The Relationship Between Neighborhood Alcohol Outlet Density and Youth Violence: a Systematic

THE RELATIONSHIP BETWEEN NEIGHBORHOOD ALCOHOL OUTLET DENSITY AND YOUTH VIOLENCE: A SYSTEMATIC REVIEW

by

Michelle Mellers

BS, Cornell University, 2007

Submitted to the Graduate Faculty of

Department of Epidemiology

Graduate School of Public Health in partial fulfillment

of the requirements for the degree of

Master of Public Health

University of Pittsburgh

2014

UNIVERISTY OF PITTSBURGH

GRADUATE SCHOOL OF PUBLIC HEALTH

This essay is submitted

by

Michelle Mellers

on

April 25, 2014

Essay Advisor:

Anthony Fabio, PhD, MPH _____________________________________

Assistant Professor

Department of Epidemiology

Graduate School of Public Health

University of Pittsburgh

Essay Reader:

Ravi Sharma, PhD _____________________________________

Assistant Professor

Behavioral and Community Health Sciences

Graduate School of Public Health

University of Pittsburgh

Copyright © by Michelle Mellers

2014

Anthony Fabio, PHD, MPH

THE REATIONSHIP BETWEEN NEIGHBORHOOD ALCOHOL OUTLET DENSITY AND YOUTH VIOLENCE: A SYSTEMATIC REVIEW

Michelle Mellers, MPH

University of Pittsburgh, 2014

ABSTRACT

Background: Violence continues to be a major public health problem in the United States. Access to alcohol has been found to cause harmful behaviors such as violence, so it has been hypothesized that higher alcohol outlet density is associated with higher rates of violence. However, the results of studies assessing this association are inconsistent. In particular, the results vary by study design, type of alcohol outlet, and severity of violence. In this review, we aim to review the literature and assess whether levels of alcohol outlet density are related to neighborhood violence.

Methods: We conducted a literature search on OVID using keywords that were related to “alcohol outlet density” or “violence”. We defined alcohol outlet density as any type of distribution center for alcohol (off-premise, on-premise, restaurant, bar, etc.) in a unit area. We defined violence as any type of violence as defined in ICD-9 or police crime statistics reports such as homicide or assault. We excluded articles that focused on: intimate partner violence, LGBT violence, or violence concentrated in a college setting.

Results: Using our inclusion/exclusion criteria we found 41 articles. The first article we found was published in 1981 and looked at violence in Cleveland, OH in 1970. The most recent articles were three articles published in 2013. The early articles tended to use linear regression and models with few covariates and later papers tended to use Bayesian statistics with more covariates. Most of the articles tended to use small spatial units such as census tracts or block groups. The articles reported finding different effect sizes with some reporting finding no effect and others finding a large effect, results varied by off vs. on premises outlets as well as severity of violence.

Conclusion: We found that the articles do not provide clear evidence of an association between AOD and violence. The replicability between the studies was low and the results of the studies are too varied to draw a conclusion. We found that some of this difference may be due to methodological weaknesses. Future research should differentiate between types of alcohol outlets and severity of violence.

TABLE OF CONTENTS

1.0 INTRODUCTION 1

2.0 METHODS 2

3.0 RESULTS 4

3.1 Study Type: Ecologic, published 1999 and earlier 4

3.2 Study Type: Ecologic, published 2000 to 2004 7

3.3 Study Type: Ecologic, published 2005 to 2009 9

3.4 Study Type: Ecologic, published 2010 and later 13

3.5 Study Type: Multi-Level 17

3.6 Study Type: Case-Control 19

4.0 DISCUSSION 20

4.1 Review of Theory Used 20

4.2 Misidentification of the Appropriate Geographic Level 21

4.3 Use of Subject Home address vs. Incident address 25

4.4 Selection and Sampling Bias 29

4.5 Residual Bias 30

4.6 Small Effect Sizes 34

4.7 Interpretation of Study Results 36

4.8 Off vs. On Premises Outlets 38

4.9 Violence Severity: Minor Violence vs. Severe Violence 40

5.0 CONCLUSIONS 44

5.1 Policy Implications 44

5.2 Future Work and Final Remarks 45

APPENDIX: TABLES 47

BIBLIOGRAPHY 59

LIST OF TABLES

Table 1: Summaries of Studies Investigating Alcohol Outlet Density and Violence 47

Table 2: Summary of outcome by outlet type for all types of violence 51

Table 3: Summary of association by alcohol outlet type and homicide 54

Table 4: Summary of association by alcohol outlet type and violent crime (murder, forcible rape, robbery, and aggravated assault and other combined crime outcomes 55

Table 5: Outcome association by alcohol outlet type and assaults using incident address 57

