Free Response Questions

lable at ScienceDirect

Social Science & Medicine 108 (2014) 34e45

Contents lists avai

Social Science & Medicine

journal homepage: www.elsevier .com/locate/socscimed

The myth of conformity: Adolescents and abstention from unhealthy drinking behaviors

Carter Rees a,*, Danielle Wallace b

aArizona State University, School of Criminology and Criminal Justice, 411 N. Central Ave., Room 600, Phoenix, AZ 85004, USA bArizona State University, School of Criminology and Criminal Justice, 411 N. Central Ave., Room 600, Phoenix, AZ 85050, USA

a r t i c l e i n f o

Article history: Available online 12 February 2014

Keywords: Adolescent drinking Friendship networks Minority influence Conformity Dissent Social influence

* Corresponding author. E-mail addresses: [email protected], carterrees@

http://dx.doi.org/10.1016/j.socscimed.2014.01.040 0277-9536/� 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Adolescent peer groups with pro-drinking group norms are a well-established source of influence for alcohol initiation and use. However, classic experimental studies of social influence, namely ‘minority influence’, clearly indicate social situations in which an individual can resist conforming to the group norm. Using the National Longitudinal Study of Adolescent Health (“Add Health”), a nationally repre- sentative sample of adolescents, we find evidence that being a non-drinking adolescent does not unilaterally put youth at risk for drinking onset when faced with a friendship network where the ma- jority of friends drink. Our results also show that a non-drinking adolescent with a majority of drinking friends is significantly less likely to initiate alcohol abuse if he or she has a minority of non-drinking friend(s). Furthermore, a drinking adolescent with a majority of friends who drink has a decreased probability of continuing to drink and has overall lower levels of consumption if he or she has a minority of friends who do not drink. Our findings recognize that adolescent in-group friendships are a mix of behavioral profiles and can perhaps help adolescents continue or begin to abstain alcohol use even when in a friendship group supportive of alcohol use.

� 2014 Elsevier Ltd. All rights reserved.

Introduction

Adolescents may be the archetypical social animals, intention- ally spending vast amounts of time together to fulfill the human need for social interaction. Friendships provide protection, knowledge, support, and behavioral guidance via in-group rules and norms. It is within these friendship groups that the ‘I’ becomes ‘we’ as adolescents become concerned with ‘us’ instead of ‘me’ (Kroger, 2004). Friendship groups wield enormous power to exact behavioral conformity. Friends can both accept and reject an adolescent, promote and restrict behavior, improve self-esteem and ridicule mercilessly (Bagwell & Schmidt, 2011). The social life of an adolescent is often better if he or she conforms because failure to do so risks ostracism from the group (Williams, Forgas, von Hippel, & Zadro, 2005).

Yet, it is within the friendship group that adolescents must also satisfy the compulsory drive for personal differentiation, but do so without becoming noxious to their friendship groups (see Snyder &

gmail.com (C. Rees).

Fromkin, 1980). This creates a tension between individuality and group membership. Adolescents have the demonstrated ability rebel, push back, and dissent from the normative influence of parents and schools while remaining students and family members nonetheless. Discussions of this ability to resist conformity and influence have yet to be adequately extended to the adolescent friendship group.

We question the mechanistic view of the all powerful group behavioral norm and the passive adolescent in discussions of social influence. We draw upon established social psychological theories of resistance to social influence (e.g., Moscovici, 1980) to assess the extent to which an adolescent is a servile recipient of friends’ in- fluence, having his or her behavior dominated by majority rule. Or, if a small collective of individuals can resist the influence of established behavioral norms within the friendship group. Furthermore, we ask if a numerical minority can also significantly influence the established behavior of a member of the majority away from the dominant behavioral position.

Our study links the presence of an in-group behavioral minority to adolescent drinking, a behavior shown to be predominately influenced by peers. For example, does having two non-drinking friends decrease an adolescent’s established drinking patterns if

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 35

the friendship group also consists of four drinking friends? This encourages the discussion of social influence to include conformity to and resistance from in-group norms. This is important because research consistently references pro-drinking group norms of an adolescent’s friends as a risk factor for the onset and continued use of alcohol (Balsa, Homer, French, & Norton, 2011; Donovan, 2004). Broadly, our goal is to advance theoretical and empirical research on interpersonal influence from a socio-psychological perspective. We examine the effect of drinking and non-drinking friends using the social network component of the Add Health data, a nationally representative sample of middle and high school adolescents in the United States.

Alcohol use, adolescents, and peers

Alcohol consumption among adolescents continues to be an issue in the United States being associated with an assortment of other unhealthy behaviors such as illicit drug use, and sex (Donovan & Jessor, 1985; Kulbok & Cox, 2002; Valois, Oeltmann, Waller, & Hussey, 1999). Adolescents who are drinkers are more likely to show signs of emotional distress (Crosnoe, Muller, & Frank, 2004). Risky drinking behavior during adolescence also leads to a variety of detrimental outcomes such car accidents (Lang, Waller, & Shope, 1997), driving while under the influence (Farrow & Brissing, 1990), emergency room visits (Meropol, Moscati, Lillis, Ballow, & Janicke, 1995), poor academic performance (Crosnoe et al., 2004), dropping out of high school (Chatterji & DeSimone, 2005), sexual victimization (Bachanas et al., 2002), and attempted suicide (Miller, Lippmann, Azrael, & Hemenway, 2007). Lastly, high risk drinking places an adolescent at greater risk for heavy drinking and alcohol dependence in adulthood (DeWit, Adlaf, Offord, & Ogborne, 2000).

