Alcohol Outlet Density and Area-Level Heavy Drinking Are Independent Risk Factors for Higher Alcohol-Related Complaints

Abstract

Alcohol outlet density has well-documented associations with social and health indicators such as crime and injury. However, significantly less is known about the relationships among alcohol-related complaints. Bayesian hierarchical Poisson regression with spatial autocorrelation was used to model the association between on- and off-premises alcohol outlet density and area-level prevalence of current drinkers and heavy drinking, and graffiti density—an indicator of physical disorder—in association with calls from civilians reporting illegal use, alcohol sales, and other alcohol-related activities (hereafter alcohol-related complaints). Complaints were separated into two groups based on whether they occurred at (a) clubs/bars/restaurants or (b) elsewhere. Alcohol-related complaints and graffiti were collected from NYC Open Data. Alcohol density data are from ESRI Business Analyst and information on the prevalence of drinking from the New York City Community Health Survey. The unit of analysis consisted of ZIP codes in New York City (n = 167), and the design was a cross-sectional analysis of aggregated data between 2009 and 2015. In multivariable models, a one-unit increase in off-premises alcohol outlet density was associated with a 47% higher risk of alcohol-related complaints at clubs, bars, and restaurants [rate ratio (RR = 1.46, 95% CI = 1.21, 1.77)]. Area-level prevalence of heavy drinking was associated with a 59% higher risk of alcohol-related complaints at the club, bars, and restaurants (RR = 1.59, 95% CI = 1.34, 1.86) and a 40% higher risk of complaints elsewhere (RR = 1.40, 95% CI = 1.20, 1.63). In New York City, area-level heavy drinking prevalence is a strong independent mechanism that links alcohol outlet density to alcohol-related complaints. Area-level heavy drinking should be investigated as a predictor of other public health problems such as drug overdose mortality.

Electronic supplementary material

The online version of this article (10.1007/s11524-018-00327-z) contains supplementary material, which is available to authorized users.

Introduction

Greater alcohol outlet density is associated with higher rates of social and health problems related to drinking [1, 2], which include violent crime [3], road crashes [4], sexually transmitted disease [5], HIV prevalence [6], child abuse and neglect [7], and emergency department injuries [8].

Despite this robust body of work, we focused on a potentially novel outcome, civilian complaints related to alcohol (e.g., underage drinking, drinking alcohol in an unlicensed establishment, and sale of alcohol to a minor, hereafter referred to as alcohol-related complaints). First, no empirical research shows that these outcomes are associated with alcohol outlet density, particularly within the USA, although there is some related international work. For example, one multilevel study from Australia showed that respondents who lived closer to rather than farther away from alcohol outlets and in areas with a high density of those outlets per 10,000 persons reported a higher number of problems in their neighborhoods concerning drunkenness, property damage, and assault victimization [9], controlling for sociodemographic covariates. One ecological study, also from Australia, found that higher rates of property damage, including graffiti, offensive behavior reported to police occurred in areas with greater sales of alcohol, controlling for relevant covariates [10].

A second reason we focused on alcohol-related complaints was because alcohol may correlate or interact with other features of the social and built environment that underlie drug and alcohol problems. Moreover, community systems’ approaches to alcohol interventions suggest that it is essential to monitor regulatory and enforcement efforts’ responsiveness to alcohol problems, and we see civilian alcohol-related complaints as a potential marker of public surveillance [11]. Therefore, alcohol-related complaints may be useful to monitor, to target, and to intervene in areas that experience related public health problems [12] such as unintentional drug overdoses.

In this study, we also examined two competing mechanisms that potentially link alcohol outlet density and alcohol-related complaints: alcohol risk behavior versus physical disorder. First, an area’s alcohol risk behavioral characteristics, such as the prevalence of binge or heavy drinkers, could potentially link alcohol outlet density with alcohol-related complaints because a high level of individual alcohol consumption [13, 14] is related to the harm of oneself and others [2, 11]. A second perspective, however, is based on amenity effects theory, which postulates that certain environmental conditions, such as alcohol outlets, attract other incivilities such as violence and other public indicators of physical disorder, such as graffiti, that may not happen elsewhere [15].

In this study, we contribute to the alcohol density research in the following ways. First, we add to the two peer-reviewed studies that investigated alcohol outlet density in association with health or social outcomes in New York City (NYC) [16, 17]. Second, we investigate whether alcohol outlet density is linked to alcohol-related complaints via two contrasting, yet plausible area-level mechanisms: heavy drinking prevalence and density of graffiti.

