Abstract
Background:
There is a need to better understand the extent to which social capital (reflected in social networks tapping friendship, financial support, advice/informational support) can aid recovery for those residents living in abstinence-based recovery homes.
Methods:
Social network characteristics of 42 recovery homes (Oxford Houses) were examined, including friendship, willingness to loan money, and advice-seeking to assess the extent to which house network patterns were related to house-level resident measures of proximal recovery outcomes of well-being (e.g. social support, self-esteem, stress) and financial health (e.g. earned wages).
Results:
We found that the density of the willingness to loan money network within a house was positively associated with house-level earned wages, social support, and self-esteem, and negatively associated with stress. Conversely, the density of house advice-seeking relationships was positively related to house-level stress.
Conclusions:
Houses in which residents are willing to share resources with other members who may be in need showed higher rates of well-being at the house-level. Advice-seeking in itself may signal stress, as stress may motivate residents to seek advice from more peers. The implications of these findings are discussed.
Keywords: Recovery homes, Addiction, Oxford house, Social networks
For some populations who share the same social environment, the relationships among all dyads of individuals are often referred to as a social network (Hahm et al., 2012). Social networks can have multiple relational dimensions, such as friendship, trust, and advice-seeking. Measuring and analyzing these important social connections using social network analytic tools has enhanced the understanding of network effects on individual members, including structural constraints on information flow and behavioral influences among connected members within a network (Dishion, 2013; Weerman et al., 2011). These methodologies move beyond an exclusive focus on individual-level variables in an effort to study transactions between persons and their environments and elucidate the relationship between individual factors and those of their social environments (Jason & Glenwick, 2016; Parkin, 2015).
Network analytic techniques utilize numerous measures to characterize specific patterns in the connections (and non-connections) in a network. These attributes can describe characteristics of individuals in a network (e.g., isolation, centrality), relational characteristics between dyads (e.g., reciprocity, transitivity), or characteristics of the entire network (e.g., density, average path length, diameter). For instance, density is a network-level measure of the proportion of ties that are present out of all possible ties, and reciprocity is either a network-level or individual-level measure of the proportion of extant ties that are mutual.
Social network measures can identify pathways through which social capital may flow. Social capital is defined as the actual or potential resources individuals can obtain through their relationships with others (Bourdeau, 1985), and is an important function of some types of social networks. In Bourdieu’s terms, social capital is, “…the aggregate of the actual or potential resources which are linked to possession of a durable network of more or less institutionalized relationships of mutual acquaintance or recognition” (1985, p. 248). Many studies have examined the benefits of social capital on individual outcomes (Kim & Kawachi, 2006) and this literature tends to demonstrate that social capital is an important contributor to overall well-being. For example, Neale and Stevenson (2195) found that homeless drug and alcohol users, living in hostels, can access opportunities for building social capital and fostering recovery capital. Panebianco, Gallupe, Carrington, and Colozzi (2016) studied the personal support networks and social capital of those who relapsed and were drug-free post-treatment. Those who were drug-free versus those who relapsed had larger, less dense, more heterogeneous, and reciprocal support networks.
The communal nature of group recovery homes suggests that social network methodology could quantify and measure whole network relationships to better understand how relationships support or challenge recovery, and also measure to what extent individuals are making use of social capital in the homes. Some environments, such as group substance use recovery homes, are especially rich in social capital, which may come in different forms. Recovery homes represent the largest recovery-specific, community-based post-treatment option for those with substance use disorders (Polcin, Korcha, Bond, Galloway, & Lapp, 2010). Previous studies suggest that individuals attain significantly better recovery outcomes if they remain in a recovery home residence for at least six months (Jason et al., 2012; Jason, Stevens, Ferrari, Thompson, & Legler, 2012), which may be due to the social capital availed by the home environment. However, only about 50% of residents stay for that long (Jason, Olson, et al., 2006; Jason, Davis, Ferrari, & Anderson, 2007). While it is not clear what causes early dropout, one reason may be social integration difficulties. Conversely, successful social integration into the house may prevent early dropout and thus facilitate a sustained recovery.
Individuals overcoming a substance use disorder can benefit from social support (Kelly, Hoeppner, Stout, & Pagano, 2012). Recovery homes, which offer a new source of friendships, support, and advice from others living sober lifestyles, should especially benefit new residents. In addition to friendship, other beneficial social linkages emerge among those in recovery homes, including a willingness to provide resources, such as short-term loans, or to provide advice. Multiple types of social linkages among residents—friendships, willingness to lend money, and advice-seeking—may serve to influence current life satisfaction and a concomitant willingness to remain in the home. Additionally, these types of supportive social relationships can benefit people in recovery and improve substance use recovery rates (Havassey, Hall, & Wasserman, 1991).