A vast body of literature establishes the importance of peer in- fluence in adolescents’ drinking behavior and substance abuse (Fujimoto & Valente, 2012a, 2012b). Adolescent exposure to sub- stance abuse by friends is a strong, positive correlate of personal substance abuse (Crosnoe et al., 2004; Kobus & Henry, 2010). Recent studies used detailed social networking data to add clarity to how and why peers matter in the explanation of drinking during adolescence. For instance, adolescents tend to drink to earn social rewards from their friends (Allen, Porter, McFarland, Marsh, & McElhaney, 2005; Diego, Field, & Sanders, 2003). Different types of friends such as a best-, close- or regular friends have separate and significant influence on increasing and adolescent’s drinking (Rees & Pogarsky, 2011; Urberg, Degirmencioglu, & Pilgrim, 1997). Reciprocal friendships amplify the effect of peer influence on adolescent alcohol (Fujimoto & Valente, 2012a), as do personal dispositions, school level factors, and relationship characteristics (Botticello, 2009; Brechwald & Prinstein, 2011; Vitaro, Brendgen, & Tremblay, 2000). In sum, health related research has empirically established peer influence and friendship network conditions as prerequisites for understanding the etiology of adolescent drinking.

Yet, focusing on the various ways peers influence and promote drinking behaviors only shows the “dark side of friendship” and ignores the “ways in which particular friendships contribute [in] both positive and negative ways to well-being and adjustment” (Bagwell & Schmidt, 2011; p. 165). Berndt & Murphy (2003) refer to this as the “myth [that] friends’ influence is predominantly nega- tive” (p. 278). Brechwald and Prinstein (2011) state “the majority of research examining peer influence effects.has focused on social- ization of anti-social, deviant, and health-risk behaviors” (p. 167) but also acknowledges that the past 10 years have resulted in “sorely needed research” related to peer influence of healthy be- haviors (see also Brown, Bakken, Ameringer, & Mahon, 2008). Integration of these two literatures is critical to recognizing adolescent friendship networkmay contain amix of both pro-social

and anti-social peers and what this means for peer influences and adolescent drinking (see Haynie, 2002; Windle et al., 2008).

Friends disapprove of and discourage drinking (Keefe,1994). Yet, little research engages these findings as possible ways in which adolescents can use in-group friendships as a means of resisting in- group norms favorable to alcohol consumption. Instead, research has envisioned adolescents in need of specialized training in resistance to social influence and conformity towards substance use through teacher, peer, or school led intervention programs (see Hawkins, Catalano, & Miller, 1992). The ability to resist social in- fluence and to rebel against group norms seems to be reserved for high-risk youth in their continuing personal attempts to resist conventional norms and their strong need for independence (see Jessor & Jessor, 1977; Paton & Kandel, 1978).

Social influence: a bias towards conformity

Despite experimental evidence to the contrary, there is a ten- dency to regard the study of social influence and how it changes behavior as equating to the study of an individual always con- forming to the group norm (Moscovici, 1976; see also Packer & Miners, 2012). For example, Asch’s (1951, 1956) famous line studies on social influence provide the foundation for many contemporary studies of social influence and conformity (Friedkin & Johnsen, 2011). His work is regularly cited as evidence that the group majority is able to cause individuals to comply or conform to its position (Levine, 1999). Yet, the original intent of Asch’s exper- iments was the examination of resistance to social influence and group suppression of non-conformity (Moscovici & Faucheax, 1972). Asch (1952) felt social influence research had “.taken slavish submission to group forces as the general fact and has neglected or implicitly denied the capacities of men for indepen- dence” (p. 451). That is, social influence is more than the individual behaviorally yielding to an established group norm. The findings of Asch’s (1951, 1956) experiments demonstrate strong support for the capacity of individuals to resist the majority group’s influence (Bond & Smith, 1996; Wood, Lundgren, Ouellette, Busceme, & Blackstone, 1994). Even having one supporter for an individual’s dissenting opinion against the majority reduced conformity to the group’s erroneous position from 32 percent to 5.5 percent, thereby almost eliminating conformity to the erroneous group norm (Asch, 1951). Asch (1956) concludes that there is a considerable effect of the majority group on the individual but “it was by no means complete or even the strongest force at work” (p. 10). That is, in- dividuals can resist the influence of the majority group and this resistance is even more pronounced with a single partner sharing their opinion.

A focus solely on conformity to group norms and how the majority obtains conformity ignores the demonstrated ability of individuals to be non-conformists, actively and successfully pro- moting within group change. This emphasis suggests only the majority can exercise social influence because they have power in numbers or status; deviance from the majority position is an in- dividual defect and not an impetus for behavioral or opinion change. Therefore, in keeping with the idea of conformity bias, an erroneous position held by the in-group majority is not correct- able by a powerless in-group minority (Moscovici, 1985); resolu- tion to in-group conflict will always be settled in favor of the majority by virtue of their implied position of overwhelming social influence.

