Table 1. Descriptive Statistics for All Model Variables 9

ALCOHOL CONSUMPTION AS A MEDIATOR BETWEEN ALCOHOL

OUTLET DENSITY AND TRAUMATIC INJURY

Emily K. Landis

BA, University of Pittsburgh, 2014

Submitted to the Graduate Faculty of

Epidemiology

Graduate School of Public Health in partial fulfillment

of the requirements for the degree of

Master of Public Health

University of Pittsburgh

2016 UNIVERSITY OF PITTSBURGH

GRADUATE SCHOOL OF PUBLIC HEALTH

This essay is submitted

Emily K. Landis

April 27, 2016

and approved by

Essay Advisor: Antony Fabio, PhD ______Assistant Professor Department of Epidemiology Graduate School of Public Health University of Pittsburgh

Essay Reader: J. Jeffrey Inman ______Associate Dean for Research and Faculty Department of Marketing and Business Economics Katz School of Business University of Pittsburgh

iii Antony Fabio, PhD

ALCOHOL CONSUMPTION AS A MEDIATOR BETWEEN ALCOHOL

OUTLET DENSITY AND INJURY

Emily K. Landis, MPH

University of Pittsburgh, 2016 ABSTRACT

Building on previous research related to alcohol outlet density and its impact on injury, this study examines whether alcohol consumption mediates the association between alcohol outlet density and traumatic injury. Data were analyzed based on Nielsen consumption variables along with injury data for census tracts in the city of Pittsburgh, Pennsylvania. Mediation analysis showed that there was no significant indirect effect of consumption as a mediator, although Pearson regression analysis showed that there was a significant association between on premise alcohol density, consumption of alcohol, and alcohol related traumatic injury. This study shows public health significance as it suggests an alternative mechanism leading to alcohol related traumatic injury so that prevention can be focused on limiting alcohol consumption instead of establishments that sell alcohol.

iv TABLE OF CONTENTS

1.0 INTRODUCTION...... 1

2.0 METHODS...... 4

3.0 RESULTS...... 7

4.0 DISCUSSION...... 12

BIBLIOGRAPHY...... 17

v LIST OF TABLES

Table 1. Descriptive statistics for all model variables...... 9

Table 2.Pearson regression correlation models related to traumatic injury (*p<0.5)...... 10

Table 3. Pearson regression correlation models related to traumatic injury (*p<.05)...... 11

vi LIST OF FIGURES

Figure 1. Pathway coefficients for mediation model related to on premise alcohol outlet density

(*p<.05)...... 9

Figure 2. Pathway coefficients for mediation model related to off premise alcohol outlet density

(*p<.05)...... 10

vii 1.0 INTRODUCTION

Alcohol has been shown to have negative effects on perception and judgment that can lead to a flawed ability to interpret the actions of others18. Over half (about 63.9%) of all violence related injuries studied across fourteen countries was attributed to alcohol use, and globally in 2012,

5.1% of all injury and disease was attributed to alcohol [1, 2]. In addition, 13.2% of disability adjusted life years related to injury have been attributed to alcohol use3. Many studies have shown a correlation between alcohol availability as a determinant for increased incidence of violence, and injury4.Alcohol consumption is thought to increase in areas where there are more available outlets. Evidence suggests that when the number of outlets within a 1-km network distance is eight or more, the risk of drinking at levels associated with short-term harm is increased25. It was found in another study that a 10% decrease in alcohol outlet density (AOD) would decrease wine consumption by 4% and other alcohol from between 1 and 3%5.In Finland, it was found that after a new law allowing the sale of more alcohol was enacted, there was a 22% increase in alcohol outlets and a subsequent increase in alcohol consumption by 56%9. In another study done among college students at eight different colleges, significant associations were found between AOD within a two mile radius and frequent and heavy drinking, along with drinking related problems even among the most experienced binge drinkers and least experienced drinkers19. Other evidence from the California 50 city study shows greater drinking frequencies being associated with greater use in home, whereas greater drinking volumes were