The largest network of recovery homes in the US called Oxford Houses (OH) are similar in terms of their house rules, but case studies suggest that they vary in the availability of recovery-related social capital (Jason et al., 2020). Recovery homes represent a significant portion of the resident’s day-to-day social experiences—certainly as it relates to recovery—and whatever sobriety-reinforcing value-added this environment provides will be in some way dependent on social processes operating between and among the residents. Additionally, this means that these homes represent a fair amount of an individual’s overall social capital. We can think of the set of all types of social relationships among residents as a social network representation of a recovery home’s social environment, one which co-evolves over time with individual recovery-related attitudes and behaviors and changing house composition. Social relationships, measured through friendship, willingness to lend resources, and advice-seeking might play large roles in terms of whether residents perceive these settings as supportive and beneficial to their sense of well-being.
The current article examines the extent to which these interpersonal relationships relate to individual-level well-being. This study closely examines four markers of an individual’s well-being while in recovery homes: how much money they make at their place of employment, how much stress they are experiencing, how stable they perceive their social support, and how positive is their self-esteem. Wage-earning indicates independence and potential economic mobility. Individuals who stay in the house longer (thus are more likely to stay clean), earn more money than those who are not provided opportunities to live in these recovery homes (Jason, Olson, & Harvey, 2015) or leave earlier (Gomez, Jason, Contreras, DiGangi, & Ferrari, 2014; Lo Sasso, Byro, Jason, Ferrari, & Olson, 2012). Stress has shown to predict negative outcomes for residents, and negatively predicts variables that aid sobriety, like the quality of life (Laudet & White, 2008). Within Oxford House (OH) recovery home studies, stress has negatively predicted self-mastery (Majer, Jason, Ferrari, Olson, & North, 2003), social support (Gottlieb, 1985), and the number of important people in one’s network (Stevens, Jason, Ram, & Light, 2015). Abstinence motivated social support, like that used in mutual-help groups (Humphreys, Makowski, Moos, & Finney, 1999), is a predictor of recovery (Longabauch et al., 1995; Zywiak, Longabaugh, & Wirtz 2002) and plays a critical role in the OH recovery home environment. Social support has been shown to predict a longer duration of abstinence within OH environments (Jason, Davis, & Ferrari, 2007). Finally, self-esteem contributes to a sense of hope during recovery (Ferrari, Stevens, Legler, & Jason, 2012), and low self-esteem predicts future substance use (Richter, Brown, & Mott, 1991) and other high-risk behaviors (Nyamathi, 1991).
There is a need to better understand the extent to which social capital (i.e. friendship, financial support, advice/informational support) aids recovery in terms of wages, stress, social support, and sense of self-esteem. The current exploratory study examined the interrelations among social network measures (i.e., nodes, mean degree, diameter, average path length, centrality, reciprocity, density, transitivity, and isolates) to identify to what extent network measures would be informative about these social-ecological variables in recovery homes. We hypothesize that the structure of social support networks as measured by the described network statistics would be related to individual indicators of recovery (i.e., wages, stress, support, self-esteem) (See Fig. 1).
Fig. 1.
Social Network and Behavioral Outcome Measures at the Individual and House Levels.
Method
Settings and participants
OHs are the largest network of recovery homes for individuals with substance use, with over 2000 homes in the US. OHs are rented, gender-specific single-family homes for 6 to 12 individuals. Each OH operates democratically with majority rule (i.e. > 80% approval rate) regarding membership and most other policies (Oxford House Manual, 2011). Residents are required to abide by three rules, namely, pay rent, abstain from using alcohol and other drugs, and contribute to the maintenance of the home. Violation of the above rules results in eviction from the house (Oxford House Manual, 2011).
Procedure
Data were collected from OHs located in North Carolina, Texas, and Oregon. OH statewide organizations are strong and well-developed in these areas, which facilitated communication with residences about possible participation. We believe that residences from different geographical regions provided some ability to address the generalizability of findings. Recruitment attempts were made to residences after state organizations helped field staff assemble lists of residences to approach. Member-elected house presidents were asked to introduce the study to the residents by reading a description of it from a project-provided script; houses were accepted into the study if the house president and all other members (or all but, at most, one member) agreed to participate. The first thirteen consenting houses from each state were accepted into the study, and three more houses were added for a total of 42.
Participants were part of a longitudinal study that collected information every four months over a 2-year period (the current study involves data from the first point of contact). Participants were recruited and interviewed by field research staff in face-to-face meetings. Participants completed demographic, stress, self-esteem, support, and social network measures. They were compensated $20 for completing these assessments. Permissions for this study were granted by the DePaul University Institutional Review Board. More descriptive data both for the house network variables and for the participants has been published elsewhere (Jason et al., 2020; Kassanits et al., 2020).
Measures
Demographic information collected included age, race/ethnicity, and gender. Participants were also asked to report their wage-earnings from their place of employment in the last 30 days. The log of the wages for the last 30 days was computed and used as a continuous variable.
Social network measures
The Social Network Instrument (SNI; Jason, Light, Stevens, & Beers, 2014) was utilized to capture the social dynamics within each OH. This instrument has been used in several investigations of the social networks of recovery home residents (Jason & Stevens, 2017; Jason et al., 2014; Jason et al., 2018; Light et al., 2016) The SNI measures relationship characteristics, but critical to this study are the friendship, loan, and advice-seeking relationships. 1 In this study, we used data from the SNI to calculate house-level indices characterizing friendship, loan willingness, and advice-seeking relationships.
