Social Network Moderators of Brief Alcohol Intervention Impact

Abstract

Objective:

This investigation examined the impact of social networks on drinking reduction efforts following a brief alcohol intervention. In a reanalysis of data from an earlier randomized controlled trial with nonstudent emerging adult drinkers (Lau-Barraco et al., 2018), we aimed to test three domains of pre-intervention social network features as potential factors influencing intervention response: (1) general network characteristics (i.e., network size, network stability), (2) general network alcohol use (i.e., network alcohol abstainers, network heavy/problem drinkers), and (3) risky peers in network (i.e., proportion of drinking buddies, presence of drinking buddies identified as heavy/problem drinker).

Method:

Participants were 164 emerging adult heavy drinkers recruited from the community (65.9% men; mean age = 21.98 years; 56.2% ethnic minority). Participants were randomly assigned to either a brief personalized feedback intervention or assessment-only control and provided data at 1-month and 3-month follow-ups.

Results:

Greater network stability and greater representation of alcohol abstainers in one’s social network were associated with improved initial post-intervention response. Heavy/problem drinkers in the network did not moderate initial post-intervention effects on drinking outcomes, but there was potentially a stronger intervention effect on risk reduction for those with higher proportions of drinking buddies in their network.

Conclusions:

Study findings provided some evidence that a personalized feedback intervention was efficacious in mitigating some risky social network influence. However, findings did not support a consistent impact across all the network variables examined. Future research is needed to further clarify social network influences and how they may be targeted to enhance intervention efficacy.

Extensive research supports the positive association between an individual’s social network and their own drinking behaviors. This relationship has been demonstrated across population groups (e.g., Barnett et al., 2014; Lau-Barraco & Collins, 2011; Leonard & Mudar, 2000; McCutcheon et al., 2014) and study designs, including cross-sectional (e.g., Mason et al., 2014; Meisel et al., 2018) and prospective (e.g., Lau-Barraco et al., 2012; Reifman & Watson, 2003; Witkiewitz et al., 2017) investigations. From the substance use treatment literature, social networks have been shown to play a key role in treatment response with riskier networks being related to poorer prognosis (Bond et al., 2003; Eddie & Kelly, 2017; Longabaugh et al., 2010). While social networks appear to impact alcohol treatment outcomes, there are surprisingly few investigations examining social network influences on brief intervention outcomes. The present investigation sought to address this gap in the literature by examining the impact of specific social network characteristics (i.e., risky network members, network size, network stability, and alcohol use) on changes in drinking in response to a brief alcohol intervention. The intervention specifically targeted a community-based sample of nonstudent emerging adult drinkers, an understudied group that is at elevated risk for alcohol-related problems and alcohol use disorders (Harford et al., 2006; Hingson et al., 2017; White et al., 2005).

Social Networks and Alcohol Use

Research among adults supporting the relevance of social networks in one’s drinking has largely centered on college student populations but has also been supported in other groups (see Knox et al., 2019 for a review). Among college samples, the influence of social network drinking on one’s own drinking behaviors is evident when considering weekly drinking quantity (Barnett et al., 2014), daily drinking quantity (Phua, 2011), risky drinking (Mason et al., 2014; Meisel et al., 2018), and drinker status (i.e., drinker vs. nondrinker; Balestrieri et al., 2018). Prospective studies support these findings and show that associating with a heavier drinking social network is related to increases in personal drinking from high school to college (Reifman & Watson, 2003) and from the beginning of college to the end of college (Phua, 2011). Among nonstudent emerging adults, drinking habits of their social network members are related to personal alcohol involvement (Lau-Barraco & Collins, 2011). In addition to emerging adult populations, peer or social network drinking and personal drinking have been found to be cross-sectionally and prospectively related in other populations, including general adults (McCutcheon et al., 2014; Rosenquist et al., 2010), married couples (Leonard & Homish, 2008; Leonard & Mudar, 2000), and dependent drinkers (Delucchi et al., 2008; Witkiewitz et al., 2017).

Social Network Features

Studies examining social networks have found associations between aspects of the social network and personal drinking. Several features of the social network that have been shown to relate to drinking are network stability (i.e., connections in one’s network over time), network size (i.e., the number of connections in one’s network), and presence of particular peers (i.e., drinking buddies). Related to network stability, knowing people in their networks longer is related to having a heavier-drinking social network (Reifman et al., 2006). However, it has also been shown that high network turnover (i.e., network instability) during the transition from high school to college is a risk factor for increased alcohol use, particularly when network turnover is the result of heavy drinker additions (Meisel & Barnett, 2017). Similarly, among college drinkers mandated to receive an intervention for violating campus alcohol policies, it was found that those incorporating more heavy drinkers into their social networks over a year also reported higher consumption and greater perceptions of peer drinking, suggesting that less stable networks driven by addition of drinking peers may be a risk factor for heavier alcohol use (DeMartini et al., 2013). While prior findings are somewhat mixed, they do provide evidence that one’s network turnover or network retention is related to alcohol use and, thus, is conceivable that less stable networks could exert impact on the individual’s intervention response.

General network size has been shown to play a role in one’s drinking, such that more peer ties (i.e., more people in one’s network) is associated with greater alcohol use by the individual (Cook et al., 2013; Leonard et al., 2000). The size of a person’s network may indicate the availability of peers or peer connections (Valente et al., 2004) and could be related to an individual’s potential exposure to alcohol use, including opportunities to join social activities where alcohol may be present; thus, larger networks may provide greater exposure to drinking behavior that serve to hinder one’s attempt to reduce their own drinking. However, research on the impact of network size, as well as network stability, on brief alcohol intervention response has been scarce. It may be that larger networks could attenuate the potential benefits of brief interventions on drinking reduction.

Certain social network members appear to have more influence on drinking than others. For instance, the subset of peers that an individual drinks with specifically may be especially impactful on their own drinking. One’s “drinking buddies”, defined as individuals from their network designated as companions for the primary purpose of drinking, are related to increased drinking. A greater presence of network drinking buddies is associated with increased individual drinking (Lau-Barraco & Linden, 2014; Lau-Barraco et al., 2012; Leonard & Homish, 2008; Leonard et al., 2000; Neighbors et al., 2019). The influence of drinking buddies remains, even after considering the impact of general or heavy drinking by peers, which lends support to the unique influence of drinking buddies in the peer-use relationship (Leonard & Homish, 2008; Reifman et al., 2006). Given drinking buddies’ distinct impact on one’s alcohol use, additional research is warranted that may be able to shed light on which type of drinking buddies are most impactful, particularly as it relates to their influence on one’s response to alcohol intervention programming. On the basis of prior research, it is reasonable to expect that this specific subset of peers designated as drinking buddies who are also heavy drinkers may compound risk for an individual and impact their drinking behaviors. However, the extent to which they exert influence on an individual’s response to alcohol intervention programming remains to be examined.

