Paws on the Street: Neighborhood-Level Concentration of Households with Dogs and Urban Crime

Abstract

The formative work of Jane Jacobs underscores the combination of “eyes on the street” and trust between residents in deterring crime. Nevertheless, little research has assessed the effects of residential street monitoring on crime due partly to a lack of data measuring this process. We argue that neighborhood-level rates of households with dogs captures part of the residential street monitoring process core to Jacobs’ hypotheses and test whether this measure is inversely associated with property and violent crime rates. Data from a large-scale marketing survey of Columbus, OH, USA residents (2013; n = 43,078) are used to measure census block group-level (n = 595) rates of households with dogs. Data from the Adolescent Health and Development in Context study are used to measure neighborhood-level rates of trust. Consistent with Jacobs’ hypotheses, results indicate that neighborhood concentration of households with dogs is inversely associated with robbery, homicide, and, to a less consistent degree, aggravated assault rates within neighborhoods high in trust. In contrast, results for property crime suggest that the inverse association of dog concentration is independent of levels of neighborhood trust. These associations are observed net of controls for neighborhood sociodemographic characteristics, temporally lagged crime, and spatial lags of trust and dog concentration. This study offers suggestive evidence of crime deterrent benefits of local street monitoring and dog presence and calls attention to the contribution of pets to other facets of neighborhood social organization.

Introduction

A voluminous body of research considers how the everyday behaviors of neighborhood (NH) residents can prevent crime. Among the most influential theories in this literature is that of Jacobs (1961), underscoring the fusion of mutual trust and local surveillance—or “eyes on the street”—among residents in deterring offenders. Residential trust is central to numerous other theories of crime and place, with research on collective efficacy, for example, finding that heightened NH rates of trust and shared norms toward intervention combine to reduce crime (Sampson 2012). However, evidence suggesting that residential monitoring can reduce crime has been less resolute. Much of this research lacks an explicit measure of resident-based NH street monitoring, as opposed to measures capturing the presence of both residents and visiting outsiders simultaneously (Browning and Jackson 2013). This distinction is important because Jacobs’ thesis is focally about the benefits of sufficiently high levels of street activity engaged in by local residents, whereas an overwhelming presence of outsider pedestrian traffic has been acknowledged to be potentially conducive to crime (Taylor 1988).

This study seeks to address this gap with NH-level dog presence data, which we argue captures NH variation in a key routine activity relevant to the presence of “eyes on the street” (Wood et al. 2015). Specifically, we assess whether census block group-level rates of households with dogs are inversely associated with block group property and violent crime rates. Consistent with Jacobs, we hypothesize that these associations will be most evident among NHs high in trust. Negative binomial models for crime are stratified by crime types—including property crimes, aggravated assault, robbery, and homicide—to ensure that any observed associations with dog presence rates are not specific to crimes potentially more attributable to household deterrence effects (e.g., burglary vs. robbery). Data for this study are drawn from a large-scale marketing survey of Columbus, OH, USA, residents measuring self-reported household dog presence (survey n = 43,078; block group n = 595), and data from the Adolescent Health and Development in Context study (AHDC) are used to measure NH-level rates of trust.

Background

We begin by introducing Jacobs’ hypotheses about crime, underscoring her primary focus on NH-level trust and monitoring in keeping NHs safe. We then review the evidence addressing these hypotheses, paying particular attention to the limited availability of measures capturing NH-level street monitoring among residents. Finally, we introduce literature suggesting that households with dogs contribute to NH surveillance and security and discuss how NH-level variation in dog presence captures a potentially important dimension of residential monitoring from the standpoint of Jacobs’ model.

Jacobs’ Crime Control Model

Jacobs' (1961),The Death and Life of Great American Cities was published as a critique of the prevailing efforts of Modernist-inspired architects and planners to do away with, what they perceived to be, the chaotic and unplanned nature of cities (Haar 1959). In their view, the uneven mixing of residential units and commercial amenities was a primary driver of anonymity and disorder within urban NHs. In contrast, Jacobs argued that mixed-use NHs draw residents and visitors out onto the streets, promoting “casual, public contact at a local level” that over time generates “a web of public respect and trust, … a resource in time of personal or neighborhood need” (Jacobs 1961, 56). Commenting on the origins of this trust, Jacobs asserted that it develops “over time from many, many little public sidewalk contacts”, such as when residents get “advice from the grocer” or “compare notes on their dogs” (Jacobs 1961, 56). Most relevant to this study are Jacobs’ remarks regarding how residents can keep NH streets safe, observing that there must be a relatively constant stream of “eyes on the street,” both to enact surveillance themselves and to keep the interest of potential onlookers in streetside buildings. For example, it is noted that “at night, with the security of the doormen as a bulwark, dog walkers safely venture forth and supplement the doormen” (Jacobs 1961, 40). Surveillance alone is not adequate to deter NH crimes, however, and must be accompanied by sufficiently high levels of local trust or “… an almost unconscious assumption of general street support when the chips are down—when a citizen has to choose, for instance, whether he will take responsibility, or abdicate it, in combating barbarism or protecting strangers” (Jacobs 1961, 56). Taken together, Jacobs’ crime control model hypothesizes inverse associations of residential trust and street monitoring with crime, but also an interaction between these measures such that higher levels of trust enhance the effect of residential monitoring on crime.

It is important to note that while much of The Death and Life of Great American Cities is dedicated to understanding how the built environment shapes urban life, Jacobs’ crime control model focuses on the proximate effects of NH surveillance and trust. This is not to say that hypotheses about land use and crime are inapparent in Jacobs’ work, however. Indeed, a now diverse body of research drawing on Jacobs’ ideas considers how the built environment—including land use compositions—relates to crime (Hipp and Williams 2020), and some work even considers how the built environment relates to NH trust and informal social control (Wickes et al. 2019). Nevertheless, available tests of Jacobs’ crime control model have yet to specifically assess the hypotheses regarding whether residential street monitoring and trust combine to deter crime.

Trust, Street Monitoring and Crime

Trust among residents is central to social disorganization perspectives on crime and collective efficacy theory in particular (Sampson, Raudenbush, and Earls,1997). Drawing on Bandura's (1982) personal and group-level self-efficacy research and the wide-reaching benefits of “weak” network ties (Granovetter 1973), collective efficacy emphasizes the importance of interpersonal trust between residents and shared expectations for intervention when challenges arise (Coleman 1988; Sampson 2012). Consistent with the theory, studies have found that NH rates of collective efficacy are inversely associated with violent crime (Sampson et al. 1997; Morenoff, Sampson, and Raudenbush 2001). Provided that trust between residents is key to the conceptualization and measurement of collective efficacy, this research largely validates Jacobs’ hypotheses regarding NH trust and crime. Considerably less attention has been paid to relationships between residential street monitoring and crime, however.

Jacobs’ hypotheses regarding NH surveillance and crime are important to numerous theoretical traditions within criminology. This is perhaps most apparent within routine activities theory (Cohen and Felson 1979). The routine activities approach stipulates that crime tends to occur in the presence of a suitable target and a motivated offender, and in the absence of a capable guardian. Most relevant to the present study is research on capable guardians and guardianship (Reynald 2011). Some of this work focuses on the identification of types of guardians—such as “place managers” who are well-positioned to watch over a location (Eck and Weisburd 2015)—while other research examines how the built environment can generate crime deterrent “defensible space” (Newman 1972).