Methods

Study Setting and Unit of Analysis

The unit of analysis in this study is ZIP codes because previous research in NYC considered them reasonable proxies for neighborhoods and because of practical reasons such as availability of public health data [18, 19]. We used the polygons of ZIP codes (N = 167), for which boundaries were constructed by the New York City Department of City Planning [20]. Other ecological data are linked to these ZIP code boundaries. Areas with no residential population, such as Prospect Park and Central Park, are not included in the study.

Alcohol-Related Complaints

NYC Open Data is an open-source online portal that contains over 1500 datasets, including those that are owned and published by the City of New York. NYC has a reporting platform called NYC 311, which residents can call to register complaints on a range of issues including sanitation, noise, crime, and road conditions in their neighborhood. Residents can also submit their complaints to this platform online. The 311 system is independent of the 911 system, which is primarily used for emergencies, and not all 311 calls require a police officer’s presence.

In this study, the 311 calls, however, did result in a police officer responding and logging the complaint. The XY coordinates of the specific complaints were available in the database and were geocoded to the ZIP code of the specific location corresponding to the complaint. Our measure of alcohol-related complaints was defined according to the 311 website’s descriptions of visible problems, which include: illegal drinking of alcohol, underage drinking outside or inside bars or clubs and unlicensed establishments including social clubs or bodegas, the sale of unlimited drinks for a fixed price, the sale of alcohol to a minor by a club, bar, or store, and the sale of alcohol to a visibly intoxicated person [21]. Alcohol-related complaints are organized in NYC Open Data into five broad categories: complaints that occurred at (1) club/bars/restaurants, (2) streets/sidewalks, (3) parks/playgrounds, (4) residential building/house, and at (5) store/commercial places. Descriptions of the reports that happened at clubs, bars, or restaurants were mostly based on underage drinking (75%) followed by after-hours sales (20%) and drinking in public (5%). For all the remaining categories (3 to 5), all (100%) were characterized by drinking in public. According to amenity effects theory, it is plausible that clubs, bars, and restaurants might attract a greater number of specific kinds of risky behaviors associated with nearby alcohol outlets than other locations such as parks or sidewalks. Given this reason, we did not examine an overall complaint variable but rather two outcome variables: (1) complaints at clubs, bars, and restaurants, and (2) complaints at other locations (hereafter, referred to as elsewhere).

We retrieved data for complaints for the period January 1, 2010 through December 31, 2015. We selected those years to allow for a 1-year lag and 2-year lead for the period (2009–2013) that we had data for current area-level drinkers and heavy drinking prevalence. A total of N = 7053 records were available for analysis across the 167 ZIP codes. We summed the events between 2010 and 2015 so that each ZIP code would have sufficient cases to ensure the stability of rates in the relative risk calculations. Moreover, there were no clear trends across the 6 years.

Alcohol Outlet Density and Related Covariates

We obtained data on alcohol outlets and other places that sold alcohol such as groceries/supermarkets and convenience stores from Business Analyst for the years 2009 through 2013 [22]. Business Analyst allows users to search the North American Industry Classification System (NAICS) Codes, which are business classifications based on economic activity [23]. We extracted data for NAICS code 722410, which is defined as establishments that include bars, taverns, nightclubs, or drinking places primarily engaged in serving alcoholic beverages for immediate consumption. Those data were classified as on-premises. Next, we extracted data for NAICS code 445310, which is defined by establishments primarily engaged in retailing packaged liquor that include beer stores, liquor stores, and wine shops. Those data were classified as off-premises. We also obtained data for the following retail establishments considered as covariates in prior alcohol outlet density research: [24, 25] wholesalers of alcohol (424810/820), groceries/supermarkets (445110), and convenience stores (445120).

We used Arc Health © [26], an add-on software developed for ArcMap [27], to create a density surface of alcohol availability, which accounts for the background population [28]. Specifically, we used kernel density with a case-side kernel and fixed bandwidth of 1000 m. In this study, we used the general population as the background population (population at risk) in calculating alcohol outlet density. The population data layer is a raster dataset generated by converting the Census 2010 data at the block level into pixels in a grid format, which regularizes the areal population units to ensure that population counts at different locations are spatially comparable (regardless of the geographic boundary). In the population data layer, a pixel value is the number of people in the small area (50 m × 50 m) represented by that pixel.

