Victims

With the Compliments of Springer Publishing Company, LLC

Violence and Victims

www.springerpub.com/vv

Violence and Victims, Volume 27, Number 5, 2012

Assessing the Relationship Between Alcohol Outlets and Domestic Violence: Routine Activities and the Neighborhood Environment Caterina G. Roman, PhD Temple University, Philadelphia, Pennsylvania

Shannon E. Reid, MS University of California, Irvine Studies have consistently found a positive relationship between alcohol outlet density and assault, but only a handful of studies have examined whether outlet density has an influence on domestic violence. Using a framework based in crime opportunity theories, this study estimates spatial econometric regression models to test whether the density of alcohol outlets across neighborhoods is positively associated with police calls for service for domestic violence. Models also were developed to test whether the relationships found were consistent across time periods associated with the use of alcohol outlets (weeknights and weekends). The findings indicate that off-premise outlets were associated with a significant increase in domestic violence, but on-premise outlets (specifically restaurants and nightclubs) were associated with a decrease in domestic violence. The risk for domestic violence in areas of high densities of off-premise outlets was found to be high during the weekend but not during the weeknight, suggesting different routine activities for domestic violence offenders during the week.

Keywords: spatial; ecology; opportunity; time of day

O

ver the years, researchers have been calling for comprehensive, theory-based studies of the influence of alcohol outlets on violence and injury (Freisthler, 2004; Lipton, Gorman, Wieczorek, & Gruenewald, 2003). Researchers also have stressed the importance of conducting place-based ecological studies that can shed light on the situational aspects of crime from the perspective of neighborhoods (Fagan & Davies, 2000; Sampson, 2001). In recent years, several studies have been published that seek to remedy the lack of theoretically informed and methodologically sound research on the effects of high alcohol outlet densities (e.g., see Freisthler, 2004; Gruenewald, Freisthler, Remer, LaScala, & Treno, 2006; Gruenewald & Remer, 2006; McKinney, Caetano, Harris, & Ebama, 2009). Although these and other studies have strong implications for reducing crime and victimization, for the most part, they have been limited to establishing links between alcohol outlets densities and assault (usually defined as incidents of bodily harm against a person) and have not focused on domestic violence (i.e., physical abuse,

© 2012 Springer Publishing Company http://dx.doi.org/10.1891/0886-6708.27.5.811

Copyright © Springer Publishing Company, LLC

811

812

Roman and Reid

persistent emotional abuse, or the threat of physical abuse against a domestic partner, other members of a household, or blood relative). Furthermore, even with current methodological advances, ecological studies often suffer from problems related to the use of large neighborhood units, such as census tracts or zip codes, which mask potentially important variation (McKinney et al., 2009; O’Campo, 2003). This study seeks to overcome limitations of past research by using a block grouplevel dataset that incorporates a vast array of environmental variables that have a sound theoretical basis for a possible association with domestic violence. This study is theoretically grounded within routine activities theories (Cohen & Felson, 1979) and related “opportunity” theories, such as social disorganization. Models are developed to capture a wide range of contextual variation in neighborhood places and processes. Over the years, ecological research in this vein has helped further the discussion that certain places and spaces have features that facilitate or hinder the opportunity for crime (Eck & Weisburd, 1995; Felson, 1994; Sherman, Gartin, & Buerger, 1989). Opportunity theories can be very generally categorized as theories that aim to explain variations in crime as caused by (a) the physical environment, (b) the predisposed structural dynamics of neighborhoods (social disorganization theory), and (c) victim lifestyles or the “routine activities” of people. Specifically, routine activity theory focuses on the conduct of daily activities or “routine” activities not only for the victim but also for the offender and guardian as well. In other words, the presence (or absence) of motivated offenders, potential targets, and guardians depends on the activities in which people are engaged and other characteristics of an area. Space, place, and time are important key considerations and guide the research questions for this study: (a) Is greater availability of alcohol associated with an increase in the rate of domestic violence across block groups? (b) Does the relationship between alcohol availability and domestic violence vary by type of alcohol establishment? and (c) Does the relationship between alcohol availability and domestic violence vary by time of day associated with the times and days when greater numbers of people patronize alcohol outlets? Most extant research has been conducted on large-scale geographical units, employed simple regression analysis, overlooked the problem of domestic violence, or failed to take into account the spatial structure of the data. Even when studies use sound analytic techniques, there remains a gap for research to be conducted in the United States because most recent work in this area has been limited to using data from one or two states or jurisdictions (e.g., California) or has been conducted outside of the United States (see for instance Livingston, 2010, 2011). Furthermore, only a small handful of studies exist that examines place-based variables at levels smaller than the census tract and none that examine domestic violence as the outcome. By incorporating a wide range of theoretically based contextual factors related to neighborhood places and processes and measured at the block group level, this study can contribute to a better understanding of the health risks and crime risks posed by alcoholselling establishments, essentially narrowing the long list of characteristics associated with high-crime neighborhoods. In turn, a nuanced understanding of the extent of neighborhood crime risk can provide the basis for community-based innovations targeted to both people and places. An integrated examination of space and time in the context of domestic violence has the potential to provide additional information on the nature of domestic violence in its relationship to alcohol availability and perhaps provide clues about the perpetrators.


