Skip to main content
PMC home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Jan 1.
Published in final edited form as: J Hum Behav Soc Environ. 2021 Jul 12;32(5):663–678. doi: 10.1080/10911359.2021.1947930

Dynamic Interdependence of Advice Seeking, Loaning, and Recovery Characteristics in Recovery Homes

Leonard A Jason 1, Gabrielle Lynch 2, Ted Bobak 3, John M Light 4, Nathan J Doogan 5
PMCID: PMC9400817  NIHMSID: NIHMS1722321  PMID: 36034076

Abstract

Recovery homes in the US provide stable housing for over 200,000 individuals with past histories of homelessness, psychiatric co-morbidity and criminal justice involvement. We need to know more about how these settings help those remain in recovery. Our study measured advice seeking and willingness-to-loan relationships and operationalized them as a dynamic multiplex social network—multiple, simultaneous interdependent relationships--that exist within 42 Oxford House recovery homes over time. By pooling relationship dynamics across recovery houses, a Stochastic Actor-Oriented Modeling (SAOM) framework (Snijders et al., 2010) was used to estimate a set of parameters governing the evolution of the network and the recovery attributes of the nodes simultaneously. Findings indicated that advice and loan relationships and recovery-related attitudes were endogenously interdependent, and these results were affected exogenously by gender, ethnicity, and reason for leaving the recovery houses. Prior findings had indicated that higher advice seeking in recovery houses was related to higher levels of stress with more negative outcomes. However, the current study found that recovery is enhanced over time if advice was sought from residents with higher recovery scores. Our study shows that social embedding, i.e. one’s position in relationship networks, affects recovery prospects. More specifically, the formation of ties with relatively more recovered residents as an important predictor of better outcomes.

Keywords: Advice seeking, Loaning, Recovery Homes, Oxford Houses, Recovery Latent Factor, Social Networks, Stochastic Actor Oriented Modeling

Recovery homes are an inexpensive and effective option for individuals who have housing instability and a need for a setting to support abstinence (Polcin, 2009). Over 250,000 individuals live in these settings each year, and most have histories of homelessness, criminal justice involvement and psychiatric co-morbidity (Jason, Wiedbusch, Bobak, & Taullahu, 2020). Individuals are more likely to remain in these community-based settings when their personal social networks include recovery home members (Brereton et al., 2014), and longer lengths of stay in these settings are positively associated with maintenance of abstinence (Jason, Olson, Ferrari, Majer, Alvarez, & Stout, 2007).

Social capital theory would suggest that social relationships provide the key facilitating resources within recovery homes (Bliuc, Best, Iqbal & Upton, 2017; Burns & Marks, 2013; Granfield, & Cloud, 2008; Laudet, & White, 2010). Hence, social relationships in recovery homes might mediate length of stay, which has been shown to be a key predictor of sustained recovery (Jason et al., 2007). The literature lacks a clear description of the dynamics and processes of social networks that characterize the transmission of recovery social capital within recovering individuals’ social environments, along with any linkage to recovery outcomes. There is a need to investigate the social dynamics that generate and maintain recovery social capital for recovery home residents. Elements of social and physical recovery capital (Cloud & Granfield, 2008) may be at least as valuable as social support, and networks that provide access to such resources (such as willingness to loan and provide advice) might be better indicators of true social acceptance compared to “friendly” relations.

It should be noted that network relations can be maladaptive too. Relations with housemates who violate rules could affect one’s own recovery in a negative way. Jason et al. (2021, a) found the density of willingness-to-loan-money relationships among house residents was positively associated with a number of recovery-positive outcomes (i.e., improved wages, social support, and self-esteem). However, density of advice-seeking relationships was related to more stress, which could lead to poorer recovery outcomes. It is plausible that this finding simply reflects the negative effect of stress on outcomes; that is, whether one seeks advice or not, being stressed is an indicator of poorer recovery prospects. Jason et al. (2021, b) also found higher density of sharing resources was predictive of lower relapse, whereas density of higher advice seeking was predictive of higher relapse rates. However, the question of primary interest is not whether stressed individuals have poorer recovery outcomes, all else being equal, than less stressed individuals. Rather, we want to know whether stressed individuals who seek advice do better than those who do not, and if so, under what circumstances. For example, it seems plausible that seeking advice from other residents would be more effective, the more stable those residents’ own state of recovery.

Typically, issues like social support have been addressed by assuming the social environment is fixed over the time frame of a study; this has even been true in network-based conceptualizations (Walker, Wasserman & Wellman, 1993). This approach may make sense when the social environment in question changes on a much slower time scale than the individual behaviors and attitudes of its members; for instance, when the social environment in question is quite large (a city or country, or sometimes even a neighborhood). Recovery house social environments, on the other hand, may change quite quickly in response to composition or individual change, because they are relatively small (5–10 individuals, generally) and, by design, highly interdependent both socially and instrumentally. Thus, a modeling framework is required that does not take either the social environment or the behaviors and attitudes of the individuals comprising them as fixed. Both longitudinal and endogenously-evolving complex systems approaches allow linkages of specific embedding in the relationship networks to behavioral changes, and vice versa.

This study extends the above-noted literature by applying Stochastic Actor Oriented Modeling (SAOM; Snijders, van de Bunt & Steglich, 2010) to questions of recovery home social dynamics and individual outcomes. SAOM provides a framework for understanding how important types of social relationships change over time within recovery homes, and how these dynamics interact with individual recovery progress and risk of early departure. This modeling framework makes it possible to describe how mechanisms of social integration and addiction recovery interact in recovery home contexts, as residents develop potentially abstinence-supportive relationships with their housemates, while at the same time being exposed to (and hopefully, adopting) attitudes and habits that other residents have found helpful in their own recovery.

Jason, Light, Stevens, and Beers (2014) developed a SAOM of a small set of sober living homes examined over a three-month period. Willingness to loan money relationships tended to be reciprocated, but reciprocation occurred less in confidant relationships where there is a confider and listener. When Light et al. (2016) used SAOM to examine three levels of resource sharing in this same dataset, they found that confidant relationships were more likely when residents were more willing to loan money to the other resident. These studies suggest that the approach of pooling network dynamics across multiple recovery homes can yield useful results (see also Wӧlfer, Faber & Hewstone, 2015).