Extending research on defensible space, Taylor (1988) proposed a territorial model of informal social control, directing attention to both physical design features and social cohesion among residents in reducing crime. Like Jacobs, Taylor argued that physical features conducive to residential surveillance are important to NH crime deterrence, particularly when “anonymity” between residents is minimal. However, Taylor also argued that local surveillance can be conducive to crime when predominately enacted by visiting outsiders, such as those frequenting commercial areas. In this view, heavy street traffic can overwhelm the capacity of locals to monitor their NH effectively and foster the anonymity that undermines informal social control efforts. This perspective contrasts with Jacobs’ in that Jacobs’ model anticipates crime deterrent benefits of outsider street traffic when trust and residential street monitoring are sufficient.

In light of this debate, substantial literature considers how NH land use features either promote or deter crime (Browning, Pinchak, and Calder 2021). Most relevant to Jacobs’ residential monitoring hypotheses is research on mixed land use NHs, or those where both residential and commercial areas are present. For example, in partial support of both Jacobs’ and Taylor’s hypotheses, Browning et al. (2010) found evidence of a nonlinear association between a measure of mixed land use density and violent street crimes. At low levels of residential and commercial density, increases in this measure are positively associated with crime but negatively associated with crime at high levels of residential and commercial density. Browning and Jackson, (2013) found a similar association between NH rates of street activity and rates of violence. In a study assessing the association between crime and NH “walkability”—a measure designed to capture “NH density and access to nearby amenities” (Carr, Dunsiger, and Marcus 2010, 460)—Lee and Contreras (2020) found that walkability is overall positively associated with property and violent crime rates.

These land-use studies provide important tests of Jacobs’ hypotheses regarding NH surveillance and crime. Even so, limited research has explicitly assessed the effects of NH-level measures of street monitoring enacted by residents, leaving tests of Jacobs’ crime control model incomplete. Though the availability of resident mobility data has increased precipitously in recent years, much of this data aims to capture mobility to locations that residents frequent beyond the NH. Moreover, whereas Global Positioning System data may overcome this limitation, these data remain costly, particularly when collected on a large enough scale to capture NH-level variation (Browning, Pinchak et al. 2021). These challenges call attention to the need for NH-level measures capturing variation in residents’ everyday activities that promote local monitoring.

Dogs, Social Organization and Crime

One routine activity that encourages NH monitoring is dog walking. Unlike most other types of household pets, dogs necessitate routine walks, most frequently carried out in the NH (Westgarth et al. 2017). Indeed, studies have found that pairing humans with a companion dog substantially increases daily physical activity (Shibata et al. 2012; Westgarth, Knuiman, and Christian 2016). Moreover, rates of households with dogs have increased sharply in recent decades, particularly amid rising rates of social isolation (Ozer and Perc 2020). A recent study assessing the demographics of households with pets underscores that dogs are remarkably common in the United States, with between 38.4% and 46.1% of households having a dog (Applebaum, Peek, and Zsembik 2020). Furthermore, although there are sizable racial disparities in dog presence—e.g., 22.8% among Black respondents compared to 53.7% among white respondents—and by family structure—e.g., upwards of 50% among married couples, 39.2% among unmarried couples, and 26.3% among single adults—there is remarkable consistency across socioeconomic categories. For example, the percentage of households with a dog in the lowest to highest income quartile is 37.6%, 46.6%, 54.1%, and 49%, respectively (based on General Social Survey [GSS] data; see Applebaum et al. 2020).

Despite this growing significance of animals within social life, research considering animals has long been on the margins of sociology (Bryant 1979). Parsons, Durkheim, Mead, and other classic sociological theorists have all been accused of underestimating the potential contribution of animals to the study of society (Ross 2017). Sociologists have been working to correct this course however, with recent research highlighting the contributions of animals to changes in the family (Laurent-Simpson 2021), non-profit and volunteer organizations (Peggs 2013), and healthcare delivery (e.g., service and therapy dogs) (Beck and Marshall Meyers 1996; Rodriguez, Bibbo, and O’Haire 2020a). The applicability of dogs has also caught the eye of law enforcement (Chapman 1990), and some police departments have even encouraged residents to walk their dogs in certain areas to deter crime (Fesperman 2014; Friedersdorf 2014).

Research drawing on various methodological approaches suggests that dog walking contributes to NH surveillance and safety. In one qualitative study of pet-facilitated interaction in a public park, Robins et al. (1991) found that “dogs expose their human companions in public places to encounters with strangers, facilitate interaction among the previously unacquainted, and help establish trust among the newly acquainted.” Anderson (1990) similarly observed that dog walkers “make Village streets safer for all kinds of people during the early morning hours before work, in the evenings before dinner, and late at night. At these times, people who have come to ‘know’ one another through their dogs form an effective neighborhood patrol” (Anderson 1990, 224). Studies affirm that dogs can serve as catalysts for social interaction, increasing occurrences of casual conversation with neighbors even when one’s dog is absent, especially among older populations (Rogers, Hart, and Boltz 1993; McNicholas and Collis 2000; Wood, Giles-Corti, and Bulsara 2005; Westgarth et al. 2017). Moreover, these benefits appear to be unique to dogs, with one study finding that individuals with dogs report having more forms of routine interactions with neighbors than individuals who have some other kind of animal companion (Wood et al. 2015). Still, some studies also find that dogs can be used to create and police socially exclusive spaces, particularly on the borders of gentrifying and racially mixed NHs (Tissot 2011; Mayorga-Gallo 2018). Research finds that NH conflict tends to be concentrated at the boundaries of such NHs (Legewie and Schaeffer 2016), suggesting that dog walking in low-trust NHs is less beneficial to crime deterrence efforts, as Jacobs (1961) would lead us to expect.

In addition to studies finding that dog walking contributes to NH surveillance, some research, albeit limited, has also considered whether dog presence is associated with crime. Grooms and Biddle (2018) found that Milwaukee, WI land parcels with versus without dogs have between 1.40 and 1.71 percentage points lower property crime rates. Miethe (1991) similarly found that homes with dogs experience less theft and vandalism than those without dogs, and Montoya, Junger, and Ongena (2016) found that this is particularly true at night. In an ethnographic study of burglars, Cromwell, Olson, and Avary (1991) additionally found that dog presence can be a property crime deterrent, and Logie, Wright, and Decker (1992) found that burglars perceive even the presence of “beware of dog” signs as an effective deterrent. These studies focus solely on property crimes that tend to occur within the home however, suggesting that dogs may only deter at-home crimes rather than contributing to street guardianship. Much of this research also pays little attention to how potentially confounding NH factors such as socioeconomic disadvantage, racial composition, or trust may influence the results, calling into question the robustness of their documented effects of dogs on crime.

The Present Study

Despite the widespread influence of Jane Jacobs' (1961) “eyes on the street” hypotheses, surprisingly little research has assessed how NH variation in residential street monitoring relates to NH crime. A lack of data suited to measure NH-level residential monitoring has been a primary reason for this gap in the literature, as state-of-the-art mobility data remain costly to collect on a large scale (Browning, Pinchak et al. 2021). However, research on household dog presence offers significant evidence that dogs encourage routine surveilling walks within the NH and facilitate social interactions between residents (Wood et al. 2015). These benefits of dog walking align with what Jacobs describes as the benefits of residential street monitoring. Indeed, some studies suggest that dogs contribute to household crime deterrence (Grooms and Biddle 2018), but research has yet to assess how NH variation in dog concentration relates to broader NH street crimes, such as robbery rates.

This study tests Jacobs’ residential monitoring hypotheses with data capturing NH variation in household dog presence and official crime data in Columbus, OH. Specifically, we assess whether NH-level (census block group) rates of households with dogs are inversely associated with NH-level rates of aggravated assault, robbery, homicide, and property crimes. Consistent with Jacobs’ hypotheses, we furthermore draw on data from a survey of Columbus residents capturing NH-level rates of trust to test whether higher rates of trust enhance the inverse association between NH dog presence and crime.

