Neighbourhood deprivation and alcohol consumption: does the availability of alcohol play a role?

Abstract

Background Previous studies suggest that the physical availability of alcohol may mediate the association between neighbourhood-level material deprivation and alcohol consumption. This study tests the relationships between neighbourhood-level deprivation, alcohol availability, and individual-level alcohol consumption using a multilevel analysis.

Methods Data are from cross-sectional surveys conducted between 1979 and 1990 as part of the Stanford Heart Disease Prevention Program (SHDPP). Women and men (n = 8197) living in four northern/central California cities and 82 neighbourhoods were linked to neighbourhood deprivation variables derived from the US census (e.g. unemployment, crowded housing) and to measures of alcohol availability (density of outlets in the respondent's neighbourhood, nearest distance to an outlet from the respondent's home, and number of outlets within a half mile radius of the respondent's home). Separate analyses were conducted for on- and off-sale outlets.

Results The most deprived neighbourhoods had substantially higher levels of alcohol outlet density than the least deprived neighbourhoods (45.5% vs 14.8%, respectively). However, multilevel analyses showed that the least deprived neighbourhoods were associated with the heaviest alcohol consumption, even after adjusting for individual-level sociodemographic characteristics (OR 1.30, CI 1.08–1.56). Alcohol availability was not associated with heavy drinking and thus did not mediate the relationship between neighbourhood deprivation and heavy alcohol consumption.

Conclusions Although alcohol availability is concentrated in the most deprived neighbourhoods, women and men in least deprived neighbourhoods are most likely to be heavy drinkers. This mismatch between supply and demand may cause people in the most deprived neighbourhoods to disproportionately suffer the negative health consequences of living near alcohol outlets.

Heavy alcohol consumption is linked to multiple health problems, including motor vehicle crash injuries, domestic violence, and chronic diseases such as cancer and liver cirrhosis.¹ Because the environments in which people live shape many health behaviours,^2–4 there has been increased attention as to how neighbourhood environments may influence alcohol consumption and how environmental changes may decrease the burden of alcohol-related health problems.¹

Although studies have shown that higher individual-level socioeconomic status/position (SES) is associated with increased rates of drinking, studies investigating the relationship between neighbourhood-level deprivation and alcohol consumption have found varying results.² Karvonen and Rimpela⁵ found that teenage boys but not girls from neighbourhoods with high rates of unemployment had increased risk of alcohol consumption. Other studies have found little evidence of neighbourhood variation in alcohol consumption before and after controlling for individual-level indicators of SES.^6,7 Although Scribner et al.⁸ found significant variation in rates of alcohol consumption between neighbourhoods, this variation did not depend upon the level of neighbourhood deprivation.

Alcohol availability, measured by access to stores and restaurants that sell alcohol, may shed light on these inconclusive findings by helping to explain the potential link between neighbourhood deprivation and alcohol consumption. Previous research has examined multiple dimensions of this argument. First, a neighbourhood's level of deprivation has been associated with the number of alcohol outlets, with more outlets located in deprived neighbourhoods.^9,10 Second, alcohol availability is likely to be associated with increased alcohol consumption.^8,11,12 Third, the density of alcohol outlets may affect alcohol-related outcomes rather than consumption per se. For example, past studies have shown that higher density of alcohol outlets is associated with increased rates of youth drinking and driving,¹³ assault violence,¹⁴ and motor vehicle crashes¹⁵ at the city level after controlling for indicators of city deprivation. It has also been shown that higher density of alcohol outlets is associated with rates of injury¹⁶ and driving after drinking¹¹ at the neighbourhood level, and with homicide¹⁷ and rates of traffic injury¹² after controlling for aspects of neighbourhood deprivation. Finally, one intervention study suggests that decreasing access to alcohol outlets has decreased injuries resulting from night-time motor vehicle crashes and emergency room visits for assaults.¹⁸ We know of only one previous study that has specifically examined the relationship between neighbourhood deprivation, alcohol availability, and individual alcohol consumption, and this research demonstrated a potential mediating effect of alcohol availability on the relationship between neighbourhood deprivation and alcohol consumption.¹⁹

