Abstract
Although theories posit that some types of local institutions will have a crime-producing influence in neighborhoods while others will have the opposite effect, the empirical evidence is far from conclusive. Previous studies are typically limited to analyzing cross-sectional data and one type of institution. Using longitudinal data of the number of employees of various institutions within census tracts across nine U.S. cities, the present study examines the longitudinal impact of four types of institutions on violent and property crime. Negative binomial regression models suggest that alcohol outlets and banking establishments increase criminal opportunities, whereas “third places” like coffee shops and cafes induce efficacious neighborhood control and social action. Civic and social organizations have no statistical relationship with crime.
Keywords
Introduction
Communities and crime research has established that crime is not randomly distributed across space; rather, crime is spatially concentrated (Baller, Anselin, Messner, Deane, & Hawkins, 2001; Messner et al., 1999; Sherman, Gartin, & Buerger, 1989). In many cities, a large percentage of all crimes occur in a small percentage of all neighborhoods (Morenoff, Sampson, & Raudenbush, 2001; Shaw & McKay, 1942). Such neighborhoods tend to be impoverished minority communities that lack the collective ability to maintain effective social controls (Sampson & Groves, 1989; Sampson, Raudenbush, & Earls, 1997). One source of neighborhood control is the institutional base of the neighborhood (Bursik & Grasmick, 1993; Hunter, 1985; Morenoff et al., 2001; Peterson, Krivo, & Harris, 2000; Sampson & Groves, 1989; Triplett, Gainey, & Sun, 2003; Wilson, 1987), broadly defined as the local organizations and businesses situated in the focal neighborhood. The institutional base provides residents a host of services and activities, which includes but is not limited to banking, entertainment, retail, politics, advocacy, religion, and social services.
The goals of local institutions are not uniform. For some local institutions, their objectives center on identifying and solving community problems (e.g., civic and crime prevention organizations); however for others, their goals have less to do with helping the larger community and more to do with generating profits (e.g., bars and banks). Accordingly, for one to assume that a strong institutional base is associated with lower crime rates is too simplistic in light of institutional diversity. Previous research has suggested that some institutions facilitate effective social controls (Beyerlein & Hipp, 2005; Lee, 2008; Papachristos, Smith, Scherer, & Fugiero, 2011; Peterson et al., 2000; Slocum, Rengifo, Choi, & Herrmann, 2013), whereas others produce criminal opportunities (Bernasco & Block, 2011; Kubrin, Squires, Graves, & Ousey, 2011; Pridemore & Grubesic, 2013; Roncek & Maier, 1991; Sherman et al., 1989).
A challenge with studying the relationship between local institutions and crime is that this is likely an endogenous process. That is, it is likely that the presence of some types of local institutions vary in response to neighborhood crime rates. If it is true that a selection process operates in which some institutions select certain types of neighborhoods to locate (with crime being a critical consideration), an implication is that cross-sectional studies may actually reflect how crime affects the presence of institutions rather than the reverse. Nonetheless, the bulk of prior research is conducted as cross-sectional analysis. A risk is that even if local institutions are found to be associated with neighborhood crime rates in cross-sectional studies, these are not necessarily capturing the unidirectional effect of institutions on crime. What is needed to tease apart this relationship is longitudinal data.
This study builds on previous research by assessing the longitudinal impact of differing types of institutions on neighborhood crime rates using a sample of census tracts (N = 1,024) located in nine U.S. cities. The results suggest that social disorganization and routine activities theories are relevant to understanding the institutions–crime nexus. In particular, the social reality that neighborhood control and criminal opportunity, emanating from institutions, simultaneously operates in neighborhoods to affect crime rates. In determining how different types of local institutions influence crime, given that previous studies have almost exclusively implemented cross-sectional analysis to model such a dynamic process, we estimate both cross-sectional and longitudinal models to determine whether the former produces institutional effects on crime consistent with the latter.
The Protective Effects of Local Institutions
Research on the protective effects of local institutions emanates from social disorganization theory. Tracing back to the “Chicago School,” neighborhood characteristics such as poverty, residential instability, and ethnic heterogeneity were posited to have a crime-producing influence (Shaw & McKay, 1942). At present date, extensive research shows that neighborhoods that suffer from these conditions lack a dynamic in which residents share common values and goals, trust one another, and are willing to intervene on behalf of the common good, which decreases the ability to informally control the behavior of residents, thereby increasing the likelihood of crime (Bursik & Grasmick, 1993; Kornhauser, 1978; Kubrin & Weitzer, 2003; Sampson & Groves, 1989; Sampson et al., 1997). A fundamental implication of the theory is that in socially disorganized neighborhoods, residents lack the local institutions that facilitate the mobilization of resources and the formulation of social ties that are necessary for informal social control (Beyerlein & Hipp, 2005; Morenoff et al., 2001; Pattillo, 1998; Peterson et al., 2000; Sampson & Groves, 1989; Slocum et al., 2013; Wilson, 1987). Consequently, for communities and crime scholars, there is a growing need to identify those institutions that stimulate social organization or collective efficacy among residents.