1 associated with greater use outside the home. This result supports the notion that greater proportions of bars among on-premise establishments are related to greater drinking frequencies, quantities, heavy drinking and volumes used22. Outlet density affects alcohol consumption by changing the physical aspect of access by decreasing the proximity to alcohol or how far someone has to travel home after drinking11. With each added alcohol outlet there is also the potential for increased business competition, which leads to reduced prices and more consumption overall. Increased clusters of outlets can have an amenity affect, where large amounts of people are drawn to a more commercialized area, leading to more consumption11.

More importantly, a link to injury has been established based on alcohol availability, where intentional and unintentional injuries were found to be more common in areas of off premise alcohol outlets, such as convenience, grocery, and liquor stores. On-premise outlets where there is alcohol available, such as bars and restaurants, can lead to increased socialization among at- risk drinkers, especially young adult men that are likely to consume more in social situations than at home[10,11]. One study found that 1.09 unintentional injuries and 0.28 intentional injuries were attributed to one added off premise outlet per year6.In a longitudinal study by Gruenewald that looked at outlet density and assault, it was found that by decreasing the number of outlets by just one in each of the studied postal codes over six years, it would prevent 290 assaults, which amounts to an average of 48 per year7, which is a small actual effect size. Another study looking at the relationship between alcohol availability and injury using a survey from a community trial found after that self-reported injury was significantly associated with both on and off premise

AOD8. A study done in Australia linked 0.8% increase in alcohol related injuries presenting to the emergency room to one on-site alcohol premise in an area of outlet clustering, along with having the one of the highest levels of alcohol related violence in the area12. Further evidence to

2 support AOD related to injury involves a longitudinal study done over a nine year period in 186 postal codes in Australia that found changes in density was associated with increased violence, specifically the addition of one on-premise liquor license accounted for an increase in 0.25 assaults per year23.

Linking alcohol consumption, outlet density, and traumatic injury suggests that the greater the alcohol availability leads to more consumption and produces a larger number of alcohol related problems, specifically, traumatic injury. However, there are considerable variations in the literature related to on and off premise outlet density and harms, potentially due to location and type of outlet[12-16].Many studies have only shown significance when relating injury and AOD when controlling for demographic factors such as race, age, and neighborhood characteristics[8,17]. Even when studies show significant associations, such as some examples mentioned above, the effect size is very small and is not considered to be a substantial increase in injuries attributed to added outlets [6, 7, 12]. In addition, many confounders are not adequately adjusted for, making the effect size shown even smaller11. A great limitation in the literature illustrates that there have been no studies that analyze alcohol consumption, AOD, and injury together in the same study. It is apparent that AOD does not present itself to be the biggest indicator of injury, meaning that consumption is accounting for the association between AOD and injury. Based on these assumptions, this study hypothesizes that alcohol consumption is a mediator between alcohol outlet density and traumatic injury.

3 2.0 METHODS

This study used data from Pennsylvania Department of Health, Bureau of Emergency Medical

Services (EMS) to determine total traumatic injury aggregated by zip code in Pennsylvania for the years 2013 and 2014. This dataset was based on the EMS provider’s primary impression upon arrival at the scene of the incident. It also examines total traumatic injury in Pennsylvania based on whether the patient had smell of alcohol on breath or the patient admits to alcohol use.

