Do Neighborhood Characteristics of Sober Living Houses Impact Recovery Outcomes? A Multilevel Analysis of Observational Data from Los Angeles County

Abstract

Objective:

To identify neighborhood factors associated with recovery outcomes for sober living house (SLH) residents.

Methods:

Six-month longitudinal data for new SLH residents (n=557) was linked with census tract data, services available, alcohol outlets, and Walk Scores^® (0–100 score indicating access to neighborhood resources) for 48 SLHs in 44 neighborhoods in Los Angeles County.

Results:

Non-significant neighborhood characteristics in separate regressions for all outcomes were residents’ ratings of perceived risk, percentage of residences with access to a car, percentage of homes over $500,000, percentage of renter-occupied units, percentage with income less than $25,000, percentage that were non-white, the density of substance inpatient within 10 miles, and transit scores from Walk Score^®. Multilevel regressions found outpatient substance abuse treatment and density of AA groups were positively associated with more abstinent days. No neighborhood variables were associated with psychiatric symptoms. Higher perceived neighborhood cohesion, lower crime ratings, and better transportation ratings were associated with higher recovery capital.

Conclusion:

Greater neighborhood densities of substance abuse services and AA groups may help residents achieve more days abstinent. While residents may achieve better substance use outcomes even with negative perceptions of the neighborhood, positive perceptions of the neighborhoods may help them acquire more recovery capital.

1. Introduction

Substance use research shows direct and indirect impacts of alcohol and illicit drug problems on mental and physical health (Chan et al., 2008; Eddie et al., 2019; Mertens et al., 2003; Whiteford et al., 2013). The public health impact is seen in higher use of health care services among individuals with substance use disorders (SUDs), and the finding that those with a SUD die at an average age of 61 while those without a SUD die around age 83 (Neumark et al., 2000). Despite increased understanding of its consequences, SUDs persist in the United States. An estimated 40,263,000 people in the U.S. had a SUD in the past year (SAMHSA, 2020), and according to the National Center for Health Statistics, an estimated 107,622 drug overdose deaths occurred in 2021 (Ahmad et al., 2021). Even after seeking treatment, an estimated 85% people will relapse within the first year (Brandon et al., 2007). Re-admission to treatment is often part of substance use recovery. An estimated 60–67% of persons entering treatment have been in treatment before (Office of Applied Studies, 2000; Rezai-Zadeh et al., 2019). On average, people with SUDs attempt treatment three to four times over eight years before reaching a stable state of abstinence defined as 12 months of not using any substances (Dennis et al., 2005).

The high incidence of relapse, need for multiple admissions to treatment programs, and continued health risks indicate that SUDs may require more of a chronic condition approach, similar to ongoing care for asthma, cancer, diabetes, and depression (McLellan et al., 2000). However, treatment programs often have time-limited aftercare programs that do not address the ongoing needs of persons with SUD, including the need for an alcohol- and drug-free living environment. If SUD is treated as an acute problem through only offering short-term treatment, then people in recovery from SUDs might lose support they still need when they leave a treatment program. Recovery residences (RRs) offer a variety of abstinent-based housing options that the National Association of Recovery Residences (NARR, 2012) describes as having the purpose of providing “a safe and healthy living environment[s] to initiate and sustain recovery – defined as abstinence from alcohol and other non-prescribed drug use and improvement in one’s physical, mental, spiritual and social wellbeing.”

1.1. Research on Sober Living Houses

Peer-operated sober living homes (SLHs) are a type of RR that offers long-term housing and social support (Wittman & Polcin, 2014). Unlike residential treatment programs, they do not provide group counseling, case management, or treatment planning, but residents can stay as long as they wish, provided they abide by house rules and pay fees (Wittman & Polcin, 2014). Residents are either encouraged or required to attend self-help meetings, such as Alcoholics Anonymous (AA). A social model of recovery encourages peer support and empowerment of residents in decisions affecting the household. SLHs with greater social model recovery activities had a positive association with 6-month outcomes (Polcin et al., 2020). SLHs provide housing and support for recovery to a variety of persons, including those attending outpatient treatment and recently released from incarceration (Polcin, 2006; Polcin & Henderson, 2008). They can also serve as an alternative recovery option to formal professional treatment programs or as a transitional option when leaving programs.

Studies have shown that SLH residents make significant improvements on a wide variety of outcomes, including alcohol and drug use, severity of alcohol- and drug-related problems, psychiatric symptoms, employment, and arrests (Polcin, et al., 2010a; Mericle et al., 2019). Factors that predict favorable outcomes include fewer alcohol and drug users in resident social networks and higher levels of involvement in 12-step groups, both of which are central components of the social model approach to recovery used by SLHs.

Rather than living in an environment that may trigger substance use, SLHs offer an abstinent-based alternative that allows people with SUDs to increase existing assets to support their long-term recovery (Best et al., 2010). These assets can be viewed as recovery capital (RC), or the different elements that may support recovery, such as social, physical, human, and cultural capital (Burns & Marks, 2013). Research shows that RC increases were associated with greater employment, higher levels of social support, higher quality of life and greater recovery group involvement (Hard et al., 2022). Cano et al. (2017) found that greater time at RRs was associated with greater RC and positive wellbeing. A study by Polcin et al. (2021) found more social model recovery activities in SLHs were associated with increased RC.