Data and Methods

Crime Data

Geocoded crime reports between 2014 and 2016 were obtained from police departments of all municipalities within the interstate 270 outer belt of Columbus, OH, USA, and linked to 2010 residential census block group boundaries within this area (n = 599). We use census block groups to operationalize NHs rather than larger census tracts because crime is often concentrated within highly specific areas of urban environments (Weisburd, Groff, and Yang 2012). We construct four separate crime count dependent variables of aggravated assaults, robberies, homicides, and all property crimes from these reports. Property crime counts are measured by combining burglaries, automotive crimes, and larcenies within block groups between 2014 and 2016. Aggravated assault, robbery, and homicide counts are measured within block groups for this same period. We focus on these specific violent crimes because they are more likely to occur outside places of residence on the street and are thus more directly relevant to Jacobs’ “eyes on the street” hypotheses. For example, in a recent study of crime in Chicago between 2017 and 2020, Kim and McCarty (2021) found that robberies are about 11.66 times more common in “public” locations—defined as those with descriptions such as “street” or “sidewalk” or at nonresidential settings (e.g., gas stations, schools)—compared to “residential” locations (e.g., apartments, residence garages). Homicides were 6.68 times more common in public locations, and assaults were 1.46 times more common in public locations. In contrast, battery and sexual assault were more common in residential locations.

Violent and property crime data between 2010 and 2012 were similarly obtained from the City of Columbus Police and linked to block groups within the city of Columbus proper. These data were used to construct respective temporally lagged crime rate measures of the dependent variables (n = 510).¹ In cases where a block group is split across municipalities, we use the 2010–2012 total population to construct the respective lagged crime rates only for the portion of the block group that is within the bounded study area. An identical block group-level measure of the total population was constructed between 2014 and 2016, and the natural log of this measure is specified as the exposure in negative binomial regression models for the respective crime count dependent variables, described in the “Analytic Strategy” section.

Household Dog Presence Data

Market survey data were used to generate a measure of 2013 census block group-level rates of household dog presence. The market survey data were provided to the Ohio State University Center for Human Resources Research by Giant Partners, Inc. The household dog presence data is compiled primarily by survey data and secondarily by product purchase and registration information. The surveys were conducted through a multi-channel approach, including direct mail, email communications, and social media marketing. The email communications and social media marketing led prospective respondents to a survey wherein they were prompted with up to twenty questions from a larger list of rotating questions related to lifestyle, ailments, hobbies, and interests. The pet-related questions allow respondents to answer “yes” to the type(s) of animals that reside in the household (e.g., dog, cat, or other). Survey responses were incentivized by either offering a sweepstakes entry or coupons and discounts on related products. Within the interstate 270 outer belt of the Columbus study area, the survey respondent sample size is 43,078.²

We measure NH rates of household dog presence (“neighborhood dog concentration”) using the block group-level estimated random effect, or empirical Bayes residual, from a two-level logistic regression model where binary responses to the dog ownership question are clustered within block groups (Raudenbush and Bryk 2002). This measurement approach takes into account differences in the reliability with which the block group-level log odds of dog presence are estimated by regressing block group-level random effects toward the mean by a factor proportional to their unreliability (Raudenbush and Bryk 2002). Because this measure is inherently based on the proportion of respondents reporting a dog, we select block groups with at least two market survey respondents (n = 595). To ensure that results are robust to the inclusion of low-sample size block groups, we conducted sensitivity analyses selecting block groups with at least 10 (Peduzzi et al. 1996) and 20 market survey respondents (Raudenbush and Sampson 1999), discussed in the “Supplemental Analyses” section.

There are two primary shortcomings of the market survey data in our study. First, because the dog question was presented to respondents as part of a rotating list of questions, the data do not precisely estimate the proportion of the Columbus, OH, households with dogs. Indeed, only 14.52% of respondents report having a dog, which is noticeably lower than results from the above reviewed national studies would lead us to expect (Applebaum et al. 2020). Furthermore, because only “yes” responses to the dog question were recorded, the denominator used in our NH proportion of households with a dog measure is likely too large. However, under the assumption that the dog question is presented to respondents nearly randomly across the study area, these data are sufficient to estimate the relative proportion of households with dogs within Columbus, OH, or analogously, estimating a ranking of the Columbus NHs by the proportion of households with dogs, which aligns with the aims of this study.

Nevertheless, we acknowledge the potential for bias in both overall sample selection and estimating block-group level dog concentration. To this end, Table 2 displays bivariate correlations between all study variables, with the block-group level number of households sampled in the market survey (with a minimum of two respondents) as the first column in this matrix. The correlations between the number of market survey households sampled with the block group population (0.41), residential instability (−0.40), and block group-level proportion of family households (0.49) are especially sizable. Although the former two correlations intuitively capture a greater likelihood of being sampled in more populous and residentially stable NHs, the latter correlation suggests a potential over-representation of households with families. All of these measures are controlled for in our analytic models, however.³

We additionally considered two tests of the validity of the market survey data. We first assessed how block group aggregations of market survey reports of socioeconomic status (SES)—including family income and educational attainment—align with our census-based measure of SES (described below). The correlations between market survey-based block group aggregations of family income and educational attainment with our census-based measure of SES were relatively high at .890 and .755, respectively. We then compared household dog presence estimates based on data from recent national studies with estimates based on market surveys and compared these estimates by sociodemographic categories of marital status, income, and age (Applebaum et al. 2020).⁴ These comparisons are displayed in Supplemental Appendix Table 1. To briefly summarize, while our data do indeed underestimate dog presence across sociodemographic variables (e.g., 25.7% of married market survey respondents versus about 52.6% of 2018 GSS respondents; and 13.9% of single market survey respondents versus 26.3% of single GSS respondents), the ratios of dog presence between categories of these variables are quite consistent. For example, GSS 2018 data suggest that the ratio of married people with dogs compared to married plus single people with dogs is 0.667. This same ratio is 0.649 in the market survey data. Similar consistency is observed across quartiles of income. However, comparison across age groups underscores that the market survey data underrepresents the presence of dogs among respondents ages 18–29. Although we control for a census-based measure of the proportion of residents ages 25+ in our analytic models, we acknowledge the potential for resulting bias in our NH dog concentration coefficients. It is important to note, however, that national estimates are not necessarily reflective of household dog presence demographics in the study area of Columbus, OH. In sum, these additional investigations suggest that the presently used market survey data reasonably correspond with crime relevant census-based measures (SES and population) and recent household dog presence estimates, giving us further confidence in the potential for the market survey data to accurately capture NH variation in rates of households with dogs.

Finally, in addition to assessing our household dog presence data, we considered alternative dog data sources, including the American Housing Survey (AHS) issued by the U.S. Department of Housing and Urban Development and official Franklin County, OH pet registration records. The AHS assessed household rates of pet presence by asking respondents whether assistance would be needed in evacuating pets in case of a disaster in select cities in 2013 and 2017, but respondents were not able to report what kind of pet (e.g., dog, cat, horse). This survey thus renders assessments of correlations between NH-level dog concentration and crime impossible. Although official Franklin County, OH, pet registration records overcome these limitations, registration data are potentially more subject to reporting bias than are self-reported data, given that pet registration requires payment. For example, in a study of household dog presence and property crime rates, Grooms and Biddle (2018) estimated dog licensure compliance rates of only 35–40% in Milwaukee, WI. Moreover, Franklin County residents can purchase permanent licenses for their dogs. These licenses are not removed from the county database when dogs die, nor are changes in address necessarily reported or checked. Taken together, in addition to investigating potential shortcomings of the presently utilized dog presence data, we furthermore have reason to believe that alternative data sources are less suited to our research questions.