However, previous research found that social networks of loan willingness facilitate positive outcomes, while the reverse appears to be true for advice seeking networks (Jason, 2021, a, b), even though one might expect that advice-seeking would be helpful. Our study adapts longitudinal network modeling methods (SAOM) to the general problem of understanding why recovery house residency “works” for many individuals, and specifically, whether advice seeking and loaning relationships are beneficial, and if so, under what circumstances. In the current study, we applied the SAOM framework (Snijders et al., 2010) to identify structural and behavior-based mechanisms describing the endogenous evolution of advice seeking and loaning relationships and recovery attitudes and behaviors of individuals within recovery houses.

Method

Settings

The study was conducted in a set of Oxford Houses (OH) that are part of a network that consists of over 3,000 recovery homes in the US. These homes are self-governed with no professional staff. Each house is rented and gender-segregated, housing about 6 to 12 individuals in recovery. All residents are required to follow three primary rules: paying their fair share of rent (which typically ranges between $100 to $125 per week), contributing to the maintenance of the home, and abstaining from using alcohol and other drugs.

Data were collected from OHs located in North Carolina, Texas, and Oregon. Member-elected house presidents were asked to introduce the study to residents by reading a project-provided script about the study; houses were accepted into the study if the house president and all, or all but one member agreed to participate. The first thirteen consenting houses from each state were accepted, and three more houses were added for a total of 42. One house dropped out completely, but another was added after wave 1 bringing the total to 43 houses. However, only 42 houses had the 2 or more waves of data available on the residents that are required for a longitudinal analysis.

Participants were part of a longitudinal study that collected information every four months over a 2-year period (from 2015 to 2018), for a total of 7 waves (thus allowing us to compare the current study’s data with another Oxford House study that only used 6 waves because of missing data issues in the seventh; Jason et al., 2021, b). Participants were recruited and interviewed by field research staff in face-to-face meetings. Each participant was compensated $20 for each completed assessment. Permission was obtained through the DePaul University Institutional Review Board.

There were 714 residents in the OHs throughout the 2-year study period, of which 666 (93%) agreed to participate in our study.[1] Of those agreeing to participate during the longitudinal study, 497 (74%) left the OHs at some time during the course of the 6-wave study period.

Measures

Resident demographic information included age, sex, race/ethnicity, education and employment. Race/ethnicity included White (78.8%), Black (8.6%), Latinx (10.1%) and all other (2.5%). Oxford Houses are gendered and accordingly gender was included as a house level predictor. Sex was coded 0 for male and 1 for female. We categorized residents as employed part time or full time, versus not (unemployed, disability, student, retirement). We classified residents in a binary fashion as either having a) a high school education or GED or less or b) some college, technology school, or higher degree. The amount of time living in the Oxford House residence prior to study participation was assessed at wave 1.

Latent Recovery Factor

The recovery factor was a factor score based on a confirmatory factor analysis of all recovery capital indicators including wages, self-efficacy, stress, self-esteem, social support, alcoholics anonymous affiliation, quality of life, and length of stay. A confirmatory factor analysis supported a single latent variable (Jason, Guerrero, Salomon-Amend, Stevens, Light, & Stoolmiller, in press).

The analysis includes the participating resident’s recovery factor (RF) score at each wave during which they were observed. This measure was calculated from the following instruments.

Wages.

Self-report wages for the last 30 days were square-root transformed to reduce right-skew and treated as a continuous variable.

Quality of Life.

The World Health Organization Quality of Life Assessment-Brief (World Health Organization Group, 1998) is a 24-item questionnaire that assesses quality of life across four dimensions: social relationships, environment, physical, and psychosocial. This scale has been validated in substance using populations (Garcia-Rea & LePage, 2010). The subscales varied in their reliability (α = .89 for social relationships, .84 for environment, .83 for physical, and .83 for psychological). The alpha for the whole measure for our sample was .89.

Self-efficacy.

The Drug Taking Confidence Questionnaire (Sklar et al., 1999) is an 8-item survey that measures self-efficacy in terms of abstinence. Participants are asked to consider themselves in eight, theoretically high-risk situations and indicate how confident they are that they could resist the urge to use a substance given the theoretical circumstances. This measure for our sample has good reliability (α =.95).

Self-esteem.

The Rosenberg’s Self-Esteem Scale (Rosenberg, 1965) was utilized to measure the participant’s positive and negative feelings about 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”. Examples of 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 for our sample was .92.

Stress.

The Perceived Stress Scale (Cohen et al., 1983) measured the degree to which situations in participants lives are appraised as stressful. PSS consists of 4-items measured on a 5-point Likert scale ranging from “never” to “very often.” Examples of 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 perceived stress scale for our sample was .73.

Social support.

The Interpersonal Support Evaluation List (Cohen & Wills, 1985; Cohen, Mermelstein et al., 1985) was utilized to measure three types of perceived social support (tangible, appraisal, belonging). Tangible support refers to instrumental aid and monetary assistance; appraisal support refers to having someone to talk to about one’s problems; and belonging support refers to the availability of people one can do activities with. The Interpersonal Support Evaluation List consists of 12-items measured on a 4-point Likert scale ranging from definitely false to definitely true. The internal reliability of the support scale for our sample was .88.

Sense of Community.

The Psychological Sense of Community (Jason. Stevens, & Ram, 2015) is a 9-item scale utilized to measure participant’s sense of community. Examples of items include “This Oxford House is important to me” and “For me, this Oxford House is a good fit”. The three subscales are Entity, Membership, and Self, and for our sample, they have Cronbach alphas of .67, .92, and .91, respectively. The SOC scale was used as a whole measure, and for our sample, α = .91.

Hope.

