Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Jun 1.
Published in final edited form as: Drug Alcohol Depend. 2015 Mar 24;151:144–150. doi: 10.1016/j.drugalcdep.2015.03.014

The roles of outlet density and norms in alcohol use disorder*

Jennifer Ahern a, Laura Balzer b, Sandro Galea c
PMCID: PMC4447528  NIHMSID: NIHMS675213  PMID: 25858787

Abstract

Background

Alcohol outlet density and norms shape alcohol consumption. However, due to analytic challenges we do not know: (a) if alcohol outlet density and norms also shape alcohol use disorder, and (b) whether they act in combination to shape disorder.

Methods

We applied a new targeted minimum loss-based estimator for rare outcomes (rTMLE) to a general population sample from New York City (n=4000) to examine the separate and combined relations of neighborhood alcohol outlet density and norms around drunkenness with alcohol use disorder. Alcohol use disorder was assessed using the World Mental Health Comprehensive International Diagnostic Interview (WMH-CIDI) alcohol module. Confounders included demographic and socioeconomic characteristics, as well as history of drinking prior to residence in the current neighborhood.

Results

Alcohol use disorder prevalence was 1.78%. We found a marginal risk difference for alcohol outlet density of 0.88% (95% CI 0.00%–1.77%), and for norms of 2.05% (95% CI 0.89%–3.21%), adjusted for confounders. While each exposure had a substantial relation with alcohol use disorder, there was no evidence of additive interaction between the exposures.

Conclusions

Results indicate that the neighborhood environment shapes alcohol use disorder. Despite the lack of additive interaction, each exposure had a substantial relation with alcohol use disorder and our findings suggest that alteration of outlet density and norms together would likely be more effective than either one alone. Important next steps include development and testing of multi-component intervention approaches aiming to modify alcohol outlet density and norms towards reducing alcohol use disorder.

Keywords: Alcohol-Related Disorders, Residence Characteristics, Alcohol Outlets, Social Norms, Semi-parametric Estimation

1. INTRODUCTION

A substantial body of research indicates that characteristics of the physical and social environment play a role in patterns of alcohol use (Becares et al., 2012; Campbell et al., 2009; Karriker-Jaffe, 2011; Lindstrom, 2008; Murayama et al., 2012; Popova et al., 2009; Renalds et al., 2010). Interventions informed by this work that target physical and social aspects of environments have been developed and shown to be efficacious (Campbell et al., 2009; Moreira et al., 2009). Alcohol outlet density and group norms around alcohol consumption are two of the features of the alcohol environment for which the best evidence exists.

Higher alcohol outlet density is related to more alcohol consumption and related harms. A recent systematic review of outlet density and hours and days of sale concluded that these aspects of alcohol access are related to drinking levels and drinking related harms (Popova et al., 2009). Interventions to reduce outlet density indicate some success; two studies have documented violence reductions following outlet density reduction and targeted outlet closures (Xu et al., 2012; Yu et al., 2008). A college based program that included alcohol access restrictions among a set of activities documented modest reductions in alcohol consumption and related problems (Weitzman et al., 2004).

Individual perceptions of norms have been consistently found to influence drinking behavior; in the college setting, individual web-based approaches to correct students’ misperceptions of how peers drink have effectively reduced drinking related harms (Moreira et al., 2009). Research on group norms in workplace and community environments has found that norms against drinking and drunkenness are related to lower levels of those behaviors (Ahern et al., 2008; Barrientos-Gutierrez et al., 2007). Interventions that target norms at the group level through broad social marketing campaigns have documented success in reducing consumption among adolescents and college students (DeJong et al., 2006; Stead et al., 2007); for general adult populations there is less evidence, but some research suggests drinking related risks, such at drinking and driving, can be reduced with social marketing campaigns (Perkins et al., 2010).

It is important to understand of how aspects of environments in combinations may contribute to alcohol consumption because many interventions include multiple components and combined relations need to be understood to inform multi-component intervention designs. For example, outlet density and norms could have a strong positive interaction such that there is no effect on alcohol use unless both are targeted. In contrast, there could be negative interaction such that once outlet density or norms are altered, the other has no additional effect on the alcohol use. This scenario would indicate that interventions could target outlets or norms, depending on which was more feasible. Two studies have examined alcohol outlet density or access together with aspects of drinking norms; in those studies both characteristics were related to alcohol use, but a possible combined relation or interaction was not examined (Paschall et al., 2012; Weitzman et al., 2003). We are aware of one study that has considered the relation of outlet density and norms in combination related to drinking behavior. A study of a social norms marketing campaign to reduce college drinking found that the effect of the campaign was modified by alcohol outlet density such that the intervention had an effect where density was low, but did not have an effect where density was high (Scribner et al., 2011). This suggests the positive interaction scenario above, with protection conferred by norms only when outlet density is low, but this question has not been examined in a general population sample.