Assessing the Relationship Between Alcohol Outlets and Domestic Violence

813

LITERATURE REVIEW The empirical literature examining alcohol outlets and violence within an ecological framework emerged in the mid-1990s. The studies at the time found that geographic factors can influence patterns of alcohol use and alcohol-related problems (Gorman, Speer, Gruenewald, & Labouvie, 2001; Scribner, MacKinnon, & Dwyer, 1994, 1995). In one of the most often cited ecological studies of the effects of alcohol availability (geographic density of outlets) on violence, Scribner et al. (1995) found that in a typical city in Los Angeles County, one additional liquor outlet was associated with 3.4 additional assaults. Other research confirms the association between outlet densities and assault (Alaniz, Parker, Gallegos, & Cartmill 1998; Gruenewald et al., 2006; Gruenewald & Remer 2006; Gyimah-Brempong, 2001; Parker & Rebhum, 1995), particularly the few research studies conducted using small units of analysis (Alaniz, Cartmill, & Parker, 2000; Costanza, Bankston, & Shihadeh, 2001; Gorman et al., 2001). Regarding domestic violence, a systematic literature review returned only three published ecological studies that examined the relationship between alcohol outlets and domestic violence. Examining municipalities in New Jersey, Gorman, Labouvie, Speer, and Subaiya (1998) found that domestic violence, measured as complaints reported to the police, was not significantly related to alcohol outlet density after controlling for the sociodemographics of neighborhoods. A national, population-based study using zip codes as the unit of analysis and survey reports of intimate partner violence found that an increase of 10 alcohol outlets per 10,000 residents was associated with a 34% increased risk of male-to-female personal violence (McKinney et al., 2009). When the authors broke down outlets into on premise versus off premise, they found, contrary to their hypotheses, the association held with on-premise outlets but not off-premise outlets. A recent Australian study (Livingston, 2010) revealed contrary findings—the author found a negative relationship between on-premise licenses (bar, restaurant, café) and domestic violence, and no relationship between packaged liquor outlets (the equivalent of off premise or carry out and liquor stores in the United States) and domestic violence. The study also found that although sociodemographics reduced the strength of the association between alcohol outlets and domestic violence, it remained strong after controls were added. However, a longitudinal study by the same author (Livingston, 2011) of 186 post codes in the greater Melbourne (Australia) area found a large positive association between packaged liquor licenses and domestic violence. The study also found small effects for pub licenses and on-premise licenses. The limited literature on this topic is surprising, given the wealth of literature showing that drinking alcohol is a risk factor for domestic violence. Furthermore, a review of published studies on drinking and domestic violence estimated that men were drinking in about 45% of the cases (estimates ranged from 6% to 57%; Roizen, 1993). A study by Caetano, Schafer, and Cunradi (2001) that examined intimate partner violence across racial groups found that rates of intimate partner violence were much higher among men who reported drinking five or more drinks per occasion at least once a week than among those who abstained from alcohol consumption. They also found that rates of intimate partner violence were two to four times higher among men with alcohol problems than among men without alcohol problems. Other research confirms these findings (Stuart et al., 2006; Walton-Moss, Mangenello, Frye, & Campbell, 2005). Although alcohol outlets have not been systematically included in ecological studies of the predictors of domestic violence, studies examining community context and domestic


814

Roman and Reid

violence are not new. Using the 2000 National Household Survey on Drug Abuse, Cunradi (2007) found that neighborhood disorder (fights, abandoned buildings, and graffiti) was significantly correlated with domestic victimization (as noted by questions about intimate partner violence). Using the National Survey of Families and Households (NSFH), Fox and Benson (2006) examined social disorganization and intimate partner violence and found that violence against women is more prevalent and more severe in socioeconomically disadvantaged neighborhoods. Another study using NSFH data (Van Wyk, Benson, Fox, & DeMaris, 2003) found that population density, percentage of single parents, percentage of nonwhite racial heterogeneity, percentage of those with low educational attainment, percentage on public assistance, percentage of those below the poverty line, and percentage of unemployed all had a significant positive relationship with domestic violence. Browning (2002), using the Chicago Health and Social Life Survey, found that female population and concentrated disadvantage were also significantly correlated with women’s self-reporting of domestic violence. In a study on femicide and social disorganization, Frye and Wilt (2001) found that lower socioeconomic status and higher community social disorganization significantly predicted intimate partner homicides.

METHODS Study Site and Unit of Analysis The study was conducted using data for District of Columbia Census block groups. The District of Columbia is a high-crime, metropolitan area that has a total area of 68.3 square miles and had an estimated 581,530 residents in 2006 (U.S. Census Bureau, 2007). The block group captures sufficient variation in the presence of alcohol-selling establishments and other independent variables hypothesized to be related to crime. Any larger level unit of analysis would mask important microlevel variation (Brantingham, Dyreson, & Brantingham, 1976; Gottfredson, 1981; Sherman et al., 1989). The District of Columbia is made up of 433 block groups as designated by the U.S. Census 2000. This study used 431 of the 433 block groups in the District. The two block groups that were excluded from the analysis consist of the National Mall (which is an open area national park) and Bolling Air Force Base. These blocks groups share police jurisdiction with other law enforcement agencies and therefore do not provide estimates of crime or provide unreliable estimates of crime. For the remaining 431 block groups in the District of Columbia, there is an average of 573 households and 1,304 residents; 20% of the residents were younger than 18 years old. Sixty percent of the residents are Black, 31% are White, and 8% are Hispanic (of any race). The average block group size is 0.14 square miles; the smallest block group is 0.02 square miles; and the largest block group is 1.87 square miles (U.S. Census Bureau, 2000).

Measures Domestic Violence. Domestic violence data were obtained from the Metropolitan Police Department (MPD) and consist of 911 calls received for domestic violence-related incidents from January 1, 2005 to December 31, 2006. There were 21,349 domestic violence calls for service that came into MPD over the 2 years and an average of 24.60 per block group. Calls were mapped using ArcMap 9.0 using a street centerline file provided by the District of Columbia’s Office of Chief Technology Officer (OCTO). The measure was