The State Hope Scale (Snyder et al., 1996) consists of 6 items that measure participants’ current state of hope. The Hope measure contains two sub-scales Agency (α = .94) and Pathways (α = .81). We included a 3-item subscale of hope that measures Environmental Context (Stevens et al., 2014) (α = .97). This 9-item hope scale was analyzed as a whole measure, and for our sample the α = .90.

Exit Status

We also included the Addiction Severity Index-Lite (ASI-Lite; McLellan et al., 1997) to assess problems in drug and alcohol use over the past 30 days. The ASI-Lite has been demonstrated to have good validity and reliability (Cacciola, et al., 2007). This instrument along with other information from the house officers was used to create the following exit status variables: positive outcomes (53% left on good terms), negative outcomes (24% either relapsed, or left for disruptive behavior or financial reasons), still a resident (22%), or left with no reason available (1%). Three 0/1 indicator variables were created from this information: still in residence, left for a positive reason, or left for a negative reason.

Social Networks

The Social Network Instrument (SNI; Jason & Stevens, 2017) was utilized to capture the social dynamics within each OH. This instrument has been used in several investigations on the social networks of recovery home residents (Jason et al., 2014; Jason & Stevens, 2017; Light et al., 2016). This type of social network instrument has been found to be a reliable measure (Hlebec & Ferligoj, 2002). The SNI used with our sample had a Cronbach’s alpha of .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). The SNI measures several relationship types, including loan, and advice-seeking. Data were also collected on friendship, help, frequency, and strength, but these relationship types were outside the scope of the current study.

Residents rated each other house member on the relationships of advice seeking or money loaning.1 We refer to the rater as “ego” and those being rated, whether we defined them as connected or not, as “alters.” An advice-seeking relationship was considered present if ego reported seeking advice from the alter very often or often, but not present otherwise (i.e., regularly, rarely, never). A money loaning relationship (which represents a willingness to lend resources) was considered present if ego reported being willing to loan an alter the sums of either $100 or $500, but was not considered present if ego reported being willing to loan alter smaller amounts of money (i.e., $0, $10, $50).

Analysis

Social network analysis of 42 OHs was conducted using the stochastic actor-oriented modeling (SAOM) framework as implemented in the R package RSiena (Ripley et al., 2020). A thorough and intuitive review and example use of the software to construct and estimate a SAOM can be found in Snijders, van de Bunt, and Steglich (2010). For this study, the software was used to model the endogenous co-evolution of social relationships and behavior among the residents of the OHs.

SAOMs apply an agent-based decision-making approach to modeling relationships and behavior. That is, individuals in the system being modeled are assumed able to make relationship choices to their outgoing ties and behaviors, and it is their actor-based choices that drive the temporal evolution of the larger system (i.e., the observed networks and behaviors).

A SAOM consists of a set of equations, one for each endogenous variable, expressed as a function of a set of predictors—referred to by the software, and in later parts of this paper as “effects.” The equations can share common endogenous or exogenous variables and can thus be interlinked in useful ways to model social evolution. Our study included two relationship equations, one each for advice seeking and loaning networks, and two behavior equations, for recovery factor (RF) and rate of exit for negative reasons (negExit). The relationship equations describe the stochastic creation, dissolution, and maintenance of ties between house members, whereas the behavior equations describe changes (up, down, or none) in recovery factor scores, and the hazard of a negative exit at any given point in time.

The equations in SAOMs can be seen as functions describing the utility the actors place on being associated with particular configurations of their local network and their behaviors. Predictors in SAOMs are the dimensions about which “preference” can vary. For example, if the researcher believes actors have a preference related to being engaged in reciprocal relationships, a relevant predictor should be included in the model. The term “preference” may be (though in general need not be) taken literally if doing so is theoretically appropriate; in the present study, we adopt such an interpretation of effects, as it appears to dovetail well with the types of choices available to recovering individuals in sober living homes. A wide variety of these predictors (or “effects”) are available in RSiena. Aside from basic mechanisms of network closure (tendencies for existing ties to promote or inhibit formation of additional ties), the effects of primary interest in this study include those that take account of characteristics of ego (tie chooser) and alter (choice recipient) as predictors of the tie formation process. “EgoX” effects allow for variation in ego’s preference for having a certain number of out-ties. “AlterX” effects allow for variation in ego’s preference about to whom they connect as a function of the alter’s value on the variable X. “SimilarityX” also allows for variation in ego’s preference for who to connect with, but it depends on the similarity of ego’s and alter’s value on X, and is often used to model homophily.

In the behavior equation for recovery factor change, we are mainly interested in whether the number of relationships of a given type (loan willingness or advice seeking), possibly including the average or total of some characteristic X of the relevant alters, predict changes in ego’s RF. Total alter advice is a score that attaches to each individual and it is measured the total recovery score of all alters that ego has nominated. In the behavior equation for rate of house exit, the same considerations apply, except that the endogenous variable is a rate of exit (time to event), so the interpretation of effects in this equation differs from the interpretation of RF changes. For more information on effects, see Snijders et al. (2010) and the RSiena Manual (Ripley, Snijders et al., 2020).

Model building used the recommended (Ripley et al., 2020) forward selection process, starting with model of just rates, outdegree (density), and reciprocity for both advice-seeking and loaning relationships, and then adding ego and alter effects. For RF, besides rates, effects included only the effect of the endogenous variable on itself (linear and quadratic components), and for house exits, only rate effects were used. We initially kept effects with t-values in the vicinity of 1 or greater. Ultimately all surviving specifications were re-estimated to determine statistical significance of effects, and only statistically significant (or nearly, if the effect seemed important for other reasons) effects were retained in the models presented below.

Fit is assessed by testing how well the model reproduces large-scale structural or behavioral features of the data. For instance, a well-fitting model should reproduce network statistics such as proportion of reciprocated relationships, triadic structures, and so on (Lospinoso & Snijders, 2019). The approach measures statistical agreement defined as Mahalanobis Distance2. The standard suite of structural features used in goodness-of-fit checking includes outdegree distribution (total number of others chosen by ego), indegree distribution (total number of alters who choose ego), and the triad census (all possible triadic structures). In general, parameter estimates were pooled across houses and waves, and missing data were in effect treated as MAR (Little & Rubin, 1987). Further details are available in the RSiena manual (Ripley et al., 2020). More details on the description of the data are included in Doogan et al. (2019).