We calculated the average of the alcohol outlet density across these years because we did not have any prior theoretical reasons to examine change, nor was there evidence of changes in alcohol policy in NYC. For instance, the average change in population for NYC overall between 2010 and 2015 was in the 0.8% (range 0.2% to 0.9 %) [29]. Moreover, our exploratory analyses found no significant trends in alcohol outlet density. Lastly, our measure of alcohol density has been validated in prior public health research [30].

Mechanisms

Area-Level Drinking Factors

Data on current drinkers and heavy drinking from the Community Health Survey (CHS) were obtained through a custom request to the NYC Department of Health and Mental Hygiene (NYCDOHMH). The CHS is an annual probability household survey of approximately 9000 residents ages 18 years and older each cycle. The survey obtains data on health and behavior outcomes such as drinking alcohol and HIV testing [31]. At the time of our study, NYCDOHMH provided 5-year age-adjusted prevalence estimates for ZIP codes for the year 2009–2013. CHS uses post-stratification weights to weight the data to the NYC adult residential population as per the American Community Survey 2009–2013. There was an average of 348 survey observations per ZIP code in the study period. Similar to alcohol outlet density, drinking/alcohol use behaviors did not vary (in other words, the behaviors showed a flat trend) in NYC including the years after the economic recession, specifically, between 2011 and 2014 [32].

Current drinking in the CHS was ascertained through the question: “A drink of alcohol is 1 can or bottle of beer, 1 glass of wine, 1 can or bottle of wine cooler, 1 cocktail, or 1 shot of liquor. During the past 30 days, how many days per week or per month did you have you had at least 1 drink of any alcoholic beverage?” The response option was the number of days per week (range 1–7) or days in the past month (range 1–30) that the respondent had at least one drink.

In this study, we operationalize area-level current drinker prevalence as the proportion of persons within a ZIP code who reported at least one drink.

The CHS defines heavy drinking as more than two drinks per day for men or one drink per day for women, informed by criteria from the Dietary Guidelines for Americans [33]. In this study, we operationalize area-level heavy drinking prevalence as the proportion of persons within a ZIP code classified as heavy drinkers, among those who reported at least one drink.

We received a modified CHS dataset in which ZIP codes with a population of less than 30,000 were combined with an adjacent ZIP code within the same United Hospital Fund (UHF). UHFs are aggregate geographic areas that range between one and nine adjoining ZIP codes that correspond to catchment areas for certain healthcare facilities [34]. Eighty-nine of the 167 ZIP codes (53%) involved some combination. ZIP codes that were combined to obtain a stable prevalence estimate were assigned the same prevalence estimate when two or more ZIP codes were combined. For example, the ZIP code 10030 was combined with 10037 and 10039, and therefore all three would receive the same estimate. Areas that involved a greater proportion of combined ZIP codes were in the borough of Staten Island, the Bronx, Manhattan, and some parts of Brooklyn and Queens.

Graffiti Prevalence

Graffiti has been documented as one valid indicator of neighborhood physical disorder [35, 36]. We retrieved reports of graffiti recorded in NYC Open Data, which is also based on 311 calls that were logged. Graffiti could have occurred at a residential, mixed-used, or commercial structures, although the type of graffiti (e.g., spray paint, etc.) was not described. We used the counts of graffiti reports for the years 2014 and 2015 only (n = 95, 761), and used the kernel density function in ArcGIS to create the density of graffiti per square mile. We then used the Zonal Statistics as table function to obtain the density per ZIP code.

Sociodemographic Variables

Based on prior research [3, 37, 38], the following variables were included in the analysis. These are the percentage of Black/African-American residents, the proportion of persons 18 to 24, and 25 to 34 years of age, the proportion of female-headed households, and the proportion of vacant households for rent. For instance, alcohol outlets are often disproportionately located in Black neighborhoods [39]. At the individual level, younger age has been associated with higher problem drinking and risk-taking [40], and the ecological argument that a higher proportion of younger-aged individuals in an area is associated with higher alcohol and drug problems appear to have some validity based on prior ecological studies [41, 42]. A higher proportion of vacant buildings presents higher opportunities for illicit drug and alcohol use behavior [43] and higher visibility of activity that includes making police calls to report narcotics use or sales [44].

Lastly, we included a socioeconomic deprivation index, which consisted of a principal components analysis (PCA) using methods described in prior ecological analyses across NYC ZIP codes [45]. The variables included in the PCA were a percentage of persons below the poverty level, the percentage of population 25 years of age with less than a high school diploma, percentage of population 16 years of age who are unemployed, median household income, and proportion of homes greater than $300,000. The latter two variables were reverse coded before the index was created. All variables were retrieved from the American Community Survey (2009–2013) 5-year estimates and downloaded online from Infoshare, a company that customizes Census-based data to multiple geographic levels, specifically for NYC [46].