815

derived using the average of the aggregate sum of the incidents or calls for service across the 2-year time span. In addition to examining total crime spanning all days and times, to account for varying crime risk by time of day, this study divided the number of incidents and calls into two different time periods: (a) weekday nights (Monday, Tuesday, Wednesday, Thursday from 3:00 p.m. to 9:59 p.m.) and (b) the weekend (Friday 10:00 p.m. to Sunday 4:59 a.m.). The choice of these time periods was based on the data collected by the study authors on hours of alcohol outlet operation and can represent leisure activity hours more closely approximating domain-specific models within routine activities. The key premise underlying routine activities is that convergence of actors in space and time—motivated offender, suitable target, no capable guardian—represents a unique interaction. Research suggests that domain-specific models will facilitate drawing a link between opportunity measures and victimization (Garofalo, Siegel, & Laub, 1987; Gottfredson, 1984; Hoyt, Ryan, & Cauce 1999; Lynch 1987). Essentially, adding a time dimension will further facilitate an understanding of the flow of offenders and targets. Hence, the selection of these periods, as guided by routine activities, is not meant to be used for strict comparisons of outcomes across “even” time periods but to provide a framework to inform prevention and intervention strategies mirroring typical patterns of human activity. Descriptive statistics for the dependent variable (and disaggregated by time periods), as well as all independent variables, can be found in Table 1. There was a yearly average of 5.90 calls for domestic violence-related offenses on weekends and 4.71 on weeknights. In addition, for comparison purposes, hourly averages also are shown in Table 1. Alcohol Outlets. Alcohol outlet data are based on license information for all establishments registered with the Alcoholic Beverage Regulation Administration (ABRA) for 2006. These data were obtained from OCTO. In 2006, there were 1,501 licenses for businesses to sell alcohol in District of Columbia. Each license location was geocoded to the business location using ArcMap. All addresses were validated, and 100% were geocoded. Of the addresses, companies licensed as caterers (8) and wholesalers (17) were dropped from this analysis. Three additional businesses were dropped when the two block groups were deleted from the sample, leaving 1,473 alcohol-selling establishments. These outlets were then grouped by license type (a field provided in the data) into the categories of restaurant (46.4%), store (32.6%), tavern (8.3%), nightclub (4.1%), hotel (4.9%), and multipurpose facilities (3.7%). Each of these outlets was then coded by where the alcohol is consumed (on premise or off premise). Stores are the only category of license that constitutes off premise; the remaining categories fall into on-premise outlets. The size of the block group in square miles is used to create the density measures for all alcohol outlet variables. Social Disorganization or Guardianship. There are four variables used to measure social disorganization constructs consistent with the literature (Bellair, 1997; Velez, 2001; Warner & Roundtree, 1997) and derived from Census 2000 block group data. Racial/ ethnic heterogeneity is calculated by one minus the sum of squared proportions of each of the four races: Black, White, Asian, and Hispanic. Values range from zero to one, where low scores indicate blocks that are racially and ethnically homogenous, and high scores represent blocks that are more heterogeneous. Concentrated disadvantage is operationalized as an index of five Census items: (a) percentage of all households receiving public assistance, (b) percentage of population with income below the federal poverty level in 1999, (c) percentage of Black (non-Hispanic), (d) percentage of civilian population aged 16 years or older in labor force who are unemployed, and (e) percentage of households


816

TABLE 1.

Roman and Reid

Descriptive Statistics Min

Med

Skew

180

0.00

16

2.28

6.85

44

0.00

4

2.22

4.71 (0.17/hr)

5.11

38

0.00

4

2.14

On-premise outlets per square miles

24.38

68.83

650

0.00

0.00

4.51

Off-premise outlets per square miles

14.89

23.08

200

0.00

5.88

2.87

Restaurant per square miles

17.08

52.24

550

0.00

0.00

5.28

Tavern per square miles

3.30

11.81

100

0.00

0.00

4.81

Nightclub per square miles

1.04

4.66

0.00

0.00

5.83

14.89

23.08

0.00

5.88

2.87 0.797

M

SD

24.60

27.25

DV—weekends (31 hr period/week)

5.90 (0.19/hr)

DV—weeknight (28 hr period/week)

Domestic violence (DV)

Max

DV by time periods

Alcohol outlet density

Store per square miles

45.45 200

Social disorganization Concentrated disadvantage

20.004

0.795

3.030 21.150

20.060

Residential stability

0.007

0.855

1.930 22.320

20.020 20.106

Racial heterogeneity

0.267

0.188

0.760

0.000

0.250

0.499

0.195

0.134

0.974

0.000

0.159

3.04

Motivated offenders Population (18–29 year olds) Adult arrests Population density

213.27

281.88

3302.00

1.00

128.00

4.82

9.32

1.06

11.40

0.52

9.51

23.17

1328.07

1075.75

6533.33

3.70

1011.11

1.56

Physical environment Physical incivilities Parcels—commercial/ retail

0.038

0.076

0.734

0.000

0.017

5.52

Parcels—vacant

0.076

0.105

0.762

0.000

0.039

2.94

Metro stop (dummy)

0.081

0.273

1

0

0

3.077

1.80

1.04

4.32

0.00

1.79

Controls Aggregated assaults 00–01 (ln)

Note. N 5 431; no missing data.


