < .05. **p < .01. ***p < .001.
We then focused on the main effects of neighborhood characteristics on crime at the business. Hypotheses 2 and 3 predicted that businesses located in busy contexts—as indicated by CBG commercial property and traffic activity—would experience more crime. As expected, commercial property and traffic activity in the CBG were both positively and significantly associated with crime at the business, net of other variables. Social disorganization theory suggests that businesses located in CBGs with high levels of concentrated disadvantage and low levels of residential stability would experience more crime. As expected, concentrated disadvantage was positively and significantly related to crime at the business. While residential stability maintained a negative relationship with crime at the business, this relationship was not statistically significant. Similarly, the coefficient for foreign born was negative and nonsignificant.
Finally, Hypotheses 4 and 5 focused on the interactive effects of busy contexts and busy businesses on crime. Specifically, we predicted that the positive relationship between busy businesses and crime would be strengthened in CBGs with more commercial property and higher levels of traffic activity. Therefore, we created two interaction terms to test for the potential moderating effects of CBG commercial property and traffic activity on the relationship between the busy business index and crime. Table 4 presents the interactive effects of busy businesses and busy contexts on crime. While both interaction terms were positive as expected, neither reached statistical significance. In sum, Hypotheses 4 and 5 were not supported, as CBG commercial property and traffic activity did not significantly enhance the effect of the busy business index on crime at the study area business parcels. 12
Interactive Effects of Busy Businesses and Busy Contexts.
Note. n = 9,028. CBG = census block group.
*p < .05. **p < .01. ***p < .001.
Supplementary Analyses Exploring Effects by Crime Type
We then examined whether the significant main effects reported above hold across crime types, or if the effects of busy businesses and busy contexts are particularly relevant for a specific subset of crimes. To this end, we explored the effects of the covariates on property and drug offenses separately from violent crimes. 13 The model predicting drug and property offenses (including robbery, see note 12) was very similar to the full model, with the busy business index (B = .18, SE = .04, p < .001), CBG traffic (B = .14, SE = .03, p < .001), and commercial property (B = .56, SE = .26, p < .05) all positively associated with property and drug offenses. The other covariates also exhibited similar effects with respect to direction and significance. A very different picture emerges, however, when predicting violent crimes (excluding robbery, see Note 12): the busy business index is no longer significant (B = −.09, SE = .06, p > .05), the effect of CBG traffic is weaker (B = .10, SE = .05, p < .05), and commercial property is nonsignificant (B = .54, SE = .39, p > .05). As with the models predicting total crime at the business parcels, the interactive effects were nonsignificant in the crime type specific models.
Discussion
The present study examined the distribution and sources of crime across freestanding businesses in San Antonio, TX. Previous work has demonstrated considerable concentration in crime across other microgeographic units of analysis, and our findings indicate that crime is also highly concentrated among freestanding businesses, with relatively few businesses accounting for a large proportion of all crimes. The degree of concentration observed in the present study is similar to what has been reported in previous research. We found, for example, that 3.97 percent of businesses accounted for 61.55 percent of crimes. This compares to 4.5 percent of street segments accounting for approximately half of all crimes in Seattle (Weisburd et al. 2004), less than 3 percent of street segments and intersections accounting for more than half of all gun assaults (Braga et al. 2010) and 5 percent accounting for approximately two-thirds of all robberies in Boston (Braga et al. 2011), 5.25 percent of blocks accounting for about half of all crimes in Vancouver (Curman et al. 2015), and approximately half of all crimes occurring at just 2 percent of all street segments in Brooklyn Park, MN (Gill et al. 2017). In short, our study adds to the growing number of studies documenting the substantial variation in crime across micro places, highlighting the need for attention at these types of units of analysis. Furthermore, the present study revealed that the variation in crime at businesses was nonrandom, thus lending support to Weisburd’s (2015) call for theoretically informed models to explain the sources of variation.