Neighborhood Trust Data

Data for our measure of NH-level trust are from Wave I of the AHDC study, a data collection effort of 1,405 Columbus, OH youth and their families focused on the consequences of everyday developmental contexts for health and well-being. Wave I of the AHDC is a representative sample of study area households with youth ages 11–17 residing within the Columbus Interstate 270, the Franklin County outer belt, including the majority of the City of Columbus and several wealthier inner suburbs. For more information on the study design of the AHDC and the representativeness of the study area, see Browning, Calder et al. (2021).

Block group-level trust is based on respondents’ trust reports in their NH and at specific routine activity locations. Participant caregivers (mean age = 45.6, 87% female) were asked to report the coordinates of various routine activity locations (e.g., schools, workplaces, grocery stores, friends’ houses) using an interviewer-assisted Google Maps-based “location generator” module (Browning, Pinchak et al. 2021). We linked the coordinates of these locations to census block groups using the R sf: Simple Features package (Pebesma et al. 2019). Respondents were then asked to report to what extent they agree that “people on the streets can be trusted” at each of their activity locations and in their NH.⁵ Responses for this question ranged from 1 (“strongly disagree”) to 5 (“strongly agree”). Respondents were asked to report on their NH both during the day and at night, while all other locations were reported on only once. We then estimated block group-level aggregations of trust (respondent n = 1,257; report n = 5,918; block group n = 565) using a linear cross-classified multilevel model where reports are clustered within respondents and within block groups (Sampson et al. 1997; Raudenbush and Bryk 2002) using the lme4 package in R (Bates et al. 2015). The estimated block group-level random effect was recovered from this model for all block groups with available reports for the corresponding measure. In order to obtain corresponding estimates for unobserved block groups, the block group-level estimated random effect was then spatially smoothed across the study area using a conditional autoregressive (CAR) model proposed by Leroux, Lei, and Breslow (2000) and implemented in R using the CARBayes: Bayesian Conditional Autoregressive Modelling package (Lee 2020). This process yielded estimates of NH trust for all 615 Columbus block groups within the outer belt boundary. The correlation coefficient between the estimated block group-level measure of trust from the CAR model with the estimated random effect from the cross-classified multilevel model exceeds .99, indicating that our approach did not spatially smooth the estimated random effect but allowed us to obtain NH measures of trust for the 50 block groups without trust reports.

Demographic Controls

Block group-level (“NH”) control measures are based on data averaged across the 2009–2013 rounds of the American Community Survey administered by the U.S. Bureau of the Census. SES is the average of six block group measures, including the proportion of residents with a college degree, the proportion of residents in a professional occupation, the proportion of households with at least $50,000 in annual income, the reverse coded proportion of residents in poverty, the reverse coded proportion of female-headed households, and reverse coded proportion of residents ages 16–64 unemployed. Proportion Black is the proportion of residents who are Black. Residential instability is the proportion of residents ages 5 and older who moved in the past 5 years or are renting. Proportion families is the number of census designated family households divided by the number of total households in the block group. Proportion young males is the number of resident males between the ages of 15–24 divided by the total population in the block group. Proportion age 25+ is the proportion of residents ages 25 or older. Proportion residential is the percent of land in each block group designated as residential (vs. commercial or park). Square miles is the sq. mile area per block group. Number of Market Survey Households is the number of households per block group queried for the dog presence market survey.⁶

Analytic Strategy

We fit negative binomial regression models to account for over-dispersion in the block-group crime counts. In our models, The natural log of block group-level 2014–2016 total population is specified as the exposure variable, allowing regression coefficients to be interpretable as associations of a unit increase in the corresponding independent variable on the log crime rate holding the other independent variables constant (Osgood 2000). Crime is highly spatially clustered within areas of cities, and research has long highlighted the potential for “spillover” or “adjacency” effects of spatially proximate NH processes on crime (Peterson and Krivo 2010). Numerous analytic approaches have been proposed to account for these effects, including the spatial generalized linear model (SGLM) framework (Kelling et al. 2021). However, recent work by Khan and Calder (2022) demonstrates the choice of approach to account for spatial effects—including SGLM approaches—can have strong consequences for inference on regression coefficients. Therefore, we present results from the following non-SGLM approach, arguably standard in the crime and place literature (Lee and Contreras 2020; Kim and Hipp 2021; Tillyer, Wilcox, and Walter 2021). In our models, we control for spatially lagged measures of block group-level dog concentration and trust (the focal independent variables). This approach allows for the estimation of an association between a focal independent variable and a suitable transformed version of the dependent variable to be assessed net of the effect of levels of the independent variable in the surrounding area (Haberman and Ratcliffe 2015; Tillyer et al. 2021). Specifically, we estimate associations of block group-level trust and dog concentration with crime outcomes while controlling for average levels of these independent variables in the surrounding block groups (using queen contiguity). To be clear, we use this strategy in order to account for potential spillover effects of our focal independent variables in adjacent block groups. However, we do not explicitly hypothesize or test the robustness of such spillover effects on our dependent variables. Doing so would necessitate controlling for adjacent levels of all our independent variables, which may lead to problems of multicollinearity.

Results

Descriptive statistics for all block-group level study variables are displayed in Table 1. Correlations between all independent study variables and log-transformed total violent and property crime rates are displayed in Table 2. To assess the potential influence of multicollinearity on the results, variance inflation factors (VIFs) were generated for all independent variables. When including temporal lags of the dependent variables, the average VIF is 2.46, and 2.23 when not considering the temporal lags. Further, our dog concentration and trust focal independent variables have maximum VIFs of 1.95 and 1.50, respectively. Allison (2012) explains that while VIFs greater than 2.5 can be problematic for inference, this is largely only the case for focal independent variables, and that larger VIFs for control variables can be “safely ignored.” Our results are presented separately for robbery, homicide, aggravated assault, and property crime rates in Tables 3–6, respectively.⁷

Table 1

Descriptive Statistics

Variables	n	Mean	Standard Deviation	Min.	Max.
Dependent variable 2014–2016 crime counts
Robbery	595	9.31	10.25	0	60
Homicide	595	0.41	0.78	0	5
Aggravated assault	595	6.85	7.84	0	47
Property crimes	595	151.51	144.39	4	1818
Block group population	595	1139.02	603.54	65	7610
2010–2012 crime rates
Robbery	507	6.36	8.56	0	133.3
Homicide	507	0.13	0.29	0	2.4
Aggravated assault	507	1.77	2.18	0	11.3
Property crimes	507	78.22	140.58	0	3000
Dog concentration	595	0.00	1.00	−3.27	2.66
Trust	595	0.00	1.00	−2.85	2.29
Adjacent dog concentration	595	0.00	1.00	−2.80	3.21
Adjacent trust	595	0.00	1.00	−4.48	7.18
Socioeconomic status	595	0.00	1.00	−3.15	2.04
Proportion black	595	0.29	0.29	0	1
Residential instability	595	-0.02	0.97	−1.76	3.27
Proportion families	595	0.56	0.19	0	0.93
Proportion young males	595	0.08	0.08	0	0.62
Proportion age 25+	595	0.66	0.13	0.02	0.95
Proportion residential	595	0.49	0.29	0	1
Square miles	595	0.31	0.38	0.02	3.00
Number of market survey households	595	70.73	42.37	2	315

Note: Dog concentration, trust, SES, and residential instability measures are z-score standardized.