Results

Network Dynamics

Table 1 shows p values on the above-noted structural features for the model. Parameter estimates are reported along with their 95% CIs (b=estimate, [95%CI lower, upper]). Parameter estimates were based on 1164 iterations during the estimation routine, with convergence diagnostics, covariance and derivative matrices based on 3003 iterations. Model convergence was very good; the overall maximum convergence ratio (a summary measure across effects) was .13 (the conventional cutoff is 0.25; Ripley et al., 2020), and all individual parameter convergence t ratios (the autocorrelation between successive iterative estimates, which ideally are near zero) were .07 or less.

Table 1.

Stochastic Actor-Oriented Model Results–maximum likelihood estimation

Parameter Estimate SE p-Value 95% Confidence Interval Convergence t-ratioa
Network Dynamics
 1. Advice rate (period 1) 2.16 0.36 <.001 (1.5, 2.9) −0.03
 2. Advice rate (period 2) 2.32 0.49 <.001 (1.4, 3.3) 0.01
 3. Advice rate (period 3) 2.43 0.44 <.001 (1.6, 3.3) 0.04
 4. Advice rate (period 4) 2.14 0.41 <.001 (1.3, 2.9) 0.01
 5. Advice rate (period 5) 6.56 1.46 <.001 (3.7, 9.4) 0.00
 6. Advice: outdegree (density) −0.63 0.10 <.001 (−0.8, −0.4) −0.02
 7. Advice: reciprocity 0.86 0.15 <.001 (0.6, 1.1) −0.01
 8. Advice: transitive triplets 0.42 0.07 <.001 (0.3, 0.6) −0.02
 9. Advice: 3-cycles −0.42 0.12 <.001 (−0.6, −0.2) −0.02
 10. Advice: Length of Stay ego 0.08 0.04 .059 (0.0, 0.2) 0.00
 11. Advice: Education similarity 0.21 0.11 .070 (0.0, 0.4) −0.01
 12. Loan rate (period 1) 3.15 0.52 <.001 (2.1, 4.2) 0.01
 13. Loan rate (period 2) 2.71 0.63 <.001 (1.5, 3.9) −0.01
 14. Loan rate (period 3) 2.50 0.41 <.001 (1.7, 3.3) 0.00
 15. Loan rate (period 4) 2.56 0.53 <.001 (1.5, 3.6) 0.01
 16. Loan rate (period 5) 3.48 0.57 <.001 (2.4, 4.6) 0.00
 17. Loan: outdegree (density) −0.76 0.15 <.001 (−1.0, −0.5) 0.01
 18. Loan: reciprocity 0.80 0.12 <.001 (0.6, 1.0) 0.02
 19. Loan: Employment alter 0.34 0.14 .017 (0.1, 0.6) 0.01
Behavior Dynamics
 20. Rate RF (period 1) 1.44 0.27 <.001 (0.9, 2.0) 0.00
 21. Rate RF (period 2) 1.21 0.24 <.001 (0.7, 1.7) 0.01
 22. Rate RF (period 3) 1.54 0.32 <.001 (0.9, 2.2) 0.00
 23. Rate RF (period 4) 1.39 0.30 <.001 (0.8, 2.0) −0.03
 24. Rate RF (period 5) 1.79 0.37 <.001 (1.1, 2.5) −0.02
 25. RF linear shape 0.37 0.11 <.001 (0.2, 0.6) 0.01
 26. RF quadratic shape −0.49 0.09 <.001 (−0.7, −0.3) 0.01
 27. RF total alter (Advice) 0.19 0.09 .045 (0.0, 0.4) 0.01
 28. RF: effect from Sex −0.39 0.15 .010 (−0.7, −0.1) −0.02
 29. RF: effect from Black 0.63 0.25 .011 (0.1, 1.1) 0.04
 30. Rate nExit (period 1) 0.34 0.13 .007 (0.1, 0.6) −0.03
 31. Rate nExit (period 2) 0.21 0.08 .011 (0.0, 0.4) −0.01
 32. Rate nExit (period 3) 0.22 0.10 .024 (0.0, 0.4) 0.00
 33. Rate nExit (period 4) 0.07 0.04 .139 (0.0, 0.2) −0.07
 34. Rate nExit (period 5) 0.02 0.02 .346 (0.0, 0.1) 0.00
 35. Outdegree effect on rate NegExit (Advice) 0.27 0.13 .033 (0.0, 0.5) −0.06
 36. Indegree effect on rate NegExit (Loan) −0.54 0.27 .046 (−1.1, 0.0) −0.05
Open in a new tab
a

Ratio of deviations of simulated vs. observed statistics for each effect, calculated in Phase 3 of the RSiena model estimation procedure. Conventionally, a value of less than 0.10 indicates good convergence (Ripley et al., 2020).

Rate parameters for advice and loan willingness are inter-wave-specific estimates of the latent amount of change in each endogenous variable, but have no substantive significance other than confirming sufficient empirical variation for a statistical model to explain. Network closure parameters for loaning and advice relationships included outdegree (for density), reciprocity, transitive triplets, and 3-cycles. These effects index structural tendencies important to predicting tie dynamics and provide useful descriptive information, but are otherwise not substantively significant here. Outdegree density for advice (b=−0.63, [−0.8, −.4]) and loaning (b=−0.76, [−1.0, −0.5]) were significant. Because the relationship is negative, it means that fewer than half of the potential advice or loan ties occur in these networks. This is an indicator of sparseness so there are not too many residents that others trust to loan at least $100 or to seek advice (i.e., a low probability of ties, which is also reflected in the low density of the network and the low average outdegree). The negative estimates also indicate that if a resident has one outgoing advice or loan tie, the chance are diminished that the resident will add another one over time.