Social Networks and Drinking Behavior Change

While there is evidence to support the benefits of social networks (Bond et al, 2003; Eddie & Kelly, 2017; Groh, 2007; Longabaugh et al., 2010; Mowbray, 2014; Zywiak et al., 2002), aspects of the network could also negatively impact drinking behavior change and treatment response. For example, poorer treatment response has been related to having more heavy drinkers (Bond et al., 2003), more general support for drinking (Zywiak et al., 2002), and greater frequency of drinking (Longabaugh et al., 2010) within the network. Similarly, the strength of pro-drinking influences within the social network is linked with a lower likelihood of abstinence (Kaskutas et al., 2002) and elevated drinking-related consequences (Wu & Witkiewitz, 2008). Among young adults, having more high-risk friends and the amount of time spent with them are negatively related to abstinence following residential substance use treatment (Eddie & Kelly, 2017).

Accrued evidence from the treatment literature has established the impact of social networks on treatment outcomes, and similar connections may be present for non-disordered but risky drinkers undergoing intervention programming. However, research regarding the degree to which networks impact response to brief interventions remains an understudied area of investigation. In studies of brief interventions for drinking, there have only been a few studies that specifically examined social network variables related to brief intervention response. These investigations varied in the populations targeted. One study specifically focused on incarcerated drinkers (Owens & McCrady, 2016), where a brief motivational intervention administered prior to release did not yield benefits over an educational intervention on post-release outcomes; however, medium to large effect sizes suggest potential benefits of the intervention, including reducing proportion of heavy drinkers and drug users in their social network.

Reid and colleagues (2015) examined the influence of two social network factors on college student drinking reduction initiation and maintenance following two alcohol interventions (counselor delivered brief motivational intervention [BMI] vs. computer delivered intervention [CDI]). Perceiving their social network as heavy drinking resulted in an initial reduction of alcohol outcomes, and perceived peer acceptability of decreasing use predicted initial reductions of drinking consequences. Interestingly, having networks less accepting of reducing drinking was associated with a faster rate of decay of intervention effects when examining maintenance of drinking reductions among CDI as compared to BMI participants. Thus, BMI was especially efficacious in maintaining reductions when individuals were in riskier networks (i.e., low peer acceptance for reducing drinking). Overall, these findings suggest that those with riskier social networks may be less likely to initiate drinking reductions. Moreover, the network’s impact on the maintenance of drinking reductions may depend on type of intervention, with only the counselor-delivered BMI approach maintaining reductions for individuals in riskier networks.

Social network moderators of a BMI were also tested by Alvarez and colleagues (2020) in a sample of heavy drinking trauma patients. In particular, the study examined percentage of heavy drinkers and percentage of abstainers in one’s social network as moderators impacting the effectiveness of a BMI, delivered with and without a telephone booster session, as compared to brief advice (BA). Results showed that abstainers in the network did not influence intervention response. However, there was some evidence of significant interactions between network heavy drinking and condition, with active intervention conditions (either BMI or BMI+booster versus BA) showing reduced effectiveness for reductions in binge drinking or at-risk drinking for those with heavy drinkers in their network. These findings may provide preliminary support for the negative impact one’s network may exert on their drinking reduction efforts following a brief intervention.

Together, findings from these brief intervention studies suggest that social networks exert influence over one’s response to intervention efforts and that social network attributes warrant further empirical consideration, especially given the paucity of research in this area. In particular, research could benefit from a closer examination of other aspects or indicators of the social network. In the aforementioned studies, social network factors were specifically constrained to network proportions of heavy drinkers (Alvarez et al, 2020; Owens & McCrady, 2016), proportions of abstainers (Alvarez et al, 2020), or average levels of peer drinking and peer support for drinking (Reid et al., 2015). Other attributes of the social network could also impact intervention response, and identification of these attributes may lead to a greater understanding of social network influence and how it may be targeted to enhance brief intervention efficacy.

Current Study

The overall goal of the present study was to test social network moderators of a personalized feedback intervention (PFI). Data for the current investigation came from a randomized controlled trial of a PFI tailored to nonstudent heavy drinkers between 18–25 years (Lau-Barraco et al., 2018). The brief intervention was found to lead to significant reductions in drinking as compared to controls one month following the intervention. In the present investigation, we sought to extend prior research by testing three groups of pre-intervention social network variables as moderators of brief intervention response: (1) general network characteristics (i.e., network size, network stability), (2) general network alcohol use (i.e., network alcohol abstainers, network heavy/problem drinkers), and (3) risky peers in network (i.e., proportion of drinking buddies, presence of drinking buddies identified as heavy/problem drinkers). It was hypothesized that individuals with greater network size, fewer network alcohol abstainers, and riskier social networks (i.e., high network heavy/problem drinkers, high network drinking buddies) would be less responsive to the intervention, but no firm predictions were made for network stability given mixed findings in prior research.

Method

Participants and Procedure

The sample consisted of 164 participants (65.9% men) with a mean age of 21.98 (SD = 2.02) years. Ethnicity of the sample were 48.3% African Americans, 40.9% Caucasians, 6.7% Hispanics, 1.2% Native American/Indians, and 3.0% “Other.” Participants were predominantly single/never married (71.3%), nonparents (66.7%), and employed (54.3%). Participants were a community-based sample recruited from a midsize, urban southeastern city in the United States through flyers, newspapers, and online advertisements (i.e., posts on Craigslist and Facebook). To meet eligibility criteria, participants were required to be 18 to 25 years-old, report no current or prior college attendance, no current enrollment in high school, and must have engaged in at least two heavy drinking episodes (i.e., 4/5+ standard drinks for women/men) in the past month. Participants were excluded if they reported consuming more than 40 drinks weekly and/or a positive history of substance use treatment. Interested individuals were assessed for study eligibility during a brief telephone screening, and eligible participants were scheduled to attend the in-person baseline/intervention session. All participants provided informed consent.

Participants were randomly assigned to one of two conditions: tailored PFI (n = 90) or assessment-only (AO; n = 74) control. The PFI consisted of personalized feedback regarding the participant’s alcohol use, alcohol-related consequences, gender-specific normative drinking comparisons, personal risk factors (e.g., family history of alcoholism, dependence symptoms), and alcohol expectancies. There was also information tailored to nonstudents related to adaptive ways of coping and managing stress, comorbid substance use, and goal attainment/achievement strivings. The personalized feedback was presented graphically in a feedback report with the individual’s information. The PFI was delivered within a 50- to 60-minute brief motivational intervention session (BMI; Miller and Rollnick, 2002; Walters & Baer, 2006) administered by doctoral-level clinical psychology student interventionists. Participants in the AO condition attended the in-person meeting with an interventionist for the baseline assessment. With the exception of the active intervention component, all study procedures were identical between the study conditions. In the original RCT trial, follow-up assessments were conducted at 1, 3, 6, and 9 months, post-intervention. However, only data from 1-month and 3-month follow ups were utilized in the current study due to a lack of intervention effect at later follow-ups (Lau-Barraco et al., 2018). A full description of the study and procedures is provided in Lau-Barraco et al. (2018). Study procedures were approved by the University’s Institutional Review Board and followed American Psychological Association (2010) guidelines.