20.12


817

with children headed by a woman. The concentrated disadvantage index is calculated as the sum of z scores for these items divided by five (the number of items). Residential stability is the sum of z scores for responses to two Census items: percentage of those living in same house since 1995 and the percentage of housing occupied by owners. The sum of these two items is then divided by two (the number of items). Population density is measured as the number of residents in the block group per square mile. The variable is logged to smooth its uneven distribution. Increased population density can be viewed as either improving guardianship (i.e., more eyes on the street) or after a certain tipping point, reducing guardianship in the sense that overcrowding increases anonymity and allows potential offenders cover. Motivated Offenders. Two variables were included to represent motivated offenders—a key element in routine activities theory: adult arrests and young adult population (Bryant & Miller, 1997; Roman, 2004). Adult arrests are the number of all arrests of adults aged 18 years and older aggregated for the calendar years 2005 and 2006. Most adults arrested (most arrests are for minor crimes, mostly drug offenses) are not incarcerated for any long periods (average time in jail after an arrest is 1 month; Cunniff, 2002; James, 2004), and hence, the variable can be viewed as representing the potential to offend again rather than representing a deterrent effect (Bernasco & Block, 2011). Young adult population represents the proportion of the population that is between the ages of 18 and 29 years within each block group as estimated by the U.S. Census 2000. Physical Place Risk. The four measures described in this section are built environment variables derived from routine activities principles that are associated with the opportunity for violence or victimization in that internal and external features of the physical environment can be conducive to or facilitate crime or the potential success for a criminal offender (Felson, 1994; Spelman, 1993). Regarding this study, the intent was to capture and control for possible environmental features related to the flow of victims and offenders. All data used to develop these variables described in the following text were provided by OCTO. Physical incivilities are operationalized using calls received by the District of Columbia Citywide Call Center (202-727-1000). The call center was designed by city administrators to be a centralized point of contact for neighborhood quality of life issues that do not need to involve the police. The calls used for this variable are calls for abandoned vehicles, graffiti removal, illegal dumping, and streetlight repair for 2005–2006. The calls were averaged over the 2-year period. Commercial/retail parcels are the percentage of parcels in 2006 within a block group that are designated retail, commercial, or motel/hotel/inn. Vacant parcels are the percentage of parcels in 2006 that are categorized as vacant and abandoned. Metro station is a dummy variable to denote whether the block group has a metro stop. Block Group-Level Control Variables. The study controls for the size of each block group in square miles because according to routine activity theory, larger block groups are hypothesized to provide more opportunity for offending. The size of the block group (in square miles), derived from 2000 Census data, is used as an offset variable to neutralize the potential impact of the different scales and the differences in the populations at risk for victimization. The study also controls for prior levels of crime by using a measure of aggravated assault incidents from January 1, 2000 to December 31, 2001. (Data for 2000 and 2001 were not available for domestic violence). Temporal lags of crime were introduced into the models as a means of capturing unobserved heterogeneity. This approach is commonly used in ecological studies of crime using a cross-sectional design (Markowitz, Bellair, Liska, & Liu, 2001; Morenoff, Sampson, & Raudenbush, 2001). The number of assaults


818

Roman and Reid

per block for those years was transformed using the natural log transformation to smooth the uneven distribution. Spatial Autocorrelation. Under the framework of routine activities, people are buying and consuming alcohol along the paths that coincide with their routine activities. As described in detail in the following text, we examined the potential for domestic violence in one block to be correlated to the domestic violence in a block group contiguous to it as an independent variable because applications of routine activity theory support its effects (Morenoff & Sampson, 1997; Roncek & Montgomery, 1995; Smith, Frazee, & Davison, 2000). Spatial autocorrelation is assessed using two different methods that coincide with the two types of modeling techniques used in this study: (a) testing for the presence of spatial autocorrelation and then upon finding it, adding a spatial lag variable as a control in our negative binomial models (i.e., a conditional spatial autocorrelation model), and (b) estimating a fully simultaneous spatial autocorrelation model to account for the joint nature of crime generation in neighboring block groups. First, using the GeoDa software (Anselin, 2003) the Moran global spatial autocorrelation test was conducted for the dependent variable. We obtained a significant and positive Moran statistic: 0.394 (p , .001) for domestic violence. To account for the presence of spatial autocorrelation, a spatial lag variable was created in GeoDa using the queen criterion. Second, spatial autocorrelation is tested for and incorporated directly within the modeling itself. As will be described in more detail in the following text, the Generalized Cross Entropy (GCE) approach (Bhati, 2008) allows one to estimate a flexible count outcome models (with overdispersion or underdispersion) and that allows for substantive spatial autocorrelation. The final model resolves the simultaneity between the observed count in one block group and its surrounding units by deriving a reduced form specification. As in the standard linear model case, the GCE reduced form captures the spatial autocorrelation in a coefficient r (Rho) that can be subject to standard statistical testing.

Analytic Strategy The distribution of domestic violence across block groups exhibits underdispersion, indicating that the variance is less than the mean. Any truncation of the distribution of the dependent variable renders ordinary least squares (OLS) estimates biased and inconsistent (Long, 1997; Tobin, 1958). Negative binomial regression models can account for the large number of zeros, particularly when overdispersion is exhibited. However, negative binomial models do not perform well when in models with underdispersed dependent variables. Given this issue, as well as the weaknesses associated with using a negative binomial framework with spatially autocorrelated data, we determined that it would be more appropriate to use an innovative spatial econometric technique that falls under the information theoretic umbrella. Models are estimated using both standard negative binomial regression and the GCE technique. We use both types of models as a way to compare the very new information theoretic models to the generally better known, but not as appropriate, negative binomial models. All GCE models are estimated using the Statistical Analysis System (SAS) software. The information theoretic approach (more specifically the GCE approach) readily handles both overdispersed or underdispersed count outcomes. Unlike more traditional approaches (including, e.g., the negative binomial model) under the GCE model, no assumptions are made about the parametric form of the mixing distribution, about the functional form of



819

the mean–variance dependence, or about the randomness stemming from a finite mixture of distributions. Rather, the restrictive dependence implied by the Poisson model is recognized and relaxed semiparametrically. The GCE model can recover both overdispersed or underdispersed random variables very accurately. Simulated evidence provided by Bhati (2008) shows that the modeling strategy is fairly robust to various types of heteroskedasticity. Moreover, given the simultaneous nature of the GCE modeling strategy, it allows one to estimate the spatial autocorrelation coefficient without the need for various transformations of the dependent variable. Comparison of the GCE models to the negative binomial regression models demonstrated the superiority of the GCE approach, and as such, only GCE models are presented.1 Four sets of models are tested. The first model adds the variables representing the different constructs in blocks to show how the results change with each block of variables. The second set of models examines the full model (with all the blocks of variables), but the dependent variable is disaggregated by the two time periods: weekends and weeknights. The third set of models examines the full model of alcohol outlets for each dependent variable but disaggregates on premise and off premise by type of outlet: stores, restaurants, taverns, and night clubs. All off-premise outlets are stores; hence, the stores measure will have the same influence as the off-premise measure. We also tested for collinearity problems, as the potential for collinearity is high in neighborhood-level studies. We calculated the variance inflation factor (VIF) defined as VIFi 5 1 / (1 2 Ri2), where Ri2 is the multiple correlation coefficient of variable Xi regressed on the remaining independent variables (Belsley, 1991). For the final regression models, we include the mean VIF and highest VIF.