All 42 zip codes were determined based on their location in the City of Pittsburgh and including five surrounding areas (Baldwin, Bethel Park, Upper Saint Clair, Mount Lebanon, and

Whitehall) to increase variability. To decrease the likelihood of gathering personal identifying information, incident zip codes with less than five cases were listed as such and did not specify individual numbers of cases. For this reason, those listed as 5 and under were classified as “5” to aid in quantitative analysis. In order to determine the measurement for alcohol consumption,

Nielsen consumption panel data from the University of Chicago Kilts Data Center graduate school of business was used for years 2013 and 201430. The alcohol consumption variable was calculated using Nielsen Panel Data, which is a survey of 40,000-60,000 U.S. households in 52 metropolitan areas that focus on consumption of household products that have Universal Product

Codes (UPC), or unique barcodes that specify a 12 digit code that is unique for a certain product and brand. Scanners are given to participants to record purchases from any outlet that are intended for in-home, personal use. The data is from 2013 and 2014, where about 80% of

4 households are retained from year to year. All zip codes were matched to the EMS data and UPC codes were gathered for all liquor, beer, wine, and cooler purchases, including liquid ounces per household, which was aggregated into the consumption variable. Because the injury data given by the Pennsylvania Department of Health was available on the zip code level, it was essential to link those to census tracts level variables, which included AOD and consumption variables.

HUD-USPS crosswalk files were utilized to allocate zip codes to census tracts which are derived from data in the quarterly USPS Vacancy Data. This data reflects the locations of business and residential addresses, both of which are being used in this analysis. Due to the fact that the data allocation method is based on addresses rather than by area or by population, it allows the spatial distribution of population and residences to be accounted for. The process of converting zip- code-level data to census tract using HUD USPS zip code crosswalk files involved weighting the estimates by the spread of the census tract population that resides in each zip code, so that if a zip code is associated with more than one census tract, it was assigned to the county with the highest proportion of all addresses in the zip code to the total number of all types of addresses in the entire zip code. In order to convert each zip code to a census tract, this methodology was used based on the 2012 quarter one boundaries for the changes related to the 2010 census. The 42 original injury zip codes were merged with the Neilsen consumption zip codes for a total of 49 unique zip codes. After converting each zip code to census tract, it was found that there were 47 unique census tracts for the 49 unique zip codes, and after merging the injury and consumption data sets, it was determined that there were 32 unique census tracts that included data from both categories. After merging the census tracts for injury and consumption with these density data, it was found that there were 31 census tracts that included all consumption, injury, and density data, making this the final sample. Data for alcohol availability was determined by summing the

5 number of establishments per census tracts that sold beer, wine, and distilled alcoholic beverages and dividing by the number of residents per capita. Two companies that do sales and marketing and provide data on businesses, InfoUSA28 and Dun&Bradstreet29, were used as database sources for data generated in 2003 and 2009 for Allegheny County, Pennsylvania. All establishments were geocoded using GIS from the Allegheny County Health Department and all classifications are based on the North American Industry Classification System (NAICS) standards. A logistic regression analysis was performed using SPSS software to investigate the extent to which AOD is related to injury and the amount of alcoholic drinks consumed per census tract while adjusting for the effects of median household income on the model. The SPSS Process Macro was used to perform the mediation analysis24.

6 3.0 RESULTS

Descriptive statistics are shown in Table 1 for all variables included in the model. The mean for alcoholic drinks consumed per census tract in Pittsburgh was 78, while there was an average of

13 on premise alcohol outlets and 2 off premise alcohol outlets per census tract. The average amount of alcohol related traumatic injuries (ARI) was 53 with a standard deviation of 41.8 and the average amount of traumatic injuries was 615 with a standard deviation of 505.8. One covariate, median household income, was adjusted for due to small sample size, which had a mean of $43,915.55. Pearson regression analysis correlations are presented in Table 2 and Table

3. Table 2 includes traumatic injury, whereas the Table 3 includes ARI. Model 1 includes density adjusting for the covariate and model 2 includes density plus consumption adjusting for the covariate. No significance was found in any of the model variables with respect to total traumatic injury, except for the covariate median household income, as seen in Table 2. Taking into account ARI, no significance was found with on premise or off premise density alone

(model 1). However, when consumption was added to the on premise model for ARI, the model became significant, as shown in Table 2 (model 2). On premise AOD (95% confidence interval

(.095, 1.408)) and consumption (95% confidence interval (-.912, 5.411)) were significantly positively correlated with alcohol related traumatic injury while adjusting for median household income, where all variables in the model had a p-value of less than .05. The relationship between these variables was tested by doing a mediation analysis using SPSS Process macro.