1.2. Neighborhood Considerations

While most recovery research examines the individual’s characteristics and their impact on recovery, socioecological approaches to public health problems have also begun to examine how environmental factors may also impact outcomes, including substance use (Karriker-Jaffe, 2013; Molina et al., 2012; Stahler et al., 2009; Swan et al., 2021). Social ecological theories of development view individuals as influenced by not only the people in their social environment, but also the larger context where these interactions occur (Bronfenbrenner, 1979; Burton, 2015; CDC, 2022; Hawkins, 1996). People with lower SES likely live in disadvantaged neighborhoods where they experience the stressors of higher rates of crime, drug use, and violence while living in an area that may also have physical disorder, such as abandoned buildings, litter, and graffiti (Latkin & Curry, 2003; Winstanley et al., 2008).

Examinations of neighborhood effects on substance use have tended to focus on risk factors that exacerbate SUD and SUD-related issues (Boardmen et al., 2001; Karriker-Jaffe, 2011; Latkin et al., 2007; Hannon & Cuddy, 2006). Molin et al. (2012) found a negative association between neighborhood affluence and past year SUD. Using a “treatment ecology” theoretical framework, Jacobson (2004) looked at how location of substance abuse treatment programs could impact attrition and found salient factors included neighborhood disadvantage, community resources, and travel time. Mericle et al. (2018) examined the effects of social network characteristics and neighborhood economic status on alcohol relapse and found the number of heavy drinkers in one’s social network increased the risk of relapse. The risk was highest among persons living in disadvantaged neighborhoods. However, a few researchers have focused on buffering or restorative factors, though most of this research focuses on in non-SLH settings, such as treatment, general population, and other types of RRs (Bryden et al., 2013; Jacobson, 2004). Examples of buffering effects for SLHs could include factors such as the availability of 12-step meetings, mental health services, and substance use treatment programs.

Findings from Mericle et al. (2020a). underscore the importance of examining both restorative and destructive aspects of SLH neighborhoods. For example, co‐ed SLHs were located in less ethnically diverse neighborhoods and farther away from recovery resources. Larger house capacity was associated with increased density of off-premise alcohol outlets but also increased proximity to treatment. Higher fees were associated with lower neighborhood disadvantage and off‐premises alcohol outlet density but also greater distances from treatment programs and other recovery resources. While Mericle et al. did not assess how resident outcomes were related to neighborhood factors, they did describe how restorative and destructive aspects of neighborhoods were not equally distributed among SLHs and how SLHs are associated with neighborhood factors that both support recovery and place residents at risk. This study aims to address the gap in research on SLHs, neighborhood factors, and outcomes.

1.3. Purpose

The first aim of this study was to describe the neighborhood characteristics where SLHs are located. The study assessed characteristics that were publicly available measures of risk and protection (e.g., alcohol outlets, proximity of AA meetings, and mental health services). Consistent with recommendations from Mahoney et al. (2021), we also obtained data about neighborhoods based on measures of SLH residents’ perceptions.

The second aim was to assess how these neighborhood characteristics might affect substance use, psychiatric severity, and recovery capital over time. The purpose of this aim is to provide SLH residents, operators, and recovery professionals with information on what neighborhood factors impact recovery outcomes. The results of these analyses could help those stakeholders determine SLH neighborhood characteristics to consider and disregard in their decisions and recommendations. For example, if the income of the neighborhood residents is not associated with recovery outcomes in that neighborhood, then operators could open an SLH in a lower-income neighborhood that might also possess some of the more significant characteristics, such as density of substance abuse services or better public transit. Since SLHs do not offer on-site services as treatment programs do, availability of services in the neighborhood are an important factor to include. Based on prior research examining how neighborhoods impact longitudinal trajectories of alcohol dependence (Karriker-Jaffe et al., 2020), we hypothesized some neighborhood characteristics would operate as protective influences, including a high density of 12-step meetings, mental health services, and substance abuse treatment programs. Other neighborhood characteristics, such as the density of alcohol outlets, high crime, low income, and low social cohesion, would increase negative outcomes.

2. Methods

2.1. Study Sites and Participants

Participants were recruited between 2018–2021 in Los Angeles County SLHs that were members of the Sober Living Network (SLN). SLN is a non-profit organization that offers guidance and support to over 550 SLHs in five Southern California counties (About Us - Sober Living Network, n.d.). Members are inspected to ensure that they meet SLN standards, attend SLH management training, meet regularly to share resources, and are listed on an online directory. We excluded SLHs that housed children, had fewer than six beds, had more than 25 beds, or advertised fees that were over $4500 per month. Most of these exclusion criteria involved recruitment considerations, while also trying to keep the definition open to a general definition of SLHs.

The SLH selection for this study purposely maximized the diversity of neighborhood SES. Using data from the US Census Bureau’s American Community Surveys, we divided SLH neighborhoods into SES quartiles based on access to a car, education, income, and unemployment. Of the 44 neighborhoods where we enrolled 48 SLHs, 27.1% were from the lowest SES quartile, 20.8% from the second, 27.1% from the third, and 25.0% from the highest. To provide a broad depiction of all residents entering the houses and maximize generalizability, we used few inclusion/exclusion criteria. Based on an a priori power analysis, we aimed for 600 participants (recruitment ended early due to COVID-19). Participants were required to be 18 years of age or older, have a history of drug/alcohol issues, provide contact information for follow-up interviews, and live in the house for less than one month to ensure that they were new residents.

