Social Network Feedback and Drinking Outcomes among Emerging Adult Risky Drinkers Living in Urban Communities

Abstract

Background.

Alcohol use and related problems often increase during emerging adulthood and are influenced by social networks. Investigating alcohol-specific feedback from network members may be useful for understanding social influences and designing interventions to reduce risky drinking among emerging adults.

Purpose/Objectives.

This study examined whether drinking practices and consequences among emerging adult risky drinkers living in disadvantaged urban communities were influenced by receipt of encouragement, discouragement, or mixed messages about drinking from network members.

Methods.

Risky drinkers ages 21 to 29 (N = 356; 228 females; mean age = 23.6 years) residing in the community were recruited using digitally implemented Respondent Driven Sampling, a peer-driven chain referral method. A web-based survey assessed drinking practices, negative alcohol-related consequences, and drinking feedback from social network members including friends, spouse/partner, and other family members.

Results.

Negative binomial generalized linear modeling showed that discouragement of drinking by friends was associated with fewer drinking days and negative consequences, whereas discouragement by family members (excluding spouse/partner) was associated with more drinks per drinking day. Mixed feedback (sometimes encouraging, sometimes discouraging drinking) from friends and spouse/partner was associated with more drinking days and negative consequences.

Conclusions/Importance.

Social network feedback had both risk and protective associations with drinking practices and problems among emerging adults, with discouragement to drink by friends appearing to serve a protective function. The findings suggest the utility of interventions delivered through social networks that amplify the natural protective function of friend discouragement of drinking, in addition to addressing established risks associated with peers.

Introduction

Emerging adulthood, the developmental transition from adolescence to adulthood that occurs during the late teens to the mid-to-late twenties, is a critical time for growth and development (Arnett et al., 2011; Scales et al., 2016). During this period, emerging adults (EAs) typically gain independence and autonomy and explore life options before completing the transition to full-fledged adult roles (Arnett et al., 2011). It is also, however, a developmental stage characterized by increased risk-taking, including harmful substance use (Arnett et al., 2011; Salvatore, 2018). National survey data show that EAs report higher rates of past-month binge drinking (34.9%), daily use of tobacco products (25.8%), and use of illicit drugs (23.9%) than any other age groups (Substance Abuse and Mental Health Services Administration [SAMSHA], 2019). Moreover, prevalence of alcohol and other substance use disorders is highest among EAs (McCabe et al., 2017). Despite their elevated risk, EAs are generally less motivated to modify their drinking practices and less likely to receive treatment compared to other age groups (Hingson et al., 2015).

Although EAs’ heightened risk-taking and its potential for long-term negative consequences are significant public health concerns, the bulk of work to date has focused on traditional college students enrolled full-time at four-year institutions of higher education and has investigated college-specific influences (Leeman et al., 2016; National Institute on Alcohol Abuse and Alcoholism [NIAAA], 2019a). A significant proportion of the EA population has been neglected, including those who live in the community, are enrolled in community colleges, are not students, or attend school part-time or intermittently (Carter et al., 2010; Pokhrel et al., 2014; Salgado García et al., 2020; Tucker et al., 2020). Studies that directly compared full-time college students and their non-college peers showed that college students drink more heavily, but EAs not in college tend to experience more alcohol-related problems (Quinn & Fromme, 2011; White, Labouvie et al., 2005). Such differences may reflect differences in environmental influences on drinking, rather than student status per se. EAs with full-time student status and those living in the community may have different responsibilities and role expectations (White, LaBouvie et al., 2005) and be influenced differently by drinking social norms (Quinn & Fromme, 2011). In addition, EAs who are not traditional four-year college students tend to be financially disadvantaged, members of ethnic minority groups, first-generation college students, and more likely to work part-time or full-time (American Association of Community Colleges, 2017; Salgado García et al., 2020). These findings suggest a need for research and interventions tailored for EAs living in different circumstances.