7 The results were presented in Figure 1, which shows that there is a significant direct effect between on premise AOD and ARI and for the indirect effect of the number of alcoholic drinks on ARI, but not for the effect of AOD on drinks. A mediation analysis was also performed with off premise AOD, drinks and ARI, of which no pathways were significant, as shown in Figure 2.

Because of these results, it was found that there was no mediation effect on the relationship between AOD and alcohol related traumatic injury and the hypothesis that alcohol consumption mediates the relationship between AOD and traumatic injury was not supported.

8 Table 1. Descriptive statistics for all model variables

Variable µ SD Traumatic Injuries 615 505.79722 Alcohol Related Traumatic Injuries 53.931 41.48143 On Premise Alcohol Density 13.399 22.3956 Off Premise Alcohol Density 2.485 2.8627 Consumption 77.924 44.76983 Median Household Income $43,915.55 $18,360.43 Total Population 4011.19 1068.554

Figure 1. Pathway coefficients for mediation model related to on premise alcohol outlet density (*p<.05)

9 Figure 2. Pathway coefficients for mediation model related to off premise alcohol outlet density (*p<.05)

Table 2.Pearson regression correlation models related to traumatic injury (*p<0.5)

Traumatic Injury Model 1 Model 2 Independent Variable Beta Confidence Beta Confidence Interval Interval On Premise Density 3.5587 (-6.868, 7.082) 0.349 (-6.108, 6.806) Consumption N/A N/A 2.249 (-.912, 5.411) Median Household Income 0.0043 (-.025, -.008)* -0.017 (-.026, -.008)*

Off Premise Density 30.184 (-23.375, 83.742) 25.22 (-25.270, 75.716) Consumption N/A N/A 2.253 (-.765, 5.271) Median Household Income -0.015 (-.023, -.007)* -0.015 (-.024, -.007)*

10 Table 3. Pearson regression correlation models related to traumatic injury (*p<.05)

Alcohol Related Traumatic Injury Model 1 Model 2 Independent Variable Beta Confidence Beta Confidence Interval Interval On Premise Density 0.697 (-.045, 1.440) 0.751 (.095, 1.408)* Consumption N/A N/A 0.26 (.002, .518)* Median Household Income -0.001 (-.002, .000)* -0.001 (-.002, .000)*

Off Premise Density 2.813 (-1.938, 7.565) 3.619 (-.756, 7.993) Consumption N/A N/A 0.295 (-.004, .594) Median Household Income -0.001 (-.002, .000)* -0.001 (-.002, .000)*

11 4.0 DISCUSSION

THE CURRENT STUDY

Alcohol is a known contributor to injury and many studies suggest that the more outlets there are, leads to more opportunity for alcohol to be consumed, which leads to more injury. It has been shown in many instances that studies supporting this theory have small effect sizes or find little change in injury as alcohol outlet density has increased over large spans of time (6, 7, 12).

There has been no known mechanism to explain why alcohol consumption may be the more important factor in the relationship between AOD and injury. This study investigates the relationship of alcohol consumption as a mediator between AOD and traumatic injury using a regression and mediation analysis for census tracts in Pittsburgh, Pennsylvania. A mediation analysis was performed to find the indirect effect of consumption on traumatic injury. The results show a significant relationship between AOD, consumption, and ARI when adjusting for median household income. However, the results do not show that AOD had a significant impact on consumption, the mediator variable, meaning there was no evidence of mediation. Despite that there was no mediation effect of alcohol consumption on traumatic injuries; there are several things to consider.