Analysis

We described the variables with mean and standard deviation. Given that all variables were continuously distributed, we investigated correlations among the predictors to assess multicollinearity for the regression models (data not shown). Four maps were created in ArcMap 10.4, created by ArcHealth©, to highlight the alcohol outlet density of locations of alcohol-related complaints, the graffiti density overlaid on the distribution of Black/African-American racial composition, choropleth maps of the prevalence of heavy drinking, and relative risk distribution of alcohol-related complaints (at clubs/bars/restaurants) adjusted for the outcome and covariates at the ZIP code level. The cut-points in each choropleth maps are based on quartile distribution (i.e., four categories).

We used Bayesian hierarchical Poisson regression analysis, which accounts for spatial autocorrelation, to assess the relationship between alcohol-related complaints and alcohol outlet density, area-level drinking factors, and sociodemographic variables. Alcohol-related complaints were assumed to follow a Poisson distribution (level 1). The models fitted include an intercept (α) that represents the (log) average relative risk over NYC, ZIP code-level covariate vectors (X_ZIP[i]), and corresponding coefficients (β), and both spatial (s_ZIP[i]) and unstructured (u_ZIP[i]) random effects (level 2). These random effects accounted for unobserved covariates and overdispersion, and uncertainty accounted for the combined ZIP codes [47]. Specifically, s_ZIP(i) is assumed to follow an intrinsic conditional autoregressive (iCAR) distribution [48] and addresses the spatial autocorrelation issue within the dataset. Model specification is available in the Appendix. The Bayesian hierarchical analysis was implemented using “runmlwin” in MLwiN 2.32 software within Stata 15.0 [49]. Two chains were fitted for each model using the default priors for the fixed and random parts of the model. The initial 20,000 iterations, on which the models converged, were discarded as burn-ins. We ran another 40,000 iterations for each chain, resulting in a total number of 80,000 samples for the posterior inference. We assessed convergence by examining trace-plots’ histories and inspected the precision estimates using the Gelman-Rubin statistics in WinBUGS 14 software. Sensitivity analysis of priors (by specifying different vague priors to unknown parameters) [50] showed that the results were not sensitive to MLwiN’s default prior specifications.

We conducted two separate analyses for: (a) complaints at clubs, bars, and restaurants, and (b) complaints elsewhere. Next, we used a sequential analysis approach by first examining the role of alcohol outlet density adjusted for covariates (model 1) and then added the area-level drinking factors and the graffiti density (model 2) (Table 2). Area-level heavy drinking had missing data for seven areas, and values were imputed via the mi impute function in Stata, which imputes values based on the other covariates in the model. Three imputed datasets were created, and the model 2 analysis was then performed using the mi estimate command. We inspected whether the alcohol density coefficients were attenuated by comparing the relative change in coefficients between the models (i.e., model 1-model 2/model 1). Relative risks (RR) and 95% credible intervals (CIs) were reported.

Table 2.

Bayesian hierarchical spatial regression predicting alcohol-related complaints in New York City, ZIP codes (N = 167), 2009 to 2015