2.2. Procedures

Figure 1 illustrates the flow of participants from being identified as new residents in SLHs, to being screened, enrolled, and interviewed. Baseline assessments (N=557) were conducted on average 16 days (SD=9) after entering the house, with a range of 0 to 42 days. At least one month after entering the house, 82% of participants (n=457) answered questions regarding their perceptions of the SLH’s neighborhood. Even if participants left the SLH, interviewers contacted participants six months after baseline, and 83% (n=462) completed this follow-up interview.

Figure 1.

Observational Study Flow Chart for Sober Living House Neighborhood Characteristics

The bottom of Figure 1 also includes the characteristics of the study participants. The sample consisted of 557 new residents to SLHs that were mostly male and users of alcohol and methamphetamine. Age ranged from 18 to 77 with a mean of 40 (SD=12.5). Participants were paid $30 for the baseline, $15 for the one-month, and $40 for the six-month follow-up. All study procedures were approved by the Public Health Institute Institutional Review Board (IRB).

2.3. Outcome Measures

2.3.1. Primary Outcome

The main goal of SLHs is to support substance use recovery. Our primary outcome was therefore related to the use of alcohol and drugs. We used the Timeline Followback (Sobell et al., 1996) method to collect daily substance use data over the prior six months. We confirmed drug use data with urine drug screens. Due to the impact of COVID precautions on study procedures and some participants’ preference for phone interviews, we were unable to collect urine samples for all interviews. Since the concordance rate was 98.2% between 657 urine drug screens and TLFB data, we chose to use percent days abstinent (PDA) for our primary outcome analyses.

2.3.2. Secondary Outcomes

Because recovery is increasingly conceptualized as including improvement in multiple areas of functioning beyond abstinence (Witkiewitz et al., 2020), we also collected data on a variety of secondary variables.

1. To measure psychiatric symptoms, we used the Psychiatric Diagnostic Screening Test (PDSQ) which in prior studies showed a Cronbach’s alpha to be greater than 0.80 for 12 of the 13 psychiatric disorders assessed (Zimmerman et al., 1999; Zimmerman & Mattia, 2001). We calculated an overall total score based on 115 items across the 13 disorders with dichotomized ratings for each symptom (yes=1/no=0). Higher scores indicate more psychiatric symptoms.

2. To assess recovery capital, we used White’s (2009) Recovery Capital Scale (RCS) which has 35 items rated on a 5-point Likert scale. Prior studies have found this scale to have a Cronbach’s alpha of 0.90 (Polcin et al., 2020), indicating a high level of internal consistency. We used the RCS total score with a maximum score of 175. Higher scores indicate greater recovery capital.

2.4. Resident-level Predictors

2.4.1. Perceptions of Neighborhood

In order to allow time for new residents to gain impressions of the SLH neighborhoods, their ratings of the neighborhood were collected at least one month after moving into the SLH during the data collection phase from 2018–2021.

1. Neighborhood Environment Index (Crum et al., 1996) consists of 6 items assessing risk characteristics of the neighborhood, such as intoxicated people on the street, abandoned buildings, and trash/broken bottles. The yes(1)/no(0) responses to items are summed for an overall score. Higher scores in the range from 0 to 6 indicate a greater risk perceived in the neighborhood.

2. Neighborhood Cohesion (Sheidow et al., 2001; Tolan et al., 2001) is a 5-item scale designed to assess the level of positive interaction and support in the neighborhood. Items are rated on a 5-point scale, and scores can range from 5 to 25. Higher scores on this assessment indicate perceptions of more positive interactions in the neighborhood.

3. Perceived Neighborhood Crime Scale (Brook et al., 2011) is a 5-item instrument that asks participants to rate their perceptions about the level of crime in their neighborhood. Items address issues such as the level of violence, drug dealing, and gang activity. Items are rated on a 4-point scale, with a possible overall range from 5 to 20. Higher scores indicate more perceived crime.

4. Public Transportation Perception is the resident’s rating of access to public transportation in the SLH neighborhood on a scale of 1–5, with 1 being “very poor” and 5 being “excellent.”

2.5. Neighborhood-level Predictors

2.5.1. Neighborhood Demographics

We downloaded American Community Service variables from Social Explorer in 2018 and merged this with participant data. These neighborhood variables can be grouped into categories that we hypothesized may have an impact on outcomes.

1. Residential Access to Car is the percent of households without access to a car. This was included as an indicator of mobility, which is important for Los Angeles County, because much of the county is car-dependent (Manville et al., 2022).

2. Education was based on educational attainment, as derived from a single question asking, “What is the highest grade of school…has completed, or the highest degree…has received?” We examined the percentage of adults over age 25 without a high school diploma.

3. Housing Value is an estimate of how much a house would sell for on the current market. Houses are limited to owner-occupied units. We examined the percentage of expensive homes with a value of at least $500,000. For this dataset, the median value of a house in Los Angeles County in 2018 was $543,500.