Across research on how the physical and social environment relate to alcohol use, it is notable that the more severe potential consequences, such as alcohol use disorder, have not been frequently studied. This is, in part, because uncommon outcomes are challenging to analyze due to data sparsity (e.g., the number of outcome events is smaller than the number of confounder strata; Harrell et al., 1996; King and Zeng, 2001; Peduzzi et al., 1996). Recently, Balzer and colleagues proposed an extension of targeted minimum loss-based estimation (TMLE) to facilitate the examination of rare outcomes (Balzer et al., under review; Balzer and van der Laan, 2013; van der Laan and Rose, 2011). TMLE is a semi-parametric algorithm, incorporating machine learning methods with an additional targeting step to reduce bias. The rare outcomes TMLE (rTMLE) further incorporates an upper bound on the risk of the infrequent outcome.

In this paper, we examine the relation of neighborhood alcohol outlet density and norms around drunkenness with alcohol use disorder using the rare outcomes analysis approach rTMLE. We consider the possible combined relation of outlet density and norms with alcohol use disorder. Specifically, we estimate the marginal risk of alcohol use disorder under exposure and no exposure to high outlet density, permissive drunkenness norms, and combinations of these exposures. If assumptions for causality are met (elaborated in the discussion), these risks can be interpreted as the counterfactual risks of alcohol use disorder under different exposure conditions. We estimate all relations on the additive scale rather than the multiplicative scale, because it is more relevant to public health and more appropriate when considering combined relations or interaction (Greenland, 1993; Poole, 2010).

2. MATERIAL AND METHODS

New York Social Environment Study (NYSES) is a multilevel study designed to examine neighborhood level exposures, including economic, social and structural characteristics, and mental health and substance use in New York City (NYC). The NYSES was conducted in 2005; we used random-digit-dial methods to contact and interview 4,000 NYC residents. Only land-line telephones were included due to rules prohibiting random-digit-dialing to cell phones at that time; only 5% of households were cell phone only in 2005 (Blumberg and Luke, 2011). One adult 18 years or older was interviewed by telephone in each household; the respondent was the person who either most recently or would next celebrate their birthday (randomly selected). Interviews were conducted in English or Spanish. The cooperation percentage was 54%. Respondents were offered $10 in compensation for their participation. Informed consent was obtained from all participants, and the study protocol was approved by the institutional review boards of the New York Academy of Medicine, the University of Michigan, and the University of California, Berkeley.

2.1 Neighborhoods

Respondents provided information about their residential address or nearest cross-streets so that their locations could be geocoded and linked to their neighborhoods of residence (Ahern et al., 2009). The neighborhood units were the 59 community districts in NYC. Community districts are recognizable neighborhood areas, each headed by an administrative community board, and many characteristics of these neighborhoods have been associated with aspects of health (see for example Ahern and Galea, 2011; Ahern et al., 2008, 2009; Galea et al., 2007, 2005).

2.2 Measures

Respondents were interviewed with a structured questionnaire that included questions on demographic and socioeconomic characteristics that were confounders of the relation of interest including age, race/ethnicity, gender, marital status, place of birth, interview language, education, income, employment status, and years lived in the current neighborhood. Details on the confounders can be found in Table 1.

Table 1.

New York Social Environment Study characteristics and bivariable associations with alcohol use disorder

Alcohol Use
Disorder
N % N % P-valuea
Total 4000 100.00 71 1.78
Age
  18–24 350 8.84 17 4.86 <0.01
  25–34 685 17.30 21 3.07
  35–44 815 20.58 13 1.60
  45–54 808 20.40 15 1.86
  55–64 612 15.45 5 0.82
  ≥ 65 690 17.42 0 0.00
Race/ethnicity
  White 1616 41.56 30 1.86 0.38
  African American 1055 27.13 17 1.61
  Asian 164 4.22 6 3.66
  Hispanic 958 24.64 18 1.88
  Other 95 2.44 0 0.00
Gender
  Male 1880 47.00 48 2.55 <0.01
  Female 2120 53.00 23 1.08
Marital Status
  Married 1632 41.39 16 0.98 <0.01
  Divorced 479 12.15 7 1.46
  Separated 208 5.28 8 3.85
  Widowed 354 8.98 2 0.57
  Never Married 1270 32.21 38 2.99
Birth Place
  NYC 1810 45.86 32 1.77 0.45
  Other US Location 731 18.52 17 2.33
  Different Country 1406 35.62 22 1.56
Interview Language
  English 3545 88.63 67 1.89 0.13
  Spanish 455 11.38 4 0.88
Education
  Less than High School 508 12.95 6 1.18 0.67
  High School/GED 923 23.53 16 1.73
  Some College 879 22.41 15 1.71
  College Graduate 883 22.51 17 1.93
  Graduate Work 730 18.61 17 2.33
Income
  ≤ $40,000 1605 40.13 31 1.93 0.14
  $40,001 – $80,000 1093 27.33 23 2.10
  > $80,000 722 18.05 14 1.94
  Missing 580 14.50 3 0.52
Unemployed
  No 3679 91.98 63 1.71 0.31
  Yes 321 8.03 8 2.49
Years Lived in Neighborhood
  <8 yrs 1330 33.39 44 3.31 <0.01
  8–21 yrs 1318 33.09 22 1.67
  >21 yrs 1335 33.52 5 0.37
Drinking before Moved to Neighborhood
  Ever Drank/Tried Drinking 706 17.65 3 0.42 <0.01
  Monthly Drinker 1948 48.70 55 2.82
  Never Drank 1346 33.65 13 0.97
Neighborhood Alcohol Outlet Density
  Low 2080 52.00 26 1.25 0.01
  High 1920 48.00 45 2.34
Neighborhood Drunkenness Norms
  Restrictive 1979 49.48 25 1.26 0.02
  Permissive 2021 50.53 46 2.28
Open in a new tab
a