Measures

Social Network

Baseline social network characteristics were assessed using a modified version of the Social Network Map (SNM; Tracy & Whitaker, 1990). Using a self-report, written format, participants were asked to list up to 10 individuals (e.g., family members, friends, boyfriend/girlfriend, co-workers, etc.) within their social network with whom they had any form of contact and were considered the most important to them during the past year. Participants then responded to detailed questions regarding each network member. In addition to basic demographic (e.g., gender, age) and relationship characteristics (e.g., length of relationship) of each member, specific questions were asked regarding the member’s alcohol use. For each member, participants indicated (1) his/her general drinking pattern during the past year (non-drinker, light social drinker, moderate social drinker, heavy social drinker, problem drinker), and (2) if he/she is considered a “drinking buddy,” defined as a person whom “you got together with on a regular basis to do activities that centered around drinking, going to bars or nightclubs.”

Based on information from the SNM, indicators of their social network were derived: social network size (i.e., the total number of individuals listed on the SNM), network stability (i.e., the proportion of individuals in each participant’s network whom they reported knowing for less than one year and the proportion they reported knowing for more than 5 years; higher proportions of individuals with a length of relationship of less than one year is considered a less stable network), network alcohol abstainers (i.e., proportion of network members who do not consume any alcohol), network heavy or problematic drinkers (i.e., proportion of network members who were either a heavy social drinker or problem drinker), and network drinking buddies (i.e., proportion of network members identified as a drinking buddy). Finally, we also coded if participants reported at least one drinking buddy in their network who was a heavy social drinker or problem drinker (1 = yes, 0 = no).

Alcohol Use

The Daily Drinking Questionnaire (DDQ; Collins et al., 1985) assessed participants’ typical weekly alcohol use at baseline and at 1- and 3-month follow-ups. Participants reported the typical number of standard drinks (e.g., 12-ounces of beer, 5-ounces of wine, or 1.5-ounces of liquor) they consume on each day of the typical week, averaged over the last three months. Weekly alcohol quantity was calculated as the typical total number of drinks per week (i.e., summed across days on the DDQ). Drinking frequency was calculated as the total number of days alcohol consumption was reported for a typical week. High versus low risk was calculated using suggested guidelines for excessive drinking (HHS & USDA, 2020). Women reporting less than or equal to 3 drinks per day and less than or equal to 7 drinks per week were coded as low risk (0). Women reporting more than 3 drinks per day and/or more than 7 drinks per week were coded as high risk (1). Men reporting less than or equal to 4 drinks per day and less than or equal to 14 drinks per week were coded as low risk (0). Men reporting more than 4 drinks per day and/or more than 14 drinks per week were coded as high risk (1).

Alcohol-related Consequences

The Brief Young Adult Alcohol Consequences Questionnaire (BYAACQ; Kahler et al., 2005) assessed participants’ alcohol-related consequences at baseline and 1- and 3-month follow-ups. Participants indicated whether each of the 24 items described something that happened to them in the past month (e.g., “’While drinking, I have said or done embarrassing things.”). Response options were “yes” (1) or “no” (0). A total score was calculated by summing endorsed items.

Demographics

Participants provided information on basic demographic characteristics including age, gender, ethnicity, employment, income, parent status, and relationship status.

Analysis Approach

To address the overall study aim (i.e., to examine social network moderators of brief intervention impact in a sample of nonstudent emerging adult heavy drinkers), piecewise latent growth models were conducted, using a structural equation modeling approach. To focus on initial intervention response and short-term maintenance, data were limited to growth from baseline through the 3-month follow-up only. The piecewise latent growth model specified an intercept and two slopes with loadings for Slope 1 specified as 0 for baseline data, 1 for 1-month data, and 1 for 3-month data to represent growth from baseline to 1 month (capturing immediate post-intervention change); loadings for Slope 2 were specified as 0, 0, 1 for baseline, 1-month, and 3-month data respectively to represent growth from 1 month to 3 months (capturing short-term maintenance). Each latent factor was regressed onto condition (dummy coded as 0 = AO control, 1 = PFI), the social network moderator (mean centered), and the interaction between the two. Each social network moderator was examined in a separate model, as was each alcohol outcome. Sex was controlled for at baseline in all analyses. Maximum likelihood estimation was used in models for the outcomes of weekly alcohol quantity, frequency of drinking, and alcohol-related problems, which is robust to minor non-normality. However, a robust weighted least squares estimator was used for models with alcohol risk as the outcome due to its binary nature, reflecting a probit model. To facilitate model convergence, correlations among slope latent factors were fixed to zero for the outcomes of frequency of drinking and low-risk versus high-risk status. Effect sizes for significant findings were calculated two ways: 1) by dividing the group difference in post-intervention growth (unstandardized b) by the pooled standard deviation for within-group growth (variability of the latent slope; more commonly reported for latent growth models), or 2) dividing that same group difference by the pooled standard deviation for the outcome at baseline (more akin to a traditional Cohen’s d; Feingold, 2009).

Results

An examination of the data revealed no outliers for alcohol-related problems or drinking frequency. Four outliers for alcohol quantity at baseline, eight at 1-month, and four outliers at 3-months were winsorized. Normality was confirmed for weekly alcohol quantity and drinking frequency. A natural log transformation was used for alcohol-related problems (natural log of the original metric + 1) due to positive skew.

At baseline, participants reported consuming an average of 24.16 (SD = 21.49) drinks per week, drinking 4.38 (SD = 1.89) days per week, and an average of 7.97 (SD = 5.32) alcohol-related problems. At the 1-month follow-up, they reported consuming an average of 15.73 (SD = 15.13) drinks per week, drinking 3.79 (SD = 2.33) days per week, and an average of 6.72 (SD = 5.89) alcohol-related problems. At the 3-month follow-up, they reported consuming an average of 13.96 (SD = 14.59) drinks per week, drinking 3.52 (SD = 2.56) days per week, and an average of 6.09 (SD = 6.01) alcohol-related problems. The number of participants considered high-risk at baseline was n = 130 (79.3%), whereas low-risk was n = 34 (20.7%). At the 1-month follow-up, n = 77 (60.6%) participants were high-risk, and n = 50 (39.4%) were considered low-risk. At the 3-month follow-up, n = 56 (56.6%) participants were high-risk, and n = 43 (43.4%) were considered low-risk. As seen in Table 1, participants reported an average of 7 important people in their social networks, with less than 10% of those network members being relatively new contacts (known less than one year) and almost 50% of network members being long established (known more than five years). On average, 25% of network members abstained from alcohol use, but 28% were heavy social or problem drinkers. Almost half (48%) of network members were drinking buddies. Among those reporting at least one drinking buddy in their network, almost one-third of the sample (33%) reported having no drinking buddies that were heavy or problem drinkers, but two-thirds of the sample (67%) reported having at least one drinking buddy who was a heavy or problem drinker. Among network members identified as a heavy social or problem drinker, 76.25% were designated as drinking buddies. For moderate, light and nondrinking network members, 63.7%, 33.3%, and 5.4% were identified as drinking buddies, respectively. An examination of the association between baseline network characteristics and alcohol use variables revealed significant associations between drinking quantity and proportion of alcohol abstainers, heavy/problem drinkers, drinking buddies, and heavy/problem drinking buddy. Drinking frequency was related to proportion of drinking buddies. Problems was associated with proportion of heavy/problem drinkers and heavy/problem drinking buddy. Lastly, risk level was related to proportion of heavy/problem drinkers, drinking buddies, and heavy/problem drinking buddy. See Table 2.