4. Housing Tenure is the percentage of renter-occupied housing units. Units are classified as “renter-occupied,” if the owner or co-owner does not live in the unit. This includes units rented for cash rent and those occupied without payment of cash rent.

5. Low Household Income is the percent of households with income less than $25,000 in the past 12 months. This includes the income of the householder and all other individuals 15 years of age and older living in the household, regardless of relation.

6. Unemployment came from the percent of people unemployed or not in the labor force in the neighborhood. Civilians 16 years of age and older are classified as unemployed if they were without a job, and those out of the labor force are neither employed nor looking for a job.

7. Racial and ethnic composition is measured using the percent of the population who self-identified as White, Black, Latinx, Asian, or Pacific Islander. The racial composition of the neighborhood was included as a proxy for the impact of other structural factors we may not have included.

2.5.2. Substance Abuse and Mental Health Services Density

Services Density consisted of several measures of substance use and mental health services in the local area using the Substance Abuse and Mental Health Services Administration’s (2018) treatment locator website. We created neighborhood variables to represent the density of treatment facilities and self-help groups (Fortney et al., 1995; Fortney et al., 2000); density is calculated as the number of facilities within a noted distance from the SLH:

1. Substance Abuse Inpatient coded for those that have the listed type of care as substance abuse (SA), with service settings identified as inpatient.

2. Substance Abuse Outpatient coded for those that have the listed type of care as SA, with service settings identified as outpatient.

3. Mental Health Inpatient coded for those that have the listed type of care as Mental Health (MH), with service settings identified as inpatient.

4. Mental Health Outpatient coded for those that have the listed type of care as MH, with service settings identified as outpatient.

5. 12-Step Group Density are variables consisting of the number of 12-step meetings within 0.5 miles. We collected recent 12-step meeting schedules for groups, including Alcoholics Anonymous, Narcotics Anonymous, SMART Recovery^®, and Crystal Meth Anonymous to develop this variable.

2.5.3. Other Neighborhood Variables

1. Alcohol Outlet Density consisted of the number of alcohol outlets within a half-mile radius of the SLH residence. In March 2018, we downloaded raw data of alcohol outlets with active sales licenses from California Alcohol Beverage Control.

2. Walk Score^® measures the walkability of an address and provides a community-level indicator of geographic access to different amenities. Walk Score^® uses a gravity-based model that awards points based on the distance to the nearest destination of each type (e.g., retail sites, recreational features) using data sources such as Google and OpenStreetMap. Points range from a score of 0–100, with 100 being the most walkable. We also included this organization’s scores for transit, Transit Score^®, and biking, Bike Score^®, to see if daily errands can be completed by transit or bike. These measures may be particularly important for people who have lost their driver’s license. These scores were downloaded in 2018.

2.6. Analysis

To achieve the first aim, analyses described the characteristics of SLH neighborhoods. Preliminary analyses also examined correlations between neighborhood characteristic predictors to assess collinearity and to inform the multilevel models (MLMs) fit for Aim 2 (Supplemental Table 1). To achieve the second aim, longitudinal MLMs tested whether neighborhood characteristics were associated with outcomes in both separate and combined regressions. MLMs included outcome data from baseline and six-month timepoints and adjusted for demographic characteristics (including gender, race/ethnicity, and age) and random effects of neighborhoods and within-subjects.

MLMs specified two levels: Level 1 for observations across interviews within residents, and Level 2 for residents within neighborhoods. We included the random intercept of the participants in our models to allow for differences in baseline values and clustered standard errors to account for correlation or non-independence of outcomes within individuals. To examine the variance in each outcome that was due to the neighborhoods, we computed the intraclass correlations (ICCs) for each outcome. These are referred to as “empty” models and can range from 0 to 1. ICCs for social science research results tend to be between .05 and .20 and can be viewed as similar to eta-squared effect sizes (Peugh, 2010). Another way to interpret ICCs from empty models is as the expected correlation for individuals within the same neighborhood. Even though residents are also nested within houses, these analyses examined the effect of neighborhood characteristics on resident outcomes, so we focused on modeling clustering effects at the neighborhood level. Also, we did not add the house level in our modeling since the 48 SLHs were in 44 neighborhoods, so this additional level would have been redundant.

For MLMs that addressed Aim 2, examining neighborhood predictors and outcomes, we first tested each of the Level-2 neighborhood characteristics individually in separate regression models (Table 1). The purpose was to examine neighborhood characteristics with slightly different variations that may apply to different stakeholders and theories. Many of these characteristics measure similar or overlapping constructs, thus evidenced in the high correlation coefficients for the neighborhood characteristics (Supplemental Table 1). We reviewed these correlations and the characteristics that were significant for predictors of dropouts. Variables with correlations> 0.5 were reviewed and representative variables were selected to be presented in Table 2. We then ran combined MLMs that included these representative neighborhood predictors that were significant at p<0.10 (Table 2). As these are exploratory analyses examining multiple neighborhood characteristics and outcomes based on multiple theories, we did not have a policy of making adjustments for multiple comparisons to avoid Type I errors (Rothman, 1990). We completed this process in Stata (StataCorp, 2015) using full information maximum likelihood estimators with robust standard errors for each outcome.

Table 1.