chi-square p-value comparing alcohol use disorder by covariate categories

Neighborhood drinking norms were measured using questions modified from the National Survey on Drug Use and Health (Ahern et al., 2008; NSDUH). Respondents were asked their opinion about adults getting drunk at least once a week, and given the response options of “acceptable”, “unacceptable” and “don’t care”. The neighborhood measures were the proportions of residents who believed it was “unacceptable” for adults to get drunk at least once a week in each neighborhood. For analysis, norms were dichotomized at the median as more permissive (exposed) and less permissive (unexposed); this avoided analysis challenges we would have faced with positivity violations (Petersen et al., 2012).

Alcohol license data were acquired from the New York State Division of Alcoholic Beverage Control/State Liquor Authority using the online public license query. We calculated the number of off-premise active liquor licenses per neighborhood; these are licenses that permit only the retail sale of alcoholic beverages directly to consumers for their consumption (i.e., alcoholic beverages may not be sold for resale). Alcoholic beverages cannot be consumed on the premises of establishments holding these licenses, nor are any open containers of alcoholic beverages allowed on the premises. License type abbreviations included in this category are: A, AX, C, DS, DX, E, L, ST, W, and AW (class 122). Density of alcohol outlets was calculated as outlets per square mile. For analysis, the density was dichotomized as above the median (exposed) or below the median (unexposed).

Drinking behavior was assessed using the World Mental Health Comprehensive International Diagnostic Interview (WMH-CIDI) alcohol module (Kessler et al., 2004; Kessler and Ustun, 2004). The outcome for this analysis was symptoms consistent with alcohol use disorder, assessed using the DSM-V criteria within the past 12 months. This is a combination of the previously distinct alcohol abuse and dependence classification, consistent with the DSM-V in which these are grouped in one diagnosis (Hagman and Cohn, 2011). The WMH-CIDI alcohol measures that include a retrospectively recalled history of alcohol use were used to capture the history of drinking prior to residence in the current neighborhood.

2.3 Analysis

For our primary analysis, we applied a targeted minimum loss-based estimation (TMLE) approach designed for examination of rare outcomes, known as rTMLE (Balzer et al., under review; Balzer and van der Laan, 2013; van der Laan and Rose, 2011). In general, TMLE is a semi-parametric substitution estimator that involves estimation of the average outcome conditional on exposure and covariates (i.e., the outcome regression) as well as the probability of being exposed, conditional on the covariates (i.e., the exposure mechanism or the propensity score). Both the outcome regression and exposure mechanism were estimated using SuperLearner, a machine-learning approach, which avoids the potential bias inherent in assuming a parametric form for the relations amongst variables (Sinisi et al., 2006; van der Laan et al., 2007). In particular, SuperLearner uses cross-validation to create the best combination of algorithm specific estimates from a library of algorithms. In this application, the library for outcome regression included the unadjusted mean, stepwise logistic regression, all logistic regressions with a single main term, and all logistic regressions with a main term for the exposure and one additional covariate. The library for the exposure mechanism included all logistic regressions with a single main term as well as stepwise logistic regression with and without interactions. These routines were selected for interpretability and to avoid instability due to data sparsity. Analyses were adjusted for demographic and socioeconomic characteristics as well as history of drinking prior to residence in the current neighborhood (confounders are listed in Table 1).

In TMLE, the initial estimator of the outcome regression is modified to reduce bias for the parameter of interest and to ensure valid inference. This modification augments the initial outcome regression with information in the estimated exposure mechanism. rTMLE further incorporates bounds on the conditional risk of the rare outcome within exposure-covariate strata (code is available from the authors upon request). In this application, the lower bound was set to the natural limit of 0 and the upper bound of 10% was selected with cross-validation from the set of possible upper bounds (2.5%, 5%, 7.5%, 10%, 12.5%). Results based on upper bounds at 7.5% and 12.5% yielded equivalent results.