Table 2

Bivariate Correlations Between Block Group-Level Measures

	# Market HHs	ln (violent crime Rate)	ln (Property crime Rate	Trust	Dog conce	Adjacent trust	Adjacent dogs	SES	%Black	Residential instability	%Families	%Young males	%Age 25+	%Residential	Sq. miles	Population
ln(# of market HHs)	1
ln(violent crime rate)	−0.06	1
ln(property crime rate)	−0.16*	0.79*	1
Trust	0.03	−0.58*	−0.48*	1
Dog concentration	−0.05	−0.32*	−0.26*	0.26*	1
Adjacent trust	0.00	−0.50*	−0.36*	0.41*	0.23*	1
Adjacent dogs	−0.06	−0.29*	−0.24*	0.23*	0.55*	0.29*	1
SES	−0.05	−0.76*	−0.54*	0.52*	0.36*	0.53*	0.31*	1
%Black	0.17*	0.53*	0.27*	−0.43*	−0.56*	−0.44*	−0.62*	−0.67*	1
Residential instability	−0.40*	0.40*	0.46*	−0.26*	−0.33*	−0.23*	−0.20*	−0.45*	0.22*	1
%Families	0.49*	−0.17*	−0.31*	0.14*	0.10*	0.10*	0.11*	−0.00	0.05	−0.56*	1
%Young males	−0.27*	0.07	0.19*	−0.03	−0.05	−0.01	−0.09*	−0.11*	−0.07	0.47*	−0.37*	1
% Age 25+	0.03	−0.20*	−0.19*	0.13*	0.25*	0.13*	0.19*	0.43*	−0.21*	−0.46*	−0.00	−0.65*	1
%Residential	−0.16*	0.06	0.04	−0.03	−0.08	−0.05	−0.08	−0.06	0.00	0.15*	−0.17*	0.14*	−0.08*	1
Sq. miles	0.22*	−0.08	0.01	0.05	0.18*	0.25*	0.17*	0.10*	−0.08*	−0.09*	0.09*	−0.09*	0.09*	−0.28*	1
Population	0.41*	−0.24*	−0.27*	0.07	0.13*	0.13*	0.20*	0.14*	−0.09*	−0.02	0.05	−0.00	−0.05	−0.13*	0.42*	1

^*p < .05; two-tailed tests.

Note: Includes block groups with two or more market survey respondents.

Table 3

Negative Binomial Regression Models for 2014–2016 Robbery Counts per Block Group

	Robbery
	Model 1	Model 2	Model 3	Model 4
Socioeconomic status	−0.729***	−0.697***	−0.330***	−0.319***
	(0.068)	(0.068)	(0.060)	(0.060)
Proportion black	−0.126*	−0.100+	−0.041	−0.030
	(0.060)	(0.060)	(0.049)	(0.048)
Residential instability	0.118+	0.120+	0.006	0.003
	(0.063)	(0.062)	(0.049)	(0.049)
Proportion families	−0.197***	−0.192***	−0.160***	−0.158***
	(0.054)	(0.053)	(0.042)	(0.042)
Proportion young Males	−0.006	−0.014	−0.029	−0.032
	(0.060)	(0.059)	(0.047)	(0.047)
Proportion Age 25+	0.171**	0.154*	0.029	0.020
	(0.063)	(0.062)	(0.049)	(0.049)
Proportion Residential	0.014	0.024	−0.070*	−0.063*
	(0.037)	(0.037)	(0.030)	(0.030)
Square miles	0.105**	0.113**	0.006	0.013
	(0.040)	(0.040)	(0.031)	(0.031)
Population	−0.000***	−0.000***	−0.000***	−0.000***
	(0.000)	(0.000)	(0.000)	(0.000)
#Market survey households	0.001	0.001	0.002+	0.002+
	(0.001)	(0.001)	(0.001)	(0.001)
Adjacent trust	−0.131**	−0.138**	0.015	0.007
	(0.045)	(0.044)	(0.036)	(0.035)
Adjacent dog concentration	−0.064+	−0.057	0.007	0.010
	(0.050)	(0.049)	(0.039)	(0.039)
Trust	−0.289***	−0.288***	−0.073*	−0.081*
	(0.041)	(0.040)	(0.034)	(0.033)
Dog concentration	−0.053	−0.078+	−0.009	−0.038
	(0.048)	(0.048)	(0.038)	(0.040)
Trust × dog concentration		−0.129***		−0.082**
		(0.037)		(0.030)
ln(2010–2012 rate-dependent variable)			0.837***	0.824***
			(0.048)	(0.048)
lnalpha	−0.632***	−0.668***	−1.456***	−1.480***
	(0.080)	(0.081)	(0.102)	(0.103)
Intercept	−4.612***	−4.574***	−6.116***	−6.071***
	(0.082)	(0.083)	(0.113)	(0.114)
Block group N	595	595	507	507

^†p < .1.

^*p < .05.

^**p < .01.

^***p < .001; two-tailed tests.

Note: All non-logged variables are standardized z-scores. The natural log of block group population is specified as the exposure with its coefficient constrained to 1.

Table 4

Negative Binomial Regression Models for 2014–2016 Homicide Counts per Block Group

	Homicide
	Model 1	Model 2	Model 3	Model 4
Socioeconomic status	−0.824***	−0.767***	−0.716***	−0.666***
	(0.148)	(0.147)	(0.153)	(0.152)
Proportion black	−0.024	−0.004	−0.040	−0.025
	(0.123)	(0.120)	(0.129)	(0.125)
Residential instability	0.247*	0.240*	0.272*	0.264*
	(0.120)	(0.118)	(0.121)	(0.119)
Proportion families	−0.053	−0.038	−0.035	−0.025
	(0.112)	(0.109)	(0.113)	(0.110)
Proportion young Males	−0.349*	−0.338*	−0.306*	−0.292+
	(0.152)	(0.150)	(0.152)	(0.149)
Proportion Age 25+	0.142	0.125	0.111	0.097
	(0.129)	(0.127)	(0.130)	(0.127)
Proportion Residential	−0.065	−0.044	−0.057	−0.035
	(0.075)	(0.074)	(0.075)	(0.074)
Square miles	0.101+	0.118+	0.097	0.118+
	(0.076)	(0.074)	(0.077)	(0.076)
Population	−0.001***	−0.001***	−0.001***	−0.001***
	(0.000)	(0.000)	(0.000)	(0.000)
#Market survey households	0.005*	0.005*	0.006*	0.006*
	(0.002)	(0.002)	(0.002)	(0.002)
Adjacent trust	−0.081	−0.095	−0.033	−0.051
	(0.089)	(0.089)	(0.092)	(0.093)
Adjacent dog concentration	−0.136+	−0.127	−0.126	−0.119
	(0.103)	(0.101)	(0.108)	(0.105)
Trust	−0.180*	−0.232**	−0.171*	−0.226*
	(0.083)	(0.087)	(0.083)	(0.088)
Dog concentration	−0.042	−0.138	−0.016	−0.118
	(0.103)	(0.112)	(0.106)	(0.117)
Trust × dog concentration		−0.186*		−0.179*
		(0.076)		(0.078)
ln(2010–2012 rate-dependent variable)			0.827**	0.779**
			(0.309)	(0.299)
lnalpha	−1.167*	−1.382*	−1.355*	−1.625*
	(0.501)	(0.595)	(0.579)	(0.732)
Intercept	−7.730***	−7.708***	−7.814***	−7.784***
	(0.238)	(0.234)	(0.253)	(0.248)
Block group N	595	595	507	507

^†p < .1.