Previous studies have typically conducted cross-sectional analysis to determine the types of local institutions that help lower neighborhood crime rates through the establishment of informal social control. A seminal study by Peterson et al. (2000) examined whether the presence of recreation centers and libraries impacted crime rates in Columbus (OH) neighborhoods. Peterson and colleagues discovered that while libraries had no significant effect on crime, the presence of recreation centers appeared to mitigate violent crime in the most disadvantaged neighborhoods. In another study, Beyerlein and Hipp (2005) investigated how religious institutions were associated with crime rates in U.S. counties. The authors found that greater numbers of congregations per capita—regardless of the denomination (mainline Protestantism, evangelical Protestantism, and Catholicism)—were associated with lower crime rates. In a third example, Lee (2008) developed a civic engagement index that not only included the number of religious congregations, but also the number of civic associations, sport leagues, and hobby and special interest groups and found that rural counties with higher levels of civic engagement had lower crime rates. Collectively, these studies indicate that civic, social, and religious institutions may help to control crime in neighborhoods.
Another type of institution that potentially contributes to lower crime rates is what Oldenburg (1999) refers to as third places: “public places that host the regular, voluntary, informal, and happily anticipated gatherings of individuals beyond the realms of home and work” (p. 16). Informal public gathering places like coffee shops, cafes, ice cream parlors, and pizza parlors are potentially the locus for the formulation and maintenance of social networks, which in turn initiate mechanisms of informal social control. We propose that these establishments present a favorable setting for residents to build their social networks. As such, when crime and other community problems arise, the neighborhood is likely better equipped to collectively solve such problems through the dissemination of information and mobilization of resources. A recent longitudinal study by Papachristos et al. (2011) demonstrated that third places can provide aggregate benefits. The authors showed that the presence of coffee shops was related to lower homicide rates in Chicago neighborhoods. However, this study showed the effect of coffee shops on neighborhood crime under the context of gentrification, thereby the crime control benefits identified by the authors may not be generalizable to more stable neighborhoods.
Despite some evidence that civic, social, and religious institutions are negatively associated with crime, it would be premature to consider these results conclusive, in part because there are a small number of studies, and more importantly, they usually involve cross-sectional analysis. For example, a recent study by Slocum et al. (2013) showed only weak to moderate evidence that civic and religious-based organizations are important for neighborhood crime reduction. Despite creating numerous measures for different types of community organizations (many civic and religious based), most of these measures failed to demonstrate significant (main) effects on violent and property crime. Furthermore, given the theoretical insights of Oldenburg (1999) and the empirical study conducted by Papachristos et al. (2011), we need more evidence to determine whether third places are important for understanding neighborhood crime.
The Adverse Effects of Local Institutions
While explanations for the protective effects of local institutions are based on social disorganization theory, the routine activities perspective sets the theoretical groundwork for why some types of local institutions might increase crime. The routine activities perspective views crime as the convergence in time and space of motivated offenders, suitable targets, and the absence of capable guardianship (Cohen & Felson, 1979; Felson, 1987; Felson & Boba, 2010). A major implication is that people’s routine activities have the capacity to influence crime rates without any significant change to their own criminal dispositions. One key to crime reduction, then, is to identify places that present an opportunity structure or situational setting in which crime is likely to occur. Two key local institutions, alcohol outlets and payday lenders, may elevate neighborhood crime rates due to the customers they attract, the services and products they provide, or the environmental features of onsite locations.
Bars and liquor stores have been posited to increase criminal opportunities in neighborhoods (Eck, Clarke, & Guerette, 2007; Felson & Boba, 2010; Parker, 1995; Roncek & Maier, 1991; Sherman et al., 1989). The reasoning hinges on the purchase and consumption of alcohol. First, when people drink, alcohol breaks down users’ normative standards of conduct and impairs cognitive decision making—“the disinhibiting effect” (for a detailed discussion see Parker, 1995). Second, individuals who drink in bars and at liquor store locations are situated in environments where intimate advances and masculine posturing is commonplace (Parker & Rebhun, 1995; Pridemore & Grubesic, 2013). The disinhibiting effect of alcohol in conjunction with the social norms of bars and liquor stores therefore elevates the risk for on-scene occurrences of crime. And lastly, patrons of these establishments not only have a tendency to leave intoxicated, but they are also known to carry cash, which leaves them vulnerable to motivated offenders. As a result, the criminal opportunity structure of bars and liquor stores extends to surrounding areas of onsite locations. Because the aforementioned features elevate the level of motivated offenders and suitable targets (assuming capable guardianship is held constant), it is not surprising that previous studies have typically found that bars and liquor stores have a positive effect on neighborhood crime (Bernasco & Block, 2011; Hipp, 2010; Livingston, 2008; Nielsen & Martinez, 2003; Peterson et al., 2000; Pridemore & Grubesic, 2013; Roman & Reid, 2012; Roncek & Maier, 1991). Yet despite this extensive body of research, only a handful of studies have assessed this relationship in a longitudinal manner (Gruenewald & Remer, 2006; Hipp, 2010; Livingston, 2008).