Overall, we estimated the marginal risk of alcohol use disorder under exposure and no exposure to high outlet density, and under exposure and no exposure to permissive drunkenness norms, controlling for individual covariates. We then estimated the risk of alcohol use disorder under the four combinations of these exposures, analogous to joint or simultaneous intervention on these two neighborhood characteristics, controlling individual covariates. Estimates were compared on the absolute scale (i.e. marginal risk differences). Confidence intervals were based on the estimated influence curve, with clustering by neighborhood (Schnitzer et al., 2014).

In recent years, a variety of methods have been developed that allow the scientific question to define the parameter of interest, in our case the marginal risk difference, rather than having the analysis approach define the parameter of interest (e.g., if logistic regression is utilized, the parameter estimated is the conditional odds ratio). For the purposes of comparison, we estimated the same relations using the other estimators that allow this flexibility, including the standard TMLE algorithm (van der Laan and Rose, 2011), inverse probability of treatment weighting (IPTW) and augmented-IPTW (A-IPTW; Hernan et al., 2000; Robins, 1999; van der Laan and Robins, 2003). The latter algorithms are based on solving an estimating equation and are thereby not substitution estimators.

3. RESULTS

Study participants were demographically similar to the overall population of New York City based on the 2000 census, with 41.6% White, 27.1% African American, 4.2% Asian, 24.6% Hispanic, and 2.4% of other racial groups. Mean age was 47 years (range 18 – 94), 53.0% of respondents were female, and 35.6% were born outside the United States. As expected, alcohol use disorder was uncommon; the past 12 months prevalence was 1.78%. A full description of the sample, and bivariable associations of covariates with alcohol use disorder are provided in Table 1. Outlet density ranged from 9.8 to 132.1 outlets per square mile, with a median of 58.3. The proportions of residents who believed it was “unacceptable” for adults to get drunk at least once a week in each neighborhood ranged from 53% to 89% with a median of 80%.

Based on our analysis with rTMLE, the marginal estimated risk of alcohol use disorder under high alcohol outlet density in all neighborhoods was 2.20%, and under low outlet density it was 1.32%. The marginal difference in the risk of alcohol use disorder associated with high alcohol outlet density in all neighborhoods compared with low density in all neighborhoods was 0.88% (95% CI 0.00%–1.77% - the lower confidence limit does not cross the null, the p-value is less than 0.05), adjusted for confounders.

In examination of alcohol norms, the marginal estimated risk of alcohol use disorder under permissive norms in all neighborhoods was 3.15%, and under restrictive norms it was 1.10%; this translated to a marginal risk difference of 2.05% in alcohol use disorder associated with permissive norms compared with restrictive norms (95% CI 0.89%–3.21%), adjusted for confounders.

When considering the combined relation of outlet density and norms, we found that under conditions of low outlet density and restrictive norms the estimated risk of alcohol use disorder was 0.70% - this was the estimated baseline level of the outcome in the absence of either exposure. Under permissive norms, leaving outlet density low, the estimated risk of alcohol use disorder was 2.55%; thus we estimated a marginal risk difference of 1.85% from a change in norms in the absence of high outlet density (the difference between 2.55% and 0.70%). Under high outlet density, leaving norms restrictive, the estimated risk of alcohol use disorder was 1.49%; this indicated an estimated marginal risk difference of 0.79% for a change in outlet density in the absence of permissive norms (the difference between 1.49% and 0.70%). Based on these values, we calculated the expected value of alcohol use disorder in the absence of additive interaction between norms and outlets. The expected value was the baseline level, plus the relation for each exposure in the absence of the other: 0.70%+1.85%+0.79% = 3.34% (Rothman et al., 2008). The estimated risk of alcohol use disorder under conditions of permissive norms and high outlet density was 3.43%, only slightly higher than that expected in the absence of interaction (p-value for the difference from the expected value = 0.22). This indicated that there was no additive interaction between outlet density and norms in relation to alcohol use disorder.

Analysis results from TMLE, IPTW, and A-IPTW were similar to those from rTMLE, with estimates of the marginal risk difference for outlet density between 0.87% and 1.00%, for norms between 2.01% and 2.10%. Differences between the expected risks in the absence of interaction and estimated risks under both exposures were smaller, there was no significant additive interaction between outlets and norms in relation to alcohol use disorder (see Supplementary Material for full results1).

4. DISCUSSION

Among a general population sample of adults in New York City, we found that both higher alcohol outlet density and permissive norms around drunkenness had substantial and statistically significant relations with alcohol use disorder, with a stronger relation for norms. We did not find evidence of additive interaction between the two neighborhood characteristics in relation to alcohol use disorder.

The findings on outlets and norms are consistent with past research that indicates these exposures are related to alcohol use and alcohol related harms, and add to that research by providing evidence that these exposures also shape alcohol use disorders (Ahern et al., 2008; Barrientos-Gutierrez et al., 2007; Popova et al., 2009). This suggests that the environment plays a role in both behaviors and in clinical outcomes, and thus effective intervention with the environment has the potential to reduce the need for treatment in the population.