^*p < .05.

^**p < .01.

^***p < .001; two-tailed tests.

Note: All non-logged variables are standardized z-scores. The natural log of block group population is specified as the exposure with its coefficient constrained to 1.

Table 5

Negative Binomial Regression Models for 2014–2016 Aggravated Assault Counts per Block Group

	Aggravated assault
	Model 1	Model 2	Model 3	Model 4
Socioeconomic status	−1.012***	−0.998***	−0.647***	−0.636***
	(0.068)	(0.069)	(0.067)	(0.067)
Proportion black	−0.152**	−0.143*	−0.189***	−0.183***
	(0.058)	(0.058)	(0.053)	(0.053)
Residential instability	0.029	0.028	−0.023	−0.026
	(0.061)	(0.060)	(0.055)	(0.054)
Proportion families	−0.129*	−0.125*	−0.081+	−0.078+
	(0.053)	(0.053)	(0.048)	(0.048)
Proportion young Males	0.016	0.014	−0.005	−0.005
	(0.059)	(0.058)	(0.053)	(0.053)
Proportion Age 25+	0.222***	0.213***	0.104+	0.097+
	(0.063)	(0.063)	(0.057)	(0.057)
Proportion Residential	−0.002	0.002	−0.046+	−0.042
	(0.036)	(0.036)	(0.033)	(0.033)
Square miles	0.129***	0.132***	0.078*	0.080*
	(0.039)	(0.039)	(0.034)	(0.034)
Population	−0.000***	−0.001***	−0.000***	−0.000***
	(0.000)	(0.000)	(0.000)	(0.000)
#Market survey households	0.002+	0.002+	0.003**	0.003**
	(0.001)	(0.001)	(0.001)	(0.001)
Adjacent trust	−0.243***	−0.247***	−0.119**	−0.122**
	(0.046)	(0.046)	(0.042)	(0.042)
Adjacent dog concentration	−0.011	−0.007	0.024	0.027
	(0.048)	(0.048)	(0.043)	(0.043)
Trust	−0.203***	−0.205***	−0.063+	−0.068+
	(0.040)	(0.039)	(0.037)	(0.037)
Dog concentration	−0.102*	−0.117*	−0.055	−0.074+
	(0.048)	(0.048)	(0.044)	(0.046)
Trust × dog concentration		−0.060+		−0.055+
		(0.037)		(0.034)
ln(2010–2012 rate-dependent variable)			0.796***	0.795***
			(0.070)	(0.070)
Lnalpha	−0.830***	−0.844***	−1.240***	−1.255***
	(0.089)	(0.090)	(0.104)	(0.104)
Intercept	−5.046***	−5.027***	−5.713***	−5.695***
	(0.084)	(0.085)	(0.104)	(0.104)
Block group N	595	595	507	507

^†p < .1.

^*p < .05.

^**p < .01.

^***p < .001; two-tailed tests.

Note: All non-logged variables are standardized z-scores. The natural log of block group population is specified as the exposure with its coefficient constrained to 1.

Table 6

Negative Binomial Regression Models for 2014–2016 Property Crime Counts per Block Group

	Property crime
	Model 1	Model 2	Model 3	Model 4
Socioeconomic status	−0.333***	−0.317***	−0.137***	−0.133***
	(0.042)	(0.042)	(0.039)	(0.039)
Proportion black	−0.275***	−0.257***	−0.147***	−0.144***
	(0.041)	(0.042)	(0.035)	(0.035)
Residential instability	0.174***	0.177***	0.068*	0.067+
	(0.041)	(0.041)	(0.034)	(0.034)
Proportion families	−0.134***	−0.132***	−0.076**	−0.074*
	(0.036)	(0.035)	(0.029)	(0.029)
Proportion young Males	0.067+	0.061+	0.083*	0.082*
	(0.040)	(0.040)	(0.033)	(0.033)
Proportion Age 25+	0.106*	0.097*	−0.010	−0.012
	(0.041)	(0.041)	(0.034)	(0.034)
Proportion Residential	−0.013	−0.007	−0.019	−0.016
	(0.024)	(0.024)	(0.020)	(0.020)
Square miles	0.216***	0.219***	0.107***	0.109***
	(0.027)	(0.027)	(0.023)	(0.023)
Population	−0.000***	−0.000***	−0.000***	−0.000***
	(0.000)	(0.000)	(0.000)	(0.000)
#Market survey households	0.002*	0.002*	0.002*	0.002*
	(0.001)	(0.001)	(0.001)	(0.001)
Adjacent trust	−0.072*	−0.076**	0.024	0.022
	(0.030)	(0.029)	(0.025)	(0.025)
Adjacent dog concentration	−0.090**	−0.087**	−0.042+	−0.041+
	(0.032)	(0.032)	(0.027)	(0.027)
Trust	−0.189***	−0.184***	−0.088***	−0.089***
	(0.027)	(0.027)	(0.023)	(0.023)
Dog concentration	−0.088**	−0.091**	−0.045+	−0.051+
	(0.031)	(0.031)	(0.026)	(0.027)
Trust × dog concentration		−0.068**		−0.026
		(0.024)		(0.021)
ln(2010–2012 rate-dependent variable)			0.633***	0.629***
			(0.037)	(0.037)
lnalpha	−1.207***	−1.220***	−1.771***	−1.773***
	(0.058)	(0.058)	(0.065)	(0.065)
Intercept	−1.702***	−1.681***	−4.420***	−4.395***
	(0.052)	(0.052)	(0.163)	(0.165)
Block group N	595	595	507	507

^†p < .1.

^*p < .05.

^**p < .01.

^***p < .001; two-tailed tests.

Note: All non-logged variables are z-score standardized. The natural log of block group population is specified as the exposure with its coefficient constrained to 1.

 

Turning first to results for robbery, Model 1 indicates that the inverse association between block group-level dog concentration and the block group-level robbery rate is statistically nonsignificant, net of demographic control variables, block group-level trust, and levels of dog concentration and trust in surrounding block groups. In contrast, a one standard deviation increase in NH trust is associated with about a 25.1 percent decrease ((exp(−0.289)−1) × 100 = −25.1, p < .001) in block group-level robbery rates. Model 2 assesses Jacobs’ core hypothesis anticipating interactive effects of local surveillance—captured by NH dog concentration—and trust. The interaction term for trust*dog concentration is negative (b = −0.129) and statistically significant (p < .001), indicating that the inverse associations of NH dog concentration and trust with robbery rates are more pronounced at higher levels of one another. To illustrate this pattern, the top left panel of figure 1 displays predicted robbery rates when z-score standardized NH dog concentration and trust are at one standard deviation above and below their respective means. Here, it is clear that the inverse association between dog concentration and robbery rates is only evident among NHs high in trust. Models 3 and 4 reduce the sample to block groups for which we were able to obtain temporally lagged crime data and controls for the natural log of block group-level 2010–2012 robbery rates to estimate associations of dog concentration and trust with changes in the dependent variable. Model 3 finds no evidence of a statistically significant average association between NH dog concentration and robbery rates, and Model 4 again finds evidence of a negative trust*dog concentration interaction (b = −0.082; p < .01).

Figure 1

Predicted robbery, homicide, aggravated assault, and property crime rates when z-score standardized neighborhood trust and neighborhood dog concentration are at −1 and +1 standard deviations below and above their respective means (from Tables 3–6, Model 2s).

