Ecological Networks and Urban Crime: The Structure of Shared Routine Activity Locations and Neighborhood-Level Informal Control Capacity.

Abstract

Drawing on Jacobs (1961), we hypothesize that public contact among neighborhood residents while engaged in day-to-day routines, captured by the aggregate network structure of shared local exposure, is consequential for crime. Neighborhoods in which residents come into contact more extensively in the course of conventional routines will exhibit higher levels of public familiarity, trust, and collective efficacy with implications for the informal social control of crime. We employ the concept of ecological (“eco-”) networks – networks linking households within neighborhoods through shared activity locations – to formalize the notion of overlapping routines. Using micro-simulations of household travel patterns to construct census tract-level eco-networks for Columbus, OH, we examine the hypothesis that eco-network intensity (the probability that households tied through one location in a neighborhood eco-network will also be tied through another visited location) is negatively associated with tract-level crime rates (N=192). Fitted spatial autoregressive models offer evidence that neighborhoods with higher intensity eco-networks exhibit lower levels of violent and property crime. In contrast, a higher prevalence of non-resident visitors to a given tract is positively associated with property crime. These analyses hold the potential to enrich insight into the ecological processes that shape variation in neighborhood crime.

Spatial characteristics of urban environments have been central to criminological theory since the early Chicago School development of the social disorganization approach to understanding variation in neighborhood crime rates (Shaw and McKay, 1969). This highly influential theoretical model focuses on the role of neighborhood structural disadvantage – in particular poverty, racial/ethnic heterogeneity, and instability of residential tenure – in reducing the capacity of neighborhoods to achieve shared goals, including the informal social control of crime. Empirical research rooted in social disorganization theory has offered robust evidence in support of the basic expectations of the theory over many decades (Kubrin and Weitzer, 2003).

Both the systemic model of community crime (Bursik and Grasmick, 1993) and, more recently, collective efficacy theory (Sampson, 2012), have offered significant refinements to the original social disorganization approach, spurring a period of intensive empirical investigation. The systemic model’s emphasis on the role of neighborhood-based social network ties in generating informal social control capacity focuses inquiry on variation in neighbor network integration as a critical component of crime-regulation capacity (Bellair, 1997; Bursik and Grasmick, 1993). Yet, empirical work rooted in the systemic model has yielded equivocal results, leading to increasing skepticism regarding the benefits of informal social networks in urban communities (Bellair and Browning, 2010; Browning, Feinberg, and Dietz, 2004; Sampson, 2012; Wilson, 1996). Collective efficacy theory offers an important corrective to the systemic model’s optimistic expectations regarding social network integration and crime control, arguing that while network ties are associated with enhanced collective efficacy, they are not a necessary condition for collective efficacy’s emergence and, under some conditions, may actually impede informal social control of crime (Sampson, 2012).

We argue that networks are, indeed, important precursors to the neighborhood capacity for crime control, but that extant approaches have neglected the ecological processes that shape weak forms of location-based interaction and, ultimately, norms regarding the use of public space. We draw on the pioneering work of Jane Jacobs to develop a theoretical approach that integrates insights from social disorganization and collective efficacy theories with a focus on routine activity patterns that promote neighborhood social organization. Specifically, we employ the concept of ecological networks – links between neighborhood residents through shared routine activity locations – to describe variation in the potential for routine activity patterns to yield resident intersection in public space. Dense networks of intersection are hypothesized to increase the likelihood of public contact and familiarity and, in turn, the development of location-based trust and expectations for pro-social action (Browning et al., 2017). Beyond expectations regarding the role of shared routines among neighbors, we also consider the impact of non-residential neighborhood use, examining the hypothesis that the increasing presence of “outsiders” enhances opportunity for crime and diminishes informal social control capacity, with implications for crime rates.

To investigate these hypotheses, we employ Mid-Ohio Regional Planning Commission (MORPC) micro-simulations of 1999 routine travel patterns for the Columbus, OH population in order to facilitate unbiased measurement of tract-level eco-network structural properties. We then examine associations between measures of eco-network structure and geo-coded administrative data on crime in Columbus using the National Neighborhood Crime Study (NNCS; Peterson and Krivo, 2000). We fit spatial autoregressive models examining links between eco-network structure and neighborhood crime rates adjusting for a host of potential confounders based on Census, land use, and prior crime rate data. To our knowledge, these analyses constitute the first investigation of eco-network influences on neighborhood crime rates.

ECOLOGICAL APPROACHES TO NEIGHBORHOOD CRIME

THE SOCIAL DISORGANIZATION TRADITION

Social disorganization theory (Shaw and McKay, 1969) has been a mainstay of macro-level criminological thinking since its emergence in the work of the Chicago School of Sociology. In its original formulation, Shaw and McKay emphasized the structural origins of neighborhood informal social control capacity in poverty, residential instability, and heterogeneity by race and ethnicity. These factors were thought to be associated with limited resources and social fragmentation which combined to obstruct the quest to achieve shared goals, including the promotion of a crime-free environment. The early empirical work of Shaw and McKay offered support for the basic structural hypotheses of the theory, spawning decades of empirical research as well as theoretical elaboration and critique.

The ebb and flow of the theory’s popularity over the course of the 20^th century has been well-documented (Bursik, 1988; Kubrin, 2009; Kubrin and Weitzer, 2003) but the late 1980s and 1990s saw a substantial revitalization of neighborhood research rooted in the disorganization tradition. Research during the 1990s in particular moved beyond an exclusive focus on structural conditions to understanding the social processes that accounted for variation in neighborhood crime rates. Developments in research on the systemic model of community (Bursik and Grasmick, 1993; Kasarda and Janowitz, 1974) focused attention on informal social and voluntary organization ties in generating friendship and acquaintanceship networks thought to anchor neighborhood informal social control capacity.

Enhanced theoretical attention to the social mechanisms responsible for informal social control capacity was followed by a period of intensive empirical investigation of network influences on crime (typically measured by survey reports of the prevalence of friendship and neighboring ties). On balance, however, extant research does not support the claim that neighborhoods characterized by more integrated informal social networks are better able to regulate the prevalence of crime (Bellair, 1997; Bellair and Browning, 2010; Simcha-Fagan and Schwartz, 1986; Warner and Rountree, 1997). Studies based on data from the Project on Human Development in Chicago Neighborhoods (PHDCN) – which remains the most comprehensive study of neighborhood social processes and crime to date – indicate that more extensive friendship and kinship ties (Morenoff, Sampson, and Raudenbush, 2001) and the prevalence of social interaction and reciprocated exchange within neighbor networks (Browning, Feinberg, and Dietz, 2004) are not directly associated with crime rates.