Table 1.

Social Network Characteristics at Baseline (Pre-Intervention)

Social Network Information	M	SE
General Network Characteristics
Size of network	7.20	0.23
Stability of network
< 1 year (proportion)	0.09	0.03
> 5 years known (proportion)	0.48	0.02
General Network Substance Use
Proportion alcohol abstainers in network	0.25	0.03
Proportion heavy/problem drinkers in network	0.28	0.03
Particularly risky peers in network (i.e., drinking buddies)
Proportion drinking buddies in the network	0.48	0.02
Presence of at least one drinking buddy in network	n	%
No drinking buddies	19	11.7%
1+ drinking buddies	143	88.3%
Presence of at least one heavy drinking (HD) drinking buddy in network	n	%
No HD drinking buddies	46	32.6%
1+ HD drinking buddies	95	67.5%

Table 2.

Bivariate Correlations for Study Variables

Variable	1	2	3	4	5	6	7	8	9	10
1. Quantity	—
2. Frequency	.58 *	—
3. Problems	.35 *	.32 *	—
4. Risk	.43 *	.33 *	.27 *	—
5. Network size	−.07	−.04	.09	.09	—
6. Low network stability	−.08	−.09	−.07	.01	−.00	—
7. Proportion of alcohol abstainers	−.22 *	−.05	−.05	−.09	.05	−.04	—
8. Proportion of heavy/problem drinkers	.34 *	.13	.21 *	.25 *	−.02	.01	−.33 *	—
9. Proportion of drinking buddies	.22 *	.21 *	.13	.18 *	−.06	−.06	−.44 *	.36 *	—
10. Presence of at least one heavy/problem drinking buddy	.23 *	.11	.21 *	.20 *	.11	−.04	−.19 *	.67 *	.18 *	—

Note. Bold text indicates significant correlations.

^*p < .05

Results of the latent growth models are presented in Table 3. For each network model, effects of predictors are listed for the latent variable reflecting baseline values, for latent Slope 1 reflecting growth from baseline to the 1-month assessment, and for latent Slope 2 reflecting growth from 1-month to 3-months. For each latent factor, the intercept reflects average outcome values for the AO control condition, having average values of the network characteristic listed. The effects of the predictors then reflect how those values change for those in the PFI group, those with more of the network characteristic, and the interaction between the two. For example, for quantity as the outcome and low network stability as the moderator, the intercept at baseline reflects that average consumption for members of the AO control condition who have known 8.7% of their network members less than a year (the average for the sample) is about 26 drinks per week. This is not significantly impacted by condition (a difference of 2.3 drinks), having more of their network be members they have known for less than a year, or the interaction between the two. The Slope 1 factor, reflecting the average change in consumption at month 1 for members of the AO control condition who have known 8.7% of their network members for less than a year, decreases by about 6.6 drinks per week, which is significantly different from zero. Growth to 1-month is not significantly impacted by condition (a non-significant further reduction of 1.7 drinks), having more of their network be members they have known for less than a year, or the interaction between the two. Finally, the Slope 2 factor, reflecting the average change in consumption from month 1 to month 3 for members of the AO control condition who have known 8.7% of their network members for less than a year, decreases by about 1.2 drinks per week, which is not significantly different from zero. Growth to the 3-month assessment is not significantly impacted by condition, having more of their network be members they have known for less than a year, or the interaction between the two.

Table 3.

Moderation of Network Characteristics of Intervention Effects on Alcohol Outcomes