RESULTS In the GCE models, a pseudo R2 measure is provided only as a rough proxy for the predictive power of the model, and we caution the readers not to compare models (whether nested or not). In addition, in the GCE models, interpretation of the size of the coefficients is not straightforward. Unlike linear models or some nonlinear models, the size of the marginal effect depends on the amount of overdispersion or underdispersion. Despite this sensitivity, the signs of the coefficient are directly interpretable. As such, only the direction of the effects of specific variables is reported and interpreted. Table 2 provides the results of the nested models examining whether alcohol outlets have an influence on domestic violence. The significance levels remain virtually unchanged as the various blocks of variables are added to the model. The density of on-premise outlets has a negative influence on domestic violence, meaning that as the density of restaurants, taverns, and nightclubs increases, the levels of domestic violence decrease. The effect is opposite (positive) for off-premise outlets: As the density of off-premise outlets increase, so too does the level of domestic violence; all other variables held constant. All variables are significant with the exception of proportion of vacant parcels. Table 3 shows the results of the two time period models for domestic violence. The significance levels of the alcohol outlet variables vary across the periods. During the weekend, off-premise alcohol outlets were positively and significantly related to domestic violence calls for service. The density of off-premise outlets does not significantly influence weeknight domestic violence. However, weeknights appear to have a cooling effect on domestic violence where there are high densities of on-premise establishments. During



— — — — — — — — — — .24***

— — — — — — — — — — .734 .39

2.003 .015

.000*** .000***

2.005 .012

.628

— — — —

— — —

.168 .034 20.357

2.215

b

.077***

SE

1.768

b

SE

.38

.027***

— — — —

— — —

.021*** .016* .069***

.000*** .001***

.110***

Model 2

.270

— — — —

21.236 .030 .628

2.006 2.050 21.286

2.002 .006

22.744

b

SE

.58

.020***

— — — —

.130*** .000*** .0178***

.020 .017*** .068***

.000*** .001***

.121***

Model 3

*p , .05. **p , .01. ***p , .001 (two-tailed tests). †Coefficients for variable have been multiplied by 100.

Pseudo Rho Overdispersion parameter

R2

Constant Alcohol outlet density On premise per square mile Off premise per square mile Social disorganization Concentrated disadvantage Residential stability Racial heterogeneity Motivated offenders 18–29 year olds Adult arrests† Population (log) Physical environment Percent vacant Physical incivilities† Metro station Parcels—commercial/retail Controls Aggregated assaults 00–01 (ln)

Parameter

Model 1

TABLE 2. Generalized Cross Entropy Results for Domestic Violence by Neighborhood Characteristics

.179

2.133 .020 2.241 1.560

21.316 .020 .520

.059 2.163 21.231

2.001 .001

21.801

b

SE

.59 .31*** .18***

.019***

.088 .0000*** .035*** .216***

.123*** .000*** .018***

.020*** .018*** .068***

.000*** .001*

.149***

Model 4

820 Roman and Reid


821

TABLE 3. Generalized Cross Entropy Results for Domestic Violence by Neighborhood Characteristics and Time Period Models Weekend Parameter

b

Constant

22.517

Weeknight

SE .309***

b

SE

23.521

.365***

2.001

.001***

Alcohol outlet density On premise per square mile

2.001

.001*

Off premise per square mile

.003

.001**

.001

.001

.058

.042*

.022

.051 .045

Social disorganization Concentrated disadvantage Residential stability

2.233

.039***

2.074


21.181

.147***

21.219

.164***

21.743

.280***

21.455

.318***

Motivated offenders 18–29 year olds Adult arrests†

.020

.00**

.030

.000***

Population (log)

.491

.037***

.552

.045***

Percent vacant

.139

.186

.276

.207

Physical incivilities†

.020

.000***

.020

.000***

2.273

.073***

2.095

1.312

.458***

1.928

.531***

.253

.040***

.270

.044***

Physical environment

Metro station Parcels—commercial/ retail

.085

Controls Aggregated assaults 00–01 (ln) R2

.55

.57

Rho

.31***

.42***

Overdispersion parameter

.17***

.048

Pseudo

*p , .05. **p , .01. ***p , .001 (two-tailed tests). †Coefficients for variable have been multiplied by 100. the weeknight and during the weekend, domestic violence is significantly and negatively related to the density of on-premise outlets. The negative relationship indicates that high densities of on-premise establishments (restaurants, bars, and nightclubs) are associated with lower levels of domestic violence. The social disorganization variables behaved similarly across the time periods, with the exception of concentrated disadvantage—the levels of concentrated disadvantage were significant predictors of domestic violence only during the weekend. For the motivated offender variables, the results are generally the same across models. The proportion of


822

Roman and Reid

18–29 years olds was negative and significant across all time periods, as was the density of the resident population (a significant positive relationship). However, adult arrests did not achieve significance in the weekend night model. The physical environment variables had a mix of positive and negative relationships (although percent vacant does not have a statistically significant association with domestic violence), and the significance level of a block group having a metro stop varied across time periods. Table 4 shows the results of the models examining whether high densities of restaurants, bars, and nightclubs independently influence domestic violence. The negative association between on-premise outlets and domestic violence appears to be driven by nightclubs and restaurants. The findings signify that areas that have greater densities of nightclubs and restaurants have lower levels of domestic violence calls for service.