Reciprocity for advice (b=0.86, [0.6, 1.1]) and loaning (b=0.80, [0.6, 1.0]) were also significant. Reciprocity is positive for both networks, implying a tendency for each relationship to be bidirectional. For advice, transitive triplets (b=0.42, [0.3, 0.6]) and 3-cycles (b= −0.42 [−0.6, −0.2]) was significant. In other words, advice was found to be both transitive (if A goes to B for advice and B goes to C for advice, A is also likely to go to C for advice) and hierarchical (as shown by the negative 3-cycle effect, suggesting that advice tends to have unidirectional nature which counteracts the reciprocal effect to some degree). These effects were not significant for loan networks, suggesting that loan willingness relationships are more dyad-specific than advice relationships.

For advice networks, residents who have been in the house for a longer period of time tended to name more housemates they go to for advice (b=0.08, [0.0, 0.2], p = .059). Also the more similar the education level between ego and alter, the more likely they are to seek advice from one another (b=0.21, [0.0, 0.4], p = .07). In the loan network, those who were employed were more likely to be nominated as potential recipients of loans than those that were not employed, and this effect was significant (b=0.34, [0.1, 0.6]).

Behavior Dynamics

The model examined predictors of RF and changes in the RF, and how these variables are inter-related to each other. The linear and quadratic shape effects represent the location and shape of the RF distribution for the referent group when all other RF model terms are set to zero, and otherwise are substantively irrelevant.

For RF, the significant and positive total alter effect from the advice seeking network implies that if the sum of the RF scores among the alters from whom ego seeks advice is greater than the RF score of ego, ego’s recovery factor will significantly improve (b=0.19, [0.0, 0.4]).3 So having more people an ego seeks advice from with higher RF scores leads to a higher recovery score for the advice seeker. In general, a positive coefficient indicates that certain types of actors tend toward higher scores when given the opportunity to change.

Additionally, as sex is coded 0 for male and 1 for female, the negative value for this estimate means that men have a tendency to improve their RF faster than women during the time that they are in the recovery house (b=−0.39, [−0.7, −0.1]). The positive significant parameter for African American ethnicity shows that African Americans had better probabilities for RF improvement than other ethnicities (b=0.63, [0.1, 1.1]).

The model also included effects of advice and loan relationships on the rate (time-specific probability) of leaving the house for negative reasons.4 Relationship embedding factors affected departure rate. We found a significant net effect of advice seeking as a risk factor for negative departure (b=0.27, [0.0, 0.5]). Residents were also significantly less at risk of negative departure when more other residents were willing to loan them money (b=−0.54, [−1.1, 0.0]).

Discussion

The current study identified several interdependencies among advice seeking and loaning relationships and the RF score of residents. The results were based on a system of stochastic difference equations, and this SOAM framework (Snijders et al., 2010) allowed us to treat two types of relationships and behaviors/attitudes as endogenous, mutually-interacting entities that co-evolve over time. A better understanding of these dynamics is necessary to understand how recovery homes, and specifically Oxford House recovery homes, promote recovery in a communal setting.

One of the major findings was that seeking advice from those who are more recovered leads to increases in the advice-seeker’s RF. More specifically, the effect in question is a “total alter” effect, defined as the sum of the recovery factor scores of all alters ego goes to for advice at any given time, that is, an RF-weighted sum of the number of such alters. Furthermore, preliminary analyses found no significant effect of average alter’s RF in the advice-seeking relationship. Hence the sheer number of such alters must be important, not just their RF. Unfortunately, attempts to pull apart the effects of number of alters and the RF of those alters were inconclusive, possibly due to data limitations (we will say more about this below, when discussing study limitations) 5.

Despite these caveats, this finding helps explain a puzzling result from a previous study. Jason et al. (2021a,b) reported that advice seeking was related to stress and less positive outcomes, even though one might expect advice seeking to be beneficial. This could be because such a relationship ignores characteristics of the advice giver. In the absence of such considerations, it is likely that stress in general is a risk for bad outcomes. However the present study did take such characteristics into account, finding that seeking advice from more individuals who are also more recovered tends to be beneficial. The SAOM framework involves a dyad-based conception of relationship formation, as well as the subsequent effects of those relationships that takes account of who, specifically, the relationship partner is. This additional detail was apparently necessary to identify the true effect of such relationships.

We also found that the more others ego named as seeking advice from, the higher the hazard for a negative exit. This result seems to be similar to results from the previously mentioned study by Jason et al. (2021, b) which found that advice seeking in general was a risk factor for poor recovery outcomes. Unfortunately at this time the SAOM does not provide a rate function relationship outdegree effect directed towards alters who have a specific value (say, a higher RF than ego’s). Adding such an effect is possible in principle, however, and we hypothesize that when such coding enhancements can be completed, we will find that seeking advice from relatively more recovered alters will predict a lower hazard rate for negative exit, as was the case for advice seeking and the RF.

Another finding of interest was the attenuating effect of willingness of fellow residents to loan ego $100 or more on risk of a negative exit. In other words, residents were less at risk of negative departure when other residents were willing to loan them money. This result suggests that the individual in question is socially accepted and trusted, which in turn is likely to increase his or her feelings of belongingness to the group. Besides the practical value of access to necessary financial aid as an incentive to remain in residence and on good terms with other residents, this type of inclusion also makes a resident more likely to accept the norms and recovery practices of other residents (e.g., Festinger, Schachter & Back, 1968), which in turn is supportive to recovery outcomes. The finding is also consistent with other recent studies using different methodologies, which found that loan relationships serve as protective factors preventing early drop out (Jason et al., 2021,b).

Our study also found that African American residents tend to improve their RFs more, all else being equal, than other ethnic groups (in this sample, primarily be Caucasians). That is, an African American resident with a given RF score is more likely to increase his/her RF over time than a white resident with the same given score. These findings are also consistent with those of Harvey (2014), who found that African Americans in recovery homes relapsed at lower rates than Non-Hispanic White Americans. In addition, Bishop, Jason, and Ferrari (1998), who found that African Americans tended to stay in recovery homes longer than Non-Hispanic White Americans. Moreover, Brown, Davis, Jason, and Ferrari (2006) found that African Americans gained more resources than Non-Hispanic White Americans when living in recovery homes. Furthermore, African Americans compared to Non-Hispanic White Americans reported significantly more employment in the past 30 days while living in a recovery home (Belyaev-Glantsman, Jason, & Ferrari, 2009). Thus, even though the current finding is based on a small sample of African Americans, the result is consistent with a variety of other studies in supporting the implication that African Americans are apparently able to gain even more than other ethnic groups from recovery home settings. Health disparities researchers should continue to explore the possible benefits that recovery homes could provide to this population.