Each social network item was measured on a 5-point Likert scale (0–4). Friendship, intended to assess non-judgmental social support, was measured by responses to “How friendly are you with this person?” Responses ranged from “close friend” to “adversary”. 2 A friendship relationship was considered present if the respondent identified another resident as a close friend or friend, but not present otherwise (acquaintance, stranger, adversary). Money loaning (which represents a willingness to lend resources) was considered present if the respondent reported being willing to loan another resident larger amounts of money ($500 or $100), but was not considered present if the respondent reported being willing to loan another resident smaller amounts of money (i.e., $0, $10, $50). Advice-seeking was considered present if the respondent reported seeking advice from another resident very often or quite often, but not present otherwise (i.e., regularly, rarely, never).3 These types of social network instruments are found to be reliable measures (Hlebec & Ferligoj, 2002). The SNI used with our sample had a Cronbach’s alpha of 0.85 and all items contributed positively. A multilevel confirmatory factor analysis of the SNI found an excellent fit and per-item contribution, and neither age nor sex significantly correlated with this instrument (Jason & Stevens, 2017).
Each resident is referred to as a node (below denoted A, B, C, etc.) and each relationship between the nodes of a dyad is referred to as an edge. Each directed edge (i.e. A->B is distinct from B->A) counts as a “degree”. Each edge acts as a degree that can either be an out-degree (a rating that the ego node makes about each alter) or an in-degree (a rating that each alter makes about the ego node). The degree of a node is the total of its indegrees plus the number of out-degrees, and the mean degree is the average for the house.
Diameter indicates the most directed edges that have to be traversed to link any pair of individuals in the network, excluding nodes with no connections of any kind (isolates). For instance, the diameter would be a measure of the maximum number of social steps a piece of information would need to travel to get from one person in the network to any other person. The average path length is the average number of steps among all possible pairs of network nodes (again excluding isolates), and it is a measure of the efficiency of information transfer within a network. A rating of one is the most efficient possible value (all connections are direct, i.e. one-step), while higher numbers indicate a network with fewer direct relationships, less density, and a larger number of steps needed for information to traverse the network. We calculated isolates as the percentage of residents that are not connected to any other house residents for the relationship in question. This network index acts as a measure of network fragmentation.
Reciprocity is a network statistic that captures whether individuals in a network have a tendency for mutual connection. Two edges are considered reciprocal if a directed edge goes from node A to node B, and another from B to A. Higher reciprocity values indicate more mutual relationships. Density measures the overall interconnectedness of the network and is the sum of the directed edges divided by the number of possible directed edges. Since it is a proportion, it is naturally bounded between zero and one.
Centrality is a measure of the relative importance of nodes in connecting a network. Because this analysis focuses on the availability of social capital, at least some of which can be transferred indirectly from one individual to another via intermediaries, we used a measure of “betweenness” centrality (Freeman, 1979). Betweenness is a concept related to “brokerage”; individuals higher on betweenness centrality are those whose removal from the network would result in the most network fragmentation. In other words, such individuals have a strong linking function in the network. For the network as a whole, betweenness centrality takes the value of 1.0 when one central person is connected to everyone else, none of whom are connected to each other, and zero when everyone serves as an intermediary for the same number of dyads. That is, it essentially measures how unevenly brokerage is distributed in the network.
Transitivity quantifies the tendency for relationship “chains” which is the prevalence of triads and thus signals the presence of network clustering or cliques. An example of a transitive triad is when A directs an edge to B, and B directs an edge to C, and A also directs an edge to C. Like reciprocity, it is a measure of the tendency for the relationships in a network to potentiate additional relationships; these are sometimes referred to as “network closure” tendencies.
Outcome measures 4
The Interpersonal Support Evaluation List (ISEL; Cohen & Wills, 1985; Cohen et al., 1986) measures three types of actual or perceived social support (tangible, appraisal, belonging). Tangible support measures instrumental aid (monetary assistance); appraisal support refers to having someone with whom to talk to about one’s problems; and belonging support refers to having someone with whom to do activities with. This instrument has been used successfully with comparable recovery home samples (Majer et al., 2020). The ISEL consists of 12-items measured on a 4-point Likert scale ranging from definitely false to definitely true. The internal reliability of the ISEL scale in the present study was 0.88.
The Rosenberg’s Self-Esteem Scale (SES; Rosenberg, 1965) measures participants’ positive and negative feelings of the self. SES is a widely used 10-item, global self-esteem scale measured on a 4-point Likert Scale ranging from “strongly agree” to “strongly disagree.” Items include “I think I have a number of good qualities,” “I take a positive attitude towards myself,” and “I feel I do not have much to be proud of.” The internal reliability of the SES scale in the present study was 0.92.