DISCUSSION This study is among the first ecological studies using small units of analysis to examine whether alcohol outlet densities are associated with domestic violence while simultaneously assessing the contribution of important neighborhood structural characteristics and other factors related to crime opportunity. The study also examined whether domestic violence varied around alcohol outlets by the time of the day and week. The analyses of models regressing calls for service for domestic violence incidents on alcohol outlets, and a theory-based suite of covariates found that the relationship between alcohol outlet density and domestic violence varies by type of outlet. More specifically, off-premise outlets have a positive relationship to domestic violence calls for service, and this relationship holds strong during the weekend. In contrast, on-premise outlets—specifically restaurants and nightclubs—have a significant negative relationship with domestic violence, and this relationship held in both time periods. It is possible that the neighborhoods that have a higher density of restaurants and nightclubs are neighborhoods where there are either few residents in those neighborhoods or nearby, or, for the most part, high densities of restaurants and clubs generate a lifestyle that contributes to people from those neighborhoods being out of their homes and, therefore, not in a situation where domestic violence is likely to occur. And although the findings of this study cannot conclude that causal processes are occurring, as routine activities theory would assert, in areas of greater densities of off-premise outlets, people who are purchasing alcohol to consume in some other place may be returning to their homes nearby with their libations, increasing opportunity for drinking in the home and possibly for violence. Time of day per week appears to matter for domestic violence. Although on-premise outlets have a consistent negative relationship with domestic violence calls for service over the different time periods, the relationship between off-premise outlets and domestic violence varies across time periods; the risk of domestic violence in areas of high densities of off-premise outlets is high the weekend time block but not during the weeknight, suggesting different routine activities for domestic violence offenders during the week. This is an important finding in light of recent research, which found that the most common time range for a domestic violence incident (across all days) was from 6 p.m. to 8:59 p.m., with the second most common time range being from 9 p.m. to 10:59 p.m. (Tennessee Bureau of Investigations, 2006)—these same hours encompass our weeknight period. More research should be done to pinpoint high-risk days and times for domestic



823

TABLE 4. Generalized Cross Entropy Results for Domestic Violence by Neighborhood Characteristics and Type of Alcohol Outlet Parameter

b

SE

Constant

21.780

.149***

Restaurants per square mile

2.001

.002***

Taverns per square mile†

2.030

.001

Nightclubs per square mile

2.009

.003***

Alcohol outlet density

Stores per square mile

.001

.001*

.060

.020***

Residential stability

2.157

.018***


21.202

.069***

21.300

.124***

Adult arrests†

.020

.000***

Population (log)

.515

.018***

Percent vacant

.148

.089

Physical incivilities†

.020

.000***

2.239

.035***

1.38

.222***

Social disorganization Concentrated disadvantage

Motivated offenders 18–29 year olds

Physical environment

Metro station Parcels—commercial/retail Controls Aggregated assaults 00–01 (ln)

.187

.019***

Pseudo R2

.59

Rho

.31***

Overdispersion parameter

.18***

Mean variance inflation factor (VIF)

2.26

Highest VIF

3.35

*p  .05. **p  .01. ***p  .001 (two-tailed tests). †Coefficients for variable have been multiplied by 100. violence and to determine whether and how neighborhoods are associated with risky times. Validation of findings that weekends are especially vulnerable periods could assist with the development of new or focused strategies that incorporate targeted risk management treatment principles for both victims and offenders that would vary in their use across the days of the week.


824

Roman and Reid

Because this study is one of a few neighborhood-level studies to examine levels of domestic violence within the lens of crime attractors, we believe that it will assist in building a foundation to further conceptualize how features of the neighborhood environment might contribute to a serious public health issue. The findings imply that there might be something inherent in neighborhoods—and in particular, high densities of off-premise outlets that influences the number of calls to police for domestic violence incidents. Multilevel research, as well as longitudinal research, should be conducted to uncover the geography of alcohol purchasing patterns, consumption, and domestic violence incidents. In addition, it is important to remember that ecological studies of alcohol outlet densities such as this one are not designed to provide insight into the social processes related to alcohol use. To date, sociological research conducted at the neighborhood level generally has contributed little to our understanding of person–environment interactions regarding alcohol research. These interactions are difficult to study using extant data or field research methods. Large-scale studies are needed that model the interaction of people with certain fixed and variable environmental risks and protective elements. Small-scale field studies— and some do exist—can contribute somewhat to generating hypotheses about person–place interactions but generally are not of the size to ensure that findings are not specific to the social context under study. As sociologists and community practitioners continue their efforts to uncover methods to inform the development of neighborhood social control, multilevel research examining the microlevel conditions of neighborhoods that support and deter violence will be essential. This study should be interpreted in light of several limitations. One key limitation is the use of official police data to measure domestic violence in that calls for service for domestic violence underrepresent the true amount of domestic violence occurring—data from the National Crime Victimization Survey (NCVS) suggest that only half of the intimate partner violence against women is reported to the police (Rennison, 2000). Generally, however, research has shown that results produced using official records are roughly consistent with results using victimization data (Bastian, 1993; Blumstein, Cohen, & Rosenfeld, 1991). In addition, the 911 call data do not specify any details about the characteristics of the victims or offenders or the relationship itself, leaving this study unable to gauge possible important factors related to who is at risk regarding both offending and victimization. Also, census data were not available at the block level for the year under study (2006), and hence, there is the possibility that some of variables used in the study are not precise measures. Although the typical econometric issues of serial error correlation and spatial autocorrelation were addressed using the GCE models, we cannot overcome the limitations associated with cross-sectional data. It is possible that reverse causation may be operating. In cases of reverse causation or simultaneous equation bias, the regression estimates will suffer from bias (either upward or downward). Hence, we cannot assert causal mechanisms from our findings. Finally, generalizability of findings is limited, but not surprisingly, generalizability in this area of research is very difficult. There may be neighborhood mechanisms specific to the geographic area of study, resulting in very different contexts—contexts that differentially influence findings—across studies. Even when taking into account these limitations, this study and other recent alcohol outlet density studies on domestic violence (Livingston, 2010, 2011; McKinney et al., 2009) provide a foundation from which to continue to hypothesize about the dynamics between and among alcohol use, neighborhood socioeconomic features, social organization, and the built environment. Most published studies examining the influence of alcohol outlets on crime has focused on how outlets influence assault, as opposed to domestic violence. We hope that this



825

study provides evidence of the need to continue to explore the multidimensional effects of alcohol availability on neighborhoods and specifically regarding domestic violence. Given that the density of off-premise outlets was associated with domestic violence, the results call for a coordinated dialog across policy areas (health and public health, business enterprise, crime, etc.) that yields targeted community-based strategies to reduce alcohol availability as well as person-based strategies to reduce consumption.