	Quantity			Frequency			Problems			Risk
Moderator	B (SE)	β	p	B (SE)	β	p	B (SE)	β	p	B (SE)	β	p
Network Size
Baseline
Intercept	26.30* (2.56)	1.229	<.001	4.86* (0.24)	3.101	<.001	1.68* (0.29)	2.423	<.001	--	--	--
Condition	−2.30 (3.33)	−0.054	.490	−0.68* (0.30)	−0.217	.020	0.01 (0.11)	0.006	.933	0.21 (0.24)	0.105	.378
Network Quality	0.06 (0.88)	0.007	.949	−0.05 (0.08)	−0.077	.564	0.01 (0.03)	0.032	.773	−0.06 (0.06)	0.157	.309
Condition X Network	−1.22 (1.23)	−0.110	.319	0.03 (0.11)	0.032	.808	0.05 (0.04)	0.128	.248	−0.02 (0.09)	−0.046	.798
Slope 1: Growth from Baseline to 1 Month
Intercept	−6.55* (1.78)	−0.464	<.001	−0.41 (0.26)	−0.341	.121	−0.40* (0.15)	−0.457	.009	−0.24 (0.19)	−0.702	.198
Condition	−1.73 (2.37)	−0.061	.465	−0.15 (0.34)	−0.062	.667	−0.06 (0.16)	−0.031	.726	−0.61 (0.24)	−0.891	.009
Network Quality	−0.13 (0.65)	−0.025	.840	0.13 (0.10)	0.285	.191	0.03 (0.04)	0.086	.519	−0.08 (0.06)	−0.602	.236
Condition X Network	0.93 (0.88)	0.128	.290	−0.11 (0.13)	−0.185	.378	−0.03 (0.06)	−0.068	.595	0.043 (0.10)	0.203	.729
Slope 2: Growth from 1 Month to 3 Months
Intercept	−1.18 (2.09)	−0.083	.571	−0.01 (0.33)	−0.005	.984	−0.15 (0.88)	−2.732	.862	0.15 (0.35)	1.049	.663
Condition	−2.63 (2.76)	−0.092	.341	−0.53 (0.43)	−0.178	.218	−0.04 (0.91)	−0.330	.968	−0.13 (0.41)	−0.435	.757
Network Quality	−0.23 (0.76)	−0.044	.759	−0.12 (0.12)	−0.211	.328	−0.02 (025)	−0.893	.941	0.03 (0.10)	0.496	.794
Condition X Network	0.53 (1.04)	0.072	.611	0.06 (0.16)	0.077	.713	−0.002 (0.33)	−0.082	.994	0.04 (0.15)	0.465	.814
Low Network Stability
Baseline
Intercept	26.04* (2.53)	1.230	<.001	4.84* (0.23)	3.094	<.001	1.66* (0.11)	2.412	<.001	--	--	--
Condition	−2.03 (3.31)	−0.048	.540	−0.71* (0.29)	−0.225	.016	0.01 (0.11)	0.005	.948	0.18 (0.23)	0.095	.433
Network Quality	25.41 (14.57)	0.203	.081	1.77 (1.29)	0.192	.171	−0.49 (0.48)	−0.121	.309	1.17 (1.66)	0.205	.483
Condition X Network	−31.74 (19.54)	−0.189	.104	−3.12 (1.73)	−0.252	.072	0.26 (0.65)	0.047	.694	−0.99 (1.93)	−0.130	.609
Slope 1: Growth from Baseline to 1 Month
Intercept	−6.56* (1.78)	−0.462	<.001	−0.36 (0.26)	−0.300	.160	−0.36* (0.16)	−0.409	.019	−0.25 (0.17)	−0.749	.129
Condition	−1.73 (2.38)	−0.061	.466	−0.23 (0.34)	−0.095	.498	−0.09 (0.16)	−0.048	.592	−0.63* (0.22)	−0.929	.004
Network Quality	−11.35 (11.35)	−0.136	.317	−1.07 (1.67)	−0.153	.520	0.71 (0.81)	0.135	.384	0.28 (1.14)	0.138	.809
Condition X Network	26.18 (14.50)	0.233	.071	5.00* (2.09)	0.529	.016	−0.75 (1.00)	−0.107	.453	0.69 (1.25)	0.259	.580
Slope 2: Growth from 1 Month to 3 Months
Intercept	−0.50 (2.00)	−0.036	.801	0.010 (0.32)	0.007	.976	−0.17 (0.89)	−1.972	.847	0.20 (0.35)	0.506	.564
Condition	−3.25 (2.64)	−0.116	.218	−0.60 (0.43)	−0.204	.159	−0.02 (0.92)	−0.103	.984	−0.10 (0.42)	−0.120	.819
Network Quality	−34.42* (11.74)	−0.417	.003	−3.20 (1.90)	−0.371	.091	0.29 (3.76)	0.573	.938	−1.89 (2.36)	−0.799	.421
Condition X Network	41.79* (15.05)	0.378	.005	4.54 (2.42)	0.392	.060	0.34 (4.98)	0.492	.946	4.56 (3.45)	1.433	.186
Proportion of Alcohol Abstainers
Baseline
Intercept	26.41* (2.51)	1.234	<.001	4.87* (0.23)	3.069	<.001	1.68* (0.11)	2.423	<.001	--	--	--
Condition	−2.60 (3.27)	−0.060	.427	−0.69* (0.29)	−0.216	.019	0.01 (0.11)	0.003	.966	0.24 (0.23)	0.122	.295
Network Quality	−19.96* (9.07)	−0.250	.028	−1.21 (0.81)	−0.204	.137	−0.004 (0.31)	−0.001	.990	0.79 (0.80)	0.214	.322
Condition X Network	5.16 (12.19)	0.048	.672	1.60 (1.09)	0.200	.144	0.11 (0.41)	0.032	.783	−2.26* (0.94)	−0.453	.017
Slope 1: Growth from Baseline to 1 Month
Intercept	−6.82* (1.74)	−0.486	<.001	−0.38 (0.26)	−0.310	.141	−0.38* (0.15)	−0.426	.014	−0.27 (0.19)	−0.726	.161
Condition	−1.49 (2.31)	0.053	.520	−0.19 (0.34)	−0.078	.573	−0.07 (0.16)	−0.042	.636	−0.64* (0.24)	−0.862	.008
Network Quality	15.54* (6.30)	0.297	.014	1.25 (0.91)	0.275	.171	−0.64 (0.42)	−0.196	.128	−0.76 (0.84)	−0.552	.369
Condition X Network	−15.29 (8.58)	−0.217	.075	−2.60* (1.25)	−0.425	.038	0.81 (0.58)	0.183	.160	1.41 (1.06)	0.766	.182
Slope 2: Growth from 1 Month to 3 Months
Intercept	−1.11 (2.05)	−0.078	.588	−0.02 (0.33)	−0.012	.956	−0.18 (0.90)	−1.507	.839	0.20 (0.38)	0.264	.597
Condition	−2.66 (2.71)	−0.092	.326	−0.52 (0.43)	−0.171	.226	−0.02 (0.91)	−0.091	.981	−0.32 (0.50)	−0.207	.524
Network Quality	−14.27 (7.78)	−0.267	.067	−1.60 (1.24)	−0.285	.195	0.55 (2.68)	1.221	.837	−2.22 (1.39)	−0.785	.111
Condition X Network	19.59 (10.21)	0.272	.055	1.75 (1.61)	0.231	.278	−0.90 (3.37)	−1.474	.791	−0.95 (2.23)	−0.250	.669
Proportion of Heavy/Problem Drinkers
Baseline
Intercept	26.66* (2.38)	1.280	<.001	4.90* (0.23)	3.127	<.001	1.70* (0.11)	2.438	<.001	--	--	--
Condition	−3.79 (3.10)	−0.091	.222	−0.84* (0.29)	−0.267	.004	−0.003 (0.11)	−0.002	.982	0.19 (0.26)	0.088	.452
Network Quality	26.86* (8.37)	0.360	.001	1.05 (0.78)	0.186	.180	0.36 (0.31)	0.145	.237	1.52* (0.65)	0.386	.019
Condition X Network	−3.75 (11.13)	−0.038	.736	−0.26 (1.04)	−0.035	.800	−0.14 (0.40)	−0.041	.731	0.21 (1.14)	0.040	.854
Slope 1: Growth from Baseline to 1 Month
Intercept	−6.95* (1.76)	−0.491	<.001	−0.36 (0.25)	−0.318	.151	−0.33* (0.16)	−0.371	.039	−0.26 (0.21)	−0.562	.203
Condition	−1.01 (2.34)	−0.036	.666	−0.10 (0.33)	−0.043	.768	−0.10 (0.16)	−0.056	.529	−0.67* (0.27)	−0.717	.012
Network Quality	−13.52* (6.56)	−0.267	.039	0.94 (0.94)	0.232	.318	0.12 (0.46)	0.037	.796	0.34 (0.67)	0.204	.615
Condition X Network	2.23 (8.79)	0.033	.800	−0.55 (1.27)	−0.103	.662	0.46 (0.60)	0.110	.442	−1.82 (1.16)	−0.825	.117
Slope 2: Growth from 1 Month to 3 Months
Intercept	−1.17 (2.09)	−0.083	.576	−0.060	−0.038	.856	−0.23 (0.90)	−0.964	.796	0.23 (0.46)	0.425	.620
Condition	−2.16 (2.77)	−0.077	.434	−0.52 (0.44)	−0.164	.233	−0.03 (0.92)	−0.070	.971	−0.20 (0.54)	−0.185	.713
Network Quality	1.42 (7.58)	0.028	.852	−0.64 (1.20)	−0.113	.595	−0.43 (2.49)	−0.495	.864	0.53 (1.34)	0.277	.694
Condition X Network	−3.44 (10.32)	−0.051	.739	−0.70 (1.63)	−0.093	.668	−0.65 (3.34)	−0.568	.846	1.94 (1.64)	0.763	.236
Proportion of Drinking Buddies
Baseline
Intercept	26.15* (2.55)	1.215	<.001	4.78* (0.23)	3.04	<.001	1.73* (0.11)	2.539	<.001	--	--	--
Condition	−2.05 (3.31)	−0.047	.536	−0.64* (0.29)	−0.201	.030	−0.03 (0.11)	−0.019	.811	0.44 (0.29)	0.182	.136
Network Quality	20.19* (9.21)	0.272	.028	1.53 (0.82)	0.281	.061	0.45 (0.30)	0.190	.140	−0.33 (0.73)	−0.080	.651
Condition X Network	−7.42 (11.69)	−0.079	.526	−0.34 (1.03)	−0.049	.745	−0.42 (0.38)	−0.141	.266	2.88* (1.14)	0.546	.011
Slope 1: Growth from Baseline to 1 Month
Intercept	−6.57* (1.76)	−0.465	<.001	−0.36 (0.26)	−0.302	.174	−0.41* (0.15)	−0.469	.007	−0.25 (0.21)	−0.376	.243
Condition	−1.70 (2.36)	−0.060	.471	−0.19 (0.35)	−0.080	.586	−0.06 (0.16)	−0.034	.699	−0.84* (0.32)	−0.638	.009
Network Quality	−11.27 (6.68)	−0.23	.092	0.03 (0.99)	0.008	.973	−0.37 (0.44)	−0.121	.407	0.41 (0.90)	0.182	.649
Condition X Network	5.53 (8.35)	0.089	.508	−0.52 (1.23)	−0.101	.672	0.44 (0.55)	0.114	.422	−2.62* (1.26)	−0.915	.037
Slope 2: Growth from 1 Month to 3 Months
Intercept	−1.25 (2.10)	−0.087	.552	−0.04 (0.33)	−0.207	.905	−0.15 (0.88)	−2.208	.864	0.21 (0.40)	0.469	.612
Condition	−2.31 (2.79)	−0.080	.408	−0.45 (0.43)	−0.155	.297	−0.02 (0.91)	−0.140	.983	−0.10 (0.48)	−0.113	.836
Network Quality	2.95 (7.91)	0.060	.710	0.33 (1.22)	0.066	.788	0.26 (2.64)	1.081	.923	1.07 (1.58)	0.710	.499
Condition X Network	−0.72 (9.89)	−0.011	.942	−0.30 (1.53)	−0.047	.846	−0.48 (3.19)	−1.580	.882	0.63 (1.90)	0.329	.740
Presence of at least one Heavy/Problem Drinking Buddy
Baseline
Intercept	27.75* (2.68)	1.288	<.001	4.90* (0.25)	2.964	<.001	1.77* (0.12)	2.614	<.001	--	--	--
Condition	−3.00 (3.55)	−0.069	.399	−0.73* (0.31)	−0.219	.020	−0.06 (0.11)	−0.043	.612	0.45 (0.27)	0.204	.093
Network Quality	12.09* (5.42)	0.263	.026	0.59 (0.48)	0.167	.216	0.09 (0.18)	0.062	.612	0.73* (0.36)	0.306	.044
Condition X Network	−4.89 (7.55)	−0.076	.517	−0.38 (0.66)	−0.077	.568	0.12 (0.24)	0.059	.624	−0.51 (0.55)	−0.154	.354
Slope 1: Growth from Baseline to 1 Month
Intercept	−7.89* (3.12)	−0.522	<.001	−0.33 (0.24)	−0.293	.171	−0.42* (0.17)	−0.485	.012	−0.17 (0.24)	−0.382	.484
Condition	−1.48 (2.55)	−0.051	.563	−0.26 (0.33)	−0.113	.432	−0.04 (0.17)	−0.025	.798	−0.86* (0.26)	−0.996	.002
Network Quality	−4.28 (3.93)	−0.140	.276	0.73 (0.51)	0.303	.150	−0.28 (0.26)	−0.151	.273	0.14 (0.19)	0.152	.781
Condition X Network	−0.62 (5.43)	−0.015	.908	−0.53 (0.70)	−0.156	.451	0.27 (0.35)	0.103	.447	−0.19 (0.66)	−0.144	.781
Slope 2: Growth from 1 Month to 3 Months
Intercept	−0.37 (2.02)	−0.029	.854	0.07 (0.32)	0.047	.824	−0.20 (0.96)	−2.441	.836	0.18 (0.44)	0.546	.683
Condition	−2.35 (2.75)	−0.091	.393	−0.50 (0.44)	−0.164	.252	−0.08 (0.98)	−0.466	.938	−0.05 (0.58)	−0.073	.934
Network Quality	−1.56 (4.32)	−0.057	.718	−0.43 (0.69)	−0.133	.533	−0.20 (1.58)	−1.184	.897	0.35 (0.83)	0.502	.671
Condition X Network	3.68 (5.90)	0.096	.533	0.33 (0.94)	0.073	.726	0.14 (2.06)	0.585	.945	0.57 (1.13)	0.577	.618