Table 4 displays results for homicide rates, with Model 1 finding no evidence of an association between dog concentration and the outcome. In Model 2, the interaction of trust*dog concentration is again negative and statistically significant (b = −.186; p < .05). Predictions from this model are displayed in the second panel of figure 1, illustrating that the inverse association between dog concentration and homicide is only apparent among NHs with higher levels of trust. These patterns are similarly evident in Models 3 and 4 when controlling for temporally lagged homicide rates. Turning to results for aggravated assault rates in Table 5, the first model offers some evidence of an inverse association of dog concentration (b = −.102, p < .05). Model 2 adds the trust × dog concentration interaction, which is negative (b = −.060) but only statistically significant at the p < .1 level. Predictions from this model are presented in the third panel of figure 1. In Model 4, the trust × dog concentration interaction remains similarly evident when controlling for temporally lagged assault rates (b = −.055; p < .1).

Results for property crime rates are displayed in Table 6. The first model indicates that a one standard deviation increase in NH dog concentration is associated with an 8.42% decrease ((exp(−.088)−1) × 100 = −8.42, p < .01) in block group-level property crime rates. In Model 2, the trust × dog concentration interaction is negative and statistically significant (b = −.068, p < .01) and is illustrated in panel 4 of figure 1. When controlling for temporally lagged property crime rates in Model 3, the average association between block group dog concentration and property crime is only statistically significant at the p < .1 level (b = −.045), while the trust × dog concentration interaction in Model 4 is statistically nonsignificant. Overall, we thus find consistent evidence of interactive effects of NH dog concentration and trust for NH robbery and homicide rates, more limited evidence of these interactions for aggravated assault, and some evidence only for average inverse associations of dog concentration with property crime rates.

Supplemental Analyses

Sensitivity analyses were conducted to ensure the robustness of these results. First, all presented analyses were replicated when using two-level negative binomial models to account for any residual clustering of block groups within census tracts (Raudenbush and Bryk 2002). Results from these models are presented in Appendix Table 2, yielding substantive conclusions nearly identical to those based on the results discussed above. The only exception is for property crime, with the replicated Model 3 offering more evidence of an average association of dog concentration when controlling for temporally lagged property crime rates.

Our approach to measuring block group-level trust draws on reports for respondents’ NHs and routine activity locations. This contrasts with conventional NH social climate measurement strategies, which typically rely solely on reports about respondents’ NHs (e.g., Sampson and Raudenbush 1999). To ensure that results are not sensitive to this measurement strategy, we replicated our analyses when using trust measures based only on reports for respondents’ home NHs. These measures are based on 2,131 reports from 1,099 respondents clustered within 424 block groups of residence and spatially smoothed with the procedure described previously. Results from these analyses are presented in Supplemental Appendix Table 3. Substantive conclusions drawn from these analyses align with those presented here, with two notable exceptions. Specifically, results for robbery find less evidence of trust × dog concentration interactions when controlling for lagged robbery rates but consistent evidence for these interactions in all aggravated assault models. Analyses were replicated once again when excluding block groups with no trust reports (i.e., no spatial smoothing), and are presented in Appendix Table 4. Conclusions from these analyses broadly align with those based on our presented results, but with less evidence of a dog concentration × trust interaction for aggravated assault rates when controlling for temporally lagged assault rates (p > .1 rather than p < .1).

Next, to ensure that our results are not sensitive to the inclusion of block groups with very low sample sizes, the presented analyses and dog measures were replicated first when selecting on block groups with at least 10 respondents (n = 564), and second when selecting on block groups with at least 20 respondents (n = 536). Ten respondents per block group was selected because it is a conventional cutoff in the literature (Peduzzi et al. 1996), whereas a cutoff of 20 respondents per block group was selected because Raudenbush and Sampson (1999) find that twenty compared to ten respondents per NH unit can considerably increase the reliability of an estimated NH-level measure, and that increases beyond 20 have more marginal returns on reliability. To assess potential differences between the three analytic samples, Appendix Table 5 displays bivariate correlations between study variables and the number of market survey respondents per block group for the ≥2, ≥10, and ≥20 respondents per block sample. Results for the ≥10 respondents per block group sample are presented in Appendix Table 6, and results for the ≥20 respondents per block group sample are presented in Appendix Table 7. Results from these analyses yield substantive conclusions identical to those based on our presented analyses, although with more evidence of trust × dog concentration interactions for aggravated assault rates and the average association between dog concentration and property crime rates.⁸

Discussion

The publication of Jacobs' (1961),The Death and Life of Great American Cities ignited sweeping urban research programs spanning the social and behavioral sciences. A significant thrust of this research focuses on crime, investigating Jacobs’ hypotheses such as how NH land use and built environment features relate to NH crime rates (Hipp and Williams 2020; Browning, Pinchak et al. 2021). Nevertheless, Jacobs’ crime control model is concerned not only with physical features of NHs but centrally with how NH rates of trust and local street surveillance among residents combine to deter crime. Research on collective efficacy partially affirms Jacobs’ model, underscoring that residential trust is core to NH crime deterrence (Sampson et al. 1997). Significantly less attention has been paid to the role of NH rates of residential street monitoring due mainly to a lack of data measuring this process.

This study proposed that NH-level rates of households with dogs capture a key routine activity contributing to rates of local street monitoring among residents. Consistent with Jacobs’ monitoring hypotheses, the reviewed literature highlights that dogs promote routine walks in the NH (Shibata et al. 2012; Westgarth et al. 2016) and interactions between NHs residents (Robins et al. 1991; Wood et al. 2015). Drawing on extensive market survey data capturing NH variation in household dog presence in Columbus, OH, USA, we assessed whether block group-level dog concentration is inversely associated with block group robbery, homicide, aggravated assault, and property crime rates. We additionally utilized data from the AHDC study to consider whether block-level rates of trust enhance the crime deterrent benefits of dog concentration. Consistent with Jacobs’ crime control model, we found that NH dog concentration is inversely associated with rates of robbery, homicide, and, to a less consistent degree, aggravated assault rates among NHs higher in local trust. In contrast, results for property crime suggest that the inverse association of dog concentration is independent of levels of NH trust. These associations were observed net of control variables for NH sociodemographic characteristics, spatially lagged measures of trust and dog concentration, and temporal lags of the dependent variables. The evidence for crime reducing associations of NH dog presence with robbery and homicide rates are especially notable, highlighting that dog concentration is relevant for understanding rates of street crimes that most frequently occur on the street beyond the home (Kim and McCarty 2021).

Jacobs’ model is closely related to social disorganization theories of crime, arguing that routine activities of residents can foster deterrence and intervention norms that reduce crime (Bursik and Grasmick 1993; Sampson 2012). The reviewed literature documents cases where NH dog concentration fosters such norms, particularly among networks of residents with dogs (Anderson 1990; Robins et al. 1991; Wood et al. 2015). However, studies have less often considered how dog presence contributes to the development of social control norms among residents more generally, such as among those without dogs. To this end, research considering how NH variation in perceptions of street safety relates to actual rates of residential street activities such as dog walking remains necessary. For instance, some research finds that overlapping routine activity patterns among residents foster social control norms (Browning et al. 2017), but studies considering a broader range of everyday residential activities are scarce.

Our findings complement recent research on NH guardianship, finding that routine street monitoring increases the capability and willingness of residents to intervene when problems arise (Reynald 2010, 2011). For instance, Reynald (2011, 137) found that “guardians who consistently monitored their space demonstrated their increased capability over those who did not by providing a behavioral profile of potential offenders who stand out as being suspicious in a residential context.” Considered alongside the results of the present study, dog walking may thus contribute to NH security by equipping residents with the familiarity to identify suspect outsiders. Reynald also found that individuals’ decision to intervene in the moment is closely related to their physical capabilities (e.g., self-defense training), which dog presence may reinforce. This study provides initial evidence of crime deterrent benefits of NH dog presence, but more research is needed to understand how dogs confer these benefits (e.g., strengthening the capacity of walkers to intervene or identify outsiders). Further, limited research has considered how motivated offenders perceive crime deterrence processes (Logie et al. 1992). Thus, research assessing whether dogs can ward off street crimes from the perspective of offenders may also enlighten our results.