Men appeared to have better potential outcomes than women on RF improvement. Davis and Jason (2005) found that social support plays a different role in women’s recovery than it does in men’s. Porcaro, Nguyen, Salomon-Amend, Chaparro, and Jason (2020) found a significant negative relationship between psychiatric severity and coping resources for male recovery houses, but for female houses this relationship was directionally positive. Although little research exists on gender differences in social support specific to abstinence or substance involvement, research on women in recovery and measures of general social support suggests that women tend to have higher levels of support from friends, while men are more likely to report family members as primary sources of support (Robles et al., 1998). Further, an investigation of gender differences of stressors and resources among problem drinkers, Skaff, Finney, and Moos (1999) found that friendships had a stronger impact on women than men and that greater support from friends had a positive impact on women’s functioning (i.e., decreased depression and alcohol consumption) whereas friendship stressors tended to have a negative impact on functioning. Based on this limited research, it appears that there may be important differences between men and women in terms of the composition and impact of support networks on attaining and maintaining abstinence. As women report that friends comprise their most important sources of support, men would report that their family members fill these roles; it is possible that living in a recovery home for men is more similar to being with family members, and for women, the friendship relationships might not be exactly similar to what occurs outside of the recovery homes. Thus, it is important to further explore whether the composition of support networks for women and men in recovery from substance abuse differ in terms of the relationships that they report to be most important and with whom they spend the most time.

This study’s results are consistent with a social capital conception of residence-based recovery. Recovery homes may provide social capital by developing social relationships of various types. This, in turn, appears to lead to improvements in resident recovery related beliefs and behaviors such as abstinence related self-efficacy, coping with stress, self-esteem and hope that increase the chances of abstinence, social integration and success. Social capital research suggests that these social relationships provide resources, but little is known about the mechanisms that create (or fail to create) such relationships. This is because many studies ignore the benefits of contextually-constrained (“socially influenced”) decision making, instead treating the recovering individual’s path to recovery as independent of his or her social environment. The present study puts these constraints at the center of recovery success, addressing both their causes and their effects.

Limitations of this study include the sample, which was obtained based on combined factors of tractability while attempting to include as much regional and demographic variability within that constraint. Additionally, although the study sample included 627 residents over as many as six waves of data collection, the network tie data obtained is much less than what one would have from a single network of 627 individuals. The latter would feature potentially 627 × 627 = 393,129 ties per wave. Because ties could only occur within residences, however, our study had only 5,389 total ties across all six wave. This probably explains why we were not able to develop more detailed models, particularly models showing how recovery may have feedback effects on social relationships. We looked for such effects, e.g. effects of the RF on loan and advice relationships, but no such effects were statistically significant. It may be that these effects do not exist, but it is not typical of social network dynamics generally (Brechwald & Prinstein, 2011), nor with some extant studies of recovery home relationship formation (Doogan et al., 2019). Further analyses may reveal more subtle such feedback mechanisms than we have been able to uncover to date. Note too that the focus of this study was primarily on instrumental types of social capital. Future studies should also consider examining more purely social relationships such as friends and close friends, as these relationships may both provide distinct benefits beyond instrumentality, as well as possibly playing important roles as moderators of mediators of other mechanisms.

These caveats aside, this study presents the most comprehensive dynamic model of recovery home social dynamics yet published, including their interrelationships with recovery outcomes. Broadly speaking, the study shows clearly that social embedding, i.e. one’s position in relationship networks, affects recovery prospects. It particularly identifies the formation of ties with relatively more recovered alters as an important predictor of better outcomes. Future studies may thus focus on the mechanisms that more precisely predict the formation of such affiliations.

Acknowledgments

The authors appreciate the financial support from the National Institute on Alcohol Abuse and Alcoholism (grant number AA022763).

Footnotes

[1]

34 residents had more than one OH exit, either from two different OH’s or from the same OH. To avoid greatly complicating our model for a very small number of individuals, we only included their first exit in all analyses.

1

Each social network relationship type was measured with a 5-point scale. Participant’s ratings were represented by an adjacency matrix with each row representing the ratings provided by an individual and each column representing the ratings received by an individual. The SAOM analytic strategy required all rating values to be dichotomized (0 = no relationship present; 1 = relationship present) and entered as a corresponding element of the matrix.

2

Mahalanobis distance is a generalization of Euclidean distance that additionally accounts for correlation among the dimensions of a multivariate statistic. This is useful here because when characterizing the network with a multidimensional statistic, a single tie change to the network could affect multiple dimensions of the statistic in different—but correlated—ways. Failing to account for correlations could bias the calculated distance in either direction depending on the nature of the correlation structure.

3

Total alter is a measure of how ego’s score is affected by whether alters’ scores are on average greater or lesser than the overall sample mean. A positive estimate indicates that egos tend to have RF scores in the same direction away from the mean as their alters, and does not differentiate whether that is an increase or decrease, or whether ego becomes more similar or more dissimilar from their alters (e.g., the latter would occur if ego moves further from the mean than alters).

4

This outcome is modeled as a generalized Cox Regression proportional hazard formulation (Cox, 1972; Greenan, 2015) with a predictor-based time-to-departure rate as the (proportional) hazard rate parameterization. That is, overall between-wave rates of change can be interpreted as non-contingent rates of house departure, which may be modified by some aspects of ego’s network embedding or behavioral/demographic characteristics.