The present study examines whether neighbourhood-level deprivation is associated with alcohol consumption after taking into account individual-level sociodemographic characteristics, including a composite indicator of SES. Furthermore, it tests whether alcohol availability, measured in three different ways (density of alcohol outlets, closest distance to respondent's home, number of outlets within a 0.5 mile buffer zone), mediates this relationship. Individual survey data on drinking and individual geocoded addresses were linked to neighbourhood deprivation data from the US census and to measures of alcohol availability to examine these questions.

Methods

Data

The analysis was based on three data sources: cross-sectional survey data from the Stanford Heart Disease Prevention Program (SHDPP) 1979–1990 was linked to census-defined neighbourhood variables, and to alcohol availability records from the California Department of Alcoholic Beverage Control. The SHDPP was a long-term field trial designed to test whether a comprehensive programme of community organization and health education produced favourable changes in cardiovascular disease (CVD) risk.²⁰ The SHDPP drew participants from two treatment cities (Monterey, Salinas) and two control cities (Modesto, San Luis Obispo) in northern/central California, ranging in population size from 35 000 to 145 000 residents in 1980. To assess change in risk factors, five separate cross-sectional surveys of randomly selected households were conducted. All persons aged 12–74 years were eligible to participate and were invited to attend survey centres located in each city. The study is known for its comprehensive and careful assessment of individual risk factors and refined sampling methodology. Detailed descriptions of the study design and methodology have been published previously.²¹

Since few significant changes in risk factors and no significant changes in morbidity and mortality resulted from the intervention in treatment compared with control cities, the data for this analysis were pooled across cities.^22,23 The sample for this analysis included one woman and/or man per household aged 25–74 years, interviewed during one of the five cross-sectional surveys (n = 8419). The lower age cut-off point of 25 years was chosen to ensure that most individuals had completed their education. The sample was stable in terms of mobility: nearly 80% had lived in their community for >5 years or longer.

A number of factors were considered in defining the neighbourhood boundaries in the four cities. In order to be able to characterize neighbourhoods using census data, we chose a priori to rely upon census-defined boundaries, i.e. tracts and/or block groups, both of which have been used as proxies for geographically based neighbourhoods in previous research.^24–26 Since our study was conducted only in four cities, we also had the opportunity to verify the census-defined boundaries with archival neighbourhood maps. A further consideration was determining whether census-defined boundaries in the 1980 census were identical to those in the 1990 census. Accordingly, site visits were made to each city to meet with key contacts in the city planning departments to obtain neighbourhood maps and solicit advice on how each city defined its neighbourhoods at the time of the SHDPP. For the large majority of neighbourhoods, the boundaries corresponded well with single census tracts or block groups. When there was a difference (n = 12), we used a combination of tracts or block groups to better represent neighbourhood boundaries. As a result, a total of 82 neighbourhoods across the four cities were defined. The same neighbourhood boundaries were used for all the survey years and covered a median of 0.8 square miles (range 0.2–116.7). At the 1980 census, these neighbourhoods each contained a median of 363 people (range 16–1308). By the 1990 census, the population had increased in each neighbourhood to a median of 4251 (range 525–12153).

Addresses of alcohol outlets were obtained from the California Department of Alcoholic Beverage Control. Only licences listed as active at the time of each survey year were included. The licences were divided into on-sale outlets that allow for alcohol consumption on the premises (e.g. bars and restaurants) and off-sale outlets that allow alcohol to be purchased for consumption off the premises (e.g. convenience and liquor stores). A total of 921 outlets were located in the 82 neighbourhoods between 1979 and 1990. Of these, 56 outlets (6%) could not be successfully geocoded and were thus excluded, resulting in a total of 865 on- and off-sale alcohol outlets.