THE RELATIONSHIP BETWEEN DENSITY AND CONSUMPTION

Many studies find a relationship between alcohol density and consumption, meaning that the more outlets there are that sell alcohol in an area, the more alcohol will be consumed. Studies

12 have shown a significant effect between AOD within a two mile radius and heavy drinking, where college students were found to drink more than five drinks19. Other studies have shown that decreasing outlets within a one mile radius had a significant effect on lower reported rates of drinking among college students26. However, these studies often use only on premise density, and do not capture the amount of consumption at home, which this study aimed to do. Other studies also target populations known to be drinking whether or not there are more outlets around, such as college students, whereas the current study aimed to get a sense of consumption around the entire city of Pittsburgh. The current study found results that opposed that of other studies, where there was a significant effect of the amount of alcoholic drinks consumed and alcohol related injury, but not between AOD and drinks consumed. This could suggest that density does not play an important role in consumption. It may also suggest that consumption is the more important predictor of injury, and not availability of alcohol, as many studies present.

THE RELATIONSHIP BETWEEN DENSITY AND INJURY

Many studies have also found differing results from on and off premise density, where off premise density is more significant than on premise density for rates of violence and injury (14, 16).

There is often much variance among studies, as some studies recommend limiting the number of bars in a neighborhood to prevent injury7, but others says it is better to reduce the amount of off premise, highly clustered alcohol markets connected to the street16. However, studies that found significance with a certain type of AOD cite small effect sizes relating to injury, where adding one bar to an area would produce only 0.17 or 0.25 assaults per year (7, 23). This study found similar results relating to AOD variability. Off premise density was not significantly correlated with ARI and had no significant pathways in the mediation model, while only one model related to on premise density was significant. Type of AOD had no significant effect on traumatic injury at all, only on ARI, which supports the notion that density is not directly related to injury and

13 that some other phenomena is accounting for the association. This could explain that on and off premise density is not a stable indicator of traumatic injury and presents too much variability due to the different atmospheres and neighborhood characteristics of each type of outlet.

THE RELATIONSHIP BETWEEN CONSUMPTION AND INJURY

The current study is unique in that it presents alcohol related traumatic injury as a separate outcome from traumatic injury, which no other study has done. This is important because it is not self reported, as the data comes from an EMS provider. It is very evident in the literature that alcohol plays a role in a large part of the injuries that present to the emergency room in the world today, where about 5% of all injuries can be attributed to alcohol use globally. One study showed that risk of injury was increased by a factor of 2.5 when there was increased alcohol consumption six hours before the injury occurred27. The current study found that alcohol related traumatic injury accounted for about 11% of all traumatic injury. This study found support that alcohol consumption was significantly associated with ARI for on premise density. Although it was not found to be significant for off premise density, the results show that it was close to being correlated (p-value .053), which is why a mediation analysis was performed for off premise density, with no significant results. Further research is needed to be able to better predict off premise consumption related to injury.

LIMITATIONS

Lack of mediation in this study is most likely attributed to small sample size. Due to the small sample size, the logistic regression analysis does not allow for adjusting for many covariates, which leads to more variability in the sample. Other limitations lie in the timeframe of alcohol consumption, where the household may buy the alcohol, but it is unclear whether or not they actually consume it before buying more alcohol. The assumption made was that each household did consume their purchases within the year time frame, but it is feasible that the alcohol was

14 bought as a gift for someone outside the household or that members of the household were restocking the item and did not actually consume it. It is also difficult to determine how much alcohol was consumed by each individual member of the household and how much was consumed at one time. One more limitation is related to traumatic injury, where most studies have only looked at intentional and unintentional injuries, and have placed drinking and driving in a separate category of injury. Traumatic injuries can range from gun related violence to car crashes, all of which have been linked to alcohol use, but there is no way of knowing the bias associated with the dataset related to a specific type of traumatic injury, as it is not specified.