Regardless of college enrollment, a core aspect of emerging adulthood involves transitions of social roles and changes in social networks (Patrick et al., 2020). The majority of EAs move out of their parents’ home for education, work, or marriage, resulting in remote and less frequent communications with parents as well as changes in peers and other socializing groups, such as religious, sport, and community organizations (Scales et al., 2016). These changes can increase reliance on feedback and support from social network members, particularly from peers including a spouse or partner (Patrick et al., 2020). In the case of alcohol use, it is well documented that EAs are influenced by the drinking behaviors and norms of their social network members (Hahm et al., 2012; Lee et al., 2015; Simons-Morton et al., 2016; Tucker et al., 2015; Tucker et al., 2016a). However, compared to college student populations, social influences among EAs living in the community are relatively understudied (Lau-Barraco & Collins, 2011; Tucker et al., 2015) and important to investigate because community-dwelling EAs are more likely to have adult roles and responsibilities (e.g., employee, spouse, or parent) and a different mix of social network members (Scales et al., 2016).

More specifically, prior research has shown that general social network features such as size, composition, and overall social support are unrelated to drinking practices or seeking help for alcohol problems, whereas alcohol-specific network feedback is differentially associated with drinking and help-seeking status depending on the type of feedback (e.g., George & Tucker, 1996; Longabaugh et al., 2010; Tucker et al., 2015). For example, a community-based study with EA drinkers living in a disadvantaged urban area (Tucker et al., 2015) found that higher substance involvement was associated with having close peer network members who used substances, whereas peer discouragement of substance use was associated with reduced risk. Family discouragement had no such protective association, and general network structure and function (e.g., size, provision of social support) were unrelated to substance use. Studies with adult problem drinker populations have found similar associations, including among alcohol-dependent adult outpatients (Longabaugh et al., 2010) and problem drinkers who recently sought help (George & Tucker, 1996). For example, George and Tucker (1996) found that problem drinkers who recently entered alcohol treatment or Alcoholics Anonymous (AA) reported less network encouragement to drink and more network encouragement to seek help compared to untreated problem drinkers. Network structure and overall social support were unrelated to help-seeking status. Further, AA participants received more conflicting or mixed messages from friends about seeking help compared to treatment participants, suggesting AA provides an alternative network supportive of sobriety.

The present study investigated whether drinking practices and consequences among EA risky drinkers living in urban communities were influenced by receipt of encouragement, discouragement, or mixed messages about drinking from network members. Given the challenges of recruiting EAs who are not traditional college students, a peer-driven chain referral recruitment method (Respondent Driven Sampling [RDS]) used successfully in recruit at-risk populations including community-dwelling EAs was implemented digitally (e.g., Bauermeister et al., 2012; Tucker et al., 2020; Zhang et al., 2017). A web-based survey assessed alcohol-related feedback received from participants’ social network members and participants’ drinking practices and consequences. Based on prior findings, EAs’ drinking practices and consequences were expected to be differentially associated with the type of drinking feedback received (i.e., encouragement, discouragement, or mixed message about drinking) and the network source of feedback (i.e., friends, spouse/partner, other family members). Friends were predicted to have both risk and protective associations with participants’ drinking depending on the nature of the feedback. Feedback from other network members (e.g., family members) was expected to be relatively less influential on EAs’ drinking patterns (Tucker et al., 2015, 2016a), with the exception that riskier drinking patterns were expected to be associated with conflicting network feedback about drinking and with family discouragement of drinking. In contrast, general social network features (e.g., size, overall emotional support and tangible aid) were not expected to show differential associations with drinking.

Materials and Methods

Sample Recruitment and Characteristics

Study procedures described in Tucker et al. (2020) are summarized here. The study received university Institutional Review Board approval and was conducted in line with STROBE (von Elm et al., 2007) and STROBE-RDS (White et al., 2015) guidelines for observational studies. “Seeds” to start RDS were recruited in person from multiple high traffic community venues (e.g., sporting events, outdoor markets, art and music festivals) in North and Central Florida with relatively high percentages of young adults based on 2010 U.S. Census tract data. In-person seed recruitment by trained research staff similar in age to the target sample served to verify that RDS was initiated by EAs with the desired target group characteristics. Eligibility criteria for seeds and their peer recruits were: (1) Men and women ages 21–29 living in Florida at enrollment; (2) past-month alcohol consumption above NIAAA (2019b) gender-adjusted thresholds for heavy drinking (4+/5+ drinks for women/men) and one or more alcohol-related negative consequences in the past 90 days; and (3) web access via smartphone or computer. Although emerging adulthood is often defined as ages 18–25 (e.g., Arnett et al., 2011), young adults in their twenties were recruited because this is a dynamic developmental period for drinking-related risks and risk reduction (Lee & Sher, 2018).