SHDPP respondents were linked to neighbourhoods and alcohol outlets based on geocoding their addresses. Following the methodology of Krieger et al.,²⁷ we tested the accuracy of the geocodes in two ways. Using the government geocoding website as the ‘gold standard’ (http://www.ffiec.gov/geocode/default.htm), we found that 95–98% (depending on the survey year) of a random sample of 173 participant records geocoded to the same 1990 census tract geocode as the geocoding service that we used. Similarly, 94% of a random sample of 34 alcohol outlets were geocoded to the same location as the geocoding service. In addition, we conducted a site visit in two of the cities with maps from the Bureau of the Census to determine the 'real world' accuracy of the geocodes using a convenience sample of 21 records. We found that 20 out of the 21 geocodes were located in the correct geographic area according to the Census maps. We excluded participants who reported an address that was not within one of the cities (n = 84 or 1.0%) and participants whose addresses could not be geocoded (n = 138 or 1.6%), resulting in a final analytical sample size of 8197. There was an average of 21 and a median of 17 participants per neighbourhood (range 1–107).

Dependent variable

The dependent variable was heavy alcohol consumption based on the total number of drinks consumed per week. SHDPP respondents were interviewed by nurses at the survey centres in each city and were asked separate questions about their consumption of beer or ale, wine, hard alcohol, and after-dinner liquors. The total number of drinks per week was then classified as heavy alcohol consumption (>7 drinks per week for females and >14 drinks per week for males) based on the increased risk of mortality associated with this level of drinking.¹

Independent variables at the individual level

The independent variables at the individual level were gender, age, race/ethnicity, marital status, and a composite measure of SES. The SES composite measure was calculated as the mean of two categorical variables, each with four levels: annual household income as a percentage of the federal poverty level (0–200%, 201–400%, 401–600%, and >600%) and educational attainment (<12, 12, 13–15, and ≥16 years of completed schooling). The Spearman correlation between income and education was 0.33. City and survey time were included as control variables.

Neighbourhood-level Townsend Material Deprivation Index

The Townsend Material Deprivation Index²⁸ was used as the measure of neighbourhood deprivation for the 82 neighbourhoods and was composed of four census variables (proportion of crowded occupied housing units, unemployed persons in the civilian labour force, tenant occupied housing units, and occupied housing units without a vehicle available). For the first survey (1979–1980), the Townsend Index was calculated from 1980 census data. For the last survey (1989–1990), the Index was calculated from 1990 census data. For surveys two through four, the Townsend Index was calculated through a linear interpolation of the four census variables using the values from the 1980 and 1990 censuses. Crowded housing and unemployment were first log transformed. Next each of the four variables were standardized separately by city and survey as a relative measure, (i.e. high deprivation in Monterey at survey 1 was considered to be qualitatively different than high deprivation in Modesto at survey 5), and then summed with equal weights. Higher numbers indicate higher levels of deprivation (mean 0, range 28.4 to 7.9). Because neighbourhood effects are thought to be non-linear, where significant effects are hypothesized to occur beyond a threshold,²⁹ the Townsend Index was categorized into three groups; below 1 SD from the mean (least deprived), above 1 SD from the mean (most deprived), and within 1 SD of the mean (moderately deprived). This categorization allowed us to assess both protective and harmful effects of living in the most and least deprived neighbourhoods in relation to respondents living in the moderately deprived neighbourhoods.

Alcohol availability

To our knowledge, no standard exists for the measurement of availability of alcohol outlets; therefore, we tested three different measures (density of alcohol outlets, closest distance to respondent's home, and number of outlets within a 0.5 mile buffer zone), with all outlets combined, and then separately for on-site vs off-site outlets. The density of alcohol outlets is a neighbourhood-level variable that represents the sum of alcohol outlets in a particular neighbourhood divided by the neighbourhood's square mileage. This variable was then dichotomized into high and low outlet densities with cut-off points specific for both survey year and city. High outlet density neighbourhoods were defined as being in the highest 25% of neighbourhoods within a particular city during a given survey year. The same method was used to define neighbourhoods as having a high density of on-sale and off-sale alcohol outlets.