Place type alone does not appear to be a sufficient explanation of crime risk across microgeographic places, as our univariate findings revealed considerable variation within business type; an overwhelming majority of business types exhibited significant nonrandom distributions, adding to Eck et al.’s (2007) findings about risky facilities and reinforcing Wilcox and Eck’s (2011) Iron Law of Troublesome Places. The fact that there was substantial variability in crime within business types suggests that additional explanations beyond business type are in order. We relied on an integrated multilevel approach to understanding crime at businesses that drew on the social disorganization tradition as well as environmental criminology, with an explicit focus on busy businesses and contexts.
The primary focus of the present study was examining Wilcox and Eck’s (2011) prediction about the criminogenic effects of high-traffic contexts and places. They have argued “busy places in general—rather than specific facility types—offer criminal opportunity. High-traffic locations are located nonrandomly across cities, and thus, they help to structure criminal opportunity ecologically” (p. 475). Consistent with their arguments, we found that “busy” businesses experienced more crime, even when accounting for business type and other covariates. Such places have a large volume of users and transactions, creating numerous opportunities for criminal behavior. That said, the significant differences across business types reported in the Appendix suggest that some business types may create or restrict criminal opportunities in ways unrelated to the busy business activities measured here. In other words, our findings suggest that both busy places and specific facility types may shape criminal opportunity.
In some ways, the association between busy businesses and crime is surprising, as we might expect busier, more successful businesses to be more effective in addressing crime problems. Yet given their primary function is not crime control, some businesses may view some degree of crime as a necessary cost of doing business, since crime appears to be in part a function of sales volume. The business, however, does not solely bear the costs of crime. As Eck and Eck (2012) point out, governments play a large role in paying for crime control. They argue for regulatory policies that shift some of this responsibility by encouraging owners of high-crime places to prevent crime. Indeed, there is research that indicates pressuring place managers to take control of crime problems can be effective (e.g., Eck 1998). Our study highlights a need to focus on the small proportion of businesses that generate an overwhelming majority of crime among this group of micro places. The fact that businesses with more employees and higher sales experience more crime is in some ways promising, as such businesses may have untapped human and financial resources that can be leveraged for crime prevention.
We also found support for the importance of busy contexts in defining aggregate criminal opportunity, as businesses located in CBGs with more commercial property had more crime. Given that commercial properties are not uniformly busy, we also examined the effects of vehicular traffic and found that businesses located in high-traffic contexts had more crime, net of commercial property in the area and population density. In other words, simply relying on residential population and land use variables to understand the spatial distribution of crime fails to capture an important aspect of “busy” contexts. High-traffic communities may have more criminal opportunities for a number of reasons including a larger supply of victims, more motivated offenders, and weaker guardianship due to the anonymity that often accompanies busy contexts.
The reported findings, however, were not consistent for all crime types. Of note, supplementary crime-type specific analyses revealed that the busy business index was not significantly related to violent crimes (with the exception of robbery, see Note 12) at the business parcel, suggesting that the context and opportunity structures for these offenses may be different from financially motivated crimes and drug offenses. A primary explanation for why busy businesses may have more crime is that they act as crime generators, drawing in large concentrations of potential targets (as in the case of robbery, shoplifting, theft, etc.), or crime attractors with known criminal opportunities for specific types of crime (as in the case of drug offenses; Brantingham and Brantingham 1995). Yet, the victim in violent offenses such as assaults is usually known to the perpetrator, with the crime resulting from interpersonal conflict or abuse, making the specific criminal opportunity structures offered by busy businesses less relevant for these types of crime. Indeed, R. B. Felson, Messner, and Hoskin (1999), analyzing National Crime Victimization Survey data from the latter part of 1992 through 1994, report that a large majority of assault incidents occur between people who know one another, with only 30.5 percent of incidents perpetrated by strangers.