Once verbal informed consent was obtained and screening criteria were confirmed, eligible seeds (n = 176) viewed video instructions on a study computer tablet or their personal smart device regarding how to recruit peers “like you” and received information about compensation. RDS procedures (Heckathorn, 2007; Gile et al., 2015; Kogan et al., 2011) were adapted to a digital platform (e.g., Bauermeister et al., 2012; Tucker et al., 2020; Wejnert & Heckathorn, 2008; Zhang et al., 2017), and all peer recruitment and data collection were conducted online. Peer recruits accessed and completed an online survey on a secure research website programmed using Research Electronic Data Capture (REDCap; Harris et al., 2019), which was maintained by the University of Florida Clinical and Translational Science Institute. To protect confidentiality, survey responses were stored separately using a numerical identifier independent of participant information necessary for research compensation. Seeds did not complete the survey.

Recruitment involved providing each seed and peer recruit with three unique numerical enrollment codes, which they could text or email to peers to invite them to participate in the study. Enrollment was limited to a maximum of three recruits per participant to ensure network branching and prevent over-recruitment of a small number of network subgroups. Once initiated, recruitment chains were allowed to develop naturally to facilitate recruiting a sample independent of seed characteristics (Heckathorn, 2007; Gile et al., 2015). RDS Coupon Manager and RDS Analysis Tool, developed by Heckathorn and colleagues (http://respondentdrivensampling.org), were used to track referral chains using the unique enrollment codes. Seeds and recruits each received $30 for their initial screening and $15 for each eligible peer recruited and enrolled up to a maximum of 3 (up to $75 total compensation). Payments were made using electronically reloadable Visa™ gift cards.

Table 1 presents the characteristics of the final sample of peer recruits (N = 356) excluding seeds per standard RDS analysis procedures (Gile et al., 2015). As desired and reported in Tucker et al. (2020), digital RDS successfully recruited a sample of EA risky drinkers who on average reported greater alcohol use than an age-matched subsample of drinkers from the 2018 National Survey on Drug Use and Health (SAMHSA, 2019). The drinking eligibility screening criteria, deliberately set low to establish a basic level of risk at enrollment while keeping screening brief, worked well to screen in participants who were risky drinkers, as verified by the subsequent survey assessment. As in our past in-person RDS research (e.g., Tucker et al., 2016b), more women enrolled than men. The sample as a whole was in their lower mid-twenties (M age = 23.6, range = 21 – 29). The great majority lived in disadvantaged urban areas and were educated beyond high school, most were employed full or part-time, but over half had annual personal incomes < $20K. Less than 10% were married or were parents.

Table 1.