Beyond the relative absence of evidence suggesting that neighborhood social network integration is beneficial for the regulation of crime, some studies have indicated that networks may actually impede informal social control. Wilson’s (1996) notion of “social isolation” points to the role of integrated social networks in disseminating, rather than regulating, crime and related behavioral orientations when network-based interactions are consistently decoupled from employment, education, voluntary organization and other institutional involvements. In what they term a “negotiated coexistence” model of community crime, Browning and colleagues (Browning, 2009; Browning, Feinberg, and Dietz, 2004) argue that the prevalence of social ties in urban communities may result in the increasing integration of community members with residents involved in illicit activities. The social obligations engendered through the interconnection of licit and illicit networks may limit the effectiveness of informal social control norms as residents chose not to sanction those engaged in criminal activity with whom they maintain direct or indirect social ties (Pattillo-McCoy, 1999; Venkatesh, 2000).

The absence of evidence supporting a protective effect of conventionally conceptualized and measured informal social networks on neighborhood crime rates alongside emerging indications that social networks may potentially promote crime poses a significant challenge to the classic systemic model of community crime. Moreover, extant research leaves open the question of what proximate neighborhood-level ecological and social processes (beyond social structural and compositional background factors such as segregation and concentrated poverty) account for variability in neighborhood informal social control capacity and crime.

Among the more prominent recent alternative approaches to the origins of effective neighborhood regulatory capacity is work examining the organizational base of urban neighborhoods (Allard and Small, 2013; Marwell, 2009; Sampson, 2012). For instance, Small (2009) argues that neighborhood-based organizations shape informal interactional processes that are potentially influential sources of social support. In his analysis of an urban child care center in a disadvantaged neighborhood, Small (2009) found that involvement in the center accrued advantages to neighborhood residents through the development of weak social ties emerging as a byproduct of center activities. In some cases, even very weak familiarity (mothers who were not even on a first name basis) could result in consequential sources of support (e.g., watching a child in an emergency). In this “organizational embeddedness” approach, Small argues that the child care center context shaped mother’s evaluation of relational dynamics with other participants in center activities, resulting in high levels of trust and pro-social expectations among mothers despite, in many cases, quite limited familiarity.

Small’s work highlights the benefits of public contact among neighborhood residents engaged in conventionally-oriented, organizationally-anchored activities. At the neighborhood level, the concatenation of innumerable small-scale trust-engendering interactions across communities is likely to have consequences for the emergence of expectations regarding public space use and the prevalence of crime. Yet, research in the social disorganization tradition has largely neglected the role of weak forms of routine activity-based public contact as a precursor to effective informal control norms. In contrast, the work of Jane Jacobs offers an important foundation for understanding the social ecological origins of neighborhood self-regulatory capacity.

PUBLIC CONTACT AND THE INFORMAL SOCIAL CONTROL OF CRIME: JANE JACOBS AND THE SOCIAL ECOLOGICAL TRADITION

In The Death and Life of Great American Cities (1961), Jacobs argued that many lower income, seemingly disordered neighborhoods actually harbor a complex ecological dynamic that fosters place-based monitoring, trust, and a willingness to share responsibility for achieving safe streets. Jacobs is most frequently associated with her emphasis on the beneficial supervisory role of “eyes on the street.” Neighborhoods that encourage a steady stream of people in public space engaged in work, shopping, leisure and other everyday routine activity naturally promote street monitoring and reduced crime. The eyes-on-the-street hypothesis has generated a substantial body of research focused on testing the link between aspects of the built environment thought to promote street activity and the prevalence of crime (Browning and Jackson, 2013). Jacobs’ work also informed Newman’s “defensible space” thesis and associated research on environmental design approaches that encourage natural surveillance (Cozens, 2008; Jeffery, 1977; Newman, 1972).

In Jacobs’ view, well-functioning neighborhoods also foster public contact among residents – a less frequently considered aspect of her social ecological model. The availability of proximate routine activity destinations brings residents together both in organizational settings as well as in surrounding public spaces, resulting in weak forms of mutual acknowledgment and superficial interaction. Cumulatively, these casual encounters – rooted in structured activities such as those required by everyday employment and family maintenance needs – yield public familiarity sufficient to engender place-based trust. Indeed, research on the social psychology of trust corroborates Jacobs’ expectation, demonstrating a powerful link between familiarity and social trust (Lawler, 2001; Lawler and Yoon, 1993). As patterns of casual encounter become regularized across a set of neighborhood-based routine activity locations – potentially linking residents at multiple places – a “web of public trust” emerges (Jacobs, 1961: 56). In a conceptual linkage that anticipates collective efficacy theory, Jacobs points to the role of trust as a foundation for the emergence of expectations for public support “when the chips are down” (Jacobs, 1961: 56) – e.g., in the event of a victimization threat.

Jacobs’ model offers an ecologically-grounded approach to understanding the establishment of neighborhood-level collective efficacy with respect to the control of public space. As residents are increasingly linked through local routine activity destinations, the potential for emergent collective efficacy – characterizing the neighborhood-anchored network of intersecting people and places as a whole – also increases. Departing from the systemic emphasis on spatially abstracted social ties, this approach incorporates ecological dynamics of resident activity directly into the process by which informal social control capacity is developed and reinforced in urban areas.

Formalizing Jacobs’ “Web of Public Trust”

Ecological (“eco-”) networks describe the structure of potential intersection among residents through locations. Conventional social network approaches are typically comprised only of individuals and the ties among them. In contrast, ecological networks assign ties between resident dyads based on a common routine activity location. In this sense, eco-network ties capture interaction potential. Shared public space heightens the probability of eventual public contact and familiarity. In the aggregate, denser ecological network structures among neighborhood residents and their activity locations are hypothesized to increase trust and informal social control expectations.

Ecological networks may be characterized by a variety of structural configurations. Here we emphasize eco-network structures that align with Jacobs’ expectations regarding the emergence of public trust. Specifically, eco-networks characterized by links among residents through multiple locations are likely to maintain the structural components that spread trust. Residents who encounter one another regularly through a single routine activity are likely to develop familiarity. However, routine contacts through multiple locations – e.g., a shared shopping location and local school – may further enhance familiarity and trust. In turn, both locations benefit from the familiarity-induced trust that emerges. The tendency of households to share routines through more than one location captures the intensity of contact potential between households in the eco-network. Eco-networks with higher levels of this network structural feature are expected to more efficiently disseminate trust, informal social control norms, and crime-control capacity throughout the network of people and places.

Extant research on the consequences of variation in eco-network structure points to the benefits of higher levels of eco-network intensity both for key features of neighborhood social organization and youth behavior. Using data from Los Angeles, Browning and colleagues (2017) found evidence of a beneficial effect of eco-network intensity on changes in measures of collective efficacy, intergenerational closure, and reciprocated exchange among neighbors across two waves. Browning, Soller, and Jackson (2015) found evidence of a protective effect of eco-network intensity on the occurrence of individual-level delinquency and substance use. To date, however, no study has examined eco-network effects on neighborhood crime rates.