The Perceived Stress Scale (PSS; Cohen, Kamarck, & Mermelstein, 1983) measures the degree to which participants appraise situations in their lives as stressful. PSS consists of 4-items measured on a 5-point Likert scale ranging from “never” to “very often.” Items include “how often have you felt that you were unable to control the important things in your life?” and “how often have you felt difficulties were piling up so high that you could not overcome them?” The internal reliability of the SES scale in the present study was 0.73.
Analytic approach
Social network analysis of 42 OHs was conducted using R (see https://www.r-project.org), which is an open-source software environment for statistical computing and graphics. We used the statistical package ‘sna’ (https://cran.rproject.org/web/packages/sna/sna.pdf ), a range of processing tools for network data, to calculate the network metrics.
Intercorrelations among the social network variables for the entire 42 house sample were computed to identify and select a smaller and more targeted group of network variables that were are general indicators of distance, linkedness, and node importance within a network. The intercorrelation matric would allow us to identify these key network variables and eliminate ones that unimportant or are just mathematically related. Correlations were then conducted between several of the promising network measures (average path length, isolates, density, reciprocity, and transitivity) and behavioral variables (i.e., wages, social support, self-esteem, and stress).
Our design is hierarchical in nature with residents nested within recovery homes, a 2-level nested design. The variance of individual measures can be separated into two components with such a design: variance between individuals within a house, and the variance of average differences between houses. This leads to data that is hierarchical with a) behavioral variables tapping resident characteristics at level 1 that have correlated observations due to shared influences in the homes but also b) network variables tapping house-level characteristics at level 2.5
In addition to the house-level correlations with individual outcomes, which are the main focus of this work, the 2-level model also provides intra-class correlations (ICCs) for the outcome variables, which are important in interpreting the extent to which such outcomes tend to covary within houses, and how much of this covariance is associated with house network characteristics.6
Results
Social network information was collected from 229 participants from 42 OHs. There were 6.14 participants per house, of which we recruited 88.8% for an average of 5.45 (range = 3 to 9) residents per house. Participants were 55% male with an average age of 38.4 years (SD = 10.8). Most participants identified as European-American (82.1%), and other ethnicities included African American (9.2%), Latinx (6.5%), Alaskan Native (6.5%), American Indian (1.3%), and Pacific Islander (0.4%). The average length of stay in an OH was 10.3 months (SD = 12.55, range from 7 days to 6.8 years).
Intercorrelations among network measures
The intercorrelations of the network measures were comparable across the three friendship, loan, and advice-seeking relationship types so only the intercorrelations for the network characteristics of the advice-seeking network is discussed (see Table 1). Transitivity, which shows the connectivity of multiple individuals, was significantly positively correlated with degree and density, but significantly negatively correlated with diameter and average path length. Average path-length was positively correlated with diameter and centrality but was not significantly related to being linked as measured by density and reciprocity.
Table 1.
Intercorrelation among house level advice-seeking network measures
| Nodes | Degree | Diameter | Reciprocity | APL | Isolates | Density | Transitivity | Centrality | |
|---|---|---|---|---|---|---|---|---|---|
| Nodes | ———– | ||||||||
| Degree a | 0.07 | ———– | |||||||
| Diameter | 0.61*** | 0.40** | ———– | ||||||
| Reciprocity | −0.28 | 0.61*** | 0.17 | ———– | |||||
| APL b | 0.59*** | 0.25 | 0.94*** | 0.04 | ———– | ||||
| Isolates | 0.05 | −0.65*** | −0.49*** | −0.40** | −0.29 | ——— | |||
| Density | −0.48*** | 0.76*** | −0.05 | 0.75*** | −0.22 | −0.59*** | ——— | ||
| Transitivity | −0.23 | 0.32* | −0.35* | 0.21 | −0.59*** | −0.10 | 0.45*** | ——– | |
| Centrality | 0.54*** | 0.45*** | 0.76*** | 0.11 | 0.86*** | −0.29 | −0.01 | −0.14 | —— |
Note:
p < 0.05,
p < 0.01,
p < 0.001.
Degree is the mean degree of the network.
APL is average path length.
Isolates were negatively correlated with density and reciprocity, however, it is mathematically necessary for isolates to be related to density, as density is a measure of the number of edges, and isolates are nodes lacking any edges. Network density was highly related to mean degree and reciprocity, however, density and mean degree are related to one another for purely mathematical reasons, and they are simply different ways to report the number of edges relative to the number of nodes.
Based on findings from Table 1, average path length, isolates, density, reciprocity, and transitivity appear to have unique contributions to theoretically important domains of distance, linkedness, and node importance. We, therefore, selected these variables for our subsequent analyses.