Separate Longitudinal Multilevel Models Assessing Relationships Between Neighborhood Predictors and Six-Month Outcomes Among Residents of Sober Living Houses (N=557)^a

	Multilevel Models for Outcomes
Level and Variable	Mean (SD)	PDA	PDSQ	RCS
Level 1 – SLH Resident Fixed Effects
Age	39.8 (12.5)	-	-	-
Gender – Male (vs. others)	66.4%	-	-	-
Race – White (vs. others)	51.0%	-	-	-
Neighborhood Environment Index	1.1 (1.6)	−0.80 (0.66)	0.37 (0.62)	−0.57 (0.53)
Neighborhood Cohesion	10.0 (3.9)	−0.06 (0.27)	−0.08 (0.23)	0.44 (0.19) **
Perceived Neighborhood Crime	8.0 (3.6)	−0.41 (0.29)	0.46 (0.24) *	−0.43 (0.18) **
Public transportation perception	4.0 (1.0)	0.66 (0.83)	−1.3 (1.1)	2.48 (0.66) **
Level 2 – SLH Neighborhood Characteristic Fixed Effects
% Household w/o access to a car	10.7%	25.91 (18.19)	−3.1 (13.2)	−10.1 (12.8)
% No HS diploma	23.4%	−2.02 (7.30)	−1.0 (6.9)	−12.2 (7.1) *
% Homes GE $500K	50.0%	3.99 (4.44)	0.40 (4.7)	7.4 (4.8)
% Renter occupied	61.0%	−4.31 (5.59)	1.6 (5.1)	−3.7 (3.7)
% Household income LT $25K	23.9%	−3.97 (10.49)	0.37 (9.5)	−13.0 (9.1)
% People unemployed/not in labor force	39.4%	−2.24 (17.19)	−6.2 (16.7)	−26.8 (14.4) *
% Non-White	49.1%	4.61 (6.69)	0.66 (4.1)	−5.6 (4.5)
SA Inpatient facilities density 10mi	10.4 (3.1)	0.33 (0.44)	0.20 (0.38)	−0.07 (0.29)
SA Outpatient facilities density 10mi	25.0 (10.4)	0.26 (0.14) *	0.01 (0.09)	−0.06 (0.11)
MH Inpatient facilities density 10mi	4.4 (2.0)	1.76 (0.76) **	−0.50 (0.45)	0.16 (0.51)
MH Outpatient facilities density 10mi	17.2 (8.6)	0.40 (0.17) **	−0.02 (0.10)	−0.03 (0.14)
Self-help groups density 1mi	3.5 (3.0)	0.62 (0.29) **	0.38 (0.31)	0.46 (0.26) *
AA groups density 0.5mi	0.5 (0.9)	1.84 (0.87) **	1.3 (.56) **	1.29 (0.92)
Alcohol Outlet Density 0.5mi	6.4 (6.3)	−0.43 (0.29)	0.24 (.14) *	0.02 (0.16)
Walk Score®	67.7 (16.5)	−0.06 (0.08)	0.12 (.07) *	−0.02 (0.06)
Bike Score®	62.4 (13.5)	−0.06 (0.15)	0.13 (.08) *	−0.03 (0.07)
Transit Score®	49.4 (10.6)	0.05 (0.11)	0.10 (.11)	−0.04 (0.10)

^aNotes: Models examine each neighborhood predictor variable separately.

Level-1 N = 557; Level-2 sample size n= 44. Values reported for models are coefficients. Robust standard errors in parentheses. For multilevel models, time, age, sex, and race were also included in the models, adjusting for random effects of neighborhoods and within-subjects. Bolded coefficients and standard errors indicates *p < .10.

PDA=percent days abstinent for prior six months, PDSQ= Psychiatric Diagnostic Screening Test (0–115) with higher scores indicating greater psychiatric symptoms, RCS=Recovery Capital Score (35–175), SLH=Sober Living House, SES=socioeconomic status (composite categorical variable of % Household w/o access to a car, % No HS diploma, % Household income LT $25K, and % People unemployed/not in labor force), %=percentage, HS=High School, GE=greater than or equal to, K=1,000, LT=less than, SA=Substance Abuse, mi=miles, MH=Mental Health.

^*p < .10

^**p < .05.

Table 2.

Combined Longitudinal Multilevel Models Assessing Relationships Between Neighborhood Predictors and Six-Month Outcomes Among Residents of Sober Living Houses (N=557)^a

	Multilevel Models for Outcomes
Level and Variable	PDA	PDSQ	RCS
Level 1 – SLH Resident Fixed Effects
Neighborhood Cohesion			0.34 (0.17) **
Perceived Neighborhood Crime		.31 (.27)	−0.38 (0.22) *
Public transportation perception			2.19 (0.75) ***
Level 2 – SLH Neighborhood Characteristic Fixed Effects
% No HS diploma
% People unemployed/not in labor force			−25.8 (18.4)
SA Outpatient facilities density 10mi	0.25 (0.14) *
MH Inpatient facilities density 10mi
MH Outpatient facilities density 10mi
Self-help groups density 1mi			0.18 (0.33)
AA groups density 0.5mi	1.71 (0.80) **	0.53 (0.51)
Alcohol Outlet Density 0.5mi
Walk Score®		0.11 (0.07)
Bike Score®

^aNotes: Models include all neighborhood predictor variables from Table 1 that are significant at p<0.10. For highly correlated variables in Supplementary Table 1, representative variables were selected.