Thework of Moscovici (1976,1980,1985; see also Nemeth,1986) challenges the assumption that individuals in the minority are only the targets of influence and not the source. Moscovici’s (1980,1985) conversion theory is a formal report on how and why ‘minority in- fluence’ occurs and is supported by a large body of empirical

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e4536

literature spanning politics, decision-making, behavior, and orga- nizations (see Maass & Clark, 1984; Martin & Hewstone, 2010). Conversion theory makes the explicit argument that the minority position does not necessarily equate to being a passive receiver of influence from the majority (Wolf, 1987); majority and minority influence differ in the processes and mechanisms that produce and maintain each type of influence. Majority influence leads to the social behavior of compliance via a comparison process. Minority influence leads to the social behavior of conversion through a vali- dation process. Moscovici (1980) integrates both modes of influ- ence into a single theoretical framework and, unlike most studies of social influence, the majority position is no longer the only source of “truth and norm, and expresses the social system as a whole” (Moscovici, 1976; p. 12; Moscovici, Mugny, & van Avermaet, 1985).

Conversion theory suggests that majority groups are interested in maintaining the power to dictate norms, rules, and social control by stabilizing behaviors and absorbing behavioral outliers leading to the reduction of inter-individual differences in order to maintain the status-quo (Mugny & Pérez, 1991). Majority influence capital- izes on people’s need to need to be liked, accepted by others, and to be correct. The majority may promote behavioral uniformity and compliance through the use of social rewards, such as prestige or respect, and social disincentives such as ridicule or ostracism (Crano, 2012).

In contrast, the in-group minority position generally lacks po- wer, does not control social rewards, and is considered to be aweak position (Wood et al., 1994). However, power must not be thought of as a necessary component of social influence. Social influence, when viewed in this limited way, leaves no room for influential dissention, difference, and defiance (Jetten & Hornsey, 2011). Con- version theory suggests it is from in-group disparities and conflict that the minority derives its ability to influence. This attracts attention because it stands in contrast to the majority. A validation process is triggered where one will “examine one’s own responses, one’s own judgments, in order to confirm and validate them.one’s main preoccupation [is] to see what the minority saw, to under- stand what it understood” (Moscovici, 1980: p. 215). This process functions on both latent and manifest levels. Latent validation re- fers to an acute internal cognitive conflict that is the result of at- tempts to understand theminority position in order to convert to it. In turn, these complex internal processes can result in public manifestations of behavioral change (Moscovici and Lage, 1976). Consistency is an important aspect of minority influence. Minority influence is maximal if the position is held across time, situations, and individuals (Gardikiotis, 2011; Papastamou & Mugny, 1990). This generates confidence in the minority position and in turn enhances influence (Maass & Clark, 1984; Moscovici & Lage, 1976).

Minority-majority positions in social networks

Minority-majority influence dynamics can help us understand adolescents’ drinking behavior. A conventional view of conformity and social influence suggests being in a non-drinking minority within a friendship network places adolescents at the highest risk for the onset of drinking. Continued dissent from the in-group drinking norm is disloyal, truculent, and ironically, could be considered deviant. Similarly, the drinking adolescent who wants to stop drinking but is part of an in-group majority of drinking friends not only faces potential ostracism from the group, but also the termination of positive reinforcement for continuing compli- ance with pro-drinking norms (see Brown et al., 2008). Conformity and compliance with group norms occur because the adolescent will not risk ridicule, being labeled as disloyal, or the loss of group membership and social rewards provided by the majority. Theo- rizing from the majority-minority perspective suggests in-group

friendships may help an adolescent continue or begin to abstain from drinking even if the majority of their friendship network drinks.

The disparity in opinions within the classic studies’ treatment groups forms the basis for defining the majority and minority groups. This variation is also found in network studies of adolescent drinking behavior (see Haynie, 2002; Weerman & Smeenk, 2005). While similarity in drinking behaviors between an adolescent and his or her friends may be the result of a form of selection, this se- lection is not perfect and a social influence process still occurs (se Bauman & Ennett, 1996; Knecht, Burk, Weesie, & Steglich, 2011; Laursen et al., 2008). Adolescents may select friends who have similar drinking habits but this does not result in behaviorally homogenous network in which the social influence process “through which individuals or groups change the thoughts, feel- ings, and behaviors of others” cannot operate (Stangor, 2004; p. 84).

We expect associations with friends who drink will increase the probability of drinking for both individuals whowould be initiating drinking behavior and for individuals who already engage in drinking. Explicitly, we expect the effect of peer drinking to be robust for all adolescents. In addition, based on classic experi- mental studies of human behavior (e.g., Asch, 1951, 1952, 1955, 1956), and social psychological studies of minority influence (e.g., Moscovici, 1976, 1980; Moscovici and Faucheax, 1972) we hypoth- esize the following:

Hypothesis 1: When faced with a majority of friends who drink, the probability of time II drinking onset among time I non-drinkers will be decreased if the friendship group consists of at least one friend sharing the respondent’s non-drinking minority position.