Sample Characteristics of Peer Recruits

Variable	Frequency (%) / Mean (SD)
Demographic characteristics
Age in years	23.63 (2.60)
Gender (% women)	228 (64.23)
Race/ethnicity
Asian	68 (19.26)
Black	22 (6.23)
White	228 (64.59)
Othera	35 (9.92)
Hispanic	60 (16.95)
Highest education   Completed high school or GED   Completed four-year college degree or higher	351 (99.43) 172 (48.73)
Student (full or part-time)	228 (64.04)
Employed (full or part-time)	274 (77.40)
Personal annual income < $20k	183 (53.51)
Married	25 (7.02)
Have children	21 (5.90)
Residential areab   Urban   Poverty rate greater than U.S. average (10.5%)	351 (98.60) 231 (64.89)
Drinking risk variables
Number of past month drinking days	9.93 (5.76)
Frequency (% participants) with typical past-month   drinking exceeding high risk drinking thresholdsc	167 (46.91)
Frequency (% participants) with typical past-month   drinking exceeding very high risk drinking thresholdsd	26 (7.30)
Drinks consumed per drinking day (past month)	4.71 (4.76)
Drinks consumed on high risk drinking dayse	7.03 (6.14)
Drinks consumed on very high risk drinking daysf	17.00 (10.96)
Alcohol-related negative consequences (BYAACQ)	8.74 (5.81)
Social network characteristics
Size of young adult online network (# members)g	27.32 (51.79)
Productive peer recruiters (≥ 1 recruit)	153 (42.98)
NSSQ total social network size (1 – 10 members)h	3.98 (0.27)
Friends	2.31 (1.01)
Spouse/partner	0.51 (0.50)
Other family members	0.83 (0.82)
Other relationships	0.33 (0.63)
NSSQ Emotional Support subscale (range: 0 – 5)i
Friends	4.20 (.07)
Spouse/partner	4.71 (.55)
Other family members	4.46 (.70)
NSSQ Tangible Aid subscale (range: 0 – 5)i
Friends in network	4.08 (.79)
Spouse/partner	4.73 (.60)
Other family members	4.54 (.80)
Network drinking feedback
Friend drinking feedback
Encouragement to drink	3.23 (1.03)
Discouragement to drink	2.09 (1.11)
Mixed messages encouraging and discouraging drinking	2.63 (1.13)
Spouse/Partner drinking feedback
Encouragement to drink	2.66 (1.29)
Discouragement to drink	2.51 (1.40)
Mixed messages encouraging and discouraging drinking	2.79 (1.33)
Other family member/Relative drinking feedback
Encouragement to drink	2.41 (1.37)
Discouragement to drink	2.52 (1.43)
Mixed messages encouraging and discouraging drinking	2.37 (1.31)

N = 356

^aIncludes American Indian/Alaska Native (.6%), Native Hawaiian/Other Pacific Islander (1.1%), and more than one race (5.4%); 3 additional participants indicated “I choose not to answer.”

^bBased on the residential zip codes reported by participants.

^c4+/5+ drinks for women/men.

^d8+/10+ drinks for women/men.

^eFor 167 participants (46.91%) who reported any high risk drinking.

^fFor 26 participants (7.30%) who reported any very high risk drinking.

^gPeer online social network size used for RDS sample weights.

^hMean number of social network members listed on NSSQ.

ⁱMean of subscales averaged across social network members. BYAACQ = Brief Young Adult Alcohol Consequences Questionnaire (past 3 months; maximum = 24). NSSQ = Norbeck Social Support Questionnaire. Descriptive statistics for sample characteristics were calculated using the unweighted data set.

Survey Measures

The web-based survey assessed recent drinking practices, alcohol-related consequences, and risk and protective factors for EA alcohol use and took an average of 30.69 minutes to complete (SD = 18.71). Measures of alcohol-specific social network feedback, recent drinking practices, and demographic characteristics that are the focus of the present study are described below. Additional measures (e.g., alcohol reinforcement value, protective behavioral strategies) will be reported elsewhere (e.g., Tucker et al., 2021).

Drinking practices and consequences

The number of drinking days and typical standard drinks consumed per drinking day in the past 30 days were assessed using the abbreviated Daily Drinking Questionnaire-Revised (DDQ-R; Collins et al., 1985; cf. Leeman et al., 2016). Widely used with young adults, the DDQ-R yields reliable reports of alcohol consumption (Kivlahan et al., 1990). Alcohol-related consequences were assessed using the Brief Young Adult Alcohol Consequences Questionnaire (BYAACQ; Kahler et al., 2005), which asks about 24 negative events over the past 3 months (e.g., neglected obligations, hangover, driving after drinking). The number of consequences was summed for analysis. The BYAACQ assesses less severe but common consequences (Kahler et al., 2005, 2008) and has high internal consistency (Cronbach’s α = .90).