Non-Residential Neighborhood Activity and Crime

Recognizing the potential impact of residential ecological dynamics draws attention to the possible role of nonresidential activity patterns in shaping neighborhood self-regulatory capacity. The presence of nonresidents as a criminogenic influence has long been acknowledged in community crime models (Taylor, 1988), crime pattern theory (Brantingham and Brantingham, 1995; Brantingham and Brantingham, 1999), and routine activity approaches (Felson, 1994; Felson and Boivin, 2015). Taylor’s (1988) expectations regarding the impact of outsider presence on neighborhood social control capacity points to the criminogenic effects of nonresident presence as well as the potential for moderating effects on the protective influence of public contact patterns. In this view, the presence of outsiders – often associated with higher concentrations of commercial activity – enhances opportunities for crime by increasing the number of potential victimization targets and leads to withdrawal among neighborhood residents from public space and “territorial” obligations, limiting their informal social control capacity. In addition, outsiders may also increase the density of activity in public space, complicating the translation of public contact among residents into familiarity and trust. Visitors may also be less likely to engage in multiple routine activities in a destination neighborhood, reducing their potential to promote familiarity and trust. Consistent with this hypothesis, Felson and Boivin (2015) found that in-flows of visitors to locations - measured using a large-scale transportation survey data from a Canadian city – were positively associated with both violent and property crime rates.

In summary, we test the hypothesis that increases in the tendency of neighborhood households to share multiple locations of routine activity – or ecological network intensity – will be associated with lower levels of violent and property crime. We also consider whether nonresident activity in a neighborhood increases crime and, finally, whether any observed beneficial effects of eco-network intensity on crime are diminished in the context of higher levels of nonresident activity. We test these hypotheses using data on neighborhood composition, resident travel patterns, and violent and property crime in Columbus, OH.

DATA, MEASURES, AND ANALYTIC STRATEGY

DATA

We employ data from the 1999–2001 National Neighborhood Crime Study (NNCS; Peterson and Krivo, 2000), the 1999 Mid-Ohio Regional Planning Commission (MORPC) Travel Demand Forecasting model (PB Consult, 2006), and 2000 Census data to investigate eco-network effects on crime prevalence at the census tract level. The NNCS data include reported crime counts from the Columbus Division of Police for all 200 tracts that are wholly or partly within the city and have a population of at least 300 not more than 50% of which lives in group quarters. There are a total of 199 contiguous census tracts in the city of Columbus that meet the above criteria, a necessary condition for the spatial regression analysis. Among the remaining tracts, four are dropped from our analysis due to missing land use estimates in the county parcel data and an additional three tracts are omitted as they do not contain a majority of the area of any traffic analysis zones (TAZ; described below). Thus, the final sample for the analyses includes 192 census tracts.

Micro-Simulation Data on Routine Activity Destinations

Network structural characteristics, including the measure of eco-network intensity of potential contact we use in the current analysis, are sensitive to the size of the network. As a result, estimating network structural characteristics based on a sample of households may result in biased estimates of eco-network intensity. Although Browning et al. (2017) found that measures of the intensity of household intersection in an eco-network constructed from samples of tract-level populations were not subject to substantial bias, their simulations assumed a threshold level of within-tract sampling that is not often achieved in standard travel surveys. Consequently, we chose to employ the population-level simulated data in order to ensure accurate representation of eco-network structures at the tract level.

Eco-network structural characteristics are constructed from the output of the MORPC activity- and tour-based micro-simulation travel demand model (PB Consult, 2006), henceforth referred to as the “MORPC travel simulator.” The MORPC travel simulator yields synthetic data for the Columbus, OH, population on mobility patterns over the course of a day. The MORPC travel simulator consists of series of sophisticated probabilistic models (described below) informed by a range of rich data sources, including the 1999 Mid-Ohio Area Household Travel survey – conducted February through June 1999 to study transportation activity and travel plans within the jurisdictions of MORPC. In the survey data, complete diary records of daily weekday travel over a 24-hour period are available for a probability sample of 5,555 households (75.8% of recruited), of which 2,705 households reside in the city of Columbus, OH. Travel diaries were completed on randomly assigned days of the week, representing 13,524 persons, 10,488 vehicles and 52,031 trips. Individual-level demographic information is available for household members, including age, gender, race education, occupation, work/school status, and relationship to the household head, as well as household-level income. These data were used to tune and validate the MORPC travel simulator.

The MORPC travel simulator is an activity- and tour-based microsimulation approach to travel demand modeling. The simulator has three essential characteristics that offer advantages in achieving our key objective of deriving valid eco-network structural measures for inclusion in spatial regression models of tract-level crime rates. First, an activity-based approach takes into account travel decisions based on behaviorally realistic daily activity patterns. Second, the simulator employs the “tour” as the basic unit of analysis – where a tour constitutes a complete sequence of trips beginning and ending at home or at the workplace. This approach departs from the previously dominant individual trip-based approach that neglects the dependence of trips within larger sequences. Third, a micro-simulation approach models activity and travel decisions at the disaggregated household and person level, allowing for construction of household-level travel profiles required to capture intersection of individual households at routine activity locations (Vovsha, Bradley, and Bowman, 2005).

A necessary preliminary step in the micro-simulation process is the construction of a “synthetic” Columbus population on which to run the simulator to produce household-level travel patterns. Using total population, number of households, total labor force size, and average household income at the Traffic Analysis Zone level (TAZ; described below) as a basis, a hypothetical population of households was generated consistent with the calculated distribution of households in each area with respect to income, household size, and number of workers of any status. Each synthetic household was then randomly matched to an actual household in the American Community Survey (ACS) Public Use Microdata Samples (PUMS) from the same area of the city that shares the same income, household size, and number of workers. The synthetic household was then assigned the number of full- and part-time workers, non-working adults, university students, and children in three age categories of the matched household, which are used as inputs to the simulator.

MORPC’s travel simulator employs the TAZ as a geographic basis for traffic flow and transportation forecasting. TAZs are designed to follow Census block group, ecological, and political boundaries where possible; hold a roughly similar magnitude of trips across zones; and have relatively homogenous development density and land use within each zone. There are 1,028 TAZs within Franklin County, Ohio (in which Columbus is predominantly located), which have an average area of 0.529 square miles, (sd=0.60), and an average total population of 1,115 people (sd=1,264). Of those, 648 TAZs fall within Columbus city, with an average size of 0.378 square miles (sd=.37) and an average population of 1,199 (sd=1,369). By comparison, the 882 block groups in Franklin County average 1.05 square miles. Each Census tract contains about 4 TAZs on average.