There are several limitations to this study. The cooperation percentage was 54%, which is consistent with many other recent telephone based studies (Galea and Tracy, 2007). However, this cooperation percentage does raise concern about how well the study sample represents the city of New York. Participants were informed that they would be participating in a “survey about the neighborhoods where New Yorkers live and what people think about their neighborhoods”, and thus they were not likely to refuse based on their alcohol use. Rules prohibited random-digit-dialing to cell phones in 2005, and thus we were unable to contact the 5% of households that were cell phone only at that time (Blumberg and Luke, 2011). The distribution of demographic characteristics such as age, race, and sex is very similar to the 2000 Census data for NYC. However, the individuals we were able to reach and who agreed to participate may still differ from those in the City overall in ways that we were unable to capture. Furthermore, the survey was only administered in English and Spanish due to cost constraints, and thus individuals who could not respond in one of these languages are not represented. Self-report is standard practice in alcohol research and telephone interviews are thought to elicit more accurate reports than in-person interviews (Midanik et al., 2001); however, there may be differences between a clinical diagnosis of alcohol use disorder and our assessment of symptoms consistent with alcohol use disorder. This study examines only off-premises alcohol outlet density. It was not possible to distinguish restaurants from bars in the on-premises data, and thus relations of this heterogeneous group of establishments with alcohol use disorder would have been difficult to interpret. Finally, we were not able to examine whether outlets and norms influence each other because we lacked time varying data on these exposures and alcohol use disorder; this is an important avenue for future research.

Among several strengths, this study included a large population based sample. Social selection has been considered one of the major barriers to determining whether the environment has an influence on people, or whether people who have worse health are unable to leave or move into worse types of environments (Oakes, 2004; Sampson, 2003). We accounted for one potential contributor to social selection by adjusting for history of drinking prior to each person’s residence in his/her current neighborhood. Associations observed were not due to an effect of people with certain drinking histories being likely to move to high alcohol outlet density or permissive drunkenness norm neighborhoods. Our application of a rare outcomes analysis approach, rTMLE, facilitated examination of alcohol use disorder despite its infrequent occurrence. Estimates from this new method were similar to those from the TMLE algorithm, IPTW and A-IPTW. Theoretical and finite sample simulations, however, suggest that the rTMLE algorithm is more robust and provides more power estimate associations with rare outcomes (Balzer et al., under review; Balzer and van der Laan, 2013). We thus focused on rTMLE results in our discussion.

The estimated associations of alcohol outlet density and drunkenness norms with alcohol use disorder do not necessarily represent how alcohol use disorder might change following a policy or intervention that reduces one or both of these exposures (Kaufman, 2007; Oakes, 2004; Sampson, 2003). The assumptions necessary for causal interpretation of associations in observational research generally and observational neighborhood research specifically have been well elaborated elsewhere (van der Laan and Robins, 2003; Vanderweele, 2008). In brief, exposure must precede the outcome (temporal ordering), all confounders must be controlled (ignorability), exposures in one neighborhood cannot affect the potential outcomes of individuals in other neighborhoods (neighborhood-level stable unit treatment value assumption), and the outcomes observed for a given exposure value must reflect those that would have been observed if the exposure had been counterfactually assigned to that value (consistency assumption). Given the cross-sectional design of our study, we cannot establish temporal ordering between the exposure and outcome. Longitudinal consideration of these relations will be necessary to establish that temporal relation and improve the potential for causal interpretation. While we controlled many confounders, notably including history of drinking prior to residence in the current neighborhood, ignorability cannot be assessed empirically and can only ever be approximated with observational data. In the analyses of outlet density and norms separately, we also have to assume the exposures are independent of each other, conditional on the confounders; analyses relaxing this assumption yielded similar results. The stability assumption may be violated since the outlet density or norms in one area could affect potential outcomes in another area, particularly near the borders of neighborhoods.

Our findings support frameworks that consider different elements of the context in which behaviors such as drinking take place, and emphasize the importance of considering these in combinations for intervention (Cohen et al., 2000; Wechsler and Nelson, 2008). Despite the lack of additive interaction between outlet density and norms in relation to alcohol use disorder, each exposure had a substantial relation with alcohol use disorder. Our results indicate that alteration of both aspects of the context would likely result in a larger reduction in alcohol use disorder than either one alone. Of course, there are substantial challenges in translating this research into intervention. While social norms marketing approaches are well developed for the college setting, they are less developed for general populations (DeJong et al., 2006; Perkins et al., 2010; Stead et al., 2007). Although there are some models for reducing outlet density, it can be done in a variety of ways, such as targeting problematic outlets, or waiting for attrition of businesses, and these may have distinct effects on outcomes. Access can also be altered by modifying pricing or hours and days of sale without outlet closures (Popova et al., 2009). Important next steps include development and testing of multi-component environment intervention approaches for general populations, building on what has been learned from intervention in college and other settings (DeJong et al., 2006; Moreira et al., 2009; Perkins et al., 2010; Stead et al., 2007; Wechsler and Nelson, 2008; Weitzman et al., 2004; Xu et al., 2012; Yu et al., 2008).