Level-1 N = 557; Level-2 sample size n= 44. Values reported for models are coefficients. Robust standard errors in parentheses. For multilevel models, time, age, sex, and race were also included in the models, adjusting for random effects of neighborhoods and within-subjects. Significant predictors from Table 2 models are included here. Variables with correlations> 0.5 were reviewed and representative variables were selected to be presented here. Bolded coefficients and standard errors indicates *p < .10.

PDA=percent days abstinent for prior six months, PDSQ= Psychiatric Diagnostic Screening Test (0–115) with higher scores indicating greater psychiatric symptoms, RCS=Recovery Capital Score (35–175), SLH=Sober Living House, SES=socioeconomic status (composite categorical variable of % Household w/o access to a car, % No HS diploma, % Household income LT $25K, and % People unemployed/not in labor force), %=percentage, HS=High School, GE=greater than or equal to, K=1,000, LT=less than, SA=Substance Abuse, mi=miles, MH=Mental Health.

^*p < .10,

^**p < .05

^***p<.01.

3. Results

We did not find differential attrition for any Level-1 characteristics or outcomes, but there was a significant difference for the Level-2 neighborhood characteristic of the percentage of people unemployed, with higher attrition in neighborhoods with more unemployment. We also found differential attrition for those from neighborhoods according to their densities of mental health outpatient facilities within 10 miles, alcohol outlets within 0.5 miles, self-help groups within 1 mile, and AA groups within 1 mile with more attrition from higher-risk neighborhoods.

3.1. Changes in outcomes over time

Differences between the interviews showed significant improvement in two of three outcomes, PDA and PDSQ. RCS did not improve over the six months assessed. The mean PDA at baseline for the prior six months was just over 70%, which improved to a mean of nearly 89% at six-month follow-up. The MLMs showed a significant fixed effect for interview (β=17.7, SE = 2.36, p<.001), indicating improvement in PDA from baseline to follow-up. Mean PDSQ scores decreased from 27.8 (SD=24.1) at baseline to 17.7 (SD=20.1) at follow-up. The MLM for interview and PDSQ showed a significant decrease in PDSQ over the interviews (β=−9.75, SE = 1.18, p<.001), indicating improvement on this summed measure.

To examine how changes in PDA varied by neighborhood, we calculated the resident-level differences in PDA between baseline and follow-up. Figure 2 shows the boxplot of these differences by neighborhood; the non-overlapping confidence intervals illustrate that changes in PDA did significantly differ across some neighborhoods. The overall mean difference in PDA was 26.5 (SD=29.2) and ranged from −2 to 100. The mean difference in PDA by neighborhood ranged from 0.0 (SD=0.0) to 63.3 (SD=21.8), F(43, 342)=1.955, p=.001, η²=.197. The differences in PDA across the 44 neighborhoods show the variability in PDA change between baseline and six months.

Figure 2.

Boxplot of Differences in Percent Days Abstinent from Baseline to 6 months by Neighborhood

Note: Percent days abstinent (PDA) is abstinence from any substance use for 180 days prior to the interview. Subtracted the PDA for the six-month interview from the baseline PDA to get the difference in PDA for each six-month interview completer (n=462). Horizontal boxplots are shown for each neighborhood (n=44) of the sober living house (SLH) where a participant lived. These values are graphed to show the different quartiles for each neighborhood to represent the distribution of differences for PDA across all study neighborhoods. The black solid line on either end depicts the lower and upper quartiles, while the thicker grey line in the middle depicts the second and third quartiles. The heavy line between the second and third quartile displays the median for that SLH neighborhood. Dots and stars represent outliers for that neighborhood. Variability in PDA differences led to further analyses of SLH neighborhood characteristics.

3.2. Neighborhood Characteristics

The first column of data in Table 1 summarizes descriptive statistics for the neighborhood characteristics. The Neighborhood Environment Index mean of 1.1 (SD=1.6) indicates that the residents on average rated the neighborhoods as low risk. With responses ranging from 5 to 23 out of a possible range from 5 to 25, the mean Neighborhood Cohesion score of 10.0 (SD=3.9) indicates a low rating of positive interactions in the neighborhood. The Perceived Neighborhood Crime scale’s range from 5 to 20. The mean was 8.0 (SD=3.6), indicating overall low perceived levels of crime in the SLH’s neighborhoods. Most resident perceptions tended to skew positively, while the other neighborhood characteristics were more in the middle range. Residents’ transportation rating was an average of 4.0 (SD=1.0) out of a possible high score of 5, while the objective rating Transit Score^® was an average of 49.4 (SD=10.6) out of a possible 100. Other objective neighborhood characteristics are summarized in the bottom half and the second column of Table 1, e.g., percentage of households without access to a car was 10.7% on average across neighborhoods.