Hypothesis 2: When faced with a minority of friends who do not drink, the probability of time II drinking desistance among time I drinkers will be increased if the friendship group consists of at least one friend in the non-drinking minority position.

Linking variation in social influence to minority-majority posi- tions in adolescents’ peer friendship networks may facilitate un- derstanding why some adolescents are able to abstain from both general and problem drinking when faced with a network that should be influencing them to conform or continue to comply with pro-drinking norms. Our study of minority influence in adolescent friendship networks moves beyond previous research regarding adolescent drinking behavior in two crucial respects. First, we tie the discussion of resistance to the group norm, adolescent friend- ship networks, and drinking to lab-based experiments of social influence and human behavior within an established theoretical framework. This provides health and social network researchers with a theoretical language currently absent in explanations of the relationship between peer influence and alcohol consumption during adolescence. Second, we explicitly acknowledge the behavioral variation amongst friends within an adolescent friend- ship network. Doing so allows us to model the positive effect that peers have on the ability to abstain alcohol use even when abstaining behaviors are not the group norm.

Data and methods

Data

The National Longitudinal Study of Adolescent Health (“Add Health”) is a school-based multi-wave panel study of how indi- vidual, family, friend, and school factors contribute to youth development. The data consist of several components, including a school administrator questionnaire, an extensive in-school survey, and an adolescent inehome interview. The school administrator questionnaire represents administrators from 132 middle and high schools in the United States. Schools were sampled to assure

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 37

representativeness on size, school type, geographic region, urban- ization, and racial composition; the selection probability was pro- portional to the size of the school. From the selected schools, a stratified sample of 90,118 students enrolled in grades 7e12 at the onset of the study were eligible to complete an extensive in-school survey in 1994 (time I). A random sample of 20,745 students was then selected to complete an inehome interview in 1995 (time II). A total of 14,396 students have the required sample weights at both times I and II. These weights ensure the proper calculation of standard errors and significance tests during analysis.

During the in-school interviews, respondents, regardless of sex, identified up to 5 female and 5 male friends from a name roster of all students enrolled in their schools. This allowed for the direct measurement of the level of alcohol use for each named friend who attended the same school as the respondent (and who, therefore, was also a respondent in the in-school dataset).1 A total of 944 students did not meet the Add Health analysis guidelines for the computation of social network measures. Therefore, the initial core sample size resulted in a sample size of 13,451 cases (see Chantala & Tabor, 2010).

We next restricted our data to respondents who completed the in-school and inehome interviews, have valid data on the dependent variables at both of these time points, have at least one nominated friend with valid alcohol use data on the in-school questionnaire, and have non-missing sample weights (n ¼ 10,945 from 125 schools). Missing data was imputed chained equations in order to maintain statistical power and avoid po- tential estimate biases resulting from the list-wise deletion of cases.2 We identified all respondents who reported not drinking in the past 12 months at the time I interview and who have at least three friends. Identifying these individuals allows us to capture possible onset of alcohol use at time II while allowing for all possible minority-majority positions (see below) at time I. This resulted in a sample size of 4008 cases within 121 schools. Additional analyses required the identification of all time I re- spondents who reported drinking in the past 12 months and met the three friend criteria. This allowed us to capture the termina- tion of alcohol use at time II. The final sample size for these particular analyses was 4765 cases within 121 schools. All analyses corrected for the design of the Add Health survey data by using the primary sampling unit, stratum, and inehome grand sample weights (see Chantala & Tabor, 1999).

Measures

Outcome variables We examined the onset and continued involvement in three

different types of alcohol consumption for time I non-drinking and drinking respondents: drinking alcohol, getting drunk, and binge drinking. At the inehome interview (time II), respondents were asked on how many days in the previous 12 months they drank alcohol (range ¼ 0 “never”e6 “every day or almost every day”); on

1 The Add Health data have two other options for identifying friends: the receive network, which consists of all individuals who named the respondent as a friend; and the nomination and receive network, which consists of all individuals who either the respondent named as a friend or who named the respondent as a friend. The ensuing analyses use the send nomination network.

2 Inm ¼ 20 imputed data sets, we imputed data for all control variables: parental attachment (6.9%), friend attachment (0.75%), school attachment (8%), GPA (12.4%), age (0.3%), parental education (3.8%), two parents at home (4.0%), public assistance (14.8%), and race (2.3%). We opted not to impute data on the dependent variables or the peer alcohol use measure, but we included these variables as well as the Add Health survey design variables in our imputation models as auxiliary variables (see Enders, 2010; Reiter, Raghunathan, & Kinney, 2006; Von Hippel, 2007).

how many days in the past 12 months they had gotten drunk or “very, very high” on alcohol (range ¼ 0 “never”e6 “every day or almost every day”); and on how many days during the past 12 months they had five or more drinks in a row (range¼ 0 “never”e6 “every day or almost every day”). Higher values on each dependent variable reflect higher levels of the measured behavior. See Appendix A for descriptive statistics of all study variables.