Supplementary Material

  • We examine the relations of outlet density and norms with alcohol use disorder.

  • A new targeted minimum loss-based estimator for for rare outcomes is applied.

  • Both neighborhood exposures have substantial relations with alcohol use disorder.

  • There is no additive interaction between outlet density and norms.

Acknowledgments

Funding for this work was provided in part by the National Institutes of Health (DA 017642, MH 082729, HD 080350).

Role of Funding Source

The funder had no role: in the study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

*

Supplementary material can be found by accessing the online version of this paper at http://dx.doi.org and by entering doi:...

Contributors

JA collaborated on the design and implementation of the study, conceptualized the analysis, conducted the literature review, and wrote the article. LB developed the analysis approach, implemented the analysis, and substantially edited all sections of the article. SG obtained study funding, collaborated on the design and implementation of the study, and substantially edited all sections of the article. All authors have approved the final article.

Conflict of Interest

No conflict declared.

Contributor Information

Jennifer Ahern, Email: jahern@berkeley.edu.

Laura Balzer, Email: lbbalzer@berkeley.edu.

Sandro Galea, Email: sgalea@bu.edu.

REFERENCES

  1. Ahern J, Galea S. Collective efficacy and major depression in urban neighborhoods. Am. J. Epidemiol. 2011;173:1453–1462. doi: 10.1093/aje/kwr030. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Ahern J, Galea S, Hubbard A, Midanik L, Syme SL. "Culture of drinking" and individual problems with alcohol use. Am. J. Epidemiol. 2008;167:1041–1049. doi: 10.1093/aje/kwn022. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ahern J, Galea S, Hubbard A, Syme SL. Neighborhood smoking norms modify the relation between collective efficacy and smoking behavior. Drug Alcohol Depend. 2009;100:138–145. doi: 10.1016/j.drugalcdep.2008.09.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Balzer L, Ahern J, Galea S, van der Laan MJ. Estimating effects on rare outcomes: knowledge is power. doi: 10.1515/em-2014-0020. (under review). [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Balzer LB, van der Laan MJ. Technical Report. Berkeley: Division of Biostatistics, University of California; 2013. Estimating effects on rare outcomes: knowledge is power; p. 310. http://biostats.bepress.com/ucbbiostat/paper310/ [Google Scholar]
  6. Barrientos-Gutierrez T, Gimeno D, Mangione TW, Harrist RB, Amick BC. Drinking social norms and drinking behaviours: a multilevel analysis of 137 workgroups in 16 worksites. Occup. Environ. Med. 2007;64:602–608. doi: 10.1136/oem.2006.031765. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Becares L, Nazroo J, Jackson J, Heuvelman H. Ethnic density effects on health and experienced racism among Caribbean people in the US and England: a cross-national comparison. Soc. Sci. Med. 2012;75:2107–2115. doi: 10.1016/j.socscimed.2012.03.046. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Blumberg SJ, Luke JV. Wireless substitution: early release of estimates from the National Health Interview Survey, January–June 2011. National Center for Health Statistics. 2011 [PubMed] [Google Scholar]
  9. Campbell CA, Hahn RA, Elder R, Brewer R, Chattopadhyay S, Fielding J, Naimi TS, Toomey T, Lawrence B, Middleton JC. The effectiveness of limiting alcohol outlet density as a means of reducing excessive alcohol consumption and alcohol-related harms. Am. J. Prev. Med. 2009;37:556–569. doi: 10.1016/j.amepre.2009.09.028. [DOI] [PubMed] [Google Scholar]
  10. Cohen DA, Scribner RA, Farley TA. A structural model of health behavior: A pragmatic approach to explain and influence health behaviors at the population level. Prev. Med. 2000;30:146–154. doi: 10.1006/pmed.1999.0609. [DOI] [PubMed] [Google Scholar]
  11. DeJong W, Schneider SK, Towvim LG, Murphy MJ, Doerr EE, Simonsen NR, Mason KE, Scribner RA. A multisite randomized trial of social norms marketing campaigns to reduce college student drinking. J. Stud. Alcohol. 2006;67:868–879. doi: 10.15288/jsa.2006.67.868. [DOI] [PubMed] [Google Scholar]