Small-Group social network variables and related outcomes
The analyses in Table 2 examine how several social network structural variables identified in the prior analysis interact with house-level average outcomes (wage, stress, social support, and self-esteem). Wage was significantly correlated with the most variables, including density for the three social network types (friendship, advice-seeking, and loan), as well as advice-seeking reciprocity, friendship APL and transitivity, and loan and advice-seeking isolates. The finding that wages were more strongly associated with reciprocity might suggest that relationships in which individuals feel comfortable exchanging advice are the most useful for finding good jobs. This is understandable as, for example, instead of one person being helped, there is the potential that two (both members of the dyad) can benefit from the interactions. Also, mutual (and hence, generally, equal-status) relationships are often the most efficient for exchange of mutually beneficial information, because both individuals have the same goals and motivations, and will understand each other’s needs better. Also, wage was negatively correlated with markers of fragmented networks such as advice-seeking isolates and loan isolates.
Table 2.
Correlations of house-level average outcome variables and house-level network measures (and house-level intra-class correlations for outcomes).
| Wages | Stress | Social Support | Self-Esteem | |
|---|---|---|---|---|
| Friendship a | ||||
| APL | −0.60*** | −0.03 | −0.21 | −0.15 |
| Density | 0.59*** | 0.07 | 0.40 | 0.25 |
| Reciprocity | 0.40 | 0.07 | 0.41 | 0.13 |
| Transitivity | 0.50** | 0.10 | 0.25 | 0.15 |
| Loan | ||||
| APL b | 0.17 | −0.52*** | 0.24 | 0.08 |
| Isolates | −0.41* | 0.35 | −0.55** | −0.47* |
| Density | 0.54*** | −0.48* | 0.75*** | 0.60* |
| Reciprocity | 0.08 | −0.49** | 0.24 | 0.13 |
| Transitivity | 0.15 | 0.13 | 0.32 | 0.41* |
| Advice-seeking | ||||
| APL | 0.18 | −0.19 | 0.09 | 0.20 |
| Isolates | −0.41* | −0.07 | −0.05 | −0.20 |
| Density | 0.48*** | 0.54*** | 0.01 | −0.07 |
| Reciprocity | 0.64*** | 0.24 | 0.26 | 0.22 |
| Transitivity | 0.07 | 0.46** | −0.08 | −0.26 |
| Intra-class Corr c | 0.24*** | 0.25*** | 0.16** | 0.15** |
Note:
p < 0.05,
p < 0.01,
p < 0.001.
As there were no Isolates within the friendship network, this variable was not included.
APL is average path length.
Intra-class correlation is the ratio of house level variance to total variance and pertains to outcome variables. ICC’s were comparable across network measures so only those for density of loan are shown.
The stress measure was negatively correlated with loan density, reciprocity, and APL, a highly suggestive indicator that better access to a network of others willing to lend money is a major stress reducer. Interestingly, stress was positively correlated with advice-seeking transitivity and density suggesting that high-stress networks produce the need for people to seek advice and support.
Our analysis found that access to the loaning network was also related to feelings of social support. The proportion of loan isolates was negatively correlated with social support indicating that when the exchange of money as a resource was low, so was perceived social support. The inverse of this is also present in the data: when loan density was high, so was perceived social support. Furthermore, the self-esteem variable was negatively correlated with loan isolates and positively correlated with loan density and transitivity. These important individual measures of social embeddedness and self-appraisal were closely associated with having access to short-term loans from other house residents.
In Table 2, all 4 survey measures had significant ICCs, but the absolute magnitudes of these ICCs were modest, ranging from 0.15 to 0.25. This finding suggests that overall, house-level structures have strong effects on the averages of house-level outcomes variables (i.e., Table 2), but that most of the individual-level variability is still unexplained. Table 3 shows correlations of the house-level network measures with individual-level outcomes. They are uniformly smaller than what is in Table 2. Part of the reason for this is simply that the modest ICC values show a persistent variability regardless of general conditions within the home; in other words, although on average it is “better” to be a member of a house with a supportive social structure, many individuals in such houses may be faring poorly in terms of practical and psychological outcome measures, and of course, the converse is true as well.
Table 3.
Correlations of individual-level outcome variables and house-level network measures.
| Wages | Stress | Social Support | Self-Esteem | |
|---|---|---|---|---|
| Friendship a | ||||
| APL b | −0.30*** | −0.02 | −0.08 | −0.06 |
| Density | 0.29*** | 0.03 | 0.15 | 0.10 |
| Reciprocity | 0.20 | 0.03 | 0.16 | 0.05 |
| Transitivity | 0.25** | 0.05 | 0.10 | 0.06 |
| Loan | ||||
| APL | 0.08 | −0.26** | 0.09 | 0.03 |
| Isolates | −0.20* | 0.17 | −0.21** | −0.18* |
| Density | 0.26*** | −0.24* | 0.30*** | 0.23* |
| Reciprocity | 0.04 | −0.24** | 0.09 | 0.05 |
| Transitivity | 0.08 | 0.06 | 0.13 | 0.17* |
| Advice-seeking | ||||
| APL | 0.09 | −0.09 | 0.04 | 0.08 |
| Isolates | −0.20* | −0.04 | −0.02 | −0.08 |
| Density | 0.24** | 0.28*** | 0.004 | −0.03 |
| Reciprocity | 0.33*** | 0.12 | 0.10 | 0.09 |
| Transitivity | 0.03 | 0.24** | −0.03 | −0.10 |
Note:
p < 0.05,
p < 0.01,
p < 0.001.