	(a) Complaints at clubs, bars, restaurants		(b) Complaints elsewhere
	Model 1a	Model 2a	Model 1b	Model 2b
Alcohol outlet density	RR (95% CI)	RR (95% CI)	RR (95% CI)	RR (95% CI)
Average density of off-premises alcohol outlets	1.59 (1.30, 2.01)	1.46 (1.20, 1.77)	1.37 (1.12, 1.66)	1.26 (1.05, 1.51)
Average density of on-premises alcohol outlets	0.80 (0.62, 0.99)	0.90 (0.72, 1.13)	0.76 (0.61, 0.94)	0.84 (0.68, 1.04)
Alcohol outlet density-related covariates
Number of wholesale alcohol outlets	0.99 (0.87, 1.11)	0.96 (0.86, 1.08)	1.04 (0.93, 1.16)	1.01 (0.91, 1.11)
Number of grocery stores/supermarkets	1.18 (1.00, 1.37)	1.05 (0.90, 1.22)	1.07 (0.92, 1.23)	1.00 (0.88, 1.14)
Number of convenience stores	0.95 (0.84, 1.07)	0.95 (0.85, 1.07)	0.98 (0.88, 1.11)	0.98 (0.88, 1.08)
Sociodemographic variables
Percentage of Black/African-American residents	0.68 (0.61, 0.76)	0.68 (0.61, 0.75)	0.94 (0.85, 1.03)	0.93 (0.84, 1.02)
Proportion of persons 18 to 24 years of age	1.24 (1.12, 1.38)	1.18 (1.07, 1.31)	1.15 (1.05, 1.27)	1.11 (1.01, 1.23)
Proportion of persons 25 to 34 years of age	0.84 (0.72, 0.98)	0.81 (0.70, 0.94)	1.21 (1.05, 1.40)	1.18 (1.03, 1.35)
Proportion of female-headed households	0.97 (0.84, 1.12)	0.84 (0.73, 0.98)	0.87 (0.75, 1.01)	0.81 (0.71, 0.93)
House rental vacancy proportion	1.21 (1.04, 1.39)	1.26 (1.11, 1.43)	1.13 (0.97, 1.31)	1.11 (0.97, 1.27)
Socioeconomic deprivation index	1.03 (0.89, 1.18)	1.18 (1.01, 1.38)	1.21 (1.06, 1.38)	1.37 (1.20, 1.58)
Mechanisms
Prevalence of current drinkers		0.88 (0.72, 1.06)		0.92 (0.77, 1.10)
Prevalence of heavy drinking		1.59 (1.36, 1.86)		1.40 (1.19, 1.62)
Graffiti density per square mile		1.06 (0.97, 1.16)		1.07 (0.99, 1.16)

All models contain a binary indicator for whether a ZIP code was combined = 1 or not = 0 to obtain the prevalence of heavy drinking as well as prevalence of current drinkers. The coefficients for that variable is not displayed because the interpretation of coefficient is not meaningful or relevant to the study

Alcohol-related events such as civilian complaints may happen at one location but be influenced by alcohol obtained at outlets and used by individuals from adjacent areas [51]. To account for this possibility, we replicated the above analyses, but this time using only spatially lagged variables for alcohol outlet density, the area-level drinking factors, and alcohol-related complaints. Spatial lag variables were created in GeoDa 1.6.7 [52] using a first-order Queen Contiguity weights matrix [53].

Lastly, we tested whether alcohol-related complaints reflected informants’ views of actual problems versus subjective criticism in which their complaints reflected their own perceptions of and feelings about their area. If the relationship between an item in one construct is strongly related to an item in another construct, that relationship indicates nomological validity. We, therefore, used GeoDa software to conduct a bivariate spatial correlation between graffiti density (an objective physical disorder indicator) and alcohol-related complaints (a social disorder indicator) [36]. A Moran’s I coefficient was produced, and a significant geographic correlation was evaluated against a pseudo-p value after running 9999 permutations of the data [54].

Results

Table 1 shows that there was a mean of 38 (standard deviation [sd] = 45) off-premises alcohol outlets, and 35 (sd = 25) on-premises outlets, and 42 (sd = 31) alcohol-related complaints across the 167 ZIP codes during the study period. The average prevalence of heavy drinking across ZIP codes was 6% (sd = 4).

Table 1.

Descriptive statistics on the alcohol outlet density, alcohol outlet covariates, and socioeconomic and demographic correlates, and alcohol-related complaints in New York City, 2009 to 2015

Per ZIP code (N = 167)	Mean	Standard deviation	Minimum	Maximum
Alcohol outlet density variables
Number of off-premises alcohol outlets	38	45	1	271
Average densitya of off-premises alcohol outlets	20	21	1	121
Number of on-premises alcohol outlets	35	25	4	138
Average densitya of on-premises alcohol outlets	18	12	2	71
Alcohol outlet density-related covariates
Number of wholesale alcohol outlets	5	6	0	32
Number of grocery stores/supermarkets	151	113	7	550
Number of convenience stores	25	18	0	94
Sociodemographic variablesb
Average population	48,902	26,017	2454	1,109,948
Percent of Black/African-American residents	23	26	.01	91
Proportion of persons 18 to 24 years of age	10	3	4	20
Proportion of persons 25 to 34 years of age	18	7	6	53
Proportion of female headed households	7	4	1	22
House rental vacancy proportion	.10	.04	.03	.33
Socioeconomic deprivation indexc	.04	.99	−2	3
Mechanism variables
Prevalence of current drinkersd	57	11	38	92
Prevalence of heavy drinkingd	6	4	0	18
Number of graffiti complaints in 2014 to 2015	134	126	4	979
311 alcohol-related complaints
Total alcohol-related complaints, N = 7053	42	31	1	146
311 alcohol-related complaints at clubs, bars, restaurants, N = 2976	18	14	0	74
311 alcohol complaints at all other locationseN = 4077	24	21	0	97