Social networks and drinking-related feedback

An expanded Norbeck Social Support Questionnaire (NSSQ; Norbeck et al., 1981; George & Tucker, 1996; Tucker, et al., 2015) was used to assess overall social support and alcohol-specific feedback from social network members. Participants were asked to identify at least 4 and up to 10 network members who “provide personal support for you or who are important to you” and indicate the type of relationship (e.g., family, friend, spouse or partner). Using a 5-point rating scale (1 = “not at all” to 5 = “a great deal”), participants separately rated the extent to which each member encouraged participants’ drinking, discouraged it, and gave mixed messages, sometimes encouraging and sometimes discouraging their drinking. They also rated the extent of emotional support (e.g., making you feel liked or loved) and tangible aid (e.g., help with money, ride to the doctor) received from each member. Network members were categorized as friends, family members and relatives (other than spouse and children), spouse/ partner, or other relationship. Scores for each type of drinking feedback, emotional support, and tangible aid received from network members were calculated separately for each relationship category by averaging the ratings across the members in each category. Of 1,408 total network members listed by participants, 57.95% were friends, 12.86% spouse/partner, and 20.95% other family/relatives. The remaining “other” network group (< 10%) were excluded from the analyses because it was a heterogeneous category that included other relationships with low frequencies (e.g., neighbors, work/school associates, health care provider).

Covariates

Demographic variables associated with drinking risk behaviors in previous studies with EAs (Tucker et al., 2015; Tucker et al., 2016b) were used in the data analyses to control for potential confounding. These included age, gender, race, education level, and annual income.

Data Analysis

Per RDS analysis procedures, the analysis sample excluded seeds, and the sample of peers was checked for recruitment bias and RDS analytic assumptions (Heckathorn, 2007; Gile et al., 2015). No evidence was found for biases of non-random recruitment (homophily) or dependence of recruit characteristics on non-randomly selected seed characteristics over recruitment waves (equilibrium), except for a slight bias in recruiting peers in same race groups that was below levels requiring sample weighting (Schonlau & Liebau, 2012). In order to account for potential recruitment bias due to variations in network size (based on the number of EAs that participants reported interacting with online at enrollment), weights were calculated using the RDS II estimator in RDS Analyst (Handcock et al., 2013) and incorporated in the primary analyses examining associations among network feedback and drinking outcomes. Descriptive statistics (Table 1) were calculated using the unweighted data set.

Generalized linear modeling (i.e., proc genmod) for negative binomial distribution was conducted using SAS EQ v8.1 to predict the count outcome variables (typical drinks per drinking day, number of past month drinking days, and number of negative consequences in the past 90 days) using the social network drinking feedback measures and demographic covariates. Separate models were estimated for different drinking outcomes and social network sources. A total of nine models were estimated for drinking feedback from network members, based on combinations of each of the three drinking outcomes with each of the three network member relationship types (friends, spouse/partner, other family/ relatives). Nine separate models also were estimated for emotional support and tangible aids from network members. To adjust for the multiple significance tests, we applied Benjamini-Hochberg’s procedure controlling for the false discovery rate (FDR), i.e., the expected proportion of erroneous rejections of true null hypotheses. Significant results were reported if FDR was ≤ 5% and the observed p-value met Benjamini-Hochberg’s adaptive FDR criterion (Benjamini & Hochberg, 2000). One outlier participant reported 300 drinks per drinking day and was excluded from the analyses. Each analysis sample varied in size based on the number of network members who belonged to each relationship category and also due to cases with missing values on select predictors and outcomes (n = 20), which were minimal and not systematically item specific.

Results

As summarized in Table 1, on average the sample reported about 10 drinking days in the past month, consuming about 5 drinks per drinking day. In addition, about 47% and 7% reported usual consumption at high risk (4+/5+ drinks for women/men) and very high risk (8+/10+) levels, respectively. Participants reported an average of about 9 negative drinking-related consequences in the past 90 days, with hangovers (83.7%), less energy/tired (73.3%), and very sick stomach/vomiting (65.2%) being the most frequent consequences. Of the 356 participants, 346 (97.2%) listed at least one friend, 181 (50.8%) listed a spouse or partner, and 214 (60.1%) listed at least one family member or relative as a close social network member.

The results from the generalized linear modeling are reported in Table 2 and indicated that social network feedback for drinking selectively predicted EAs’ drinking practices and consequences. As hypothesized, friends’ discouragement of drinking was associated with fewer past month drinking days and fewer negative consequences. Mixed feedback from friends that sometimes encouraged and sometime discouraged participant drinking was associated with more drinking days and more negative consequences. Friends’ encouragement of drinking was not related to drinking outcomes, after controlling for their discouraging and mixed feedback. With regard to spouse/partner feedback, only mixed drinking feedback was associated with more consequences. Drinking feedback from family members and relatives was not protective, and their discouragement of participant drinking was associated with more drinks per drinking day.