5

Note that it would be possible for all of ego’s peers to have lower recovery factor scores than ego. But because the sum of their recovery factors is higher than ego’s recovery score, ego’s RF should improve. This could mean that, for example, ego gets different value from each individual alter, and it may not matter whether alter has a higher recovery factor at all. In other words, the effects depend on one’s alters’ recovery factor scores relative to one’s own. Even if not a single alter has a higher recovery factor than ego, the model indicates that ego may still increase the recovery factor as a result of the distribution of alters’ recovery factors. This is intriguing especially since the average-alter influence effect was not significant, as noted earlier. Nevertheless, it is fair to say that if one affiliates with alters whose recovery factor is better than their own, theirs will improve, all else being equal.

Contributor Information

Leonard A. Jason, DePaul University

Gabrielle Lynch, DePaul University.

Ted Bobak, DePaul University.

John M. Light, Oregon Research Institute

Nathan J. Doogan, Ohio State University

References

  1. Belyaev-Glantsman O, Jason LA, & Ferrari JR (2009). The relationship of gender and ethnicity to employment in recovery homes. In Jason LA, & Ferrari JR (Eds.). Recovery from addiction in communal living settings: The Oxford House model [Special Issue]. Journal of Groups in Addiction Recovery, 4, 92–99. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bishop PD, Jason LA, Ferrari JR, & Huang CF (1998). A survival analysis of communal-living self-help, addiction recovery participants. American Journal of Community Psychology, 26, 803–821. [DOI] [PubMed] [Google Scholar]
  3. Bliuc A, Best D, Iqbal M, & Upton K (2017). Building addiction recovery capital through online participation in a recovery community. Social Science & Medicine, 193, 110–117. [DOI] [PubMed] [Google Scholar]
  4. Brechwald WA, & Prinstein MJ (2011). Beyond Homophily: A Decade of Advances in Understanding Peer Influence Processes. Journal of Research on Adolescence (Blackwell Publishing Limited), 21(1), 166–179. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Brereton KL, Alvarez J, Jason LA, Stevens EB, Dyson VB, McNeilly C, & Ferrari JR (2014). Reciprocal responsibility and social support among women in substance use recovery. International Journal of Self-Help & Self-Care, 8, 239–257. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Brown JT, Davis MI, Jason LA, & Ferrari JR (2006). Stress and coping: The roles of ethnicity and gender in substance abuse recovery. Journal of Prevention & Intervention in the Community, 31, 75–84. [DOI] [PubMed] [Google Scholar]
  7. Burns J & Marks D (2013) Can recovery capital predict Addiction Problem Severity? Alcoholism Treatment Quarterly, 31, 3, 303–320. [Google Scholar]
  8. Cacciola JS, Alterman AI, McLellan AT, Lin Y-T, & Lynch KG (2007). Initial evidence for the reliability and validity of a “Lite” version of the Addiction Severity Index. Drug and Alcohol Dependence, 87, 297–302. doi: 10.1016/j.drugalcdep.2006.09.002 [DOI] [PubMed] [Google Scholar]
  9. Cloud W, & Granfield R (2008). Conceptualizing recovery capital: Expansion of a theoretical construct. Substance Use & Misuse, 43(12–13), 1971–1986. [DOI] [PubMed] [Google Scholar]
  10. 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]
  11. Cohen S, Mermelstein R Kamarck T, & Hoberman H (1985). Measuring the functional components of social support. In Sarason IG & Sarason BR (Eds.), Social support: Theory, research and application (pp. 73–94). The Hague, Holland: Martinus Nijhoff. [Google Scholar]
  12. Cohen S, & Wills TA (1985). Stress, social support, and the buffering hypothesis. Psychological Bulletin, 98, 310–357. doi: 10.1037//0033-2909.98.2.310 [DOI] [PubMed] [Google Scholar]
  13. Cox DR (1972). Regression models and life tables (with discussion). Journal of the Royal Statistical Society, Series B, 74, 187–220. [Google Scholar]
  14. Davis MI, & Jason LA (2005). Sex differences in social support and self-efficacy within a recovery community. American Journal of Community Psychology, 36, 259–274. [DOI] [PubMed] [Google Scholar]
  15. Doogan NJ, Light JM, Stevens EB, & Jason LA (2019). Quality of life as a predictor of social relationships in Oxford House. Journal of Substance Abuse Treatment, 101, 79–87. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Festinger L, Schachter S, & Back K (1968). Operation of Group Standards. In Cartwright D & Zander A (Eds.), Group Dynamics: Research and Theory (3rd ed., pp. 152–164). New York: Harper & Row. [Google Scholar]
  17. Garcia-Rea EA, & LePage JP (2010). Reliability and validity of the World Health Organization quality of life: Brief version (WHOQOL-BREF) in a homeless substance dependent veteran population. Social Indicators Research, 99(2), 333–340. [DOI] [PubMed] [Google Scholar]
  18. Granfield R, & Cloud W (2008). Conceptualizing recovery capital: Expansion of a theoretical construct. Substance Use & Misuse, 43, 1971–1986. [DOI] [PubMed] [Google Scholar]
  19. Greenan CC (2015). Diffusion of innovations in dynamic networks. Journal of the Royal Statistical Society, Series A (Statistics in Society), 178(1), 147–166. [Google Scholar]
  20. Harvey R (2014). Factors relating to substance abuse relapse: A survival analysis of adults living in Oxford House recovery homes. Doctoral Dissertation. DePaul University, Chicago, Il. [Google Scholar]
  21. Hlebec V, & Ferligoj A (2002). Reliability of social network measurement instruments. Field methods, 14(3), 288–306. [Google Scholar]
  22. 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]
  23. Jason LA, Guerrero M, Salomon-Amend M, Lynch G, Stevens EB, Light JN, Stoolmiller M, & Doogan NJ (2021,a). Network measures of advice-seeking and resource sharing are related to well-being in recovery homes. International Journal of Drug Policy. 10.1016/j.drugpo.2020.102970 [DOI] [PMC free article] [PubMed]