Measures of minority-majority positions We counted the number of each respondent’s friends who re-

ported having engaged in drinking at least once during the past 12 months and the number of friends who reported abstaining from drinking at time I. This measure is comparable to those used in socio-psychological studies and consistent with prior studies of adolescent delinquency and alcohol use (see Haynie, 2001, 2002; Payne & Cornwell, 2007).

Studies of adolescent alcohol use have tended to capture the social norms of the peer drinking behavior as the number of friends using, the proportion of friends using (Windle et al., 2008), or the average level of alcohol use amongst friends (Ali & Dwyer, 2010; Crosnoe et al., 2004; Rees & Pogarsky, 2011). These measures implicitly ignore the behavioral heterogeneity found within friendship networks and assume equal influence and norm consensus (Burt & Rees, 2014). For example, the following network scenarios each with 3 total nominated friends: 3 friends who each report having 2 drinks in the past 30 days (2 þ 2 þ 2/3 ¼ network avg. of 2 drinks); a friend who had 1 drink, another friend at 2 drinks, and a friend at 3 drinks (1 þ 2 þ 3/3 ¼ network avg. of 2 drinks); 2 friends with 0 drinks, 1 friend at 6 drinks (0 þ 0 þ 6/ 3 ¼ network avg. of 2 drinks); 1 friend with 0 drinks, another at 4 drinks, and a friend at 2 drinks (0 þ 4 þ 2/3 ¼ network avg. of 2 drinks); and 1 friend with 0 drinks, another at 3 drinks, and a friend at 3 drinks (0 þ 3 þ 3/3 ¼ network avg. of 2 drinks). It is clear that there is disagreement amongst group members about the appro- priateness of drinking even though the intensity of peer drinking remains the same in each context.3 The result of capturing the group norm in this manner is the appearance of complete unifor- mity in adherence to norms within a network (Rossi & Berk, 1985). Other scenarios are possible across different sizes of networks.

This leads us to our operationalization. We constructed a series of dichotomous variables, each capturing one of the following six minority-majority positions an adolescent could face within his or her friendship group. The majority positions include: (1) Fully united non-drinking friends: the adolescent and his or her friends are all non-drinkers, the majority behavior is that of non-drinking (n ¼ 694); (2) In majority with friends: the adolescent has a ma- jority of friends who are non-drinkers but aminority of friends who drink (n ¼ 1610). Next, the minority positions include: (1) Stands alone: the adolescent is the only non-drinker, the adolescent alone faces a unanimous majority of drinkers (n ¼ 340); (2) In Minority with a single friend sharing their behavioral position: the adolescent has one friend who is also a non-drinker and thus represents the minority while the rest of the group is a drinking majority (n ¼ 567); (3) In Minority with two or more friends sharing their behavioral position: the adolescent has two or more non-drinking friends in the minority but the majority of friends are drinkers (n ¼ 698). Our last dichotomous variable is that there are no ma- jority or minority positions in the network: the adolescent faces an even balance of drinkers and non-drinkers (n ¼ 645). These terms match the experimental situations faced by the targets of influence in the classic lab experiments of social influence (Asch, 1956).

3 See Burt and Rees (2014) for similar arguments.

4 Furthermore, the regression coefficients in the binary portion of zero-inflated count models provide inference for the odds of observing an excess zero not the decision to engage or not engage in a particular behavior (see Loeys et al., 2012). Lastly, we note that a negative binomial hurdle model is inappropriate because of under-dispersion in the outcomes indicated by the unit variance to mean ratio.

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e4538

For our time I drinking sample, the majority positions include: (1) Fully united with drinking friends: the adolescent and his or her friends are all drinkers, the majority behavior is that of drinking (n ¼ 1033); (2) In Majority of drinkers but with a single non-drinking friend: the adolescent has one friend who is a non-drinker who represents the minority while the respondent is in the majority with the rest of his or her friends (n ¼ 1220); (3) In Majority of drinkers but with at least two non-drinking friends: the adolescent has at least two non-drinking friends who represent the minority while the respondent is in the majority with the rest of his or her drinking friends (n ¼ 1167). Next, the minority positions include: (1) Stands alone: the adolescent is the only drinker, the adolescent alone faces a unanimous majority of non-drinkers (n ¼ 116); (2) In the Minority with at least one drinking friend: the adolescent has two or more drinking friends in the minority but the majority of friends are non-drinkers (n ¼ 849). Our last dichotomous variable is that there are no majority or minority positions in the network: the adolescent faces an even balance of drinkers and non-drinkers, there is no clear behavioral majority or minority (n ¼ 380).

Other variables The following demographic variables were measured: sex

(1 ¼ Male), age in years, and race/ethnicity (White, Black, and “Other”). Family-related variables include parental attachment, family structure, and parental education, measured at time I, and receipt of public assistance (1 ¼ Receiving public assistance), measured at time II. Parental attachment (Hirschi, 1969) is the average of respondents’ responses (on a scale from 1 “not at all” to 5 “very much”) to two questions measuring how much their mothers and fathers cared about them (i.e., “How much do you think he/she cares about you?”). Family structure is a dichotomous variable rep- resenting the presence of two parental figures (versus one parental figure) in the home (biological mother/father or biological parent/ stepparent). Parental education is a respondent-reportedmeasure of whether the respondent’s mother completed high school. Public assistance represents the link between household socio-economic status and deviance (Elliott, Huizinga, & Ageton, 1985), reflects receiving at least one formof household public assistance during the month preceding the time II interview: social security or railroad retirement, supplemental security income (SSI), Aid the Families with Dependent Children (AFDC), food stamps, unemployment or worker’s compensation, or some form of housing subsidy.