Payday lenders may also increase criminal opportunities. Although these establishments are legal, their exorbitant interest rates and fees mean they exploit the financial stability of many patrons, and in some cases, create debt-traps for customers who take out loans to payback previous loans. Accordingly, the individual-level impact of payday lending institutions translates to a broader impact on community crime rates—a recent study conducted by Kubrin et al. (2011) revealed that Seattle, Washington, neighborhoods with a greater concentration of payday lenders had higher crime rates. From a routine activities approach, Kubrin et al. (2011) hypothesize that the availability of large sums of cash at late hours during the evening, as well as on the weekend, is advantageous to offenders victimizing patrons of payday lenders.
A question that emerges is whether more traditional banking establishments are also associated with higher crime rates. Depending on the theory adopted, traditional banking can be posited to have a positive or negative effect on neighborhood crime. For example, the routine activities perspective suggests that traditional banking establishments increase criminal opportunities, thereby leading to higher crime rates. However, social disorganization theory implies that these establishments increase neighborhood control through the provision of loans, mortgages, and financial relief services, which results in lower crime rates.
Arguably, the most significant limitation of prior studies is the predominance of cross-sectional analysis. A natural concern is that these studies do not capture how local institutions affect neighborhood crime, but rather, they demonstrate the social process in which the presence of institutions varies in response to crime. It is plausible that owners of bars, liquor stores, or payday lending establishments choose to develop in neighborhoods that are already characterized by crime, disorder, and poverty, raising the methodological issue of temporal precedence (a selection effect may also operate for those institutions that control crime in neighborhoods). Comparatively, a longitudinal study could reveal how certain local institutions are related to subsequent change in neighborhood crime rates.
Goals of the Present Study
Previous studies have posited that some types of local institutions increase crime, while others have the opposite effect. Instead of adopting the approach that either neighborhood control or criminal opportunity is the dominant mechanism that operates in neighborhoods, the present study bridges past studies by studying the longitudinal impact that four types of local institutions, collectively embodying both mechanisms, have on crime.
This study also expands the institutions–crime literature in several methodological ways. First, we use a longitudinal data set that covers an 11-year period (1998-2008). Given the critical concern for endogeneity, longitudinal data provide greater ability to establish temporal precedence—our longitudinal models capture the unidirectional influences of local institutions on crime. 1 Second, previous studies are limited to examining neighborhoods within a single city at a single time point, raising the question whether findings can be generalized from neighborhoods of a particular city to neighborhoods of other cities, as well as from a particular time point to other points in time. We address this concern by not only using longitudinal data, but also by incorporating a sample of census tracts located across nine U.S. cities. Finally, as most research is limited to examining one type of local institution, we follow the lead of Peterson et al. (2000) and Slocum et al. (2013) and consider the effects of multiple institutions. As some types of local institutions cluster in space, it is methodologically important to minimize the risk that an unmeasured institution explains the influence that the focal institution has on crime.
Data
This study integrates data from the U.S. Census and County Business Patterns, with official crime data reported by police departments. The models include data for tracts (N = 1,024) in nine U.S. cities from 1998 to 2008. 2 These cities were not selected at random, but rather are a convenience sample of cities with available data for as many years of the study period. Therefore, this study does not generalize to the population of U.S. cities, but rather highlights the differences over time in tracts within particular cities of particular years by conditioning out the differences across cities and years, as discussed in the analytic strategy section. In addition to most of the data being situated at the tract-level of aggregation, the advantages of using census tracts are that past studies have frequently used them as a proxy for neighborhoods (for example, see Baumer, 2002; Hipp, 2007; Hipp & Yates, 2011; Peterson & Krivo, 2010)—tracts contained a mean of about 4,300 residents in 2000 with 95% of the tracts containing between 1,400 and 8,000 persons (Hipp, 2010)—and they were initially created by the Census Bureau to be relatively homogenous entities (Green & Truesdell, 1937; Lander, 1954).
Dependent Variables
The dependent variables in the analyses are based on crime reports officially coded and reported by police departments (Table 1). These data pertain to the nine cities of the study, and are aggregated to census tracts. Although these data are limited to capturing crimes that have been recognized by the police via self-determination or citizen reports, we have no reason to suspect that these data are any less valid than other official crime data sources. Accordingly, the estimated models use the number of violent crimes (murder, aggravated assault, and robbery) and the number of property crimes (burglary, larceny, and motor vehicle theft) as outcome measures.
Descriptive Statistics of Variables Used in the Longitudinal Models.
Note. Descriptive statistics are for all cities and years combined. The lagged crime rate is based on the number of crimes divided by the tract population. For descriptive purposes, the lagged crime rates and institutional variables have not been logged transformed. Number of cases = 6,327; number of tracts = 1,024.