As there were no Isolates within the friendship network, this variable was not included.
APL is average path length.
We note as well that loan network density appeared to have the most pervasive association with individual outcome measures, but this relationship disaggregated differently than for other such associations. Specifically, although it’s level 2 correlations were bigger with social support and self-esteem compared to wages and stress, the ICC’s for social support and self-esteem were smaller compared to wages and stress. In other words, social support and self-esteem were considerably more variable within houses than wages and stress, but this similarity component—smaller though it was—was quite well explained by structural conditions, especially the existence of a more robust loan network. Wages and stress, in contrast, tended to be more similar within houses, but this similarity was not as well-explained by the nature of the loan network.
Discussion
The current study explored the relationship between house-level social network characteristics and house-level outcomes in recovery homes. Several social network variables, average path length, isolates, density, reciprocity, and transitivity, were used here to represent a variety of indicators of network connectivity and closeness. These indices were strongly associated with several important individual-level variables known to be important predictors of successful recovery, i.e., wage, stress levels, social support, and self-esteem. A major finding was that the density of the loan networks was positively associated with house-level earned wages, social support, and self-esteem in that house, and negatively associated with stress. Further, dense advice-seeking networks were associated with higher stress houses, although it is important to point out that this is an association and not necessarily causal.
In our analyses of the social network data and recovery outcomes, we found that most consistently significant correlations with positive indicators of recovery were related to the loan networks, somewhat less strongly to advice networks, and least of all to friendship (see Table 2). This suggests that both job-related and self-appraisal factors are quite strongly affected by access to practical resources—especially short term loans–rather than simply socially-supportive/friendly ones. We found that loan density was significantly related to higher wages, perceived social support, and positive self-esteem. Also, willingness to loan was negatively related to overall stress, so when stress was low, residents were more comfortable with lending. If a resource-rich network was willing to democratically disperse their resources, network stress could be relieved, and, in turn, cultivate positive outcomes.
As discussed in the results, wages were positively correlated with network density and reciprocity measures suggesting that financially stable individuals are embedded in more highly connected networks. This is supported by wage’s negative correlation with network measures like friendship average path length and loan and advice-seeking isolates. The key importance of wages for residents of these recovery homes has clear implications for those who live in these settings, and it is critical to better understand efforts to encourage residents to both find and maintain employment. Having a source of income appears to be a dynamic and stabilizing force for those living in these recovery homes.
The finding that individuals in houses with a higher density of advice-seeking also had higher levels of stress seems counterintuitive at first, but maybe a result of high stress motivating individuals to seek advice. The positive relationship between stress and advice-seeking network density could serve as a signal of house health, as opposed to being interpreted as an effect of advice-seeking. It may be that advice-seeking is not necessarily “mentorship” per se, but instead—especially early in recovery—could indicate attempts on the part of a resident to find his or her way. Moreover, individual characteristics (e.g., stress being experienced, the status of other recovery skills) may be as critical in these relationships as the availability of mentors; perhaps not all residents are in a good position to make use of advice from other residents. Ultimately, one would expect that stressed individuals seeking advice will experience lower stress if they can obtain such advice and make productive use of it, a question that needs to be explored in a longitudinal framework.
Although we report strong and consistent relationships between overall relationship structures and individual outcomes in this article, we also found that these individual outcomes are not highly correlated within houses (though their ICCs are all statistically significant). Thus there is only so much association available, so to speak, for a house-level characteristic to explain. This remaining unexplained within-house variance could depend on several different factors; for example, different types of individuals may be affected differently by house-level structural conditions, or an individual’s particular position within a house social structure might be as or more important compared to the nature of the overall structure. These and other possibilities will need to be explored in future studies.
We were surprised that social support as measured by the ISEL appeared to have no relationship to friendship or advice network factors, and we had thought that there would be a stronger relationship in Table 2. It is possible that these two measures are tapping different domains, with the ISEL measuring general social support, which included friends outside the recovery house as well as family members, whereas our measures of social networks only assess relationships that are within the house (Jason & Stevens, 2017). This needs to be explored in more detail in future research.
There are several limitations to the current study. First, although our sample of 42 recovery homes is quite large compared to other network studies, it is still somewhat limited from a statistical modeling point of view; if we were describing a study of only 42 individuals, it would be considered quite small. But practical considerations limited both the number of homes that could be studied as well as their geographic distribution. Thus, results might not generalize to all OH’s, and of course non-OH settings as well. Also, we anticipate that friendship, willingness to loan money, and advice-seeking are likely to be multidimensional, but the current study only employed them under single item measurements. This under specification problem occurs in much of the social network literature as investigators are careful not to overburden participants, whose tasks can grow quickly with the size of the whole network being rated. Additionally, the current study is observational, one that allows the data to be analyzed in a way that allows us to draw conclusions from the ground up, versus with a particular theoretical lens that might be imposed on the data. Also, this article reports on cross-sectional data, and this does not allow for causal inferences to be made. Furthermore, we do not report upon differences in the findings based on in-degree versus out-degree, which will be the topic of future articles.