3.3. Factors associated with substance use

The empty PDA model had an ICC of .071, meaning 7.1% of the variability in PDA was at the neighborhood level. Table 1 shows the results of the separate MLMs for neighborhood characteristics associated with PDA. MLMs indicated none of the Level-1 residents’ perceptions of their SLH neighborhood had a p<0.10 association with PDA. Level-2 objective neighborhood characteristics that did have a p<0.10 association in separate regressions were the densities of mental health inpatient treatment facilities within 10 miles (β=1.76, p=.020), mental health outpatient treatment facilities within 10 miles (β=.40, p=.020), all self-help groups within 1 mile (β=0.62, p=.030), and AA meetings within 0.5 miles of the neighborhood (β=1.84, p=.034). All of these significant associations were positive, indicating that higher densities of mental health facilities and self-help groups were related to improved PDA. However, Table 2 shows that in the combined model that adjusted for all p<.10 predictors from Table 1, only the neighborhood predictor of AA meetings within 0.5 miles was significant at the traditional p<.05 level. For density of substance abuse outpatient facilities within 10 miles, p=.067.

3.4. Factors associated with psychiatric symptoms

The ICC for the empty model was .034, meaning 3.4% of PDSQ’s variability was accounted for by neighborhood differences. Table 1 shows that none of the neighborhood substance abuse or mental health measures were associated with PDSQ in separate regression models, but the density of AA groups within 0.5 miles (β=1.3, p =.018) was associated with more psychiatric symptoms. Other potential predictors with p<0.10 were perceived neighborhood crime (β=0.46, p=.059), the density of AA groups within 1 mile (β=0.65, p=.059), the density of alcohol outlets within 0.5 miles (β=0.25, p=.079), Walk Score^® (β=.12, p=.080), and Bike Score^® (β=0.13, p=.088). Thus, the more of these categories of services were available, then the more psychiatric symptoms were reported. Table 2 shows that none of these variables remained significant in the final regression model that adjusted for all p<0.10 representative predictors from Table 1.

3.5. Factors associated with recovery capital

The ICC for the RCS empty model was .084, meaning 8.4% of RCS’s variability was accounted for by neighborhood differences. The separate MLMs for RCS in Table 1 show the p<0.10 predictors to be the neighborhood cohesion score (β=0.44, p=.018), perceived neighborhood crime (β=−.43, p=.020), and the transportation rating (β=2.48, p<.001). Thus, neighborhood ratings of cohesiveness, lower ratings of crime in the neighborhood, and better perceptions of transportation were related to recovery capital. Other potential predictors from separate regressions (Table 1) were percent without a high school diploma (β=−12.2, p=.087), percent unemployed (β=−26.8, p=.062), and self-help groups density within 1 mile (β=0.46, p=.076). These MLMs indicate that SLH residents report more recovery capital in neighborhoods with a population with fewer high school graduates, more people employed, and greater availability of self-help groups. Table 2 shows that neighborhood cohesion, perceived neighborhood crime, and the transportation rating all remained significant predictors of RCS at the traditional p<0.05 level in the combined model (Table 2), suggesting that the association between these neighborhood characteristics and RCS is particularly robust.

3.6. Sensitivity analyses

For post hoc sensitivity analyses, we replicated the models from Tables 1 and 2 using a Tobit specification, which accounts for inherent bounding in outcomes (e.g., PDA is bounded from 0–100%); point estimates and confidence intervals did not change substantially, reflecting robustness of results (Supplemental Tables 2 and 3).

4. Discussion

This study focused on neighborhood predictors of three SUD recovery outcomes, PDA, psychiatric severity, and recovery capital, and found distinctly different neighborhood characteristics related to each outcome. Since this is the first paper to examine how neighborhood characteristics of SLHs impact recovery outcomes, we examined many characteristics based on these factors’ impacts in similar settings.

For recovery capital, resident perceptions of neighborhoods were particularly robust predictors. To the extent that residents perceived the neighborhood to be low on crime, high on cohesion, and high on accessible public transportation, they indicated higher levels of RCS. This finding is intuitively appealing and consistent with hypotheses. If the neighborhood is perceived to be a high crime area, some residents might feel hesitant to go out and seek the services they need or socialize in the community. That might be particularly the case for persons who feel vulnerable and may be in high need of support. If public transportation is not easily accessible, residents may not travel to services or community social events. Residents might also be unwilling to use public transportation if they need to wait for long periods in unsafe areas, particularly during evening hours. On the other hand, perceived cohesion was associated with higher RCS. The perceived cohesion measure assesses the extent to which neighbors know each other, report positive interactions, and feel supported by neighbors. These characteristics likely impact recovery capital by creating an environment where residents feel more motivated to engage in activities that enhance recovery.

Many objective neighborhood variables were not associated with RCS, including household income, housing prices, home ownership, and proximity of most services. Reasons for the limited association between objective measures and RCS require more research. However, one factor might be the RCS measure that we used examined a broad array of intra- and inter-personal assets that may not have been tapped by the neighborhood characteristics examined. A previous study found a variety of individual and house factors predicted RCS (Polcin et al., 2021), and these may have been more important influences than the neighborhood influences assessed here, as reflected in the low ICC. Another perspective on RCS is that it measures the resident’s perception of their current recovery assets which can fluctuate early in recovery. Recovery capital may have increased during the first few weeks; this would already be reflected in the baseline data, thus resulting in the lack of significant change in RCS and the lack of associations of RCS with objective neighborhood characteristics.