Note. SE = standard error, quantity = weekly alcohol quantity, frequency = number of weekly drinking days, problems = alcohol-related problems, risk = gender-based cutoffs, with more than 3/4 drinks per day and/or 7/14 drinks per week for women/men being scored as high risk, and drinking below those criteria being scored as low risk. Low network stability reflects proportion of network members the participant has known for less than one year. Condition was coded as 0 = assessment-only control, 1 = personalized feedback intervention. Bold text indicates significant effects.

^*p < .05

The overall aim of the current project was to examine social network moderators of brief intervention impact in a sample of nonstudent emerging adult heavy drinkers. As such, we focused on the Slope 1 latent factor (immediate change in drinking), and the interaction term between condition and network characteristic (i.e., if the social network quality moderates the effect of intervention condition on growth). Significant interactions would indicate if the social network quality moderates the initial intervention effect (Slope 1) or short-term maintenance (Slope 2). Focusing on the interaction term’s impact on Slope 2, the only significant moderation effect was for low network stability on alcohol quantity, where the reduction in drinking for those with greater proportions of their network they have known for less than one year is buffered for those in the PFI group. Simple slopes revealed that for those with higher proportions of new network members (25%)/less stable networks, the effect of condition on reduction in drinking quantity at three months was not significant, b = 3.58, p = .310, whereas for those with no new network members/more stable networks, the effect of condition on reduction in drinking quantity at three months was significant, b = −6.87, p = .022. No other moderation was significant for short-term maintenance from 1 month to 3 months.

Focusing on the interaction term’s impact on immediate post-intervention change (Slope 1), we saw significant moderation for network stability on frequency, proportion of alcohol abstainers on frequency, and proportion of drinking buddies on risk. No other moderation was significant for initial post-intervention change.

Significant moderation of post-intervention change (Slope 1) is shown in Figures 1–3. Simple slopes revealed that for those with higher proportions of new network members (25%)/less stable networks, the effect of condition on drinking frequency reduction at one month was not significant, b = 0.59, p = .221, whereas for those with no new network members/more stable networks, the effect of condition on frequency reductions at one month approached significance, b = −0.66, p = .082. Figure 1 shows drinking frequency change from baseline to one month, broken down by condition and proportion of new network members. The dashed lines (those with less stable networks) demonstrate a minor reduction in frequency for those in the AO control condition, simple b = −0.53, p = .162, but a minor increase for those in the PFI condition, simple b = 0.06, p = .850, both non-significant, suggesting a weak intervention effect among those with less stable network. The solid lines (more stable networks) indicate both the AO control condition (black line) and PFI condition (grey line) reduce their frequency of drinking. However, this reduction is very minor and non-significant for those in the AO control condition, simple b = −0.26, p = .357, and stronger and significant for those in the PFI condition, simple b = −0.93, p < .001, indicating stronger intervention effects for those with relatively stable networks (no new members).

Figure 1: Frequency of Drinking by Condition and Stability of Social Network.