The other measures of alcohol outlets were calculated at the individual-level using geographic information system software (ArcView version 3.3, ESRI). For multilevel analyses these variables were used in both a dichotomized and continuous form. One measure was the closest straight-line distance between a respondent's home and an alcohol outlet. Living near an alcohol outlet was defined as the 25% of respondents in a given city for a specific survey year who lived closest to an outlet. The other measure was the number of alcohol outlets within a circular 0.5 mile buffer zone around the respondent's home. This distance represents a 10–15 min walk which, for most people, is considered a maximum walking distance.³⁰ The 25% of respondents who had the highest number of alcohol outlets in their buffer zone (based on city and survey year) were classified as living in a high concentration buffer zone. Similar methods were used to determine the closest distance to on-sale and off-sale alcohol outlets and the number of on-sale and off-sale outlets in one's buffer zone.

Analysis

To examine the extent to which alcohol consumption varied by individual-level and neighbourhood-level factors, cross-tabulations with chi-square statistics were initially employed using SPSS, version 11.0. A series of multilevel logistic regression models with random intercepts were then examined using the SAS GLIMMIX macro (SAS Institute, Cary, NC). City, survey year, and the Townsend Deprivation Index were first entered into a model (model 1); then gender, age, race/ethnicity, and marital status were added (model 2); next the composite SES measure was added (model 3). The next set of models tested the potential mediating effect of alcohol availability with each of the measures added one at a time (model 4, with density of alcohol outlets shown). A final set of models tested the independent effect of alcohol outlets with only city and survey year included (data not shown). We also tested for random slopes for the composite SES measure and a cross-level interaction between the composite SES measure and the Townsend Deprivation Index.

Results

The nested structure of the dataset showed 8197 individuals living in 82 neighbourhoods in four different cities at five different time points. Table 1 presents the distribution of alcohol consumption per week for men and for women, and shows that women were more likely to abstain from alcohol. Table 2 presents characteristics of the sample population, with the percentage of respondents with high alcohol consumption. In bivariate analyses, it can be seen that men, and people who were white, non-Hispanic, previously or never married, in the highest SES group, and living in the least deprived neighbourhoods were significantly more likely to be heavy alcohol consumers than their counterparts. The percentage of heavy drinkers also varied by city and by survey year. The three measures of alcohol availability were not significantly related to high alcohol consumption; these measures were also not significantly related to high consumption when stratified by on-and off-site outlets (data not shown).

Density of alcohol outlets over time

The neighbourhoods showed an increasing density of alcohol outlets over time. In the 1979–80 survey, the number of outlets per square mile ranged between 0 and 7.9 (with 81.7% of the neighbourhoods having 0 outlets). By the final survey in 1989–90, the range was between 0 and 87.2 outlets per square mile, with only 4.9% of the neighbourhoods having no outlets. The other measures of alcohol outlets did not follow a clear trend over time. The distance to the nearest alcohol outlet ranged from an average of 0.66 miles in survey 1, to 0.43 miles in survey 2. Similarly, the average number of alcohol outlets in an individual's buffer zone was lowest in survey 1 (1.5 outlets) and highest in survey 2 (3.6 outlets).

Alcohol outlets and neighbourhood deprivation

Table 3 shows the distribution of alcohol outlets by the Townsend Deprivation Index. The most deprived neighbourhoods had the highest density of alcohol outlets, the highest percentage of individuals living near outlets, and the highest percentage with a high number of outlets in their buffer zone (P values ≤ 0.03). The most deprived neighbourhoods also had a significantly higher density of off-sale outlets; no significant differences were found for on-sale outlets.

Alcohol consumption, individual SES, and neighbourhood deprivation

Figure 1 shows heavy alcohol consumption as grouped by both individual SES and neighbourhood deprivation. In both the lowest and the highest SES groupings, people were significantly more likely to be heavy drinkers when they lived in the least deprived areas.