Independent Variables: Local Institutions
The County Business Patterns is a national data source that provides zip code information on U.S. businesses. As the data are originally situated at the zip code level of aggregation, the data are collapsed according to the proportion of the zip code population contained by each of the tracts comprising the focal zip code. 3 The study then uses the North American Industry Classification System (NAICS), a system that categorizes businesses by six-digit codes, to create four local institution variables: alcohol outlets, civic and social organizations, third places, and banking establishments. Specifically, as the County Business Patterns reveal the number of employees by each unique NAICS code, 4 each institutional variable is created by taking the sum of the number of employees of the corresponding NAICS codes. 5
In this study, we create an inclusive measure of alcohol outlets that purportedly increase crime. The alcohol outlets index is based on NAICS codes that highlight the number of employees of bars, liquor stores, taverns, nightclubs, and cocktail lounges. We predict that alcohol outlets will have a crime-producing influence in neighborhoods.
Scholars contend that civic and social organizations facilitate informal social control. In line with past studies, we construct a civic and social organizations index that incorporates NAICS codes highlighting the number of employees of civic associations, social organizations (e.g., clubs), youth organizations, public safety organizations, and neighborhood development associations. Our expectation is that these organizations will negatively influence crime rates.
“Third places” have been posited to be crucial for the vitality of neighborhoods (Oldenburg, 1999), and that their benefits may extend to crime control (Papachristos et al., 2011). Consequently, we construct an index of third places by drawing on NAICS codes highlighting the number of employees of coffee shops, cafes, bagel and doughnut shops, pizza parlors, ice cream parlors, diners, and snack and beverage shops. Because these types of places are likely to engender networks of effective social action, we suspect that third places will negatively impact crime rates.
Although Kubrin et al. (2011) investigated the effect of payday lenders on crime rates, such establishments capture non-traditional banking. The impact of more “legitimate” banking establishments remains to be determined. Therefore, we construct an index of banking establishments, which is derived from NAICS codes highlighting the number of employees of banks, depository trust companies, federal savings and loan associations, and credit unions. Prior to our analysis, we did not posit a hypothesis, as theories suggest that banking establishments may increase criminal opportunities or neighborhood control.
Independent Variables: Other Neighborhood Characteristics
Measures from the U.S. Census and American Community Survey are used to examine the influence of other neighborhood characteristics on crime. These measures are incorporated because they signify critical differences across neighborhoods and embody standard correlates of neighborhood crime. We construct the following variables: population of tract, percent of tract that lives in poverty, the average length of residence in tract, percent of tract represented by immigrants, percent of tract represented by Blacks, percent of tract represented by Latinos, percent of tract represented by Asians, percent of tract represented by Whites, and percent of tract represented by other races.
Census data are provided for the time points, 1990 and 2000. The American Community Survey provides corresponding data for the time period, 2005-2009; consequently, we maintain that this period is appropriate to be used as the time point for 2007. 6 For each of the neighborhood characteristic variables, standard linear interpolation is performed with three time points (1990, 2000, and 2007) to provide data for each of the remaining years that cover the time period, 1998 to 2008.
Spatially Lagged Measures
Research has emphasized the spatial dependence of neighborhoods in relation to the distribution of crime (Anselin, 1988; Anselin, Cohen, Cook, Gorr, & Tita, 2000; Deane, Messner, Stucky, McGeever, & Kubrin, 2008). It follows that a potential problem arises when performing regression analyses; the residuals of estimated crime outcomes may be spatially correlated (spatial autocorrelation). It is thus important that communities and crime studies address potential spatial autocorrelation because failing to do so may result in biased coefficient estimates, false indications of significance, and misleading suggestions of model fit (Messner et al., 1999). To account for this potential problem, we implement spatially lagged exogenous measures. Spatial lags capture the spatial impact that community factors have on the focal neighborhood, which in turn can control for spatial autocorrelation.
Spatially lagged measures are created for all of the independent variables. These measures are generated based on an inverse distance decay function with a cutoff at two miles (beyond which the neighborhoods have a value of zero in the W matrix), and the resulting spatial weights matrix (W) is then row-standardized. We then multiply this matrix by the matrix of values of our exogenous variables in the census tracts in the study. Although there can be ambiguity concerning which tracts are physically “close,” a cutoff of two miles is reasonable given prior studies that have suggested that the effect of neighborhood factors (on the focal neighborhood) can extend out 0 to 2 miles, 0 to <1.5 miles, or 0 to <1 mile (Bielefeld, Murdoch, & Waddell, 1997; Hipp & Yates, 2011; Peck, 2008). Moreover, the median census tract in 2000 was about 1.4 miles across, which translates to 1.95 square miles (Hipp, 2010).