NOTE 1. Results for comparisons between model types can be obtained directly from the lead author.

REFERENCES Alaniz, M.L., Cartmill, R.S., & Parker, R.N. (2000). Community-identified alcohol issues in the Mexican American community. Substance Use and Misuse, 35, 157–69. Alaniz, M. L., Parker, R. N., Gallegos, A., & Cartmill, R. S. (1998). Immigrants and violence: The importance of neighborhood context. Hispanic Journal of Behavioral Sciences, 20, 155–174. Anselin, L. (2003). GeoDa 0.9 user’s guide. Urbana-Champaign, IL: Spatial Analysis Laboratory, University of Illinois. Bastian, L. D. (1993). Criminal victimization 1992. Washington, DC: Bureau of Justice Statistics. Bellair, P. E. (1997). Social interaction and community crime: Examining the importance of neighborhood networks. Criminology, 35, 677–703. Belsley, D. A. (1991). Conditioning diagnostics. Collinearity and weak data in regression, Wiley series in probability and mathematical statistics. NewYork, NY: Wiley. Bernasco, W., & Block, R. (2011). Robberies in Chicago: A block-level analysis of the influence of crime generators, crime attractors, and offender anchor points. Journal of Research in Crime and Delinquency, 48, 33–57. Bhati, A. S. (2008). A generalized cross-entropy approach for modeling spatially correlated counts. Econometric Reviews, 27, 574–595. Blumstein, A., Cohen, J., & Rosenfeld, R. (1991). Trend and deviation in crime rates: A comparison of UCR and NCS data for robbery and burglary. Criminology, 29(2), 237–263. Brantingham, P. J., Dyreson, D. A., & Brantingham, P. L. (1976). Crime seen through a cone of resolution. American Behavioral Scientist, 20, 261–273. Browning, C. R. (2002). The span of collective efficacy: Extending social disorganization theory to partner violence. Journal of Marriage and Family, 64(4), 833–850. Bryant, K. M., & Miller, J. M. (1997). Routine activity and labor market segmentation: An empirical test of a revised approach. American Journal of Criminal Justice, 22, 71–100. Caetano, R., Schaffer, J., & Cunradi, C. (2001). Alcohol-related intimate partner violence among White, Black and Hispanic couples in the U.S. Alcohol Research and Health, 25, 58–65. Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588–605. Costanza, S. E., Bankston, W. B., & Shihadeh, E. (2001). Alcohol availability and violent crime: A spatial analysis. Journal of Crime and Justice, 24(1), 71–83. Cunniff, M. A. (2002). Jail crowding: Understanding jail population dynamics. Washington, DC: U.S. Department of Justice, National Institute of Corrections. Cunradi, C. B. (2007). Drinking level, neighborhood social disorder, and mutual intimate partner violence. Alcoholism: Clinical and Experimental Research, 31(6), 1012–1019. Eck, J. E., & Weisburd, D. (Eds.). (1995). Crime places in crime theory. In Crime and place. Crime prevention studies (Vol. 4, pp. 1–34). Monsey, NY: Criminal Justice Press.


826

Roman and Reid

Fagan, J., & Davies, G. (2000). Crime in public housing: Two-way diffusion effects in surrounding neighborhoods. In V. Goldsmith, P. G. McGuire, J. H. Mollenkopf, & T. A. Ross (Eds.), Analyzing crime patterns: Frontiers of practice (pp. 121–136). Thousand Oaks, CA: Sage. Felson, M. (1994). Crime and everyday life: Insights and implications for society. Thousand Oaks, CA: Pine Forge Press. Fox, G. L., & Benson, M. L. (2006). Household and neighborhood contexts of intimate partner violence. Public Health Reports, 121, 419–427. Freisthler, B. (2004). A spatial analysis of social disorganization, alcohol access, and rates of child maltreatment in neighborhoods. Child and Youth Services Review, 26, 803–819. Frye, V., & Wilt, S. (2001). Femicide and social disorganization. Violence Against Women, 7, 335–351. Garofalo, J., Siegel, L., & Laub, J. (1987). School-related victimization among adolescents: An analysis of National Crime Survey (NCS) narratives. Journal of Quantitative Criminology, 3, 321–338. Gorman, D. M., Labouvie, E. W., Speer, P. W., & Subaiya, A. (1998). Alcohol availability and domestic violence. American Journal of Drug and Alcohol Abuse, 24, 661–72. Gorman, D. M., Speer, P. W., Gruenewald, P.J., & Labouvie, E. W. (2001). Spatial dynamics of alcohol availability, neighborhood structure and violent crime. Journal of Studies on Alcohol, 62, 628–636. Gottfredson, M. R. (1981). On the etiology of criminal victimization. Journal of Criminal Law and Criminology, 72, 712–726. Gottfredson, M. R. (1984). Victims of crime: The dimensions of risk. London, United Kingdom: Her Majesty’s Stationery Office. Gruenewald, P. J., Freisthler, B., Remer, L., LaScala, E. A., & Treno, A. (2006). Ecological models of alcohol outlets and violent assaults: Crime potentials and geospatial analysis. Addiction, 5(101), 666–677. Gruenewald, P. J., & Remer, L. (2006). Changes in outlet densities affect violence rates. Alcoholism: Clinical and Experimental Research, 30(7), 1184–1193. Gyimah-Brempong, K. (2001). Alcohol availability and crime: Evidence from census tract data. Southern Economic Journal, 68(1), 2–21. Hoyt, D. R., Ryan, K., & Cauce, A. M. (1999). Personal victimization in a high risk environment: Homeless and runaway adolescents. Research in Crime and Delinquency, 36, 371–391. James, D. J. (2004). Profile of jail inmates. Bureau of Justice Statistics special report (NCJ 201932). Washington, DC: U.S. Department of Justice. Lipton, R., Gorman, D. M., Wieczorek, W. F., & Gruenewald, P. (2003). The application of spatial analysis to the public health understanding of alcohol and alcohol-related problems. In O. A. Khan & R. Skinner (Eds.), Geographic information systems and health applications (pp. 57–79). Hershey, PA: Idea Group. Livingston, M. (2010). The ecology of domestic violence—the role of alcohol outlet density. Geospatial Health, 5, 139–149. Livingston, M. (2011). A longitudinal analysis of alcohol outlet density and domestic violence. Addiction, 106, 919–925. Long, J. S. (1997). Regression models for categorical and limited dependent variables. Thousand Oaks, CA: Sage. Lynch, J. P. (1987). Routine activity and victimization at work. Journal of Quantitative Criminology 3, 283–300. Markowitz, F. E., Bellair, P. E., Liska, A. E., & Liu, J. (2001). Extending social disorganization theory: Modeling the relationships between cohesion, disorder and fear. Criminology, 39, 293–320. McKinney, C. M., Caetano, R., Harris, T. R., & Ebama, M. S. (2009). Alcohol availability and intimate partner violence among US couples. Alcoholism: Clinical & Experimental Research, 33,169–176.