Multilevel analyses

In the multilevel analyses, model 1 showed that the odds of heavy alcohol consumption was significantly higher for people living in the least deprived neighbourhoods (OR 1.33) compared with those living in the moderately deprived neighbourhoods (Table 4). In model 2, which included demographic variables, the odds associated with living in the least deprived neighbourhood remained significant. Model 3 added individual-level SES. With the exception of Salinas, each variable that was significant in models 1 and 2 remained significant in model 3. In addition, people with the highest individual SES were significantly more likely to be heavy drinkers than people with the lowest SES. In each model, the survey year was significant, indicating that heavy alcohol consumption varied over time even after accounting for compositional factors. In the next set of models, each measure of alcohol availability was added to the previous model in order to test for potential mediating effects of neighbourhood deprivation on alcohol consumption. Only the model for outlet density is displayed because no measures related to alcohol availability (including both the dichotomized and continuous variables) reached statistical significance. In model 4, none of the alcohol availability measures appeared to have a mediating effect on the deprivation index. Similarly, none of the measures of alcohol availability reached significance in a model containing only city and survey year (data not shown).

Checking for random slopes of the individual-level composite SES measure did not find significant results (not shown), indicating that the effect of individual-SES did not vary across neighbourhoods. In addition, there was no significant cross-level interaction between the composite SES measure and the Townsend Deprivation Index (not shown), indicating that the effect of individual SES did not vary across neighbourhood deprivation.

Conclusion

The first question asked by this study was whether neighbourhood deprivation was related to alcohol consumption. Although the most deprived neighbourhoods had the highest density of alcohol outlets, living in the most deprived neighbourhoods was not related to heavy drinking. Rather, respondents who lived in the least deprived neighbourhoods had the highest levels of heavy alcohol consumption, even after controlling for a range of individual sociodemographic characteristics. The second question examined was whether alcohol availability, as measured by alcohol outlets, mediated any association between neighbourhood deprivation and heavy alcohol consumption. Alcohol availability was not associated with heavy consumption and, thus, did not mediate the relationship.

Why would living in the least deprived neighbourhoods be associated with greater alcohol consumption? One plausible explanation is that because heavy alcohol consumption is associated with higher individual SES, neighbourhood deprivation may be a proxy for unmeasured individual SES. This explanation seems unlikely given the finding that even respondents with the lowest individual SES were significantly more likely to be heavy drinkers if they lived in the least deprived areas (Figure 1). Instead, the social and cultural climate in the least deprived neighbourhoods may make alcohol consumption more acceptable.³¹ Such norms may be influenced by the rates and types of advertising that occur in different communities (depending, in part, on the sociodemographic profile of the community).³²

Another explanation that did not find support in our study is that neighbourhood deprivation is related to consumption through alcohol availability. Our findings using our measures of alcohol availability require a deeper examination of the ways in which people access alcohol and the extent to which access has an impact on behaviour. For example, the use of cars and public transportation may extend the distances people are willing to travel to purchase alcohol. This may be of greater importance in small city urban environments and suburbs, like the ones in the present study, where people may be more likely to drive than in the centre of larger cities.³³ This may be especially true when people commute to work in another neighbourhood or city. A better understanding of the division between time spent at home, at work, and in other environments, as well as how people move between these spaces, may better elucidate the relationship between alcohol availability and consumption. Furthermore, other factors may influence alcohol purchases including whether off-sale outlets sell other goods (i.e. groceries), taxes on alcohol, hours of store operation, and the concentration of outlets in surrounding neighbourhoods.³⁴ Unfortunately, these factors were not available for analysis in the present study.