Individual-level, school-related variables measured at time I include school attachment and GPA. School attachment is measured as the average response to three questions on a five- point scale from 1 (“not at all”) to 5 (“very much”): “I feel close to people at this school,” “I feel like I am part of this school,” and “I am happy to be at this school.” GPA, calculated using the average of a respondent’s English/Language Arts, Mathematics, Social Sci- ence/History, and Science grades from the most recent grading period at the time of the in-school interview, ranges on a continuous and roughly normally distributed scale from 1 “D or lower” to 4 “A.”

We also control for friend attachment at time II, whichmeasures each respondent’s response to the question: “How much do you feel your friends care about you”? Finally, we control for friendship group size and days between the in-school (time I) and inehome (time II) surveys.

Statistical analysis The outcomes of drinking, getting drunk, and binge drinking are

highly skewed and contain a large proportion of zeros. We there- fore use Poisson-Logit hurdle regression models to deal appropri- ately with these distributional characteristics. Hurdle models are similar to zero-inflated count models; they are both two part

models that express a binomial probability and a count model (McDowell, 2003). However, unlike zero inflated count models that assume there is a group of individuals who are never at risk of experiencing the outcome of interest, hurdle models, instead, as- sume that all members of the sample are “at-risk” for the outcome event (see Bandyopadhyay, DeSantis, Korte, & Brady, 2011; Loeys, Moerkerke, De Smet, & Buysse, 2012). As such, hurdle models are both a binary probability model and a truncated-at-zero count model. When applied to the current study, this assumes time I non- drinkers are at-risk of becoming drinkers at time II. Similarly, time I drinkers are at risk of continuing to consume alcohol. We find this conceptually practical given our sample respondents are school based adolescents with at least three friends. Even if a non-drinker had only non-drinking friends, friends of friends can have a sig- nificant impact on an adolescent’s delinquent behavior (Payne & Cornwell, 2007; Urberg, 1992).

Hurdle models assume a two-stage process on the part of the respondent making the decision to drink (see Cameron & Trivedi, 1998). The first step is a binary decision between no drinking (0) and some drinking (1) modeled as a logit regression. This step can be considered the decision to participate or not participate in drinking alcohol. The second step assumes the participation “hurdle” (i.e., deciding to drink) has been crossed. A zero- truncated count model then focuses on those respondents with non-zero counts and models the amount of consumption condi- tional upon the participation decision. That is, the process only begins generating positive counts after crossing a zero barrier or hurdle (Hilbe, 2011).4 For example, one of our dependent variables is how many days in the previous 12 months the respondent drank alcohol coded 0 “never” to 6 “every day or almost every day”. A two-part hurdle model would estimate first the binary outcome represented by 0 (i.e., never drink) and 1 (i.e., any level of drinking). The second part of the hurdle model, or the Poisson, would model the outcome as counts, specifically coded as 1e6, or very little drinking to drinking every day. Thus, we gain an un- derstanding of what social network contexts influence the deci- sion to drink as well as what social network contexts influence how much someone drinks, once they become drinkers. This logic will be applied to our other two outcomes getting drunk and binge drinking.

Results

Table 1 shows the most common behavioral position for time I non-drinking adolescents is in the majority with non-drinking friends (40%). These individuals may have a few friends who drink, but they are not exposed to a majority of drinkers. Similarly, we see that an undisputedmajority of non-drinkers is also common (12%). Notably though, the three minority positions in the data make up nearly 41% (combined) of all the behavioral positions we typify. First, individuals who stand alone in a unanimous network of drinkers consist of 5.81% of the respondents. Also, nearly 14% of the respondents are in the non-drinkingminority positionwith one friend. Lastly, being in the non-drinking minority positionwith two or more friends occurs among approximately 7% of the re- spondents. With these last two categories, we see that the minority positions most common are those that include a small group of people. Consequently, this shows that dissention from the majority

Table 2 Poisson-logit hurdle models for time I non-drinkers: time II respondent self- reported delinquency regressed on time I social scenario and controls.