827

Morenoff, J., & Sampson, R. J. (1997). Violent crime and the spatial dynamics of neighborhood transition: Chicago, 1970–1990. Social Forces, 76, 31–64. Morenoff, J., Sampson, R. J., & Raudenbush, S. (2001). Neighborhood inequality, collective efficacy, and the spatial dynamics of urban violence. Criminology, 39, 517–560. O’Campo, P. (2003). Invited commentary: Advancing theory and methods for multilevel models of residential neighborhoods and health. American Journal of Epidemiology, 157, 9–13. Parker, R. N., & Rebhum, L. A. (1995). Alcohol and homicide: A deadly combination of two American traditions. Albany, NY: State University of New York Press. Rennison, C. M. (2000). Intimate partner violence. Bureau of Justice Statistics special report (NCJ 178247). Washington, DC: U.S. Department of Justice. Roizen, J. (1993). Issues in the epidemiology of alcohol and violence. In S. E. Martin (Ed.), Alcohol and interpersonal violence: Fostering multidisciplinary perspectives, NIAAA Research Monograph No. 24. Bethesda, MD: National Institute on Alcohol Abuse and Alcoholism. Roman, C. G. (2004). Schools, neighborhoods, and violence: Crime within the daily routines of youth. Lanham, MD: Lexington Books. Roncek, D. W., & Montgomery, A. (1995). Spatial autocorrelation revisited: Conceptual underpinnings and practical guidelines for the use of the generalized potential as a remedy for spatial autocorrelation in large samples. In C. R. Block, M. Dabdoub, & S. Fregly (Eds.), Crime analysis through computer mapping (pp. 99–110). Washington, DC: Police Executive Research Forum. Sampson, R. J. (2001). How do communities undergird or undermine human development? What are the relevant contexts and what mechanisms are at work? In A. Booth & A. C. Crouter (Eds.), Does it take a village? Community effects on children, adolescents, and families (pp. 3–30). University Park, PA: Pennsylvania State University Press. Scribner, R. A., MacKinnon, D. P., & Dwyer, J. H. (1994). Alcohol outlet density and motor vehicle crashes in Los Angeles County cities. Journal of Studies on Alcohol, 55, 447–453. Scribner, R. A., MacKinnon, D. P., & Dwyer, J. H. (1995). Risk of assaultive violence and alcohol availability in Los Angeles County. American Journal of Public Health, 85(3), 335–340. Sherman, L., Gartin, P., & Buerger, M. (1989). Hot spots of predatory crime: Routine activities and the criminology of place. Criminology, 27, 27–56. Smith, W. R., Frazee, S. G., & Davison, E. L. (2000). Furthering the integration of routine activity and social disorganization theories: Small units of analysis and the study of street robbery as a diffusion process. Criminology, 38(2), 489–523. Spelman, W. (1993). Abandoned buildings: Magnets for crime? Journal of Criminal Justice, 21, 481–495. Stuart, G. L., Meehan, J. C., Moore, T. M., Morean, M., Hellmuth, J., & Follansbee, K. (2006). Examining a conceptual framework of intimate partner violence in men and women arrested for domestic violence. Journal of Studies on Alcohol, 67, 102–112. Tennessee Bureau of Investigations. (2006). Domestic violence research brief. Retrieved from http:// www.tbi.state.tn.us/divisions/isd_csu_tibrs.htm Tobin, J. (1958). Estimation of relationships for limited dependent variables. Econometrica, 26, 24–36. U.S. Census Bureau. (2000). Census 2000 Summary File 4: Households and families: 2000. Retrieved from http://factfinder.census.gov U.S. Census Bureau. (2007). Annual county population estimates and estimated components of change: April 1, 2000 to July 1, 2006. (CO-EST2006-alldata) (CSV). 2006 Population Estimates. Washington, DC: U.S. Census Bureau, Population Division. Van Wyk, J., Benson, M. L., Fox, G. L., & Demaris, A. (2003). Neighbourhood disadvantage, individual economic distress and violence against women in intimate relationships. Journal of Quantitative Criminology, 19(3), 207–235. Velez, M. (2001). The role of public social control in urban neighborhoods: A multilevel analysis of victimization risk. Criminology, 39, 837–866.


828

Roman and Reid

Walton-Moss, B., Manganello, J., Frye, V., & Campbell, J. (2005). Risk factors for intimate partner violence and associated injury among urban women. Journal of Community Health, 30(5), 377–389. Warner, B., & Roundtree, P. (1997). Local social ties in a community and crime model: Questioning the systemic nature of informal social control. Social Problems, 44, 520–536. Wilcox, P., Quisenberry, N., Cabrera, D. T., & Jones, S. (2004). Busy places and broken windows? Toward defining the role of physical structure and process in community crime models. Sociological Quarterly, 45, 185–207. Correspondence regarding this article should be directed to Caterina G. Roman, PhD, Department of Criminal Justice, Temple University, 1115 Polett Walk, Gladfelter Hall, 5th Floor, Philadelphia, PA 19122. E-mail: [email protected]