Table 6: Outcome association by alcohol outlet type and minor violence 58

LIST OF FIGURES

FIGURE 1: ARTICLE SELECTION AND REVIEW PROCESS FOR ALCOHOL OUTLET DENSITY AND VIOLENCE 2

viii

1.0 INTRODUCTION

Since alcohol has a history of causing harmful behaviors such as violence, it is thought that controlling access to alcohol can affect violence rates in a neighborhood.(1) One way that may control access to alcohol for an individual is to change the alcohol outlet density (AOD) that the individual is exposed to. Alaniz et al. describes the effect as: “In astronomy, a great attractor is an immense region in the universe of known space so full of matter that all other galaxy clusters and individual galaxies are drawn toward the attractor by the physical force of gravity. In this approach to understanding the outlet density-violence relationship, places with outlet concentrations are social great attractors that magnetically draw young people to these locations.”(2)

Because of the large number of alcohol outlets in neighborhoods, policy makers potentially can have a big effect on violence by controlling AOD in neighborhoods. We aim to review the literature to find if we can find a clear association for any individual type of alcohol outlet or all types of alcohol outlets are related to violence. After conducting a systematic review, we found 41 studies.

We found that the results of published studies are inconsistent, yield small effects sizes, and do not always assess an appropriates geographic level or address (e.g., incident location vs. home address) that fits within reasoned etiological theories.

2.0 METHODS

We conducted a search of OVID Medline in November 2013 for peer-reviewed articles that included keywords related to “alcohol outlet density” and “violence.” For a complete list of keywords used, see Figure 1. Our outcome of interest is violence, specifically youth violence, which we defined as homicide or assault in youth age 10-24. We found keywords related to violence by using words in our search to capture this definition. In our keywords, we did not include any terms related to youth because we found that such terms would needlessly restrict our search. The exposure we are interested in is alcohol outlet density which we defined as any type of distribution center for alcohol (off-premise, on-premise, restaurant, bar, etc) whose position is defined in an area and from this position the authors calculated a density. We defined off-premise alcohol outlets as any type of retail establishment where any type of alcoholic beverage can be purchased to be consumed elsewhere. We defined on-premise outlets as any type of retail establishment where alcoholic beverages can be purchased to be consumed on-site. Bars are a type of on-premise outlet where food service is limited and serving alcohol is the primary purpose of the establishment. Restaurants are another type of on-premise outlet where business is concentrated on food and alcoholic beverages are secondary. Using these definitions, we were able to generate keywords related to AOD. Using these OVID search terms we found 57 articles. From these articles, we reviewed citations and found an additional 33 articles that met our criteria for further review.

We next subjected these 90 articles to more stringent inclusion/exclusion criteria to find the articles included for our review. Because we are interested in controlling alcohol and not eliminating alcohol, we excluded articles that studied prohibition of alcohol. We excluded studies outside of the United States because violence rates in the United States are much higher than other developed countries such as Australia or European Union countries(3). The effect of changing AOD may not be robust in countries with different violence rates so we chose to only include studies conducted inside the United States. We excluded all articles not published in English. We also excluded articles focused primarily on intimate partner violence (IPV) and LGBT violence, because research finds that IPV and LGBT may be a qualitatively different form of violence. Since the AOD and violence relationship may be different in a college setting, we excluded articles that focus on college students around colleges. We did not exclude articles that included college students along with other youths in the community as long as the study took place in the general community and not exclusively in a college setting. Since we are interested the relationship between any type of alcohol outlet and violence, we did not exclude any article based on which alcohol outlet types were included as long as a density was calculated connected with an outlet type. After applying these inclusion and exclusion criteria, a total of 41 articles were deemed relevant for inclusion in the current review.

3.0 RESULTS

3.1 Study Type: Ecologic, published 1999 and earlier

The first study looking at the link between AOD and violence was published by Roncek and Bell in 1981.(4) The authors looked at bars and violent crime (murders, rapes, assaults, and robberies) in Cleveland, OH in 1970. The authors found that an additional bar per city block would lead to an additional 1.2 incidents of violent crime per year (b=1.209, p<0.05).

Roncek and Pravatiner performed a similar study except they looked at San Diego, CA in 1970.(5) The authors again looked at bars, but found that only 2.2% of blocks studied had at least one bar on the block vs. 14.2% of blocks had bars in Cleveland. Using multivariate regression analysis the authors found that the increase of a single bar on the block would result in an increase of 0.6 violent crimes per year. (b=0.6072, b<0.05).