Drinking Getting drunk Binge drinking

B B B

Model type: Logit Male �0.162 �0.206 �0.092 Age 0.067 0.150* 0.141* Black �0.353* �0.789* �0.677 Parental attachment �0.158 �0.389* �0.236 Two parents at home �0.054 �0.227 0.055 Parental education 0.096 0.057 �0.474 Public assistance �0.513 �0.008 �0.388 Respondent GPA �0.249** �0.326* �0.375* School attachment �0.106 �0.119 �0.250* Friend attachment 0.178* 0.229 0.135 Network size 0.082* 0.057 �0.012 All friends non-drinkers �2.145*** �3.690*** �2.885*** 50% friends have drank;

50% have not �1.075** �1.343** �1.085*

Majority friends have drank; 1 has not

�0.391 �0.988** �0.985*

Majority friends have drank; 2þ have not

�0.758** �1.122** �0.874*

Majority friends non-drinkers; minority have drank

�1.120*** �1.501*** �1.192**

Constant �5.302*** �4.779*** �4.220*** Model type: Poisson Male 0.11 0.063 0.008 Age �0.024 �0.084 0.059 Black �0.014 0.205 �0.262 Parental attachment 0.045 �0.007 �0.05 Two parents at home �0.082 0.184 0.02 Parental education 0.033 �0.18 �0.317 Public assistance 0.247 �0.264 0.14 Respondent GPA �0.156 �0.245 �0.241 School attachment �0.001 �0.165 0.223 Friend attachment 0.12 0.278 0.077 Network size �0.038 �0.064 �0.067 All friends non-drinkers �0.588 1.014 0.611 50% friends have drank;

50% have not 0.247 0.398 �0.166

Majority friends have drank; 1 has not

�0.047 �0.123 �0.151

Majority friends have drank; 2þ have not

0.003 0.123 �0.117

Majority friends non-drinkers; minority have drank

�0.125 0.04 �0.027

Constant �5.453*** �5.046*** �4.614***

Notes: N ¼ 4008 for all models, which also control for time between surveys and “Other” race. The reference category for the social scenario variables is “Respondent is lone non-drinker”. Abbreviations: B ¼ Log expected count of response for drinking, getting drunk, and binge drinking models; GPA ¼ Grade Point Average. *p < 0.05, **p < 0.01, ***p < 0.001.

Table 1 Descriptive statistics for the minority influence categories.

Frequency %

Minority influence categories: T1 non-drinkers In unanimous majority of non-drinkers 498 12.43% Respondent is lone non-drinker 233 5.81% 50/50 split 402 10.03% In non-drinking minority with

1 non-drinking friend 567 14.15%

In non-drinking minority with 2þ non-drinking friends

698 17.42%

In non-drinking majority; drinking friends minority

1610 40.17%

N 4008 100 Minority influence categories: T1 drinkers In unanimous majority of drinkers 1033 21.68% Respondent is lone drinker 116 2.43% 50/50 split 380 7.97% In drinking majority; 1 non-drinking friend 1220 25.60% In drinking majority; 2þ non-drinking friends 1167 24.49% In drinking minority; non-drinking friends majority 849 17.82% N 4765 100

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 39

is frequent (nearly 41%) and occurs more frequently in small groups.

Conversely, the most common behavioral position for time I drinking adolescents is those adolescents in the majority with drinking friends but with a few non-drinking friends or fully united with other drinkers (a combined percent of 71%). Each of these categories represent almost a quarter of the time I drinkers. Alter- natively, the twominority positions in the datamake up only 20% (a combined percent) of all the behavioral positions we typify. First, very few drinkers are behaviorally isolated within their friendship network; lone drinkers are approximately 2.43% of the sample. Next, we see that being in network situation where there are behavioral minorities regarding drinking is also common, with 17.82% of the drinking respondents falling into this category. Similar to time I non-drinkers, we see that the minority positions most common are those that include a small group of people. This variation in behaviors is supported in other peer network research (Haynie, 2002) and may be expected given that selection of friends can be based on characteristics other than peer behavior (McPherson, Smith-Lovin, & Cook, 2001).

Table 2 presents estimates from hurdle models for time I non- drinkers from three models regressing self-reported drinking, getting drunk, and binge drinking on the social scenario variables, controlling for individual-, family-, and school-related covariates. Thus these models will help us to understand the social network contexts that facilitate non-drinkers to either (a) decide to drink and (b) the level at which they decide to drink for our three outcomes. We note here that we also conducted sensitivity ana- lyses (see Supplemental Materials) that also controlled for the level at which drinking friends drink at time I. These results parallel closely with those presented in the manuscript in Tables 2and 4 in terms of substantive interpretations and signif- icance tests.

Demographic and control variables in the logit portion of the hurdle model show that age is positively related to the onset of getting drunk and binge drinking (0.150, p < 0.05; 0.141, p < 0.05) and Black youths have lower odds of beginning to drink or getting drunk (exp(�0.353)¼ 0.70, p< 0.05; exp(�0.789) ¼ 0.45, p< 0.05) than are White youths. For the family-related variables, the results show that respondents are less likely begin getting drunk (�0.389, p < 0.05) when they have higher levels of parental attachment. Parental education and public assistance are generally unrelated to

the outcomes. Of the school-level variables, respondent grade point average is inversely associated with onset of alcohol use and abuse (�0.249, p < 0.01 for drinking; e0.33, p < 0.05 for getting drunk; and�0.37, p< 0.05 for binge drinking). Finally, friend attachment is unrelated to the onset of alcohol use, however, school attachment decreases the likelihood of binge drinking onset (�0.250, p < 0.05). Overall, the findings conform to previous research (Elliott et al., 1985; Hirschi, 1969; Maguin & Loeber, 1996; Nagin & Paternoster, 2000; Sampson & Sharkey, 2008; Warr, 2002).