Whole numbers above one are rounded up

^aAlcohol density calculated using ArcHealth©

^bEstimates derived from the American Community Survey 5-year estimates. The current drinker and heavy drinking data are based on the NYC Community Health Survey 5-year estimates (2009–2013)

^cBased on a principal component index of the following variables: poverty level, unemployment status, educational attainment, and median household income

^dThis is not a citywide estimate but the mean of the weighted ZIP code-level prevalence data specifically created to use in this study

^eIncludes complaints recorded at streets, sidewalks, parks, playgrounds, residential houses, and stores

Areas with a higher density of off-premises alcohol outlets were in Manhattan, followed by southern Brooklyn and Queens. The prevalence of heavy drinking was higher in ZIP codes in Queens while a higher proportion of graffiti reports occurred in a few ZIP codes in Brooklyn, then the south Bronx and Manhattan (Fig. 1). Alcohol density and alcohol-related complaints did not appear systematically concentrated in neighborhoods within the highest quartile of area-level Black/African-American racial composition. The relative risks for alcohol-related complaints after adjusting for covariates from model 2 appeared highest in some areas of southwest Queens, Rockaways, southern Staten Island, and some areas in the south Bronx (Fig. 1).

Fig. 1.

Choropleth maps of the prevalence of heavy drinking, relative risk distribution of alcohol-related complaints, and selected covariates, NYC, N = 167 ZIP codes

Regarding whether alcohol-related complaints reflected informants’ reporting of actual problems versus subjective reports of criticism, we found a strong significant positive correlation with graffiti density: Moran’s I = 0.16, p = 0.003 (results now are shown).

Table 2 contains the results of the multivariable Bayesian hierarchical Poisson regression analysis models both without (model 1a and 1b) and with the mechanisms (model 2a and 2b).

Results for Alcohol-Related Complaints at Clubs, Bars, and Restaurants

A one-unit increase in off-premises alcohol outlet density was associated with a 55% increase in alcohol-related complaints at clubs, bars, and restaurants (RR = 1.59, 95% CI = 1.30, 1.73, model 1a). Adding the area-level prevalence of current drinkers, heavy drinking, and graffiti density into the model reduced the impact of off-premises density on the outcome by 8% (1.59 − 1.46 / 1.59 × 100). The further adjusted association of off-premises alcohol outlet density was associated with a 46% increase in the alcohol-related complaints (RR = 1.46, 95% CI = 1.21, 1.77, model 2a). In model 2a, heavy drinking prevalence was statistically significant. A one-unit increase in the prevalence of heavy drinking was associated with a 59% increase in alcohol-related complaints at clubs, bars, and restaurants (RR = 1.59, 95% CI = 1.35, 1.86, model 2a). Graffiti density per square mile was not significantly associated with alcohol-related complaints. Areas with a higher proportion of Black/African-American racial composition had a lower risk of alcohol-related complaints at clubs, bars, and restaurants (RR = 0.68, 95% CI = 0.61, 0.81, model 2a). Higher socioeconomic deprivation increased the relative risk of alcohol-related complaints (RR = 1.18, 95% CI = 1.01, 1.38, model 2a). On-premises outlet density was not significantly associated with alcohol-related complaints at clubs, bars, or restaurants after adjusting for mechanisms.

Results for Alcohol-Related Complaints Elsewhere

A one-unit increase in off-premises alcohol outlet density was associated with a 37% increase in alcohol-related complaints elsewhere (RR = 1.59, 95% CI = 1.12, 1.65, model 1b). Off-premises alcohol outlets were associated with lower rates of reported problems elsewhere (RR = 0.76, 95% CI = 0.61, 0.94, model 1b). Adding the area-level prevalence of current drinkers, heavy drinking, and graffiti density into the model reduced the impact of off-premises density on the outcome by 8% (1.37 − 1.26 / 1.37 × 100). The further adjusted association of off-premises alcohol outlet density was associated with a 26% increase in the alcohol-related complaints elsewhere (RR = 1.26, 95% CI = 1.05, 1.51, model 2b). In model 2b, heavy drinking prevalence was statistically significant. A one-unit increase in the prevalence of heavy drinking was associated with a 40% increase in alcohol-related complaints elsewhere (RR = 1.40, 95% CI = 1.19, 1.63, model 2b). Graffiti density per square mile and prevalence of current drinkers were not significantly associated with alcohol-related complaints elsewhere. On-premises alcohol outlet density was no longer significantly associated with alcohol-related complaints elsewhere.