Once a synthetic population was constructed and TAZ boundaries delimited, the simulator used a sequence of eight core components, or submodels, to generate travel patterns for each synthetic household: 1) a preliminary household car ownership model (given the centrality of this travel mode for subsequent travel decisions); 2) individual daily activity pattern type and number of “mandatory” tours (e.g., for work or school, around which non-mandatory activities are organized); 3) household non-mandatory “joint” (multi-person) tours for maintenance activities (e.g., shopping, eating out); 4) household individual tours for maintenance and discretionary activities (e.g., at work sub-tours); 5) tour-level destination choice; 6) tour-level time of day choice; 7) tour-level travel mode choice; 8) tour-level frequency and location of stops. Each of the submodels were run in sequence allowing for intermediate outcome data (e.g., household car ownership) from previous submodels to serve as inputs for subsequent submodels (e.g. household travel profiles). Each component submodel is comprised of a discrete choice model² that probabilistically assigns the set of possible outcomes, from which realistic travel profiles for households in the synthetic population are ultimately simulated. This process yields a dataset of 1,322,347 individuals within 644,488 households with an associated 2,378,573 trips.

The resulting population-level simulated travel data were validated for consistency and adjusted using multiple sources of corroborating information. The submodels were initially tuned and validated on the Household Travel Survey. The full simulator was then applied to the synthetic population data and validated against the Household Travel Survey, data on year 2000 annual average daily traffic counts, and Central Ohio Transit Authority ride-check weekday transit boardings. Based on model validation results, some adjustments were made to the destination choice and travel mode submodels to ensure that the simulated data aligned with all three validation data sources.

We employ the simulated data at the household level to generate estimates of eco-network structure for Columbus census tracts. Households are assigned to TAZs based on residence. We then assign each TAZ to the tract in which the largest portion of land area falls, thus assigning households to a neighborhood census tract. Within each census tract, the eco-network is constructed by connecting the synthetic households assigned to the tract to the set of TAZs that contain the destinations of household members’ simulated travel paths. By employing the TAZ as the unit of analysis for destinations, we capture shared location at a relatively large scale. We emphasize, however, that eco-network structures constructed from data on shared spatial affiliations at the TAZ level are intended to capture meaningful aggregate variation in the structure of casual interaction potential among neighborhood residents (Browning et al., 2017). In this approach, neighborhoods with residents exhibiting low probabilities of sharing TAZ locations will have lower public interaction potential and diminished likelihood of maintaining beneficial social organizational features relevant to crime control.

MEASURES

Table 1 reports descriptive statistics for variables in the analysis. Dependent variables include indices of violent and property crime from the NNCS data. Violent crime is the rate per 1000 population of reported incidents of aggravated assault, rape, murder, and robbery averaged across the three year (1999–2001) period. Property crime rates include instances of burglary, larceny-theft, and motor vehicle theft. Both outcomes are logged in the spatial regression analyses. Prior crime rates are constructed from 1989–1991 crime data on the same tract boundaries, obtained directly from the Columbus Division of Police. We impute 1989–1991 crime rates for 7 census tracts that are missing on this covariate.

Table 1.

Census Tract Descriptive Statistics for Variables in the Analysis (N=192)

	Mean	Standard Deviation
Dependent Variables
1999–2001 violent crime rate	9.936	10.626
1999–2001 property crime rate	88.389	74.523
Lagged Dependent Variables
1989–1991 violent crime rate	10.680	11.666
1989–1991 property crime rate	90.035	60.220
Control Variables
Socioeconomic disadvantage	0.000	0.351
Proportion married couple families	0.631	0.180
Residential instability	–0.030	0.764
Percent young males ages 15–24	7.971	5.946
Percent new immigrants	4.034	5.354
Racial/ethnic heterogeneity	0.356	0.149
Log population density	8.145	0.940
Percent Black	28.204	28.520
Percent commercial area	20.341	15.307
Number of TAZ per tract	3.906	4.985
Variables Constructed from Simulated Travel Data
Log number of synthetic households	7.289	0.606
Number of destinations (simulated)	793.510	131.912
Median trip distance to TAZ destination, miles	4.231	0.668
Log global attractiveness	–4.812	0.767
Log eco-network intensity	–4.087	0.349
Intensity numerator: 4* Number of 4 cycles	34,886	97,118
Intensity denominator: Number of 3 paths	1,403,439	3,610,930

ABBREVIATIONS: TAZ= Traffic Analysis Zone

Sources: National Neighborhood Crime Study (NNCS), U.S. Census Bureau, and the Mid-Ohio Regional Planning Commission (MORPC) Travel Simulator

Independent Measures

Among households that reside in the same neighborhood, eco-network intensity of potential contact captures the tendency for two households who encounter one another at one routine activity location to also share another location. The clustering coefficient I (Robins and Alexander, 2004) formally captures this tendency within ecological networks. The coefficient I takes as its denominator the number of 3-paths (labeled L₃), that occur in a two-mode network (networks that contain two sets of nodes – in the eco-network case, people and places – and ties are only established between nodes belonging to different sets). The numerator (4×C₄) represents four times the number of 3-paths that are closed by being part of a 4-cycle (C₄), or a relation where two households are connected through two distinct activity clusters. C₄ is multiplied by 4 because every C₄ configuration contains four L₃ configurations. Calculated for a single neighborhood, j, I_j measures the overall tendencies for 3-paths to become closed 4-cycles and is formally defined as:

$$I_j=\frac{4\times C_{4_j}}{L_{3j}},$$

where L_3j is the number of 3-paths in the jth neighborhood’s eco-network and C_4j is the number of 3-paths that are closed. This measure varies between 0 and 1, with 1 indicating that all 3-paths in the network are closed (i.e., are 4-cycles).

In Figure 1, we illustrate the construction of the networks from the simulated travel data and provide a visual representation of network intensity. In Panel A, we depict two households, A and B, located in census tract #1200 (tracts are numbered 0100, 0200, … 1200 counterclockwise from the top left). For example, both households have a simulated trip to TAZ #1204 which lies within their home census tract. The resolution of the simulated trip data is at the level of the TAZ, so that households A and B are connected through TAZ #1204 even though they may not have visited exactly the same address within that TAZ. If Household A also visits TAZ #301 (hypothetical visit indicated by the dashed line), the two households would be connected through both TAZ #1204 and TAZ #0301.

Figure 1.