Although the area and population weighting techniques used to estimate the geography changes between the 2000 and 2010 census are very accurate, there is always the possibility that when part of a tract in 2000 is reallocated into a new tract in 2010, the composition of different tract fragments accounts for the dissimilarity in people across the tracts, when it should be assumed that each tract has residents that are alike21. There could potentially be another issue dealing with blocks in 2000 are sometimes split into different portions that are assigned to different 2010 tracts, which is a small difference and should not account for much error in census geography21.

FUTURE RESEARCH

Despite these limitations, the findings have important implications in further research. Based on this research and the significance of consumption on traumatic injury, policymakers should be focusing on limiting consumption and the amount of alcohol establishments sell instead of the number of alcohol selling establishments in an area to better prevent traumatic injury. Future research on mediation based on alcohol consumption could better define consumption by using a user input tracking application. It is also important to have a larger sample size in order to get results that can adjust for demographic characteristics of the community. This study emphasizes the need for future research involving all three variables of AOD, consumption, and injury and

15 provides evidence that alcohol consumption may be more of a risk factor than AOD for traumatic injury.

16 BIBLIOGRAPHY

1. Cherpitel, Cheryl J., et al. "Attribution of alcohol to violence-related injury: self and other’s drinking in the event." Journal of studies on alcohol and drugs 73.2 (2012): 277-284.

2. World Health Organization. Global status report on alcohol and health, p. XIV. 2014 ed. http://www.who.int/substance_abuse/publications/global_alcohol_report/msb_gsr_2014_ 1.pdf?ua=1

3. Rehm, J., Mathers, C., Popova, S., Thavorncharoensap, M., Teerawattananon, Y., & Patra, J. (2009). Global burden of disease and injury and economic cost attributable to alcohol use and alcohol-use disorders. The Lancet, 373(9682), 2223-2233.

4. Livingston, M., Livingston, M., Chikritzhs, T., Livingston, M., Chikritzhs, T., Room, R., ... & Room, R. (2007). Changing the density of alcohol outlets to reduce alcohol-related problems. Drug and alcohol review, 26(5), 557-566.

5. Gruenewald, P. J., Freisthler, B., Remer, L., LaScala, E. A., Treno, A. J., & Ponicki, W. R. (2010). Ecological associations of alcohol outlets with underage and young adult injuries. Alcoholism: Clinical and Experimental Research, 34(3), 519-527.

6. Morrison, C., Smith, K., Gruenewald, P. J., Ponicki, W. R., Lee, J. P., & Cameron, P. (2016).

Relating off‐premises alcohol outlet density to intentional and unintentional injuries.

Addiction, 111(1), 56-64.

7. Gruenewald, P. J., & Remer, L. (2006). Changes in outlet densities affect violence rates. Alcoholism: Clinical and Experimental Research, 30(7), 1184-1193.

8. Treno, A. J., Gruenewald, P. J., & Johnson, F. W. (2001). Alcohol availability and injury: the role of local outlet densities. Alcoholism: Clinical and Experimental Research, 25(10), 1467- 1471.

17 9. Mäkelä, P. (2002). Whose drinking does the liberalization of alcohol policy increase? Change in alcohol consumption by the initial level in the Finnish panel survey in 1968 and 1969. Addiction, 97(6), 701-706.

10. 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, 101(5), 666-677.

11. Campbell, C. A., Hahn, R. A., Elder, R., Brewer, R., Chattopadhyay, S., Fielding, J., ... & Task Force on Community Preventive Services. (2009). The effectiveness of limiting alcohol outlet density as a means of reducing excessive alcohol consumption and alcohol-related harms. American journal of preventive medicine, 37(6), 556-569.

12. Hobday, M., Meuleners, L., Liang, W., Gilmore, W., & Chikritzhs, T. (2015). Associations between alcohol outlets and emergency department injury presentations: effects of distance from the central business district. Australian and New Zealand journal of public health.