Note. AO = assessment only, PFI = personalized feedback intervention, new members = network members the participant has known less than one year. Stability of social network is reflected as no new members (0%) or high proportion of new members (25%), based on a mean of 9% and standard deviation of 17% for proportion of members within social network the participant has known for less than a year. Effect size of interaction was calculated using both variability of growth slope (d = 4.334, 95% CI [0.791, 7.876]) and variability of drinking frequency at baseline (d = 2.692, 95% CI [0.491, 4.892]).

Figure 3: Risk Category Membership by Condition and Proportion of Drinking Buddies in Social Network.

Note. AO = assessment only, PFI = personalized feedback intervention, DBs = drinking buddies. Proportion of drinking buddies in social network is reflected as low proportion (25%) or high proportion (75%), based on a mean of 48% and standard deviation of 29% for proportion of drinking buddies within social network. Effect size of interaction was calculated using both variability of growth slope (d = −3.163, 95% CI [−6.136, −0.190]) and variability of risk scores at baseline (d = −6.484, 95% CI [−12.580, −0.389]).

Simple slopes revealed that for those with no alcohol abstinent social network members, the effect of condition on drinking frequency reduction at one month was not significant, b = 0.46, p = .320, whereas for those with high proportions of alcohol abstinent network members (50%), the effect of condition on frequency reductions at one month approached significance, b = −0.84, p = .068. Figure 2 shows drinking frequency change from baseline to one month, broken down by condition and proportion of alcohol abstinent social network members. The solid lines (no alcohol abstinent social network members) demonstrate a minor non-significant reduction in frequency for the PFI condition (grey line), simple b = −0.23, p = .446, and a slightly stronger reduction for the AO control condition (black line) with borderline significance, simple b = −0.69, p = .050, suggesting a weak intervention effect among those with less stable network. However, the dashed lines (high proportions of alcohol abstinent network members) indicate no change in frequency of drinking for those in the AO control condition (black line), simple b = −0.07, p = .843, and a relatively strong significant reduction for those in the PFI condition (grey line), simple b = −0.90, p = .004. This indicates a stronger intervention effect for those with higher proportions of network members who do not consume alcohol.

Figure 2: Frequency of Drinking by Condition and Alcohol Abstinence of Social Network.

Note. AO = assessment only, PFI = personalized feedback intervention, SN = social network, abstainers = network members who do not consume alcohol. Alcohol abstinence of social network is reflected as low proportion abstinent (0%) or high proportion abstinent (50%), based on a mean of 22% and standard deviation of 26% for proportion of alcohol abstinence within social network. Effect size of interaction was calculated using both variability of growth slope (d = −2.251, 95% CI [−4.379, −0.124]) and variability of drinking frequency at baseline (d = −1.398, 95% CI [−2.720, −0.077]).

Simple slopes revealed that for those with lower proportions of drinking buddies (25%), the effect of condition on reductions in risky drinking at one month was not significant, b = 0.22, p = .556, whereas for those with high proportions of drinking buddies (75%), the effect of condition on reductions in risky drinking at one month was significant, b = −1.54, p = .003. Figure 3 shows change in likelihood of risky drinking from baseline to one month, broken down by condition and proportion of social network members who are drinking buddies. The solid lines (lower proportions of drinking buddies) demonstrate a similar reduction in risky drinking for the PFI condition (grey line), simple b = −0.57, p = .009, and the AO control condition (black line), simple b = −0.34, p = .316. The dashed lines (high proportions of drinking buddies) indicate a weak non-significant reduction in risky drinking for those in the AO control condition (black line), simple b = −0.14, p = .609, and a much stronger significant reduction for those in the PFI condition (grey line), simple b = −1.67, p < .001. This indicates a stronger intervention effect for those with higher proportions of network members who are drinking buddies. It is worth noting, however, that the end risk level is similar across conditions at 1-month, and this stronger reduction in risk reflects starting at a much higher risk level at baseline for those with more drinking buddies in their network.

Discussion

It is well-established that social networks and social ties are influential in the drinking of emerging adults. However, research into whether social network characteristics moderate intervention impact in brief intervention remains relatively scarce. In the present study, we sought to understand the impact of one’s social network on drinking behavior change following a brief alcohol intervention. In a reanalysis of data from an earlier randomized controlled trial with nonstudent emerging adult drinkers (Lau-Barraco et al., 2018), we aimed to test three domains of social network features that have been supported or postulated to relate to drinking as potential factors influencing intervention response. These social network moderators included general network characteristics (i.e., network size, network stability), network composition of alcohol user (i.e., percentage of abstainers and heavy/problem drinkers), and risky drinking peers in network (i.e., drinking buddies). We anticipated that greater network size, fewer network alcohol abstainers, and riskier social networks (i.e., high network heavy/problem drinkers, high network drinking buddies) would negatively impact intervention response while direction of impact for network stability was not hypothesized. Overall, we found support for the influence of several social network factors on behavior change from baseline to 1-month post-intervention, particularly in reducing drinking frequency and achieving low-risk drinking. However, we did not find evidence to support network size, network heavy/problem drinker composition, or presence of at least one heavy/problem drinking buddy to moderate intervention impact on any drinking outcomes for immediate post-intervention change or short-term maintenance.

With regard to immediate post-intervention change for general network characteristics, a significant interaction emerged for network stability. Having a more stable network, measured by the percentage of network members known for one year or more, was related to stronger intervention effects on frequency of drinking at the 1-month follow-up. A more stable network was also related to stronger maintenance of intervention effects on quantity of drinking at the 3-month follow-up. One reason for these findings could be that stable networks reflect relationships with family/relatives and household members among nonstudent young adults (57% family/household vs. 33% peers; Lau-Barraco & Collins, 2011) rather than primarily peers, as has been found in college student social networks (9% parents vs. 75% peers; Meisel & Barnett, 2017). Nonstudents have not gone through the transition from high school to college, which is a period linked to greater social network turnover (more/new friendships replacing parent/family members) that is consequently related to drinking risk (Meisel & Barnett, 2017). The inclusion of additional heavy drinking peers in one’s network over time is also linked to greater consumption in college drinkers (DeMartini et al., 2013; Meisel & Barnett, 2017). Having more family members within the social network serves as a protective function that has been related to lower heavy drinking and problematic drinking (Reifman et al., 2006). In fact, from a prevention perspective, involvement of parents and sustained parenting have been found to be beneficial in reducing alcohol misuse among young adults in their transition to college (Turrisi & Ray, 2010; Turrisi et al., 2010). For nonstudents, it is likely that the retention of family member ties and potentially less network turnover (reflecting social stability) during this developmental period may help mitigate risks. More stable social ties may reflect close relationships that provide support to promote behavior change following the brief intervention. Social network support is associated with lower relapse risk and less drinking in treatment seeking populations (Bond et al., 2003; Groh et al., 2007; Kaskutas et al., 2002; Litt et al., 2016). Consistent with this notion, our study findings support the benefits of stable social network ties and their implications for those receiving brief interventions for their drinking.