Counter to our predictions, we did not find evidence that busy contexts and busy businesses interact to increase crime at the parcel beyond their main effects. One possibility for the nonsignificance of the interaction terms might be our focus on crime at the parcel. This differs from Deryol et al.’s (2016) study, which found that the effects of crime generators on crime at nearby addresses were stronger in CBGs with a higher density of commercial land use. It is possible that busy contexts amplify the effects of busy businesses on crime for nearby places. That is, the high volume of customers that busy businesses attract may be particularly at risk of criminal victimization in high-traffic communities, but this amplified risk exists off-site as they travel to and from the business, where customers may be more likely to encounter potential offenders. It is also important to note that the interactive effects between nearby crime generators and CBG commercial density observed in the Deryol et al. (2016) study are limited to specific types of places (i.e., carryout liquor establishments, on-premise drinking establishments, and bus routes). Alcohol establishments, more than other types of places, may provide more potential targets and offenders due to the alcohol consumption of its customers, and these customers then encounter each other or others when traveling to or from the establishment in a commercially dense area. Similarly, nearby bus routes suggest a pedestrian population that travels from home to the bus stop, potentially serving as a pool of suitable targets. In short, the diverse set of businesses in the present study may also explain the unexpected nonsignificant interactive effects.
Our findings confirm the importance of other place-level characteristics in understanding crime at businesses. Given the theoretical perspectives we draw from, we thought it was important to account for place management in our models, and we found that code violations were related to more crime at the business. While we do think code violations are likely associated with weak place management, we recognize that this measure does not capture the range of management practices related to organization of space, regulation of conduct, control of access, and acquisition of resources that might influence crime at a business (Madensen 2007). For example, signage, on-site security personnel, and video surveillance are just a few tools that may help to control crime at businesses. Furthermore, code violations are a function of both business and enforcement behavior. Unfortunately, we do not have data on management practices and this remains a limitation of the present study. Future data collection efforts should operationalize specific place management practices designed to address crime problems as well as local law enforcement practices.
We found partial support for the social structural influence of the broader environment on crime at businesses, as businesses located within CBGs with high levels of concentrated disadvantage experienced more crime. This finding demonstrates that the well-established linkage between concentrated disadvantage and neighborhood crime includes crimes at businesses within those neighborhoods. In other words, businesses do not appear to be immune from the broader social structure in which they are embedded. That being said, the negative association between residential stability and crime at the business was not statistically significant, indicating that there may be limits to the influence of neighborhood variables on crimes occurring at businesses, as opposed to those that occur on the street or within residences. If we assume that residential stability operates through some form of collective informal social control to reduce crime in neighborhoods, it stands to reason that residents either do not activate such control to address crime at businesses or, when they do, their efforts are unsuccessful. Perhaps resident-based control is only effective among locally owned businesses that are more aware of community concerns and susceptible to related pressures. This highlights the fact that the concentrated disadvantage and the residential stability variables do not directly measure community processes. We encourage future research utilizing more direct measures of resident-based control that examine whether and how resident behavior can influence crime at businesses and whether this influence varies depending on type of ownership.
Our units of analysis—freestanding businesses—have implications that warrant further discussion. Beyond the conceptual and methodological justifications described above, freestanding businesses are consistent with a problem-solving approach to crime prevention that prefers small units of analysis over larger spatial units that may actually contain a collection of different crime problems in need of different interventions (Clarke and Eck 2005). As Weisburd (2012) argues, lowering the scale of interventions to specific places with crime problems makes interventions more relevant to those who practice crime prevention. Freestanding businesses, unlike businesses located in strip malls or shopping centers, do not share responsibility of the parcel with other entities, perhaps making the implementation of regulatory policies aimed at encouraging place management more feasible (Eck and Eck 2012). Even street segments—a common unit of analysis in micro place studies—may have several businesses and owners. This is particularly relevant for understanding and addressing crime problems, as M. Felson (1995) has argued that crime discouragement—including in the form of place management—weakens as responsibility becomes more diffuse. Studying crime at units of analysis that have practical relevance broadens the utility of our findings.