13. Connor, J. L., Kypri, K., Bell, M. L., & Cousins, K. (2010). Alcohol outlet density, levels of drinking and alcohol-related harm in New Zealand: a national study. Journal of epidemiology and community health, jech-2009.

14. Gorman, D. M., Gorman, D. M., Zhu, L., Gorman, D. M., Zhu, L., Horel, S., ... & Horel, S. (2005). Drug ‘hot-spots’, alcohol availability and violence. Drug and alcohol review, 24(6), 507-513.

15. Scribner, R., Cohen, D., Kaplan, S., & Allen, S. H. (1999). Alcohol availability and homicide in New Orleans: conceptual considerations for small area analysis of the effect of alcohol outlet density. Journal of studies on alcohol, 60(3), 310-316.

16. Branas, C. C., Elliott, M. R., Richmond, T. S., Culhane, D. P., & Wiebe, D. J. (2009). Alcohol consumption, alcohol outlets, and the risk of being assaulted with a gun. Alcoholism: Clinical and Experimental Research, 33(5), 906-915.

17. Escobedo, L. G., & Ortiz, M. (2002). The relationship between liquor outlet density and injury and violence in New Mexico. Accident analysis & prevention, 34(5), 689-694.

18. Resko, S. M., Walton, M. A., Bingham, C. R., Shope, J. T., Zimmerman, M., Chermack, S.

T., ... & Cunningham, R. M. (2010). Alcohol Availability and Violence among Inner‐City

18 Adolescents: A Multi‐Level Analysis of the Role of Alcohol Outlet Density. American

journal of community psychology, 46(3-4), 253-262.

19. Weitzman, E. R., Folkman, A., Folkman, M. K. L., & Wechsler, H. (2003). The relationship of alcohol outlet density to heavy and frequent drinking and drinking-related problems among college students at eight universities. Health & place, 9(1), 1-6. 20. Logan, John R., Zengwang Xu, and Brian Stults. 2014. "Interpolating US Decennial Census Tract Data from as Early as 1970 to 2010: A Longitudinal Tract Database" The Professional Geographer 66(3): 412–420 21. US2010. (n.d.). Retrieved March 30, 2016, from http://www.s4.brown.edu/us2010/Researcher/BoundaryChange.htm 22. Gruenewald, P. J., Remer, L. G. and LaScala, E. A. (2014), Testing a social ecological model of alcohol use: the California 50-city study. Addiction, 109: 736–745. doi: 10.1111/add.12438 23. Livingston, M. (2008). A longitudinal analysis of alcohol outlet density and assault. Alcoholism: Clinical and Experimental Research, 32(6), 1074-1079. 24. HUD USPS ZIP Code Crosswalk Files. (n.d.). Retrieved March 31, 2016, from https://www.huduser.gov/portal/datasets/usps_crosswalk.html 25. Kavanagh, A. M., Kelly, M. T., Krnjacki, L., Thornton, L., Jolley, D., Subramanian, S. V., ...

& Bentley, R. J. (2011). Access to alcohol outlets and harmful alcohol consumption: a multi‐

level study in Melbourne, Australia. Addiction, 106(10), 1772-1779. 26. Wechsler, H., Davenport, A., Dowdall, G., Moeykens, B., & Castillo, S. (1994). Health and behavioral consequences of binge drinking in college: A national survey of students at 140 campuses. Jama, 272(21), 1672-1677. 27. Watt, K., Purdie, D. M., Roche, A. M., & McClure, R. J. (2004). Risk of injury from acute alcohol consumption and the influence of confounders. Addiction, 99(10), 1262-1273. 28. InfoUSA Consumer. (n.d.). Retrieved April 28, 2016, from http://www.infousa.com/. 29. Dun & Bradstreet. (n.d.). Retrieved April 28, 2016, from http://www.dnb.com/. 30. Calculated based on data from The Nielsen Company (US), LLC and marketing databases provided by the Kilts Center for Marketing Data Center at The University of Chicago Booth School of Business.