Panel A: Travel Destinations for Two Hypothetical Panel B: A Closed Three-Path

Households Residing in the Same Tract Linking Households A and B

In the example represented in Figure 1 Panel B, we construct the network for all households in this tract, by connecting the households (square nodes) to the TAZs, (circle nodes) which are destinations of simulated trips for these households. A 3-path, a sequence of three ties (or edges) which pass through distinct households (or vertices), connects household A to TAZ #1204 to household B to TAZ #0301. If household A also visits TAZ #0301 (indicated by a dashed line in the figure), this 3-path becomes a 4-cycle (a cycle is a path which begins and ends at the same vertex). Note that a 4-cycle is the smallest possible cycle in a two-mode network, since vertices of the same type cannot connect.

The average eco-network intensity characterizing Columbus census tracts is .02 (sd=.007) (see Table 1 for descriptive statistics on the numerator and denominator of the intensity measure). Although the proportion of all possible three-paths that are closed (4 cycles) is low, variability in this measure is nevertheless hypothesized to gauge consequential differences in the extent to which residents encounter neighbors at multiple routine activity locations. To measure nonresident activity within the focal neighborhood, we construct a measure of global attractiveness, defined as the percent of households that live outside the focal tract that visit that tract. The average global attractiveness of Columbus tracts is .01 (sd=.01). In order to adjust for skewness in these measures, we take the natural logarithm of both measures.

A number of control variables are included in the models based on prior theory and research on the structural antecedents of crime (Land, McCall, and Cohen, 1990; Morenoff, Sampson, and Raudenbush, 2001). Socioeconomic disadvantage combines measures of the percent of the population age 16–64 who are unemployed or out of the labor force (joblessness), percent of employed persons age 16 and over who are working in professional or managerial occupations (reverse coded), percent of the population age 25 and over who are college graduates (reverse coded), percent of households who are female-headed families, percent of the employed civilian population age 16 and over who work in the six occupations with the lowest average incomes (secondary sector workers), and percent of the population below the poverty line (α=.93). The proportion married families is the proportion of family households in the tract that are comprised of a married couple (compared to single male- or female-headed family households). Residential instability combines measures of the percent of occupied housing units that are renter occupied, percent of housing that is vacant, and the percent of residents age 5 or over who lived in a different dwelling in 1995 (α =.69). Percent young males is the percent of the population that is male and aged 15–24. Percent new immigrants is the percentage of the total population that is foreignborn and entered the United States in 1990 or later. Racial/ethnic heterogeneity is the sum of the squared proportions of non-Hispanic Whites, non-Hispanic Blacks, non-Hispanic American Indians and Alaska natives, non-Hispanic Asians, Native Hawaiians, or other Pacific Islanders, non-Hispanics of some Other Race or two or more races, and Hispanics or Latinos. The population density of the tract is the total population of the tract divided by the tract area in square miles (logged). We also include a measure of commercial activity in the tract based on the percentage of the tract area that is dedicated to commercial land uses (logged percent commercial area). These data are based on the Franklin County Auditor’s Office tax parcel data. Based on prior findings indicating that commercial activity in urban areas may exhibit a curvilinear association with crime (Browning et al., 2010), we include both linear and quadratic terms for this variable.

Comparing Networks of Different Sizes

The generation of ecological network measures across multiple census tracts for the purposes of estimating network effects on crime introduces complexities associated with network comparison (Smith, Calder, and Browning, 2016). Estimates of eco-network intensity are confounded with network size and density, possibly biasing the effects of intensity in its original metric. In order to address this concern, we include structural measures that may influence both values of ecological network intensity as well as crime rates in baseline models. These include the total number of households in the neighborhood based on the synthetic population data (log number of synthetic households), the number of unique TAZ destinations (logged) visited by tract residents based on the simulated data (number of destinations), and the number of TAZs in each tract (logged number of TAZs).

We also constructed measures of eco-network intensity based on the deviation of the raw measure from what would be expected based on the Erdos-Renyi random graph with the same number of individuals, TAZs, and ties in the network. Specifically, we subtracted expected intensity under the Erdos-Renyi model (estimated based on 1,000 networks simulated from the model) from the observed intensity and divided by the standard deviation of the intensities derived from the simulated networks. Analytic models substituting the standardized eco-network intensity measure for the unadjusted values of the measure are also presented to corroborate findings based on the latter.

ANALYTIC STRATEGY

Evidence suggests that crime rates in urban areas exhibit non-trivial spatial dependence that may bias both coefficient estimates and standard errors. In order to account for the spatial clustering of crime rates, we estimate mixed regressive - spatial autoregressive models (Anselin, 1988) given by:

$$\mathbf{y}=\rho \mathbf{W}\mathbf{y}+\beta \mathbf{X}+\epsilon \epsilon ~\mathrm{N}(0,{\sigma }^{2}I)$$

where y is an N × 1 vector of observations on the outcome measure, W is a first order contiguity neighborhood matrix³, X is an N × K matrix of independent variables, β is a K × 1 vector of regression coefficients, ρ is the spatial lag operator, and ε is a vector of error terms that are assumed to be independent and identically distributed with mean 0 and constant variance, σ². We present the results of spatial lag models below.⁴

We include a range of control variables, including lagged dependent variables, in our primary models of violent and property crime. Key issues influencing the selection of control variables include the potential for endogeneity due to the influence of prior crime on the tendency to withdraw from shared public space (influencing eco-network structure) and the concern that models of travel destination choice incorporate TAZ level covariates such as socioeconomic status and land use covariates⁵ that may bias estimates of eco-network effects if not effectively controlled in models of crime. Appendix Table 1 presents correlation coefficients for variables included in the analysis. Correlations of eco-network intensity with structural variables demonstrate relatively strong positive correlations with proportion married, population size, and number of destinations; the latter two are expected based on the behavior of eco-network intensity estimates as network size and density increases (Smith, Calder, and Browning, 2016). In contrast, intensity is negatively correlated with tract disadvantage, tract population density, percent Black, and all measures of crime. Figures 2–4 present maps of eco-network intensity, 2000 violent crime, and 2000 property crime for Columbus census tracts. This inverse bivariate relationship between intensity and both measures of crime is visible by corresponding darker shading representing areas of the city with lower levels of eco-network connections and higher levels of violent and property crime.

Table A1.