Analytic Strategy
As described earlier, crime is an environmental signal that can dictate where entrepreneurs locate their business establishments. The methodological implication is that the presence of an institution may vary in response to crime, inducing a potential reciprocal relationship, which in turn jeopardizes the ability to demonstrate the unidirectional influences of local institutions on crime. Using longitudinal data allows us to account for the likely endogeneity in the placement of local institutions in relation to neighborhood levels of crime. To establish temporal precedence, each of the institutional variables are time-lagged by 1 year. The institutional variables are then log transformed to address the potential for outliers. 7 To reduce the risk of spuriousness, the crime rate for both violent and property crime are lagged by 1 year and included in the longitudinal models as separate predictors; 8 previous studies have shown that previous crime rates exert especially strong effects on current levels of crime (Hipp, 2010; Kirk & Papachristos, 2011; Morenoff et al., 2001; Papachristos et al., 2011).
Given the dynamic nature of neighborhood processes and the data being longitudinal, a salient concern is that an unobserved factor/process is both driving crime trends and institutional development. Consequently, we adopt a fixed-effects approach to estimate the institutional effects on neighborhood crime. Fixed-effects models essentially eliminate time-stable unobserved neighborhood processes and thereby reduce the risk of spuriousness (Allison, 2005; Halaby, 2004). Dummy control variables are created for each of the cities and years that are reflected by the data, and subsequently included in the models shown. Such additions allow the opportunity to assess changes in crime between those tracts of the same city and same year.
The dependent variables indicate the number of violent and property crimes, respectively. Given that the dependent variables reflect overdispersion, we estimate the number of violent and property crime events using negative binomial regression—a variant of Poisson regression that effectively deals with overdispersion (Hilbe, 2007; Osgood, 2000). 9
Because tracts differ in their exposure to risk with respect to crime, the tract (logged) population is used as an exposure term in all models, and the coefficient is constrained to equal 1. The inclusion of the exposure term effectively translates to crime rates. We estimate the longitudinal model below:
where y is the number of crime events that year, α is an intercept, lny is the natural logarithm of the crime rate of the previous year, ln
The analysis takes the following course. The cross-sectional models estimate crime using a pooled estimator in conjunction with the institutional variables of the current year (logged transformed), neighborhood characteristic variables, along with the dummy variables for cities and years. 12 Comparatively, as described above, the longitudinal models estimate crime using a longitudinal estimator with the crime rate of the previous year (logged transformed), institutional variables of the previous year (logged transformed), neighborhood characteristic variables, and dummy variables for cities and years. In addition to reporting the findings of the institution and neighborhood characteristic predictors, we compare the results between the cross-sectional and longitudinal models to demonstrate the importance of longitudinal analysis. Finally, we provide ancillary models that estimate neighborhood crime using the variables (i.e., within-tract and spatial lag) for each type of local institution, separately.
Results: Cross-Sectional Models
Using a cross-sectional framework for analyzing the relationship between local institutions and neighborhood crime, we find that nearly all of the institutional predictors are significant for both violent and property crime. In the cross-sectional models (Table 2), the number of alcohol outlet employees has a significant and positive association with violent and property crime. In addition, the models suggest that alcohol outlets have a spatial effect on crime; the spatially lagged measure of the number of alcohol outlet employees is significantly and positively associated with violent crime, whereas this measure is marginally significant (p = .053) with respect to property crime. The cross-sectional models also demonstrate that the number of banking establishment employees exerts a significant positive effect on crime. However, the spatially lagged measure of banking establishments has a significant and negative effect on property crime, and has no effect on violent crime. The models also show that the independent effects of third places and civic and social organizations operate similarly: these two institutions have a negative effect on the estimated outcomes, whereas the spatially lagged measure indicates a positive effect on crime.
Results From Negative Binomial Regression: Cross-Sectional and Longitudinal Models.
Note. Institutional variables are lagged by 1 year in the longitudinal models. Dummy variables for cities and years are included in the models, but not shown. Number of cases = 6,327; number of tracts = 1,024.
p < .10. *p < .05. **p < .01.
Results: Longitudinal Models
The longitudinal model features both the crime rate and institutional predictors of the previous year (Table 2). The question that emerges is whether the significant institutional predictors of the cross-sectional models remain in the longitudinal models?
The results confirm the expectation that crime of the previous year exerts a strong significant influence on crime of the current year. The coefficients estimating both violent and property crime are greater than .80, and as a result, not surprisingly, the magnitude of the other coefficients have declined in comparison with the cross-sectional models.
For the institutional predictors, there are substantive differences in findings between the cross-sectional and longitudinal models. First, we find that the number of alcohol employees has a crime-producing influence for violent, but not for property crime. The models also show that the spatially lagged measure does not have a statistical relationship with crime. These findings of the longitudinal models are substantively different in comparison with the cross-sectional models, suggesting that the cross-sectional model overestimates the size of these effects. The longitudinal models provide evidence that alcohol outlets do in fact have a crime-producing influence in relation to violent crime, yet there is no evidence that this type of institution influences property crime or that they have a spatial impact on any of the outcomes.