In past studies, we have indicated that length of time living in the recovery home is related to outcomes, and it is clear that those living there for a short period of time may have had insufficient time to build the relationships being studied. Given the importance of this topic, future investigations will be focused both on length of time as well as positive and negative reasons for leaving the home, as well as outcomes following leaving the recovery homes.
Our article suggests that social network methodology could quantify and measure whole network relationships to better understand how relationships support or challenge recovery, and this methodology also measures to what extent individuals are making use of social capital in the homes (Jason et al., 2020). Caution needs to occur with adopting a wholly quantitative approach to measure something that is qualitative, especially since the successful application of social capital requires a consideration of the nature and value of social connections as well as the social contexts within which they are developed (Pettersen et al., 2019). Furthermore, the correlation between recovery and loan networks suggesting that physical capital is equally as important as social capital. For example, using a recovery capital framework, Palombi, Hawthorne, Irish, Becher, and Bowen (2019) enhanced our understanding of the barriers related to transportation and access to sober meetings that rural participants experienced in their recovery. Therefore, there is a need to focus on different types of capital and not simply rely on the formation of social capital, as it is physical capital that may well facilitate positive forms of social capital.
Our study found that the density of loan willingness relationships within a house was positively associated with several recovery-supportive individual outcomes (i.e., wages, social support, and self-esteem). Higher house-level density of advice-seeking relationships was related to more individual stress, suggesting that residents go to others for guidance during stressful times. There is now a need to address the causal dynamics that result in these relationships, e.g., does stress lead to advice-seeking, does advice-seeking normally lead to reduced stress, or are there common circumstances where there is no effect or stress is even increased (for instance, in the case of poor advice-giving)? These types of relationships may provide those who establish and live in recovery homes a better understanding of important recovery home dynamics that might be related to successful stays in these settings.
These types of investigations can have real-world policy implications for best housing practices for those with substance use disorders. For example, this type of research can contribute to better patient-centered decision making that will be helpful for the needs of policymakers, clinicians, and patients. For example, this information could aid patient-center decision making as well as treatment centers providing discharge planning and advice to those with housing insecurity seeking safe post-treatment settings. Such a model could lead to recommendations for better ways to organize houses, including screening methods to ensure “readiness”, a formula for an optimal mix of residents, onboarding procedures for new residents, and potentially many more efficacious adaptations to current practices.
Acknowledgments
The authors appreciate the financial support from the National Institute on Alcohol Abuse and Alcoholism (grant number AA022763). We also acknowledge the assistance from members of the Oxford House organization, and in particular Paul Molloy, Alex Snowden, Casey Longan, and Howard Wilkins.
Footnotes
Declarations of Interests
None.
Data were also collected on help, frequency, and strength, but these measures are not included in the current study. Strength and frequency are not as informative since they do not represent a theoretical relationship but rather the frequency of contact and the overall perception of the relationship.
Participant’s friendship nominations were transformed into an adjacency matrix with rows signifying an ego (senders of friendship nominations) and a column with alters (receivers of friendship nominations). If a nomination was present between an ego and an alter, it would represent a degree. All values were dichotomized (0 = no degree; 1 = degree) and entered into the corresponding element of the matrix.
We also explored each of these variables to determine the network threshold that seemed most meaningful as a predictor (such as defining friendship as only those who selected a close friend, or defining loaning as just willing to loan another resident $500, or mentorship as only seeking advice from another resident very often), and we found that our selected threshold was best in this regard.
The interrelationships of these variables is the subject of another article that employed a confirmatory factor analysis (Jason, Guerrero, Salomon-Amend, Stevens, Light, & Stoolmiller, 2020)
There are several important substantive and statistical advantages to using a 2-level model in this context. The level 2 sample sizes in our data vary from 3 to 9. Even at 9 subjects in a house, the number of subjects upon which to base a level 2 mean is limited. The modest sample sizes and the fact that the level 2 sample sizes vary makes the correlations between observed level 2 means less efficient and hence less desirable than corresponding estimates from a 2-level model. For these reasons, we chose to use 2-level models to compute the various correlations. To summarize, a two-level or mixed-effect model (e.g. Gelman & Hill, 2007) provides a realistic and statistically efficient framework for examining relationships among house-level and individual-level variables for such a design.
The cross-level correlation is the product of the house-level correlation times the square root of the ICC for a particular combination of network and survey measures and represents the indirect effect of the network variable at level 2 on the survey measure at level 1 mediated through the latent survey variable at level 2. Given the standard definition of an indirect effect, if both of the constituent effects are significant, the cross-level correlation is also significant and conversely, if either is not significant, then neither is the cross-level correlation significant. Residual diagnostics were used to check all 2-level models to assess background assumptions about the distributions of the latent variables and screen for high leverage outliers. Models were fit using Mplus (version 8.3) and employed the default, robust ML estimator.