Table 2.

Social network feedback and drinking outcomes among emerging adult risky drinkers

Social Network Relationship Subgroup	Predictors	Drinking Outcomes
		# Drinking Days (past month)	Drinks Per Drinking Day (past month)	Consequences (past 3 months)
		B (SE)a	B (SE)a	B (SE)a
Friends	Encouragement of drinking	0.008 (0.031)	0.034 (0.037)	0.055 (0.034)
	Discouragement of drinking	−0.103 (0.035)**	0.033 (0.042)	−0.095 (0.040)*
	Mixed drinking feedback	0.123 (0.037)***	0.005 (0.045)	0.113 (0.041)**
	Age	0.025 (0.007)***	0.012 (0.008)	0.006 (0.007)
	Gender	0.212 (0.059)***	0.099 (0.069)	−0.081 (0.067)
	Education Level	0.039 (0.020)	−0.091 (0.025)***	0.022 (0.023)
	Yearly Income	−0.024 (0.013)	0.049 (0.016)**	−0.004 (0.015)
	Race	0.009 (0.060)	−0.279 (0.068)***	−0.199 (0.066)**
Spouse/ Partner	Encouragement of drinking	0.014 (0.038)	−0.022 (0.047)	0.014 (0.042)
	Discouragement of drinking	−0.047 (0.035)	0.044 (0.041)	0.069 (0.039)
	Mixed drinking feedback	0.039 (0.042)	−0.014 (0.053)	0.119 (0.048)*
	Age	0.023 (0.008)**	0.020 (0.010)*	0.003 (0.008)
	Gender	0.192 (0.078)*	−0.083 (0.098)	−0.191 (0.093)*
	Education Level	0.067 (0.028)*	−0.078 (0.037)*	0.014 (0.034)
	Yearly Income	−0.007 (0.016)	0.050 (0.020)*	0.013 (0.019)
	Race	0.149 (0.082)	−0.318 (0.097)**	−0.135 (0.093)
Other Family Members/ Relatives	Encouragement of drinking	−0.018 (0.032)	−0.033 (0.039)	0.018 (0.032)
	Discouragement of drinking	0.032 (0.031)	0.097 (0.035)**	0.057 (0.033)
	Mixed drinking feedback	−0.033 (0.039)	−0.033 (0.046)	0.007 (0.040)
	Age	0.023 (0.011)*	0.010 (0.013)	0.023 (0.012)*
	Gender	0.065 (0.077)	0.009 (0.091)	−0.135 (0.082)
	Education Level	0.017 (0.025)	−0.087 (0.031)**	−0.034 (0.026)
	Yearly Income	−0.024 (0.019)	0.076 (0.022)***	−0.002 (0.020)
	Race	−0.031 (0.075)	−0.090 (0.090)	−0.112 (0.081)

^*p<.05

^**p<.01

^***p<.001.

^aUnstandardized coefficients and standard errors obtained from proc genmod in SAS based on the weighted data set. Dichotomous predictor reference groups: gender (male =1, female = 0) and race (white = 1, other = 0)

Among the covariates, older age was associated with more drinking days and more drinks per drinking day. Males reported more past-month drinking days but fewer alcohol-related consequences. Higher education was associated with more drinking days but fewer drinks per drinking day. Higher income was associated with more drinks per drinking day. Being white was associated with fewer drinks per drinking day and fewer consequences.

Finally, emotional support from friends and spouse/partner was associated with fewer consequences (B = −.190, SE = .070, p < .01) and fewer drinks per drinking day (B = −.290, SE = .104, p <.01), respectively. However, emotional support from other family members was not associated with any drinking outcome, and no significant associations were found between tangible aid from any relationship source and any drinking outcome.