Second, the number of banking establishment employees exhibits a significant and positive influence on property crime, and a similar influence with violent crime, albeit a marginally significant one (p < .10). We also find no evidence of any spatial impact on crime. When comparing the cross-sectional models with the longitudinal models, the spatial impact of banking establishments on property crime is no longer significant in the longitudinal analysis. Furthermore, in the longitudinal model, the within-tract banking measure is no longer statistically significant for violent crime. In terms of banking establishments, the longitudinal analysis demonstrates that such establishments have a crime-producing influence on property crime, and a marginally similar influence on violent crime.
Third, the longitudinal models suggest that the number of third place employees is predictive of lower crime rates; the within-tract measure is significant for property crime and marginally significant for violent crime. The spatially lagged measure of third places is found to have a significant and positive impact on property crime, and no impact on violent crime. When comparing the findings of the cross-sectional models with the longitudinal models, the within-tract measure of third places is now marginally statistically significant with violent crime, whereas the spatially lagged measure no longer exerts a significant positive impact on violent crime. Nevertheless, the longitudinal analysis still lends support for conceptualizing third places as conducive to crime reduction in the focal tract.
Last, in the longitudinal models, the measures for the number of civic and social organization employees predominantly fail to show a significant influence on neighborhood crime. The lone exception is that the spatially lagged measure has a positive impact on property crime. When assessing the impact of civic and social organizations, the potential differences between cross-sectional and longitudinal analysis are emphasized, as the cross-sectional models communicate a message of crime control benefits (within-tract and spatial lag) whereas the longitudinal models suggest null findings.
We now turn our attention to the results for the neighborhood characteristic variables. In general, the longitudinal results parallel the findings of the cross-sectional models. In the longitudinal models, the percent poverty exhibits a significant and positive influence on violent crime, but no influence on property crime. The spatially lagged measure of poverty indicates a significant and positive impact on the estimated outcomes. Although the findings from the longitudinal models closely align with those of the cross-sectional models, the longitudinal analysis importantly provides evidence that the effects of poverty mainly extend to violent crime.
The average length of residence demonstrates a significant and negative influence on both violent and property crime, whereas the spatially lagged measure has a positive impact on the stated outcomes. These results parallel the findings of the cross-sectional models. For percent immigrants, the within-tract measure suggests a negative influence on neighborhood crime; the coefficient for property crime is significant and the coefficient for violent crime is marginally significant (p < .10). The spatially lagged measure indicates a significant and positive impact on violent crime, but no impact on property crime. Also, the race measures show some significant and positive influences on violent crime (in relation to the percent White), whereas we predominantly find non-significant influences in terms of property crime.
Sensitivity Analyses
When evaluating the institutional effects on crime, an alternative modeling strategy involves estimating models for each type of institution, separately. However, such a strategy downplays the spatial co-location among different types of local institutions. Consequently, as we contend that the spatial co-location among institutions is in fact a fundamental feature of the institutions–crime nexus, we estimate a series of ancillary models using the alternative modeling strategy, and compare the results of the institutional effects on crime (Appendix) with those of our longitudinal models that simultaneously account for the four institutions (Table 2). What we find is that many institutional effects found to be statistically significant by the ancillary models are not supported by the referent models, thereby cautioning the practice of estimating models in which each type of institution is inserted separately.
Finally, we tested for numerous interaction effects; nearly all of these possibilities proved to be statistically non-significant. We also conducted other sensitivity procedures, which ultimately provided evidence that our longitudinal models are robust (Table 2). 13
Discussion
An important question for criminologists is whether local institutions really matter for understanding variation in neighborhood crime? Previous studies have faced difficulty addressing this question, most notably due to concerns over endogeneity, institutional diversity, and generalizability. This study addresses these concerns by performing a longitudinal analysis using four types of local institutions in tracts from multiple U.S. cities. Although we conclude that local institutions have meaningful influences on crime outcomes, the differences in findings between our cross-sectional and longitudinal models suggest that previous studies using cross-sectional analysis likely overestimate institutional effects.
In this study, the longitudinal models demonstrate that local institutions have differential effects on crime, operating as crime attractors, effective facilitators of informal social control, or having no relationship with crime. It turns out that the results are not always consistent with prior research. Nevertheless, our findings help to inform a diverse audience.
While public discourse tends to focus on the disinhibiting effects of alcohol in an individual or small group context, our longitudinal models show that the aggregate effects of alcohol outlets are equally problematic in relation to violent crime rates. In particular, we conclude that such outlets increase the probability for the convergence in time and space of motivated offenders, suitable targets, and the absence of capable guardianship (Cohen & Felson, 1979; Felson, 1987; Felson & Boba, 2010). It follows that violent crimes will cluster at the location of alcohol outlets as well as the places surrounding these locations (Blose & Holder, 1987; Parker & Rebhun, 1995; Pridemore & Grubesic, 2013; Sherman et al., 1989). In light of our findings, we recommend that law enforcement consider increasing the number of patrol officers and the level of surveillance in areas in which many alcohol outlets are located. Similarly, policy makers may want to consider the enactment of zoning laws that limit the number (or density) of new alcohol outlets.