References
- Cohen S, Kamarck T, & Mermelstein R. (1983). A global measure of perceived stress. Journal of Health and Social Behavior, 24, 385–396. [PubMed] [Google Scholar]
- Cohen S, & Wills TA (1985). Stress, social support, and the buffering hypothesis. Pschological Bulletin, 98, 310–357. 10.1037//0033-2909.98.2.310. [DOI] [PubMed] [Google Scholar]
- Dishion TJ (2013). Stochastic Agent-Based Modeling of influence and selection in adolescence: Current status and future directions in understanding the dynamics of peer contagion. Journal of Research on Adolescence, 23, 596–603. [Google Scholar]
- Freeman LC (1979). Centrality in social networks: I. Conceptual clarification. Social Networks, 1, 215–239. [Google Scholar]
- Gelman A, & Hill J. (2007). Analysis using regression and multilevel/hierarchical models. New York, NY: Cambridge: University Press. [Google Scholar]
- Jason LA, Davis MI, Ferrari JR, & Anderson E. (2007). The need for substance abuse after-care: Longitudinal analysis of Oxford House. Addictive Behaviors, 32, 803–818. [DOI] [PubMed] [Google Scholar]
- Jason LA, & Glenwick DS (Eds.). (2016). Handbook of methodological approaches to community-based research: Qualitative, quantitative, and mixed methods. New York: Oxford University Press. [Google Scholar]
- Jason LA, Guerrero M, Lynch G, Stevens E, Salomon-Amend M, & Light JN (2020). Recovery home networks as social capital. Journal of Community Psychology, 48, 645–657. 10.1002/jcop.22277. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jason LA, Guerrero M, Salomon-Amend M, Lynch G, Stevens E, & Light JM (2020). Advice seeking and loaning of money related to relapse in recovery homes. Journal of Community & Applied Social Psychology. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jason LA, Guerrero M, Salomon-Amend M, Stevens E, Light JN, & Stoolmiller M. (2020). Context matters: Home-level but not individual-level recovery social capital predict residents’ relapse. Manuscript submitted submitted for publication. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jason LA, Light JM, Stevens EB, & Beers K. (2014). Dynamic social networks in recovery homes. American Journal of Community Psychology, 53 (3–4), 324–334. 10.1007/s10464-013-9610-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jason LA, & Stevens E. (2017). The reliability and reciprocity of a social network measure. Alcoholism Treatment Quarterly, 35, 317–327 PMCID:PMC6145821. 10.1080/07347324.2017.1355220. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jason LA, Stevens E, Ferrari JR, Thompson E, & Legler R. (2012). Social networks among residents in recovery homes. Advances in Psychology Study, 1, 4–12. [PMC free article] [PubMed] [Google Scholar]
- Kassanits J, Bobak T, Stevens E, Guerrero M, Light J, & Jason LA (2020). The relationship of Oxford Houses across heterogeneous house and setting characteristics. American Journal of Orthopsychiatry, 90 (3), 324–327. 10.1037/ort0000437. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kelly JF, Hoeppner B, Stout RL, & Pagano M. (2012). Determining the relative importance of the mechanisms of behavior change within Alcoholics Anonymous: A multiple mediator analysis. Addiction, 107 (2), 289–299. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kim D, & Kawachi I. (2006). A multilevel analysis of key forms of community-and individual-level social capital as predictors of self-rated health in the United States. Journal of Urban Health, 83 (5), 813–826. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Majer JM, Jason LA, Hickey P, Norris J, Jeong H, & Bobak TJ (2020). Social support among Oxford House residents utilizing medication assisted treatments. Alcoholism Treatment Quarterly, 38 (2), 199–214. 10.1080/07347324.2019.1678445. [DOI] [Google Scholar]
- Palombi L, Hawthorne AN, Irish A, Becher E, & Bowen E. (2019). “One out of ten ain’t going to make it”: An analysis of recovery capital in the rural upper Midwest. Journal of Drug Issues, 49 (4), 680–702. [Google Scholar]
- Panebianco D, Gallupe O, Carrington PJ, & Colozzi I. (2016). Personal support networks, social capital, and risk of relapse among individuals treated for substance use issues. International Journal of Drug Policy, 27, 146–153. 10.1016/j.drugpo.2015.09.009. [DOI] [PubMed] [Google Scholar]
- Parkin S. (2015). Salutogenesis: Contextualising place and space in the policies and pol itics of recovery from drug dependence (UK). International Journal of Drug Policy, 33, 21–26. [DOI] [PubMed] [Google Scholar]
- Pettersen H, Landheim A, Skeie I, Biong S, Brodahl M, & Oute J. (2019). How social relationships influence substance use disorder recovery: A collaborative narrative study. Substance Abuse: Research and Treatment. 10.1177/1178221819833379. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Polcin DL, Korcha R, Bond J, Galloway G, & Lapp W. (2010). Recovery from addiction in two types of sober living houses: 12-month outcomes. Addiction Research and Theory, 18, 442–455. 10.3109/16066350903398460. [DOI] [Google Scholar]
- Rosenberg M. (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. [Google Scholar]