Correlation Matrix of Eco-network structural characteristics and Census tract level controls (N=192)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)	(16)	(17)	(18)	(19)
(1) Log 1999–2001 violent crime rate	1.00
(2) Log 1999–2001 property crime rate	.86	1.00
(3) Log 1989–1991 violent crime rate	.84	.75	1.00
(4) Log 1989–1991 property crime rate	.69	.81	.81	1.00
(5) Socioeconomic disadvantage	.55	.43	.39	.62	1.00
(6) Proportion married couple families	−.72	−.55	−.52	−.75	−.79	1.00
(7) Residential instability	.43	.51	.52	.43	.48	−.52	1.00
(8) Percent young males ages 15–24	.17	.24	.26	.13	.31	−.23	.44	1.00
(9) Percent new immigrants	−.01	−.01	.04	−.01	.16	−.10	.32	.32	1.00
(10) Racial/ethnic heterogeneity	.39	.30	.26	.31	.16	−.39	.31	.13	.37	1.00
(11) Log population density	.37	.19	.22	.37	.36	−.38	.11	.28	.07	.15	1.00
(12) Percent black	.48	.30	.25	.60	.63	−.74	.16	−.12	−.06	.29	.14	1.00
(13) Log percent commercial area	.34	.37	.28	.13	.15	−.26	.53	.31	.37	.34	.09	−.03	1.00
(14) Log percent commercial area2	−.37	−.22	−.12	−.09	−.14	.18	−.05	−.05	.02	−.07	−.22	.04	−.60	1.00
(15) Log number synthetic households	−.51	−.38	−.36	−.55	−.39	.51	−.11	.00	.04	−.18	−.37	−.45	.05	.02	1.00
(16) Number of destinations	−.43	−.35	−.32	−.44	−.31	.42	−.14	−.01	−.06	−.21	−.38	−.34	−.05	.02	.95	1.00
(17) Log number of TAZ	−.02	.18	.15	−.05	.01	.11	.25	−.01	−.05	−.10	−.55	−.14	.17	−.09	.51	.50	1.00
(18) Log eco-network intensity	−.57	−.55	−.46	−.61	−.40	.39	−.15	.08	.05	−.03	−.32	−.37	−.09	.20	.49	.47	.05	1.00
(19) Log global attractiveness	−.11	.16	.10	−.17	−.10	.26	.18	.07	.11	−.05	−.44	−.29	.28	−.06	.60	.50	.73	.15	1.00

ABBREVIATIONS: TAZ= Traffic analysis zones.

Sources: National Neighborhood Crime Study (NNCS), U.S. Census Bureau, and the Mid-Ohio Regional Planning Commission (MORPC) Travel Demand Model

Figure 2.

Quintiles of Intensity of Eco-Network Connections between Households in Columbus Census Tracts

Source: authors’ calculations based on the Mid-Ohio Regional Planning Commission (MORPC) Travel Simulator Simulated Data. Note: Tracts in the city of Columbus which do not contain the largest portion of any Traffic Analysis Zones (TAZ) are dotted to represent missing. White space represents area that is not part of the city of Columbus.

Figure 4.

Quintiles of Logged Property Crime Rate in Columbus Census tracts, 1999–2001

Source: The National Neighborhood Crime Study (NNCS). Note: Census tracts in the city of Columbus with a population smaller than 300 or more than 50% group quarters residents are excluded from the NNCS data, and are dotted to represent missing. White space represents area that is not part of the city of Columbus.

RESULTS

Table 2 presents results of spatial regression models of 1999–2001 log violent crime rates. We begin with preliminary models reporting the effects of eco-network intensity of potential household contact (model 1) and both intensity and global attractiveness (model 2) incorporating all controls except prior crime. In model 1, eco-network intensity is negatively associated with violent crime (p < .001). A one standard deviation increase in intensity leads to a 28 percent decrease in violent crime. The observed effect of eco-network intensity is net of a host of structural and network control variables including significant negative effects of proportion married, percent commercial area (quadratic term), and the log number of synthetic households in the tract. Positive and significant effects are observed for racial/ethnic heterogeneity, the log number of destinations, and the log number of TAZs in the tract.⁶ The significant effect of eco-network intensity is also observed adjusting for the expected statistically significant spatial dependence term (p < .001).

Table 2.

Spatial Lag Regressions of 1999–2001 Log Violent Crime Rate on Eco–Network Structural Characteristics and Controls (N=192)

	Model 1			Model 2			Model 3			Model 4			Model 5
	Coef.		SE	Coef.		SE	Coef.		SE	Coef.		SE	Coef.		SE
Socioeconomic disadvantage	–.271		(.262)	–.477	†	(.260)	–.270		(.226)	–.274		(.226)	–.257		(.227)
Proportion married couple families	–2.350	***	(.644)	–2.528	***	(.626)	–1.484	**	(.555)	–1.510	**	(.558)	–1.465	**	(.560)
Residential instability	.135		(.106)	.193	†	(.104)	.047		(.092)	.050		(.092)	.039		(.092)
Percent young males ages 15–24	.009		(.010)	.007		(.010)	.010		(.008)	.010		(.008)	.010		(.008)
Percent new immigrants	–.016		(.011)	–.014		(.011)	–.019	*	(.009)	–.019	*	(.009)	–.018	†	(.009)
Racial/ethnic heterogeneity	1.688	***	(.384)	1.596	***	(.373)	1.214	***	(.325)	1.208	***	(.328)	1.185	***	(.326)
Log population density	.096		(.078)	.135	†	(.076)	–.014		(.068)	–.016		(.068)	–.008		(.068)
Percent Black	–.003		(.003)	–.002		(.003)	–.003		(.003)	–.003		(.003)	–.003		(.003)
Log percent commercial area	.060		(.091)	–.020		(.091)	.002		(.078)	.000		(.079)	.000		(.079)
Log percent commercial area2	–.078	**	(.026)	–.093	***	(.025)	–.103	***	(.022)	–.103	***	(.022)	–.105	***	(.022)
Log number of TAZ	–1.119	**	(.332)	–1.584	***	(.347)	–.344		(.336)	–.345		(.339)	–.162		(.342)
Log number of synthetic households	.003	*	(.001)	.005	***	(.001)	.001		(.001)	.001		(.001)	.001		(.001)
Number of destinations	.220	*	(.101)	.039		(.110)	–.012		(.095)	–.012		(.095)	–.001		(.095)
Log eco-network intensity	–.955	***	(.173)	–.902	***	(.168)	–.383	*	(.158)	–.381	*	(.161)	—
Log global attractiveness	—			.373	***	(.104)	.159	†	(.093)	.155		(.095)	.155	†	(.094)
Eco-network intensity * global attractiveness	—			—			—			–.030		(.145)	—
Standardized log eco–network intensity	—			—			—			—			–.235	*	(.116)
1989–1991 log violent crime rate	—			—			.590	***	(.072)	.586	***	(.073)	.601	***	(.072)
Intercept	6.823	***	(1.430)	9.088	***	(1.524)	3.463	*	(1.483)	3.498	*	(1.503)	1.924		(1.574)
ρ	.186	**	(.063)	.158	*	(.062)	.043		(.057)	.044		(.057)	.042		(.057)

ABBREVIATIONS: TAZ= Traffic Analysis Zone; Coef.=Coefficient; SE=standard error.

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001 (two–tailed).