The relationship between banking establishments and neighborhood crime has not received much scholarly attention. In this study, we find that traditional banking establishments have a crime-producing influence in terms of property crime, and that they have a modest influence on violent crime. Although the results of the longitudinal models cannot identify the exact mechanisms that underlie the effect of banking establishments, we cautiously propose that such establishments increase criminal opportunities. Specifically, they increase the number of attractive targets in neighborhoods (i.e., people and merchandise). Previous studies have found that various measures of retail and commercial businesses were associated with higher crime rates (Bernasco & Block, 2011; Smith, Frazee, & Davison, 2000; Wilcox, Quisenberry, Cabrera, & Jones, 2004). Banking establishments are similar to retail and commercial businesses in the sense they both have clientele known to carry money and other valuable merchandise.
Consistent with the limited theoretical and empirical research concerning the communal benefits of third places (Oldenburg, 1999; Papachristos et al., 2011), our longitudinal results suggest that third places may precipitate avenues of effective social control, thereby leading to lower crime rates. Specifically, when crime problems emerge, such businesses may become the informal setting in which residents utilize their social ties to disseminate information and mobilize resources. The implication is that although the primary function of these businesses is to generate financial profits, they may also possess a latent function favorable to efficacious neighborhood control and social action. In the future, we hope that more studies implement qualitative methods as a way to elaborate on the protective effects of third places.
The finding that civic and social organizations have no crime-reducing impact is inconsistent with social disorganization theory as well as the findings of some earlier studies (Beyerlein & Hipp, 2005; Lee, 2008; Peterson et al., 2000; Putnam, 2000; Rosenfeld, Messner, & Baumer, 2001). To understand the incongruence between theory and empirical findings, we put forth three possible explanations. First, civic and social organizations influence residents’ perception of neighborhood levels of crime, but not actual crime rates. Second, they attract people with a criminal history (e.g., gang members, parolees, individuals on probation, ex-convicts). Third, previous studies have predominantly performed cross-sectional analysis, thereby overestimating the “true” effect of these organizations. Although the results fail to show any crime control benefits of civic and social organizations, this does not mean that these organizations are fundamentally incapable of mitigating crime within neighborhoods; simply put, these organizations reflected in our data did not affect crime over the course of the study period. As a result, future research needs to further investigate how different community contexts may produce varying institutional effects on crime with respect to civic and social organizations.
This study is not without limitations. First, although the models identify institutional effects on crime, the exact mechanisms that bring about such effects are beyond the scope of this article. As a result, we can only posit that social control and criminal opportunities are relevant here. Second, there are other types of measures that can be used to evaluate the institutions–crime nexus. For example, researchers could create a measure that captures the percentage of a neighborhood’s economic base that is represented by a specific type of local institution. Moreover, studies could also implement measures that capture the density and revenue of particular institutions. Although we believe that implementing such measures will produce similar results to those of the present study, it is important that future scholarship consider the robustness of identified effects. Third, it is challenging to explain that in some cases the results of the within-tract measure differ in comparison with the spatially lagged measure. We suspect that in some instances in which there is high neighborhood control in the focal neighborhood, offenders may choose to offend in adjacent or surrounding neighborhoods. As a result, there is a negative effect for the within-tract measure and a positive effect for the spatially lagged measure. Using the same logic, we would obtain the opposite findings for the within-tract and spatially lagged measures in some instances in which there is low neighborhood control in the focal neighborhood. Lastly, this study does not assess possible differences between tracts from different cities. An obvious extension is for future studies to use random effects models with a larger sample of cities.
Conclusion
Although previous studies have provided valuable insight into the institutions–crime nexus, such studies have almost exclusively examined the effect of one type of local institution using cross-sectional analysis. This not only ignores the wide diversity of institutions, but also the likely endogeneity in the placement of local institutions in relation to neighborhood crime. This study contributes to the existing research by investigating how four types of local institutions longitudinally impact crime rates. We find evidence that local institutions can elevate both social control and criminal opportunities in neighborhoods. The results from the longitudinal models show that alcohol outlets and banking establishments act as crime attractors, whereas third places may operate as a crucial setting for the formulation and enhancement of informal social control. Equally important, we compare these findings with those of our cross-sectional models. The cross-sectional models not only show a greater number of significant institutional effects, but that these effects are stronger in magnitude. Such striking differences underscore the necessity of applying longitudinal analysis to the institutions–crime nexus.