While perceptions of the neighborhood had significant relationships with RCS, they were mostly unrelated to PDA and psychiatric severity. Significant predictors for PDA included the proximity of AA groups, outpatient substance abuse treatment, and mental health services. These findings were consistent with previous studies of SLHs showing involvement in 12-step groups (Polcin, et al. 2010a, 2010b) and level of psychiatric severity (Polcin & Korcha, 2017) were associated with improved substance use outcomes. The finding that the density of outpatient substance abuse treatment was associated with PDA was supported by a previous study that reported favorable outcomes for residents required to attend outpatient substance abuse treatment while they resided at the SLH (Polcin, et al., 2010b).

The findings for predictors of psychiatric severity were more complicated. We expected to find the proximity of mental health services would be associated with better psychiatric severity outcomes. However, neither density of inpatient nor outpatient mental health services was associated with psychiatric severity. Significant predictors of psychiatric severity were AA meeting density, alcohol outlet density, and walkability measures (each associated with increased severity), though the magnitude of the coefficients were not large. When we conducted post hoc analyses to determine the impact of COVID, these effects disappear. Future research should examine whether AA meetings are more likely to be held in certain types of neighborhoods that present increased risks to people early in recovery and how COVID impacted how people interact with their neighborhoods.

The finding that mental health services were associated with PDA, but not psychiatric severity, combined with the finding that the strongest predictors of psychiatric severity were factors ostensibly related to substance use (i.e., proximity of AA and alcohol outlets) supports a complex view of recovery. The effects of different services on distinct problems might be complex and possibly the effects vary at different time points. For example, some persons attending AA may at one point benefit primarily from reduction or abstinence of substance use. At another point, they might use peer support at AA meetings or 12-step principles discussed there to cope with psychiatric distress. Most SLHs mandate self-help meetings. Given the finding from the current study, as well as findings from previous studies (e.g., Polcin, et al., 2010a) showing strong associations between 12-step involvement and outcomes, recovery home organizations might consider implementing standards mandating self-help group involvement or offering resources that increase access, such as arranging on-site meetings, providing transportation to meetings or computers for attending on-line.

4.1. Implications for Providers

SLH providers may want to locate their residences in “more affordable” neighborhoods; others may be otherwise relegated to “less desirable” neighborhoods. Of greater importance to outcomes than a neighborhood’s socioeconomic context, especially in terms of substance use (PDA), are the availability of mental health services, AA groups, and outpatient substance abuse services. For neighborhoods that might not have a high density of these services, operators could coordinate transportation to these services. While we did find that residents who report positive perceptions of neighborhoods have higher scores of RCS, those who have negative perceptions of their neighborhoods can nevertheless improve substance use and mental health issues. The association between RCS and PDA in SLHs requires more research, but a variety of studies have shown RCS to be an important indicator of overall quality of life (e.g., O’Sullivan et al., 2019).

Standards from NARR on the neighborhood level have focused on the SLH being a “good neighbor,” such as policies around interacting with neighbors, smoking, and parking (National Association of Recovery Residences, 2018). Social model theory has theorized that individuals are impacted by their environment, thus having implications for SLHs and their neighborhoods (Polcin et al., 2014). Both this theory and study findings recommend a focus for SLHs on positive, supportive neighborhood characteristics. As SLHs aim to build a supportive recovery environment, these organizations could recommend that SLHs could look for resources in their community and supplement what the neighborhood lacks (Mericle et al., 2022a).

4.2. Limitations

The study was limited to Los Angeles County. Neighborhood characteristics might vary in different locations and in ways that affect outcomes. The study of the neighborhood characteristics in a less densely populated area or a larger variability of neighborhood density characteristics might yield different results. Many of the neighborhood characteristics were highly correlated. The current study focused on neighborhood characteristics separately. Future analyses will examine how the neighborhood factors work in combination with each other and include individual- and house-level predictors and their influences on other outcomes, such as employment, HIV risk, and legal problems. The study was conducted in part during the COVID-19 pandemic, so many neighborhood-level factors and residents’ neighborhood perceptions may have been impacted by COVID precautions, such as spending more time at home, not having in-person AA meetings, or being less likely to take public transportation. The study population was largely cis-gender, male, and White, thus impacting the generalizability of the findings.

4.3. Conclusions

SLHs offer an important option for people in recovery who want to live with others in recovery from drugs and alcohol. Though individual factors carry a large influence on outcomes, characteristics of SLH neighborhoods deserve continued attention because they are associated with substance use, psychiatric symptoms, and recovery capital. Importantly, though a resident might have negative perceptions about neighborhood crime and cohesion, operators can still offer an environment that offers an increase in abstinence and a decrease in psychiatric severity. The findings from this study regarding mental health and substance abuse treatment and self-help groups being positively associated with increased percent days abstinent may help guide those involved with sober living houses to seek out these neighborhood features.

Highlights.

• Sober living residents report more days abstinent and fewer psychiatric symptoms after six months.

• Even if sober living residents have negative perceptions of the neighborhood, they can still have positive outcomes.

• Some neighborhood characteristics may help residents achieve better abstinence and greater recovery capital.

• Positive perceptions of the sober living house’s neighborhood may help residents acquire more recovery capital.