In the separate analyses using the spatially lagged variables, neither off-premises nor on-premises alcohol outlet density in adjacent neighborhoods was associated with alcohol-related complaints at clubs, bars, and restaurants or elsewhere (results not displayed).

Discussion

In this study, we examined whether there was an association between alcohol outlet density and alcohol-related complaints at locations that considered clubs, bars, and restaurants separately from complaints at all other locations in NYC. Alcohol-related problems remain a public health priority. There were three gaps this study sought to address. First, there is a paucity of alcohol density–related studies in NYC, a large urban metropolitan area with more than 8 million residents. Second, few studies have examined alcohol-related complaints as an outcome of alcohol density. Third, it is not known whether the area-level prevalence of heavy drinking connects alcohol outlet density to alcohol-related complaints.

We find that that off-premises alcohol outlet density was associated with higher rates of alcohol-related complaints at clubs, bars, and restaurants, and at a larger magnitude than an association between complaints elsewhere. The differences in size of association are partially supported by what amenity effects theory would predict, that certain places have higher likelihoods of alcohol-related incivilities as a function of alcohol density. Importantly, the off-premises association remained robust to adjustments for area-level prevalence of heavy drinking, current drinkers, and graffiti density per square mile—our indicator of physical disorder.

We did not find evidence that alcohol outlet density in adjacent neighborhoods was associated with the alcohol-related complaints in another neighborhood. While NYC’s population is highly mobile because of the city’s strong transit system, this finding does not seem to support a possibility that alcohol-related problems reported in one area could be a function of alcohol availability in neighboring ZIP codes. However, the lack of finding could also be because the geographic scales of ZIP codes are too large for such spillover effects to be observed, and the effect size may also be different at a lower geographic scale, such as census tracts—a phenomenon known as the modifiable areal unit problem [55].

On-premises alcohol outlet density was associated with a lower risk of complaints at clubs, bars, and restaurants, perhaps because of the higher probability of regulation within those places. For example, clubs and some restaurants where alcohol is served have security guards. In addition, the divergent direction of associations (e.g., higher risk of off-premises alcohol-related complaints compared to a lower risk of on-premises alcohol-related complaints) could possibly be explained by alcohol prevention strategies. One study [56] showed that off-premises outlets are more likely than on-premises outlets to have some protective practices in place (e.g., age screening, signs posting about sales to minors). Therefore, it is possible that higher reports could reflect higher visibility of these issues at clubs, bars, and restaurants given that more complaints from these locations were about underage drinking. The important finding, however, is that the lower association between on-premises outlets with both outcomes ceased to be significant when heavy drinking and the other mechanisms were added into the model.

The next findings add significantly to the literature on alcohol outlet density research. In adjusted models, the impact of off-premises alcohol density on alcohol-related complaints at clubs, bars, and restaurants and elsewhere was reduced by 8 % after adding area-level drinking factors and indicators of neighborhood social disorder. This finding suggests a small impact but nevertheless an important role for the area-level prevalence of heavy drinking because it was the only significant mechanism among the three, and the effect sizes were larger than the effect sizes for off-premises outlet density. This study’s findings are timely in the context of historical public health policies in NYC. One recent report based on 2014 data showed that alcohol use patterns, including binge drinking, were associated with multiple health-related behaviors (e.g., multiple sexual partners and hypertension) among adults in NYC [32]. From a historical perspective, the former Mayor of NYC Michael Bloomberg and Health Commissioner Thomas Farley considered legislation to decrease the city’s number of alcohol retail outlets and advertising and promotion of alcohol as a strategy to reduce public health consequences associated with heavy drinking [57, 58].