Given that freestanding businesses are conceptually distinct from businesses located in strip malls and shopping centers, the generalizability of the findings is limited to freestanding businesses. It is unknown whether the relationships observed in the present study would operate similarly across these different units. The operationalization and measurement of key constructs in the current study are not easily applied to businesses in strip malls and shopping centers. The traffic at these businesses, for example, may be more closely tied to the characteristics of a nearby “anchor” business than the characteristics of the business itself. Furthermore, attributing crime in a shared parking area to a particular business would require a different methodology than that employed in the current study.
Beyond the issues discussed above, there are additional limitations of the present study to consider when interpreting our results. First, random assignment was not possible given the variables of interest in the current study. While we made every effort to develop rigorous models informed by prior theory and empirical research, it remains possible that one or more of the observed relationships in the present study is spurious. Given the existing literature hypothesizing the effects of high-traffic places and high-traffic contexts on crime (Wilcox and Eck 2011), the present study focused on indicators of busy businesses and busy CBGs. The available data, however, did not allow us to differentiate among the types of people who frequent the businesses and the surrounding neighborhoods and the differences in crime risk posed by various individuals and groups. Second, our study is limited to freestanding businesses in San Antonio, TX. As noted, the relationships reported here may operate differently for businesses located in strip malls and shopping centers, as well as freestanding businesses in other cities. For example, using vehicular traffic activity as a measure of busy contexts may be less appropriate for more densely populated areas with lower rates of vehicle ownership. In 2016, there were approximately 1.71 vehicles per household in San Antonio according to Census survey estimates, which is fairly close to the 1.8 vehicles available per U.S. household. Compare this to New York and Washington, DC, which only had 0.63 and 0.86 vehicles per household, respectively. We encourage researchers to test our hypotheses using other study areas to determine the generalizability of our findings. Third, it is possible that some business activities are not conducted on-site, and thus, crimes related to the business activities would not be recorded at the parcel. Unfortunately, the available data do not allow us to examine the extent to which this occurs.
Finally, our study relies on official data provided by government agencies, including the SAPD, the U.S. Census Bureau, and the EPA, all of which have limitations. Because the police data only contain crimes reported to the police, the extent to which crimes occurring at businesses go unreported to the police and are instead handled internally by management is unknown. There could also be imprecision in recording crime location, though improvements in recent years to SAPD’s geo-validating protocols minimize such concerns (see Note 5). Furthermore, freestanding businesses represent discrete and defined places; there is likely less error when reporting location for the crimes that are the focus of the present study. The Census and EPA data were only available for Census defined geographic units. To minimize the modifiable areal unit problem, we used CBGs to approximate neighborhoods because they represent the lowest level of aggregation at which the Census variables and traffic activity indicator were available. That said, CBGs impose boundaries that may not be representative of how neighborhoods are defined by municipalities and residents (see Hipp and Boessen 2013). For these reasons, we encourage future research that examines the impact of busy contexts and busy businesses on crime using data sources that measure crimes unreported to the police and alternative measures of busy contexts that do not rely on Census defined geographic units.
In conclusion, the present study contributes to the crime and place literature by examining the distribution and sources of crime at a specific microgeographic unit of analysis, that is, freestanding businesses. Crimes were highly concentrated at relatively few businesses, and this variation could not be explained away by business type alone. Instead, our multilevel analysis revealed that both neighborhood and business characteristics are associated with crime at businesses, lending support to using a multilevel theoretical approach to understanding crime at small spatial units. Among other things, busy businesses and those located in busy contexts experienced more crime, suggesting there may be unintended consequences associated with strategies designed to maximize traffic at and around businesses. Future studies should continue to examine why crime concentrates at a relatively few places.