Though Jacobs’ (1961) crime control model is primarily concerned with residential trust and local street monitoring, hypotheses about land use features and crime are also central to her work. This study did not consider the potential role of land use features in understanding how residential street monitoring and dog presence relates to crime, but this will be an important next step for future research. In particular, while NH street walkability and proximity to parks are overall positively associated with crime (Boessen and Hipp 2018; Lee and Contreras 2020), these measures may furthermore be important moderators of associations between residential dog presence and crime. On the one hand, walkability and park presence may enhance street monitoring among dog walkers. However, these features may also attract outside traffic, heightening local anonymity and reducing crime deterrent benefits of residential dog presence (Jacobs 1961; Taylor 1988). Future research is also needed to determine whether NH variation in other self-reported street monitoring behaviors (e.g., jogging) may complement our results, as little research has considered how everyday activities contribute to NH social organization or crime deterrence. The discovery of such benefits of specific routine activities is particularly important in light of efforts by police to direct street surveillance to specific areas to reduce crime (Fesperman 2014; Friedersdorf 2014). In sum, more research examining whether and how specific routine activities of residents can contribute to NH guardianship remains necessary (Reynald 2011).

Despite being among the first studies finding evidence supporting Jacobs’ residential street monitoring and trust hypotheses regarding crime, this study is not without limitations. Notably, though we conducted numerous checks on the viability of the presently used household dog presence data, we acknowledge the potential for bias in these data to influence our results. We nevertheless maintain that our data offer numerous advantages over alternative dog data sources and official pet registration data in particular, which Franklin County, OH, is not obliged to keep up to date. We thus join others in calling for more data collection efforts to measure NH variation in the presence of companion animals (Grooms and Biddle 2018; Applebaum et al. 2020; Laurent-Simpson 2021). In addition, this study only considered NH-level associations of household dog presence and trust with crime in a single city. Therefore, additional research spanning multiple cities remains necessary to validate our results and implications for policy fully. Finally, although we find some evidence that NH dog concentration is associated with changes in crime over time (i.e., by controlling for temporally lagged dependent variables), we cannot rule out the influence of reverse causality in these relationships without more robust longitudinal data on dog presence. For example, increasing crime rates may also lead residents to acquire dogs to bolster personal and NH safety, potentially biasing our associations of dog concentration downward. Consequently, we urge future studies to assess whether changes in NH rates of households with dogs over even short periods are associated with changes in crime.

Notwithstanding these limitations, the present study overcomes substantial shortcomings of previous research pursuing these research questions. It gives further credence to the proposed use of dog walking in the NH to prevent crimes (Friedersdorf 2014), particularly when trust among residents is sufficient (Jacobs 1961). Though a now extensive body of research validates that dogs are beneficial to the health and well-being of their human companions (Wells 2007; Rodriguez, Greer et al. 2020b), how pets shape the social lives of humans more generally remains significantly understudied. Thus, beyond investigations of crime, we hope that this study and our recommendations motivate additional research considering how pets contribute to and structure social processes as diverse as human mobility patterns, social network formation, and political involvement.

Funding Sources

The study is based upon work supported by the National Science Foundation Graduate Research Fellowship Program under Grant No. DGE125583. The Adolescent Health and Development in Context study is funded by the National Institute on Drug Abuse (Browning, 1R01DA032371); the Eunice Kennedy Shriver National Institute on Child Health and Human Development (Calder, R01HD088545; the Ohio State University Institute for Population Research, P2CHD058484; the University of Texas at Austin Population Research Center, P2CHD042849) and the W.T. Grant Foundation. Opinions and conclusions expressed herein are solely those of the author(s) and should not be construed as representing the opinions or policy of any agency of the Federal government.

About the Authors

Nicolo Pinchak is a Ph.D. candidate in the Department of Sociology at the Ohio State University. He is primarily interested in how features of neighborhoods, schools, and activity spaces shape individual and community well-being. His recent work examines racial inequalities in activity space exposures, measurement of residents’ neighborhoods, activity spaces and residential segregation, and how neighborhood and school socioeconomic resources interact to shape adolescent violence.

Christopher R. Browning is an Arts and Sciences Distinguished Professor in the Department of Sociology and an affiliate of the Institute for Population Research at the Ohio State University. His research focuses on neighborhood and activity space influences on health and development, emphasizing the causes and consequences of social processes such as collective efficacy and routine activities.

Bethany Boettner is a senior research associate in the Institute for Population Research at the Ohio State University. Her research interests include adolescent health, neighborhood segregation, and the structure of shared routine activity locations.

Catherine A. Calder is a professor of statistics and chair of the Department of Statistics and Data Sciences at the University of Texas at Austin. She is also an associate director of UT Austin’s Population Research Center. Her research interests include spatial statistics, Bayesian modeling, and network analysis. Her recent work has focused on identifying community structure in ecological networks and the implications of spatial confounding in spatial generalized linear mixed models.

Jake Tarrence is currently a postdoctoral fellow with the Department of Statistics and Data Sciences at the University of Texas at Austin. His research seeks to understand the causes and consequences of social inequality/heterogeneity. His ongoing projects examine the health consequence of social mobility and racial/ethnic inequality in health and space.

Footnotes

1

Crime data for the period between 2010 and 2012 were not able to be obtained for other greater Columbus, OH, municipalities.

2

Because the data are proprietary, the response rate for this survey could not be obtained.

3

We also assessed the potential for the relationship between crime outcomes and neighborhood dog concentration to be dependent on the proportion of family households within a block group by assessing the interaction between these measures in supplemental analytic models (not shown). This interaction term was never significant however, offering no evidence of this dependence.

4

We also considered comparisons by race, but market survey respondents were not asked to report race.

5

Respondents were asked to provide four street intersections or landmarks that they “think of as the boundaries of [their] neighborhood” before being asked about perceptions of trust. To ensure consistency with the larger neighborhood social climate literature, we geocode these reports of “neighborhood” trust to respondents’ block groups of residence (Raudenbush and Sampson 1999).

6

Controlling for population density rather than total population and the square mile area yields substantively identical results.

7

Our modeling approach assumes that the errors are independent over space (i.e., no spatial autocorrelation). To assess whether this assumption holds, we calculated the value of Moran’s I, a measure of spatial autocorrelation, for the residuals from each model using the spdep package in R. The estimated Moran’s Is were always positive and statistically significant at the .05 level, with estimates ranging from .06 to .25 in models without temporally lagged dependent variables, and .07 to .12 in models controlling for temporally lagged dependent variables. These results indicate that the independent errors assumption may not be appropriate for our data. As discussed by Khan and Calder (2022), spatial regression models (i.e., regression models with spatially-correlated errors) can sometimes produce wildly misleading inferences on regression coefficients when the independent variables are spatial structured, as they are in our case. For this reason, we choose to report inferences from the non-spatial models and note the evidence of model misspecification as a (minor) weakness of our approach. To the best of our knowledge, there is no statistical solution to addressing the model misspecification concern that would produce more accurate inferences on regression coefficients than what we report.

8

In models not shown we additionally assessed whether trust mediates the association between dog concentration and crime outcomes, finding no evidence of this mediation.