Footnotes
Appendix
Estimating Crime Using the Variables for Each Type of Institution, Separately
| Violent |
Property |
Violent |
Property |
Violent |
Property |
Violent |
Property |
|
|---|---|---|---|---|---|---|---|---|
| B (SE) | B (SE) | B (SE) | B (SE) | B (SE) | B (SE) | B (SE) | B (SE) | |
| Violent crime rate of previous year | 0.8694** | 0.8760** | 0.8715** | 0.8730** | ||||
| (0.0118) | (0.0113) | (0.0118) | (0.0115) | |||||
| Property crime rate of previous year | 0.8832** | 0.8834** | 0.8807** | 0.8788** | ||||
| (0.0104) | (0.0099) | (0.0104) | (0.0105) | |||||
| Alcohol employees | 0.0328** | 0.0033 | ||||||
| (0.0097) | (0.0067) | |||||||
| Spatial lag: Alcohol employees | 0.0534** | 0.0248* | ||||||
| (0.0151) | (0.0100) | |||||||
| Banking employees | 0.0156** | 0.0109* | ||||||
| (0.0055) | (0.0043) | |||||||
| Spatial lag: Banking employees | 0.0121 | 0.0110* | ||||||
| (0.0077) | (0.0054) | |||||||
| Third place employees | 0.0107 † | −0.0006 | ||||||
| (0.0065) | (0.0048) | |||||||
| Spatial lag: Third place employees | 0.0312** | 0.0261** | ||||||
| (0.0106) | (0.0074) | |||||||
| Civic and social employees | 0.0281** | 0.0085 | ||||||
| (0.0104) | (0.0077) | |||||||
| Spatial lag: Civic and social employees | 0.0382* | 0.0392** | ||||||
| (0.0155) | (0.0110) | |||||||
| Percent poverty | 0.0021** | 0.0005 | 0.0021** | 0.0005 | 0.0019** | 0.0004 | 0.0020** | 0.0005 |
| (0.0005) | (0.0004) | (0.0005) | (0.0004) | (0.0005) | (0.0004) | (0.0005) | (0.0004) | |
| Spatial lag: Percent poverty | 0.0034** | 0.0025** | 0.0044** | 0.0029** | 0.0050** | 0.0031** | 0.0035** | 0.0022** |
| (0.0009) | (0.0007) | (0.0009) | (0.0007) | (0.0010) | (0.0007) | (0.0010) | (0.0007) | |
| Average length of residence | −0.0074** | −0.0048** | −0.0076** | −0.0047** | −0.0081** | −0.0052** | −0.0081** | −0.0052** |
| (0.0021) | (0.0015) | (0.0021) | (0.0015) | (0.0021) | (0.0015) | (0.0021) | (0.0015) | |
| Spatial lag: Average length of residence | 0.0092** | 0.0032 | 0.0075* | 0.0044 † | 0.0093** | 0.0045 † | 0.0069* | 0.0039 |
| (0.0034) | (0.0024) | (0.0035) | (0.0024) | (0.0036) | (0.0025) | (0.0034) | (0.0024) | |
| Percent immigrants | −0.0014 | −0.0020** | −0.0015 | −0.0023** | −0.0017 † | −0.0023** | −0.0013 | −0.0020** |
| (0.0010) | (0.0007) | (0.0010) | (0.0007) | (0.0010) | (0.0007) | (0.0010) | (0.0007) | |
| Spatial lag: Percent immigrants | 0.0033** | 0.0007 | 0.0026** | 0.0005 | 0.0028** | 0.0007 | 0.0029** | 0.0008 |
| (0.0010) | (0.0007) | (0.0010) | (0.0007) | (0.0010) | (0.0007) | (0.0010) | (0.0007) | |
| Percent Black | 0.0019** | 0.0001 | 0.0016** | 0.0001 | 0.0020** | 0.0003 | 0.0017** | 0.0001 |
| (0.0003) | (0.0002) | (0.0003) | (0.0002) | (0.0003) | (0.0002) | (0.0003) | (0.0002) | |
| Percent Latino | 0.0021** | 0.0008 † | 0.0018** | 0.0010* | 0.0023** | 0.0011** | 0.0019** | 0.0009* |
| (0.0006) | (0.0004) | (0.0006) | (0.0004) | (0.0007) | (0.0004) | (0.0006) | (0.0004) | |
| Percent Asian | 0.0002 | 0.0007 | −0.0002 | 0.0008 | 0.0003 | 0.0010 | −0.0003 | 0.0006 |
| (0.0009) | (0.0006) | (0.0009) | (0.0006) | (0.0009) | (0.0007) | (0.0009) | (0.0006) | |
| Percent other race | 0.0030 | 0.0024 | 0.0020 | 0.0024 | 0.0029 | 0.0025 | 0.0022 | 0.0025 |
| (0.0023) | (0.0017) | (0.0023) | (0.0017) | (0.0023) | (0.0017) | (0.0023) | (0.0017) |
Note. Institutional variables are lagged by 1 year. Dummy variables for cities and years are included in the models, but not shown. Number of cases = 6,327; number of tracts = 1,024.
p < .10. *p < .05. **p < .01.
Acknowledgements
The author would like to thank John Hipp, Charis Kubrin, and Ron Huff for reviewing earlier drafts and for providing invaluable feedback.
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) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No. (DGE-0808392).