  24. Jason LA, Guerrero M, Salomon-Amend M, Lynch G, Stevens EB, Light JN, Stoolmiller M, & Doogan NJ (2021, b). Advice seeking and loaning of money related to relapse in recovery homes. Journal of Community & Applied Social Psychology. 31, 39–52. 10.1002/casp.2486 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Jason LA, Guerrero M, Salomon-Amend M, Stevens E, Light JN & Stoolmiller M (in press). Context matters: Home-level but not individual-level recovery social capital predict residents’ relapse. American Journal of Community Psychology. 10.1002/ajcp.12481 [DOI] [PMC free article] [PubMed]
  26. Jason LA, Light JM, Stevens EB, & Beers K (2014). Dynamic social networks in recovery homes. American Journal of Community Psychology, 53, 324–334. 10.1007/s10464-013-9610-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Jason LA, Olson BD, Ferrari JR, Majer JM, Alvarez J, & Stout J (2007). An examination of main and interactive effects of substance abuse recovery housing on multiple indicators of adjustment. Addiction, 102(7), 1114–1121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Jason LA, & Stevens E (2017). The reliability and reciprocity of a social network measure. Alcoholism Treatment Quarterly, 35, 317–327. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Jason LA, Stevens E, & Ram D (2015). Development of a three‐factor psychological sense of community scale. Journal of Community Psychology, 43(8), 973–985. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Jason LA, Wiedbusch E, Bobak T, & Taullahu D (2020). Estimating the number of substance use disorder recovery homes in the United States. Alcoholism Treatment Quarterly, 38(4), 506–514. 10.1080/07347324.2020.1760756 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Laudet AB, & White W (2010). What are your priorities right now? Identifying service needs across recovery stages to inform service development. Journal of Substance Abuse Treatment, 38(1), 51–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Light JM, Jason LA, Stevens EB, Callahan S, & Stone A (2016). A mathematical framework for the complex system approach to group dynamics: The case of recovery house social integration. Group Dynamics: Theory, Research, and Practice, 20(1), 51–64. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Little RJA, & Rubin DB (1987). Statistical analysis with missing data. New York: Wiley. [Google Scholar]
  34. Lospinoso J, & Snijders TAB (2019) Goodness of fit for stochastic actor-oriented models. Methodological Innovations, 12 (3):205979911988428 [Google Scholar]
  35. Mao X (2007). Stochastic Differential Equations and Applications. Cambridge, UK: Woodhead Publishing. [Google Scholar]
  36. Polcin DL (2009). A model for sober housing during outpatient treatment. Journal of Psychoactive Drugs, 41(2), 153–161. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Porcaro A, Nguyen R, Salomon-Amend M, Chaparro J & Jason LA (2020) Developing a latent coping resources factor for recovery from substance use disorder. Addiction Research & Theory. 29(3), 239–246, 10.1080/16066359.2020.1807959 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Ripley RM, Snijders TAB, Boda Z, Voros A, & Preciado P (2020). Manual for RSiena. Oxford, UK: Department of Statistics, Nuffield College. Available at: http://www.stats.ox.ac.uk/~snijders/siena/RSiena_Manual.pdf [Google Scholar]
  39. Robles RR, Marrero CA, Matos TD, Colon HM, Cancel LI, & Reyes JC (1998). Social and behavioral consequences of chemical dependence. (pp. 355–364). Drug Addiction Research and the Health of Women. National Institute on Drug Abuse; NIH Publication. [Google Scholar]
  40. Rosenberg M (1965). Society and the adolescent self-image. Princeton, NJ: Princeton University Press. [Google Scholar]
  41. SAMHSA (2012). SAMHSA’s working definition of recovery: 10 Guiding Principles of Recovery. Rockville, MD: Substance Abuse and Mental Health Services Administration, Center for Substance Abuse Treatment. Available at: https://store.samhsa.gov/system/files/pep12-recdef.pdf [Google Scholar]
  42. Simpson DD (2004). A conceptual framework for drug treatment process and outcomes. Journal of substance abuse treatment, 27, 99–121. doi: 10.1016/j.jsat.2004.06.001 [DOI] [PubMed] [Google Scholar]
  43. Skaff MM, Finney JW, & Moos RH (1999). Gender differences in problem drinking and depression: Different “vulnerabilities?” American Journal of Community Psychology, 27(1), 25–54. [DOI] [PubMed] [Google Scholar]
  44. Sklar SM, Annis HM, & Turner NE (1999). Group comparisons of coping self-efficacy between alcohol and cocaine abusers seeking treatment. Psychology of Addictive Behaviors, 13, 123–133. doi: 10.1037//0893-164X.13.2.123 [DOI] [Google Scholar]
  45. Snyder CR, Sympson SC, Ybasco FC, Borders TF, Babyak MA, & Higgins RL (1996). Development and validation of the State Hope Scale. Journal of Personality and Social Psychology, 70(2), 321–335. [DOI] [PubMed] [Google Scholar]
  46. Snijders TAB, van de Bunt GG, & Steglich CEG (2010). Introduction to stochastic actor based models for network dynamics. Social Networks, 32, 44–60. 10.1016/j.socnet.2009.02.004 [DOI] [Google Scholar]
  47. Stevens EB, Buchannan B, Ferrari JR, Jason LA, & Ram D (2014). An investigation of hope and context. Journal of Community Psychology, 42(8), 937–946. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Wasserman S, & Faust K (1994). Social Network Analysis: Methods and Applications. Cambridge, UK: Cambridge University Press. [Google Scholar]
  49. Walker ME, Wasserman S, & Wellman B (1993). Statistical models for social support networks. Sociological Methods & Research, 22(1), 71–98. [Google Scholar]
  50. Wölfer R, Faber NS, & Hewstone M (2015). Social network analysis in the science of groups: Cross-sectional and longitudinal applications for studying intra- and intergroup behavior. Group Dynamics: Theory, Research, and Practice, 19, 45–61. [Google Scholar]
  51. World Health Organization Group (1998). The World Health Organization quality of life assessment (WHOQOL): development and general psychometric properties. Social Science & Medicine, 46(12), 1569–1585. [DOI] [PubMed] [Google Scholar]

RESOURCES