Roncek and Maier expanded the Roncek and Bell study by looking at on-premise outlets in Cleveland in 1979 to 1981.(6) The authors accounted for the effect of areas around a neighborhood on the neighborhood itself by including “population potential”, the population density in surrounding blocks, and “crime potential”, the density of crimes, as covariates. The authors found that an increase of one off-premise outlet in a block was significantly associated with an increase of 0.940 violent crime incidents per year (b=0.940, p<0.05), an increase of 0.027 homicides per year (b=0.027, p<0.05), and an increase of 0.336 assaults per year (b=0.336, p<0.05).

Scribner et al. looked at 74 contiguous cities within Los Angeles County with an average size of 50,000 people.(7) Using multivariate regression, the authors found that a one percent increase of AOD in an average Los Angeles city of 50,000 people was significantly associated with an increase of 0.62% in the rate of yearly assaultive violence (b=0.62, p<0.01). An increase of on-sale alcohol outlet density by one percent was associated with a 0.36% increase in assaultive violence (p<0.01) and an increase of density of all types of alcohol outlets by one percent was associated with an increase of 0.56% of assaultive violence (p<0.05).

Unlike the Scribner et al., Alaniz et al. concentrated on smaller cities in Northern California with a higher Latino and immigrant population.(2) Alaniz et al. chose three smaller cities in CA in different geographic areas: rural, suburban, and urban. To analyze their data, Alaniz et al. decided to compare two analytic strategies: a custom coded matrix regression program and a spatial model. Alaniz et al. found that an increase of one off-premise alcohol outlet per 1,000 people in a census tract was associated with 0.398 (p<0.05) additional violent crime incidents per 1,000 in the youth population (ages 15-24).

Gorman et al. attempted to repeat the Scribner et al. study above by looking at 223 municipalities in New Jersey.(8) The authors only included violence data from the summer months in 1993 and 1994. Gorman et al. were not able to replicate the Scribner et al. results and did not find that all types of alcohol outlets were significantly associated with violent crime (b=0.05, p>0.05).

Speer et al. aimed to answer why Scribner et al.’s results were not replicated by Gorman et al.(9) Speer et al. looked at census blocks and census tracts in Nework, NJ. Using bivariate regression analysis, the authors found that increasing AOD by 1% was significantly related to an increase in violence of 1.10% in census tracts (b=1.10, p<0.0001) and an increase of 1.29% in census block groups (b=1.29, p<0.0001).

Scribner et al. looked at the effect of different AOD units of exposure, alcohol outlets per square mile and alcohol outlets per person, and youth homicide.(10) To understand these two different exposures the authors used 155 census tracts in New Orleans. The authors found off-premise outlets were significantly related to increased homicide (outlets per sq. mile: b=0.211, p<0.05; outlets per 1000 people: b=0.244, p<0.05), while on-premise outlets (outlets per sq. mile: b=0.001, p>0.05; outlets per 1000 people: b=0.011, p>0.05) and total alcohol outlets (outlets per sq. mile: b=0.098, p>0.05; outlets per 1000 people: b=0.144, p>0.05) were not statistically significantly related to homicide.

3.2 Study Type: Ecologic, published 2000 to 2004

Peterson et al. looked only at the effect of bars on different types of violent crime in Columbus, Ohio in 1990.(11) Peterson et al. included the effect of another neighborhood gathering place in their model by including recreation centers. Using ordinary least squares regression, the authors found that bars were statistically significantly associated with higher levels of violent crime (b=1.290, p<0.05), homicide (b=0.042, b<0.05), and aggravated assault (b=0.419, p<0.05).

Costanza et al. looked for an association between off-premise outlets and bars (taverns) and assault in the 278 census tracts from 1989 to 1991 in Baton Rouge, LA.(12) The authors found that higher density of off-premise outlets per 100 households significantly increased the arrest-rate for assault (b=56.37, p=0.025). Bars per 100 households were not significantly associated with assault (b=0.7.7, p>0.05).

Gorman et al. aimed to expand and correct limitations of their earlier study published in 1998.(13) Gorman et al. corrected their earlier study by including crime data for the entire year rather than just the summer months, expanding the number of neighborhood structural covariates, and accounting for spatial autocorrelation in the analysis. Gorman et al. used block groups in Camden, New Jersey. Using a spatial model, the authors found that all types of alcohol outlets were significantly associated with violent crime (b=1.303, p<0.001) .

Using data from Detroit, MI, Gyimah-Brempong looked for an effect between all types of alcohol outlets and violent crime and homicide.(14) The author accounted for other retail establishments in the census tracts by including gas stations in their model. Using an instrumental variable estimation technique, the author found that a 10% increase in the number of alcohol outlets in a census tract was significantly associated with an increase of 8.2% in the violent crime rate (p<0.01). Using an instrumental variable estimation technique, the author found that a 10% increase in the number of alcohol outlets in a census tract was significantly associated with an increase of 1.2% in the homicide rate (p<0.01).