  12. Galea S, Ahern J, Nandi A, Tracy M, Beard J, Vlahov D. Urban neighborhood poverty and the incidence of depression in a population-based cohort study. Ann. Epidemiol. 2007;17:171–179. doi: 10.1016/j.annepidem.2006.07.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Galea S, Ahern J, Rudenstine S, Wallace Z, Vlahov D. Urban built environment and depression: a multilevel analysis. J. Epidemiol. Community Health. 2005;59:822–827. doi: 10.1136/jech.2005.033084. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Galea S, Tracy M. Participation rates in epidemiologic studies. Ann. Epidemiol. 2007;17:643–653. doi: 10.1016/j.annepidem.2007.03.013. [DOI] [PubMed] [Google Scholar]
  15. Greenland S. Additive risk versus additive relative risk models. Epidemiology. 1993;4:32–36. doi: 10.1097/00001648-199301000-00007. [DOI] [PubMed] [Google Scholar]
  16. Hagman BT, Cohn AM. Toward DSM-V: mapping the alcohol use disorder continuum in college students. Drug Alcohol Depend. 2011;118:202–208. doi: 10.1016/j.drugalcdep.2011.03.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat. Med. 1996;15:361–387. doi: 10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4. [DOI] [PubMed] [Google Scholar]
  18. Hernan MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology. 2000;11:561–570. doi: 10.1097/00001648-200009000-00012. [DOI] [PubMed] [Google Scholar]
  19. Karriker-Jaffe KJ. Areas of disadvantage: a systematic review of effects of area-level socioeconomic status on substance use outcomes. Drug Alcohol Rev. 2011;30:84–95. doi: 10.1111/j.1465-3362.2010.00191.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kaufman JS. Making causal inferences about macrosocial factors as a basis for public health policies. In: Galea S, editor. Macrosocial Determinants of Population Health. New York: Springer; 2007. [Google Scholar]
  21. Kessler RC, Abelson J, Demler O, Escobar JI, Gibbon M, Guyer ME, Howes MJ, Jin R, Vega WA, Walters EE, Wang P, Zaslavsky A, Zheng H. Clinical calibration of DSM-IV diagnoses in the World Mental Health (WMH) version of the World Health Organization (WHO) Composite International Diagnostic Interview (WMHCIDI) Int. J. Methods Psychiatr. Res. 2004;13:122–139. doi: 10.1002/mpr.169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kessler RC, Ustun TB. The World Mental Health (WMH) Survey Initiative Version of the World Health Organization (WHO) Composite International Diagnostic Interview (CIDI) Int. J. Methods Psychiatr. Res. 2004;13:93–121. doi: 10.1002/mpr.168. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. King G, Zeng L. Logistic regression in rare events data. Polit. Anal. 2001;9:137–163. [Google Scholar]
  24. Lindstrom M. Social capital and health-related behaviors. In: Kawachi I, Subramanian SV, Kim D, editors. Social Capital and Health. New York: Springer; 2008. [Google Scholar]
  25. Midanik LT, Greenfield TK, Rogers JD. Reports of alcohol-related harm: telephone versus face-to-face interviews. J. Stud. Alcohol. 2001;62:74–78. doi: 10.15288/jsa.2001.62.74. [DOI] [PubMed] [Google Scholar]
  26. Moreira MT, Smith LA, Foxcroft D. Social norms interventions to reduce alcohol misuse in University or College students. Cochrane Database Syst. Rev. 2009 doi: 10.1002/14651858.CD006748.pub2. CD006748. [DOI] [PubMed] [Google Scholar]
  27. Murayama H, Fujiwara Y, Kawachi I. Social capital and health: a review of prospective multilevel studies. J. Epidemiol. 2012;22:179–187. doi: 10.2188/jea.JE20110128. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. NSDUH. National Survey on Drug Use and Health. [accessed on February 4, 2008]; http://www.oas.samhsa.gov/nhsda.htm.
  29. Oakes JM. The (mis)estimation of neighborhood effects: causal inference for a practicable social epidemiology. Soc. Sci. Med. 2004;58:1929–1952. doi: 10.1016/j.socscimed.2003.08.004. [DOI] [PubMed] [Google Scholar]
  30. Paschall MJ, Grube JW, Thomas S, Cannon C, Treffers R. Relationships between local enforcement, alcohol availability, drinking norms, and adolescent alcohol use in 50 california cities. J. Stud. Alcohol Drugs. 2012;73:657–665. doi: 10.15288/jsad.2012.73.657. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J. Clin. Epidemiol. 1996;49:1373–1379. doi: 10.1016/s0895-4356(96)00236-3. [DOI] [PubMed] [Google Scholar]
  32. Perkins HW, Linkenbach JW, Lewis MA, Neighbors C. Effectiveness of social norms media marketing in reducing drinking and driving: a statewide campaign. Addict. Behav. 2010;35:866–874. doi: 10.1016/j.addbeh.2010.05.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Petersen ML, Porter KE, Gruber S, Wang Y, van der Laan MJ. Diagnosing and responding to violations in the positivity assumption. Statistical methods in medical research. 2012;21:31–54. doi: 10.1177/0962280210386207. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Poole C. On the origin of risk relativism. Epidemiology. 2010;21:3–9. doi: 10.1097/EDE.0b013e3181c30eba. [DOI] [PubMed] [Google Scholar]