Footnotes
Appendix
Freestanding Businesses.
| Business Type | N | % | Mean # of Crimes | One-Sample Kolmogorov- Smirnov Testa | Incident Rate Ratiob |
|---|---|---|---|---|---|
| Hospitals/surgical and emergency centers | 29 | 0.32 | 0.21 | 0.16 | |
| Youth and educational services | 8 | 0.09 | 0.13 | 0.34 | |
| Colleges/universities/tech/trade schools | 18 | 0.20 | 1.22 | ** | 0.47 |
| Auto dealer | 70 | 0.78 | 1.27 | *** | 0.53 |
| Funeral homes/services and cemeteries | 35 | 0.39 | 1.09 | 0.59 | |
| Mental health | 40 | 0.44 | 0.48 | 0.82 | |
| Other professional services | 1,037 | 11.49 | 1.06 | *** | 0.87 |
| Investment | 48 | 0.53 | 0.67 | * | 0.90 |
| Banks/credit unions/lenders/ATMs | 96 | 1.06 | 1.16 | *** | 0.92 |
| Medical/health offices | 534 | 5.91 | 1.25 | *** | 0.93 |
| Home health-care services | 44 | 0.49 | 0.43 | 0.95 | |
| Contractor and building services | 1,902 | 21.07 | 1.03 | *** | — |
| Used car dealer | 226 | 2.50 | 0.89 | *** | 1.05 |
| Video rental | 10 | 0.11 | 0.60 | 1.07 | |
| Legal services | 162 | 1.79 | 1.41 | *** | 1.13 |
| Self-storage units | 111 | 1.23 | 3.68 | *** | 1.13 |
| Child day care services | 169 | 1.87 | 1.48 | *** | 1.14 |
| Entertainment/recreation | 48 | 0.53 | 4.58 | *** | 1.17 |
| Automotive | 790 | 8.75 | 1.14 | *** | 1.33 |
| Consumer goods rentals | 23 | 0.25 | 1.70 | ** | 1.35 |
| Retail | 900 | 9.97 | 2.86 | *** | 1.36 |
| Accountant/tax prep | 129 | 1.43 | 1.89 | *** | 1.38 |
| Beauty/nail salons/barber shops | 189 | 2.09 | 0.85 | *** | 1.39 |
| Car washes | 56 | 0.62 | 0.84 | ** | 1.50 |
| Insurance and real estate | 377 | 4.18 | 2.20 | *** | 1.53 |
| Assisted living/nursing/retirement communities | 34 | 0.38 | 5.21 | *** | 1.75 |
| Dry cleaning and laundry services | 63 | 0.70 | 1.35 | ** | 1.79 |
| Liquor Stores | 24 | 0.27 | 1.25 | † | 1.92 |
| Employment/vocational rehab agencies | 16 | 0.18 | 5.38 | *** | 2.08 |
| Payday loans and check cashing services | 35 | 0.39 | 2.23 | *** | 2.11 |
| Pharmacies and drug stores | 10 | 0.11 | 1.30 | 2.34 | |
| Fitness/sports | 77 | 0.85 | 4.09 | *** | 2.47 |
| Pawn shops | 43 | 0.48 | 2.53 | *** | 2.83 |
| Restaurants and coffee shops | 839 | 9.29 | 3.35 | *** | 2.83 |
| Grocery and food stores | 139 | 1.54 | 4.99 | *** | 3.16 |
| Hotels/motels | 243 | 2.69 | 5.07 | *** | 3.26 |
| Fast food | 253 | 2.80 | 3.08 | *** | 3.36 |
| Bars | 89 | 0.99 | 2.79 | *** | 3.57 |
| Gas stations | 51 | 0.56 | 5.16 | *** | 6.24 |
| Escort services/tattooing/psychics | 18 | 0.20 | 5.11 | *** | 6.51 |
| Convenience stores | 43 | 0.48 | 8.26 | *** | 6.71 |
| Total | 9,028 | 100.00 | 0.19 | *** |
Note. n = 9,028.
aNull hypothesis: Within-place type crime is a Poisson distribution.
bIncident rate ratios for the business types in the multilevel negative binomial regression predicting crime counts at businesses (rank ordered for ease of comparison). Contractor and building services were the most prevalent business type and thus the omitted reference category in the analysis. VIFs = Variance Inflation Factors.
†p < .10. *p < .05. **p < .01. ***p < .001.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