This study highlights one finding that warrants further analysis. Higher area-level Black/African-American racial composition was associated with lower risk of alcohol-related complaints at clubs, bars, and restaurants. We are unsure why we found a lower risk since other empirical research shows that alcohol outlets are disproportionately concentrated in Black neighborhoods and that alcohol-related outcomes are higher in Black neighborhoods [39, 59, 60]. Alternately, some potential reasons could be that in our study, off-premises alcohol density and alcohol-related complaints did not appear to be systematically concentrated in neighborhoods with higher Black racial concentration. Another possibility is that African-Americans are less likely to perceive disorder even when there are objective markers of disorder (e.g., graffiti, alcohol events) [61]. However, testing that hypothesis would require multilevel data about individual respondents’ races, which we do not have. Nevertheless, we observed a lower association of contextual-level racial composition with alcohol complaints independent of the observer’s race. There could be potential indicators other than race in these neighborhoods, such as individual characteristics, including age or sex, and neighborhood characteristics, including affluence [62], which are avenues to pursue in subsequent studies.

This study’s several limitations should be noted. In NYC Open Data, an alcohol-related event would be recorded only when a police officer responded to the complaint from the 311 dispatchers, and there is no way to identify potential under-coverage (e.g., if there were 311 calls about alcohol-related problems that a police officer did not respond to and log the complaint). In addition to the aforementioned possible differential reporting by individuals, the number of complaints (also relative to other types of 311 complaints) could reflect differences in resources across police departments to respond to citizens’ potential complaints. Another alternate explanation could be that alcohol-reported problems occur in areas already high in the prevalence of other problems such as noise or fights. Related, we do not know whether the problem is occurring at the club, or bar, etc., or whether the respondent is at that location because that level of detail is not available in the data. Next, although our study is possibly the first to demonstrate alcohol-related complaints as a potentially validated marker of alcohol-specific social disorder, this measure has not been sufficiently validated with other established correlates related to the consequences of alcohol, such as physical assault or alcohol-related vehicle accidents.

While our analysis makes use of a novel variable, our models may have been mis-specified, yet we are unsure of the potential impact on our study because of methodological differences in prior studies. Specifically, prior research showed that higher alcohol pricing and alcohol advertisements are associated with higher adverse drinking outcomes among adolescent individuals [63, 64], yet other research conceptualizes advertising and alcohol pricing as mediators [65] and thus would not control for them in the analysis. Nevertheless, none of those studies analyzed the ecological association between alcohol pricing and advertisement, but perhaps adjusting for these variables in our study could have reduced the impact of area-level prevalence association with the outcome. We could not address those limitations in this study because ESRI Business Analyst data did not contain these indicators, and those data are difficult to obtain.

Although we controlled for socioeconomic neighborhood effects, our PCA-based measure did not account for spatial correlation effects, and so the point estimates in the Poisson regression model may be biased. However, the impact of spatially-based socioeconomic indices versus PCA-based measures has not been sufficiently explored in alcohol-research, so we are unable to quantify the extent of potential bias. Next, alcohol density was derived from ESRI Business Analyst data—a secondary source, which may be an underestimate compared to what data might possibly have been obtained from the New York State Liquor Authority, which we did not have for this study.

Although we examined contextual-based variables such as the percentage of Black/African-American racial composition in an area, only multilevel models can decompose the contextual effects of between-place differences from the compositions of individuals [66]. Compositional effects cannot be controlled for or investigated in ecological studies [67] and remain a challenge to alcohol density research [68]. Finally, while we had some ZIP code data on the exposure and outcome across multiple years, there were no trends, and so our analysis was cross-sectional, which limited our ability to draw causal inferences from these findings.

Despite those limitations, our study provided new empirical evidence about the association between alcohol outlet density and alcohol-related complaints, a potentially novel marker that can be used in alcohol research in NYC, in the USA, and internationally. Many cities now have open-source 311 databases that log resident complaints, and so our analysis can be replicated elsewhere. We potentially validated alcohol-related complaints as a construct of social disorder through establishing a high correlation with graffiti—one observable marker of physical disorder. We also show that area-level heavy drinking is an important mechanism with stronger independent associations than alcohol outlets and structural factors such as vacant buildings or physical disorder. We recommend that further research continue to investigate whether area-level heavy drinking prevalence—a modifiable factor [69, 70]— is independently associated with health outcomes in NYC such as drug overdoses and HIV-related mortality.

Funding

Y. Ransome received funding from the Alonzo Smythe Yerby Postdoctoral Fellowship at Harvard T.H. Chan School of Public Health, and the National Institute of Mental Health K01MH111374. Support for data collection and analysis came from pilot funding from the Robert Wood Johnson Foundation Health and Society Scholars Program at Harvard University. D. Duncan was supported by grants from the National Institutes of Health and the Center for Disease Control and Prevention, including R01MH112406, U01PS005122, R21MH110190, and R03DA039748.