Sources: National Neighborhood Crime Study (NNCS), U.S. Census Bureau, and the Mid–Ohio Regional Planning Commission (MORPC) Travel Demand Model

Model 2 adds a measure of global attractiveness to the model, which is positively and significantly associated with violent crime (p < .001), indicating that the popularity of a census tract within an urban area contributes to increased violent crime. For a one standard deviation increase in global attractiveness, violent crime increases by 33 percent. The coefficient for eco-network intensity, however, remains relatively unchanged. Models 1 and 2, however, do not account for the potentially important effect of prior crime rates in residents’ tendencies to share public space. Consequently, model 3 incorporates a lagged measure of 1989–1991 violent crime in order to estimate the effects of eco-network intensity and global attractiveness net of the likely effect of prior crime context on the mobility patterns of neighborhood residents. This approach reduces bias in the estimates of eco-network intensity and global attractiveness; however, we note that the correlation between 1989–1991 and 1999–2001 violent crime is .84, leaving relatively little residual variability to explain. Although the coefficient for the effect of eco-network intensity is diminished in magnitude in model 3, it remains a statistically significant predictor of crime (p < .05); a one standard deviation increase in intensity corresponds to a 13 percent decrease in violent crime in this model. The effect of global attractiveness becomes marginally significant in this model specification (p < .10). The associated increase in violent crime for a one standard deviation increase in global attractiveness is 13 percent.

Model 4 includes the interaction effect between eco-network intensity and global attractiveness in order to consider whether the effects of intensity are attenuated when focal tracts are more frequently visited by outsiders. The interaction term does not achieve significance in model 4, offering no evidence of a moderating effect on intensity. Finally, model 5 is comparable to model 3 but employs a measure of eco-network intensity based on the standardized deviation of the observed level from what would be expected by chance (i.e., under the Erdos-Renyi model). Model 5 indicates that this alternative measure of eco-network intensity is also significantly negatively associated with violent crime, while the global attractiveness measure remains marginally significant (p < .10).⁷

Table 3 reports the results of analyses of tract-level property crime. In model 1, eco-network intensity is a negative and statistically significant (p < .001) predictor of property crime, consistent with the results observed for violent crime. A one standard deviation increase in intensity leads to a 21 percent decrease in property crime. The significant effect of intensity is observed net of comparable structural and eco-network controls to those included in models for violent crime. For property crime, we observe significant negative effects of proportion married, percent new immigrants, percent black, the number of TAZs per tract⁸, and the number of synthetic households in the tract. Positive and significant effects are observed for percent young males, racial and ethnic heterogeneity, and the linear commercial area term. Model 2 introduces global attractiveness which is a positive and significant predictor of property crime (p < .001); a one standard increase in global attractiveness is associated with a 32 percent increase in property crime.

Table 3.

Spatial Lag Regressions of 1999–2001 Log Property Crime Rate on Eco–Network Structural Characteristics and Controls (N=192)

	Model 1			Model 2			Model 3			Model 4			Model 5
	Coef.		SE	Coef.		SE	Coef.		SE	Coef.		SE	Coef.		SE
Socioeconomic disadvantage	–.171		(.153)	–.373	**	(.143)	–.144		(.129)	–.156		(.129)	–.131		(.131)
Proportion married couple families	–1.064	**	(.375)	–1.223	***	(.342)	–.745	*	(.307)	–.794	*	(.307)	–.754	*	(.312)
Residential instability	.055		(.063)	.117	*	(.058)	.033		(.052)	.044		(.052)	.024		(.053)
Percent young males ages 15–24	.014	*	(.006)	.013	*	(.005)	.009	*	(.005)	.008	+	(.005)	.010	*	(.005)
Percent new immigrants	–.019	**	(.006)	–.017	**	(.006)	–.016	**	(.005)	–.016	**	(.005)	–.016	**	(.005)
Racial/ethnic heterogeneity	.699	**	(.225)	.605	**	(.205)	.482	**	(.181)	.438	*	(.182)	.463	*	(.183)
Log population density	–.005		(.045)	.032		(.042)	–.017		(.037)	–.016		(.037)	–.009		(.037)
Percent Black	–.004	*	(.002)	–.003	+	(.002)	–.002		(.002)	–.002		(.002)	–.002		(.002)
Log percent commercial area	.157	**	(.053)	.076		(.050)	.072		(.044)	.074	+	(.044)	.069		(.045)
Log percent commercial area2	.022		(.015)	.007		(.014)	.002		(.012)	.001		(.012)	.001		(.012)
Log number of TAZ	–.415	*	(.196)	–.877	***	(.192)	–.438	*	(.179)	–.466	**	(.179)	–.207		(.188)
Log number of synthetic households	.001		(.001)	.002	**	(.001)	.001		(.001)	.001	+	(.001)	.001	+	(.001)
Number of destinations	.224	***	(.059)	.047		(.060)	.004		(.053)	.002		(.053)	.015		(.054)
Log eco-network intensity	–.682	***	(.102)	–.634	***	(.093)	–.471	***	(.085)	–.443	***	(.086)	—
Log global attractiveness	—			.365	***	(.057)	.236	***	(.053)	.221	***	(.053)	.232	***	(.054)
Eco-network intensity * global attractiveness	—			—			—			–.128		(.080)	—
Standardized log eco–network intensity	—			—			—			—			–.318	***	(.063)
1989–1991 log property crime rate	—			—			.449	***	(.060)	.437	***	(.060)	.456	***	(.060)
Intercept	6.057	***	(.930)	8.470	***	(.932)	4.884	***	(.951)	5.142	***	(.959)	2.862	**	(.993)
ρ	.204	**	(.071)	.151	*	(.067)	.068		(.061)	.056		(.062)	.067		(.062)

ABBREVIATIONS: TAZ= Traffic Analysis Zone; Coef.=Coefficient; SE=standard error.

^†p < .10;

^*p < .05;

^**p < .01;

^***p < .001 (two–tailed).

Sources: National Neighborhood Crime Study (NNCS), U.S. Census Bureau, and the Mid–Ohio Regional Planning Commission (MORPC) Travel Demand Model

Model 3 adjusts for 1989–91 property crime. As with violent crime, the correlation between property crime rates across the decade is high (.81), resulting in limited residual variability to explain.⁹ Although the coefficients for both eco-network intensity and global attractiveness are reduced somewhat in magnitude, the coefficients for both measures remain statistically significant (p < .001). In this model, a one standard deviation increase in intensity corresponds to a 15 percent decrease in property crime; the corresponding increase for a one standard deviation increase in global attractiveness is 20 percent. Introduction of the interaction term between eco-network intensity and global attractiveness in model 4 does not offer evidence of a significant effect, also consistent with the results for violence. Finally, estimates of the effect of the intensity measure deviated from the random eco-network in model 5 corroborate findings from the raw metric models.