The clustering of alcohol outlets in more deprived neighbourhoods warrants particular attention in light of the lack of association between alcohol availability and heavy drinking. Although residents of more deprived areas were less likely to be heavy drinkers, studies have shown that alcohol-related outcomes such as homicide, motor vehicle accidents, and assault are often clustered near alcohol outlets and in poorer communities. Thus, alcohol availability may not reveal how social disparities impact health through access to goods and services. Instead, alcohol availability may impact health by creating more hazardous physical areas in more deprived neighbourhoods. For example, a concentration of outlets may create areas of decreased monitoring and relaxed social restrictions where people congregate to drink. This may, in turn, give rise to a greater number of accidents and/or criminal offences and increased health risks.³⁵

From both a health and social justice perspective, the high level of alcohol outlet density, especially off-sale outlets, in the more deprived neighbourhoods requires careful examination of zoning practices, licensing procedures, and other system level policies. Revisions of licensing procedures must focus not only on ways of reducing the deleterious effects of heavy alcohol use (binge drinking, public disorderliness, etc.) but also the distribution of these effects across communities. Although debates over hours of operation and proof-of-age requirements, for example, are likely to be important,³⁶ they fail to address neighbourhood-level disparities that may continue to exist.

This study has a number of limitations. As it is cross-sectional in design, no causal inference can be drawn about the association between neighbourhood deprivation, alcohol availability, and alcohol consumption. Data from this study are from the 1980s and cannot be extrapolated to the current environment. The five different time periods suggest that over time the number and density of alcohol outlets increased, that outlets tended to become more concentrated in more deprived areas (analyses performed by each survey year, not shown), and that the percentage of heavy alcohol drinkers decreased. This finding is consistent with the decline of alcohol consumption in the United States that has been documented by previous studies.³⁷ Although the cause of this decline in consumption remains unclear, the concomitant decrease in consumption and increase in outlets strengthens the null findings in this study. Though restricted in its geographic scope of four cities in northern California, the data facilitated comparison among the cities and showed that rates of heavy drinking varied across cities. This suggests that place effects, where they exist, are likely to be specific for the locale and time period. Though the majority of participants in this study lived in their ‘community’ for 5 years or more, this study did not specifically address length of exposure because we cannot assume that participants' definition of community is the same as their definition of neighbourhood; future studies therefore need to include more refined assessments of exposure to neighbourhoods. Another limitation was the lack of more in-depth information about alcohol consumption and availability, including an assessment of binge drinking, places where alcohol was purchased, and alcohol availability in surrounding neighbourhoods.

Two further methodological concerns warrant attention. First, sampling was by household with more than one member of each household included. To counter the potential correlation between household members, ancillary analyses were conducted in which one member of the household was randomly selected for inclusion. The only appreciable difference in results was that the town of Monterey became a significant variable in all multilevel models (OR 1.28, CI 1.02–1.60 in model 1). Second, this study relied upon geographically defined census boundaries, which were verified with site visits, interviews, and archival data. Nonetheless, census boundaries may not accurately conform to neighbourhood boundaries as defined by residents. Moreover, it has been argued that neighbourhoods should be defined on the basis of patterns of social interactions.

The strengths of this study include its use of comprehensive and careful assessment of individual risk factors in face-to-face interviews conducted by health professionals, its careful construction of neighbourhood boundaries, accurate coding of residences and alcohol outlets, and multiple measures of alcohol availability.

In conclusion, we found that living in the least deprived neighbourhoods was associated with heavier alcohol consumption compared with living in the most deprived neighbourhoods, and that this relationship was not mediated by the availability of alcohol. In addition, we found a clustering of alcohol outlets in the most deprived neighbourhoods. Careful attention by researchers and policy makers is warranted to better understand why alcohol availability is concentrated in poorer neighbourhoods although residents in these areas are less likely to be heavy drinkers. This seeming mismatch between supply and demand may cause people living in the most deprived neighbourhoods to disproportionately suffer the negative health consequences of living near alcohol outlets.

The authors thank Naomi Kawakami for her technical assistance in geographical information systems and Alana Koehler for her technical assistance in preparing the tables and figures. This work was co-funded by the National Institute of Environmental Health Sciences and the National Heart, Lung, and Blood Institute: Grant RO1 HL67731 to Dr. Winkleby.

References