Discussion

The findings indicate that drinking feedback from close social network members is an important factor associated with drinking practices and consequences among EAs residing in the community and that the type of feedback received from different members has variable associations with drinking. As hypothesized, lower levels of alcohol use and fewer negative consequences were reported when EAs received discouraging feedback about their drinking from their friends. Unexpectedly and discrepant with prior research (Tucker et al., 2015), friend encouragement of drinking was not significantly associated with increased drinking practices and consequences, perhaps because the sample as a whole evidenced risky drinking patterns that were higher, on average, compared to drinking among age-matched peer drinkers from a national representative survey (Tucker et al., 2020). For this group of EAs, friends, as well as their spouse/partner, appeared to functionally encourage drinking by virtue of drinking with the participant (i.e., 90% of friends and 94% of spouse/partner). As such, encouragement of drinking from close friends may not be salient or influential enough to elevate drinking levels even higher than the sample mean drinking frequency and quantity (approximately 10 drinking days and 5 drinks per drinking day during the past month).

Whereas friends’ discouragement of drinking appeared to be a protective factor, discouragement of drinking by other network members (i.e., spouse/partner, family members/relatives) was not. On the contrary, family member discouragement of drinking was associated with greater participant drinking, as found in previous research (Tucker et al., 2015). Given the cross-sectional nature of the study, the causal direction of the association is ambiguous. It may reflect family members’ negative reaction to and concern about EAs’ risky alcohol use, EAs’ oppositional response to family members’ criticism of their drinking, or a combination of both. Regardless of causality, these findings suggest interventions for EAs should be cautious about using family messages about drinking.

One unique study finding concerned the associations observed between mixed drinking feedback from friends and spouse/partner and higher levels of participant drinking and negative consequences. This association was more consistent than any risks associated with encouragement of participant drinking by these network members. Thus, interventions aimed at promoting consistent discouragement of drinking by significant network members may be beneficial for reducing EAs’ drinking-related risks. This is in addition to interventions aimed at reducing the proportion of drinking buddies and perceived drinking norms in EAs’ social networks that prior studies have suggested are an effective intervention element (e.g., Lau-Barraco & Collins, 2011). In contrast to the significant drinking-specific feedback associations, measures of overall network support had non-significant or limited associations with drinking outcomes, as found previously (e.g., Longabaugh, 2010; Tucker et al., 2015). This suggests the value of targeting drinking-specific social network characteristics rather than attempting to increase overall network social support in interventions aimed at reducing drinking risks.

The current study has limitations. Due to the cross-sectional nature of the design, causal relations cannot be claimed. It is unclear whether network feedback influenced participants’ drinking or whether it occurred in response to participants’ drinking, or both. In addition, potential reporting biases resulting from the use of self-report measures, which were necessary for this web-based survey, cannot be ruled out. Nevertheless, the findings are generally consistent with prior research, which suggests their validity. Furthermore, the RDS study sample unexpectedly included some college students due to the RDS recruitment method that allowed recruitment chains to develop naturally. However, the majority of the sample were not undergraduate students currently enrolled full-time at four-year colleges and living on campus; most resided off-campus, and many were employed full- or part-time. Therefore, the EA sample as a whole differed from traditional college EAs residing on campus. Finally, the community sample was recruited in a particular region of a Southern state and had characteristics reflecting varying degrees of disadvantage. Further, the sample race/ethnicity composition, while diverse, included an unexpectedly larger proportion of Asians and lower proportion of Blacks compared to the North and Central Florida EA population. Whether results would generalize to other populations of EAs requires additional study, but the commonalities in findings observed among this community sample and earlier studies suggest the social network-drinking associations are robust.

In conclusion, the study replicated prior findings and contributed new knowledge regarding drinking-specific social network associations with drinking practices and consequences among EAs. Heretofore, much social network research on EA drinking has focused on descriptive and injunction norms about drinking (e.g., Borsari & Carey, 2003; Quinn & Fromme, 2011), whereas this study examined drinking-specific communications from network members defined by the nature of their relationship with participants. The role of mixed drinking feedback has been particularly understudied and appears to play a role in drinking risk among EAs as well as among persons with an alcohol use disorder (e.g., George & Tucker, 1996). The current study also was innovative in demonstrating the utility of digital RDS as a recruitment and intervention delivery channel for EAs living in the community who are harder to reach and have more diverse lifestyles than college campus student residents. Accessing their social networks to deliver targeted interventions with network members, particularly friends, may offer new prevention and intervention approaches for this under-served young adult population of risky drinkers.