Greater representation of alcohol abstainers in one’s social network at pre-intervention was associated with improved immediate intervention response. In particular, we found that PFI participants with a greater proportion of alcohol abstainers in their networks drank less frequently compared to controls. Abstainers’ impact on other drinking outcomes, including drinking quantity and reduction of risk status was marginally significant (p = .073-.075). Our findings are in contrast to those from Alvarez and colleagues (2020) showing that percent abstainers in their network did not moderate effects of a BMI. However, our findings are congruent with the literature on treatments with alcohol outpatient treatment-seekers (Zywiak et al., 2002) and those involved with twelve step facilitation groups (Bond et al., 2003). Akin to the explanations supported by the alcohol treatment literature, the protective impact of abstainers in one’s social network for brief interventions may be attributable to greater opportunities for or encouragement of alcohol-free activities and less exposure to alcohol-related cues or prompts for drinking (Fox et al, 2007; Kelly et al., 2011; Litt et al., 2009).

Interestingly, proportion of heavy/problem drinkers in the network at pre-intervention did not moderate drinking outcomes at immediate post-intervention. This is in contrast to prior research showing that having one or more heavy drinkers in the network was related to poorer response to a BMI in trauma patients (Alvarez et al., 2020). One potential explanation for the lack of a moderated effect could be that our intervention, a personalized feedback intervention that included correcting misperceived peer drinking norms, may have sufficiently suppressed the influence of these heavy drinkers in their social network. In other words, while heavy drinking peers exert influence on personal drinking (Rosenquist et al., 2010), likely through informing drinking norms (Reid & Carey, 2018), the feedback intervention was successful in changing these normative perceptions or attitudes that weakened the potential influence of their heavy drinking network members. Indeed, one active ingredient and hypothesized mechanism of behavior change of the current feedback intervention is norms correction (Lau-Barraco et al., 2018). We found that the intervention reduced drinking indirectly through correcting misperceived drinking norms, such that those in the PFI condition significantly reduced their perceptions of how much and how frequently their close friends drink, and this in turn, accounted for decreased drinking at follow-up. Given these findings, one speculation may be that the drinking norms correction mitigated any potential impact the heavy/problem drinking network members could have exerted on the individual’s drinking following the brief intervention. In other words, the fact that the intervention worked equally well, regardless of the number of heavy/problem drinkers in their pre-intervention network, suggests that it is plausible that the intervention suppressed the influence of those particular risky network members on the person’s response to the intervention.

In contrast to the general representation of heavy/problem drinkers in one’s network with whom individuals may or may not drink, the representation of drinking buddies (i.e., a subset of peers from one’s network designated as friends for the primary aim of drinking or attending bars/clubs together) did impact immediate intervention response. These particular social network members are perceived to provide emotional support and are perceived to be close relationships (Lau-Barraco & Linden, 2014), which may contribute to why they are especially and uniquely influential in one’s drinking (Leonard et al., 2000; Neighbors et al., 2019; Reifman et al., 2006). Our findings demonstrated a potentially stronger intervention effect on risk reduction for those with higher proportions of drinking buddies in their network. That is, the intervention was especially efficacious in helping drinkers achieve low risk drinking if they had more drinking buddies in their network at the start of the intervention. Given that drinking buddies predict subsequent use and problems, the stronger PFI effect observed suggests that the intervention is promising in countering these negative drinking influences from particularly risky network members. However, it is important to note that the improved PFI response observed for this group could have been spurious due to the group starting at a much higher risk level at baseline relative to other groups. Replication of this study finding is necessary to allow more definitive conclusions regarding the impact of the PFI for those with especially risky drinking networks characterized by high representation of drinking buddies.

Within a clinical context, our study findings may inform intervention work with young adults in several ways. In light of the fact that several aspects of one’s social network impact their response to intervention programs, this suggests the potential benefits of incorporating social network interventions to strengthen the efficacy of the personalized feedback intervention. Social network interventions use members of a social network or alter the composition of networks to promote change (Hunter et al., 2019; Valente, 2012). Social network interventions may be particularly useful for hard-to-reach and at-risk populations, like nonstudent emerging adult heavy drinkers (Hunter et al., 2019). Based on our findings, employing network intervention techniques, including deliberately altering the structure of one’s network by establishing new network ties (e.g., adding alcohol abstinent peers) or by strengthening existing supportive network ties (e.g., promoting network stability with family and household members), could facilitate more rapid or enduring behavior change. An emphasis on social network modification is common across treatment models (e.g., Beck et al., 2001; Kelly et al., 2011), and the extent to which treatment-seekers are able to modify their social ties is predictive of long-term recovery outcomes (Eddie & Kelly, 2017; Zywiak et al., 2002). Based on our findings for the impact of abstinent members on intervention response, a similar approach could be emphasized with emerging adult risky drinkers. They could be encouraged to increase abstinent members as to provide more opportunities for rewarding and alcohol-free activities and provide an environment that better promotes behavior change as well as enhance their protective influence in the intervention process. Drinkers may also be encouraged to maximize time spent with existing low-risk members. Thus, existing interventions could systematically encourage participants to make adaptive social network changes and provide feedback focusing on known risks conferred by one’s social network on personal alcohol use. Such efforts may prolong and enhance the benefits of personalized feedback interventions. However, despite evidence supporting the potential influence of several network factors in moderating intervention response, it is important to acknowledge that we did not find a consistent impact across all the network variables examined or consistently for a single drinking outcome. The current investigation represents one of only a small handful of studies that have examined the influence of social networks in brief intervention response. Future research is needed to clarify if and how social networks exert impact over one’s drinking reduction change efforts.

Our findings should be interpreted in light of several limitations. Social networks were assessed only at baseline before the intervention. Social networks are not necessarily static over time and may have changed in composition after the intervention. Future research should consider network changes over time and examine the influence of these network changes on intervention outcomes. It may be that that changes in network ties (through reducing contact with high risk members or increasing contact with abstinent/low risk members) partially explain the impact of the intervention. Moreover, the type of network changes (e.g., reduction of drinking buddies in network) observed could be related to how well they respond to the intervention. Assessing changes in network could also permit examination of whether the PFI successfully weakened the relationship between risky social network ties and the participant’s own alcohol use. It may be that the strength of association between network drinking and personal drinking was diminished following the intervention, such that peer drinking exerts less influence on the participant’s own drinking. Another study limitation is that information regarding social network drinking was based on participant reports and was not subject to independent verification with network members. Future research may wish to go beyond relying on egocentric network methods that collect information from only one member of the network and include data from the participant’s defined social network. Finally, the current study focused on heavier drinkers during emerging adulthood. It is likely that social network dynamics change over the life course (Wrzus et al., 2013), and thus, generalizability of our findings to those further into adulthood remains to be tested.

Overall, our study contributed new knowledge regarding the impact of social network factors in behavior change following a brief alcohol intervention. We provided support for the protective impact of non-drinkers and stable ties in one’s network in promoting behavior change. Moreover, our findings provided partial evidence that a personalized feedback intervention tailored for nonstudents may have been efficacious in mitigating the influence of risky social networks. However, the relationship between social networks and intervention effects is a complex one given the differential moderation or impact of particular social network variables on outcomes. Future work continues to be warranted to further understand the influence of social networks on intervention response.

Public health significance:

Our findings suggest that a brief alcohol intervention diminished some influence from risky social networks on alcohol consumption among emerging adult heavy drinkers from the community. This study contributed new knowledge regarding the impact of social network ties on behavior change in a vulnerable and understudied group of drinkers that will help guide the development of future interventions.