ADDITIONAL TESTS

Construction of eco-network measures includes ties in the network that occur due to shared routines both within and outside the residential tract. An advantage of this approach is that residents of neighborhoods may encounter one another at locations near their homes as well as locations that are more distal but may nevertheless still be within the “spatial community” of the neighborhood (Browning et al., 2017). Neighbors who see one another at both types of locations may experience a reinforced sense of familiarity, trust, and shared norms. An alternative hypothesis might expect that neighbors who encounter one another at increasingly distal locations from their residential neighborhood area may experience diminished social organizational benefits from these encounters. In order to examine the robustness of the observed eco-network intensity effects to the distance of network locations from the tract, we constructed a measure of distance (in miles) from residential to destination TAZ centroids for all trips and averaged this measure at the tract level. We then interacted the average distance measure with eco-network intensity and estimated the effect of this interaction term in model 3 of Tables 2 and 3. The interaction term did not achieve significance in either case and the incorporation of average trip distance did not influence the magnitude or significance of the eco-network intensity effects.

CONCLUSION

The current study offers an eco-network model of neighborhood crime rates, integrating classic and contemporary social disorganization approaches (Sampson, Raudenbush, and Earls, 1997; Shaw and McKay, 1969) with the social-ecological approach of Jacobs (Jacobs, 1961). The eco-network approach acknowledges the role of routine mobility patterns in shaping everyday familiarity, public trust, pro-social norms and, ultimately, reduced crime. Our central hypothesis focused on the expected benefits of the intensity of potential contact among households within neighborhood ecological networks – as captured by the tendency for households to share multiple routine activity locations – for the neighborhood-level capacity to control crime (Browning and Soller, 2014).

We constructed measures of census tract-based eco-network structure using simulated population co-location data and investigated the association between eco-network intensity and crime in Columbus, OH. We found evidence of a protective effect of eco-network intensity on both violent and property crime. The consistency of the results points to the importance of eco-network structures for understanding variation in neighborhood crime rates.

The findings have a number of implications for research on neighborhood context and crime. First, the eco-network approach highlights the importance of incorporating shared routine activity patterns into both theory and data collection on neighborhood crime. Daily mobility patterns and activity spaces are a significant feature of a number of well-known models of crime, including routine activity (Felson, 1994), crime pattern (Brantingham and Brantingham, 1981) and Situational Action Theory (Wikström et al., 2012) approaches. However, the integration of emergent ecological structures of shared routines into neighborhood theory has been neglected in extant research. Second, the findings shed light on the ongoing debate regarding the impact of community-level networks on crime (Bursik and Grasmick, 1993). The traditional emphasis on social networks, absent information on spatial processes that bring neighborhood residents into proximity has not yielded significant explanatory advances (Browning, Feinberg, and Dietz, 2004; Morenoff, Sampson, and Raudenbush, 2001). The current analyses, however, indicate that spatially embedded networks may offer greater insight into social organization processes relevant for crime control. Third, identifying protective eco-network structures may translate into tractable policy recommendations to the extent that urban planning decisions regarding the location of amenities and other destinations within neighborhoods have important implications for the routine activity patterns of residents.

We also found that the popularity of the tract as captured by a measure of global attractiveness was positively associated with crime (more consistently in the case of property crime). However, the introduction of global attractiveness did not diminish the association between eco-network intensity and crime, nor was the interaction with intensity a significant predictor of either outcome. The models offer no evidence that the addition of “outsiders” attenuates the benefits of public contact potential for crime regulation. Outsiders may weaken the informal social control capacity of neighborhoods in other respects. However, the positive effect of global attractiveness on crime may also be a function of increased exposure to the risk of victimization.

These analyses are characterized by a number of limitations, particularly with respect to issues of data and measurement. First, it is important to acknowledge potential concerns regarding eco-network estimates based on simulations from a travel demand model. Although the travel simulator is a highly sophisticated effort to infer population mobility patterns from survey and other data resources, we nevertheless cannot claim that our eco-network structural estimates are generated from observed data on the population of Columbus residents. Moreover, the micro-simulation approach employed by MORPC to generate the simulated data yields a single estimated travel profile at the household level. This approach translates the model-based probabilities associated with different possible activity patterns and destinations choices at the household level into the most probable set of choices. The simulated data thus offer outputs equivalent to what one would observe from a travel survey, facilitating the construction of eco-networks from a specific set of routine activity destinations assigned at the household level. Ideally, however, we would base our analyses on a large number of plausible simulations, producing a distribution of possible eco-network solutions and associated parameter estimates for the effects of eco-network intensity on crime. Unfortunately, we are limited to a single, carefully evaluated simulation provided by MORPC.

Second, our eco-network estimation approach defined shared location at the Traffic Analysis Zone level – a relatively large unit for estimating public contact potential. Though this strategy likely biases estimates of eco-network structural effects on crime downward, future data collection efforts will benefit from more precise information on routine activity intersection in space as well as time (e.g., employing mobile technology to track respondents’ travel patterns).

Third, analyses of eco-network effects should be combined with data on other aspects of place-based social climates such as street activity and monitoring; public familiarity; collective efficacy, and social and physical disorder. Data on these dimensions of social climate have been collected at the neighborhood (typically census tract) level, but more fine-grained estimates of these processes at the location level are needed to determine how locations (nodes) in eco-networks vary with respect to important aspects of social organization (St. Jean, 2007). Ultimately, moving toward a conceptualization of “eco-communities” as clusters of people who share similar routine activity locations (and the locations themselves) may better capture consequential contexts for understanding the concentration and diffusion of crime (Browning and Soller, 2014). Finally, analyses of eco-network effects would benefit from data including a larger number of neighborhoods across multiple urban areas.

Jacobs’ powerful arguments regarding the nature and consequences of urban activity patterns provide key insights into the processes that reinforce or inhibit the emergence of place-based social organization. The durability of her work and its deep impact on the practice of urban planning are testament to the force and generativity of her ideas. Yet, to date, few efforts to formalize or test her approach to the benefits of social ecologies that promote public contact have been offered. The current analysis operationalizes Jacobs’ notion of public contact employing the eco-network concept, offering the first empirical assessment of the consequences of shared routines for neighborhood crime rates. The strength and consistency of the protective effects of eco-network structure indicate the need for further exploration of this largely neglected aspect of urban social systems.

Figure 3.

Quintiles of Logged Violent Crime Rate in Columbus Census tracts, 1999–2001

Source: The National Neighborhood Crime Study (NNCS). Note: Census tracts in the city of Columbus with a population smaller than 300 or more than 50% group quarters residents are excluded from the NNCS data, and are dotted to represent missing. White space represents area that is not part of the city of Columbus.