  35. Popova S, Giesbrecht N, Bekmuradov D, Patra J. Hours and days of sale and density of alcohol outlets: impacts on alcohol consumption and damage: a systematic review. Alcohol Alcohol. 2009;44:500–516. doi: 10.1093/alcalc/agp054. [DOI] [PubMed] [Google Scholar]
  36. Renalds A, Smith TH, Hale PJ. A systematic review of built environment and health. Fam. Community Health. 2010;33:68–78. doi: 10.1097/FCH.0b013e3181c4e2e5. [DOI] [PubMed] [Google Scholar]
  37. Robins JM. Robust Estimation In Sequentially Ignorable Missing Data And Causal Inference Models. Proceedings of the American Statistical Association on Bayesian Statistical Science. 1999;2000:6–10. [Google Scholar]
  38. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. Philadelphia: Lippincott-Raven Publishers; 2008. [Google Scholar]
  39. Sampson RJ. Neighborhood-level context and health: lessons from sociology. In: Kawachi I, Berkman LF, editors. Neighborhoods and Health. Oxford: Oxford University Press; 2003. pp. 132–146. [Google Scholar]
  40. Schnitzer ME, van der Laan MJ, Moodie EEM, Platt RW. Effect of breastfeeding on gastrointestinal infection in infants: a targeted maximum likelihood approach for clustered longitudinal data. Ann. Appl. Stat. 2014;8:703–725. doi: 10.1214/14-aoas727. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Scribner RA, Theall KP, Mason K, Simonsen N, Schneider SK, Towvim LG, DeJong W. Alcohol prevention on college campuses: the moderating effect of the alcohol environment on the effectiveness of social norms marketing campaigns. J. Stud. Alcohol Drugs. 2011;72:232–239. doi: 10.15288/jsad.2011.72.232. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Sinisi SE, Petersen ML, van der Laan MJ. Super Learning: An Application To Prediction Of HIV-1 Susceptibility. 2006 doi: 10.2202/1544-6115.1240. bepress, http://www.bepress.com/ucbbiostat/paper206. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Stead M, Gordon R, Angus K, McDermott L. A systematic review of social marketing effectiveness. Health Educ. 2007;107:126–191. [Google Scholar]
  44. van der Laan MJ, Polley EC, Hubbard AE. Super learner. Stat. Appl. Genet. Mol. 2007;6 doi: 10.2202/1544-6115.1309. [DOI] [PubMed] [Google Scholar]
  45. van der Laan MJ, Robins JM. Unified Methods For Censored Longitudinal Data And Causality. New York: Springer Verlag; 2003. [Google Scholar]
  46. van der Laan MJ, Rose S. Causal Inference for Observational and Experimental Data. New York: Springer; 2011. Targeted Learning. [Google Scholar]
  47. Vanderweele TJ. Ignorability and stability assumptions in neighborhood effects research. Stat. Med. 2008;27:1934–1943. doi: 10.1002/sim.3139. [DOI] [PubMed] [Google Scholar]
  48. Wechsler H, Nelson TF. What we have learned from the Harvard School of Public Health College Alcohol Study: Focusing attention on college student alcohol consumption and the environmental conditions that promote it. J. Stud. Alcohol Drugs. 2008;69:481–490. doi: 10.15288/jsad.2008.69.481. [DOI] [PubMed] [Google Scholar]
  49. Weitzman ER, Nelson TF, Lee H, Wechsler H. Reducing drinking and related harms in college evaluation - evaluation of the "A Matter of Degree" program. Am. J. Prev. Med. 2004;27:187–196. doi: 10.1016/j.amepre.2004.06.008. [DOI] [PubMed] [Google Scholar]
  50. Weitzman ER, Nelson TF, Wechsler H. Taking up binge drinking in college: the influences of person, social group, and environment. J. Adolesc. Health. 2003;32:26–35. doi: 10.1016/s1054-139x(02)00457-3. [DOI] [PubMed] [Google Scholar]
  51. Xu Y, Yu Q, Scribner R, Theall K, Scribner S, Simonsen N. Multilevel spatiotemporal change-point models for evaluating the effect of an alcohol outlet control policy on changes in neighborhood assaultive violence rates. Spat. Spatiotemporal Epidemiol. 2012;3:121–128. doi: 10.1016/j.sste.2012.04.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Yu Q, Scribner R, Carlin B, Theall K, Simonsen N, Ghosh-Dastidar B, Cohen D, Mason K. Multilevel spatio-temporal dual changepoint models for relating alcohol outlet destruction and changes in neighbourhood rates of assaultive violence. Geospat. Health. 2008;2:161–172. doi: 10.4081/gh.2008.240. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS675213-supplement.docx (71KB, docx)

RESOURCES