Non-drinkers: resisting majority influence

In Table 2, shows the relationship between our six minority- majority positions faced by non-drinking adolescents across our

Table 3 Social situation coefficient differences for time I non-drinkers.

Drinking Getting Drunk

Binge Drinking

B-difference B-difference B-difference

All friends non-drinkers versus: 50% friends have drank; 50% have not

�1.07** �2.34** �1.80**

Majority friends have drank; 1 has not

�1.75*** �2.70*** �1.90**

Majority friends have drank; 2þ have not

�1.39*** �2.57*** �2.01**

Majority friends non-drinkers; minority have drank

�1.04** �2.19** �1.69**

50% friends have drank; 50% have not versus: Majority friends have drank; 1 has not

�0.68* �0.35 �0.09

Majority friends have drank; 2þ have not

�0.32 �0.22 �0.21

Majority friends non-drinkers; minority have drank

0.03 0.158 0.11

Majority friends have drank; 1 has not versus: Majority friends have drank; 2þ have not

0.37* 0.13 �0.11

Majority friends non-drinkers; minority have drank

0.71*** 0.51 0.21

Majority friends have drank; 2þ have not versus: Majority friends non-drinkers; minority have drank

0.34* 0.38 0.32

Table 4 Poisson-logit hurdle models for time I drinkers: time II respondent self-reported delinquency regressed on time I social scenario and controls.

Drinking Getting drunk Binge drinking

B B B

Model type: Logit Male �0.191 �0.094 0.201 Age 0.166*** 0.187*** 0.210*** Black �0.737*** �0.932*** �1.177*** Parental attachment �0.016 0.072 �0.062 Two parents at home �0.16 �0.328** �0.277* Parental education 0.355* 0.328* 0.113 Public assistance �0.181 �0.054 �0.106 Respondent GPA 0.169** �0.038 �0.08 School attachment �0.074 �0.087 �0.006 Friend attachment �0.03 0.072 0.074 Network size 0.012 0.032 �0.004 T1 Respondent drinking 0.396*** 0.534*** 0.523*** Lone drinker in network �1.128*** �1.646*** �0.789* 50% friends have drank;

50% have not �0.898*** �0.896*** �0.763***

Majority friends drank; 1 has not

�0.249 �0.485*** �0.454***

Majority friends have drank; 2 þ have not

�0.392** �0.725*** �0.435***

Majority friends non-drinkers; minority have drank

�1.020*** �1.328*** �1.214***

constant �4.712*** �6.328*** �5.896*** Model type: Poisson Male 0.118** 0.182** 0.200*** Age 0.044*** 0.015 0.013 Black 0.073 0.03 0.153 Parental attachment �0.011 �0.023 0.035 Two parents at home �0.048 �0.008 �0.028 Parental education �0.035 0.006 �0.095 Public assistance 0.02 0.187* 0.17 Respondent GPA 0.006 �0.074* �0.06 School attachment 0.016 0.03 0.014 Friend attachment �0.022 �0.064 �0.041 Network size 0.022* 0.01 0.026* T1 Respondent drinking 0.206*** 0.206*** 0.176*** Lone drinker in network 0.202 0.522 0.002 50% friends have drank;

50% have not �0.180* �0.224 �0.111

Majority friends drank; 1 has not

�0.103** �0.095 �0.079

Majority friends have drank; 2þ have not

�0.181** �0.104 �0.169*

Majority friends non-drinkers; minority have drank

�0.271*** �0.319** �0.269**

constant �5.220*** �5.093*** �5.207***

Notes: N ¼ 4765 for all models, which also control for days between surveys and “Other” race. The reference category for the social scenario variables is “In Unani- mous Majority of Drinking Friends”. Abbreviations: B ¼ Log expected count of response for drinking, getting drunk, and binge drinking models; GPA ¼ Grade Point Average. *p < 0.05, **p < 0.01, ***p < 0.001.

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e4540

three alcohol use and abuse outcomes. Remember that the logit models estimate never drank versus. any level of drinking while the Poisson models estimate drinking levels from 1 to 6, once an in- dividual has decided to become a drinker. Additionally, note that non-drinking adolescents with a friendship group consisting of a unanimous majority of drinkers is the reference category so that we could determine if being in one of the five social scenarios predicts a different alcohol use or abuse onset level than being in the reference category.

We begin by interpreting the coefficients in the logit models our two focal minority categories: (a) majority of friends have drunk, 1 has not and (b) majority of friends have drank, 2þ have not. The odds of participating in any drinking for a non-drinking adolescent with a majority of friends who drink but have one non-drinking friend is not significant; being in the minority for this drinking behavior does not impact the likelihood of taking up drinking at time II. However, consistent with our motivating hy- pothesis, the odds of participating in getting drunk for a non- drinking adolescent with a majority of friends who drink but have one non-drinking friend are decreased by 63% (100(1�exp(�0.99))) compared to non-drinkers with a unani- mous majority of drinking friends. Similarly, the odds of binge drinking are decreased by 62% (100(1�exp(�0.98))). For the category of a non-drinking adolescent with two or more non- drinking friends but a majority of drinking friends, we see that for these individuals, the likelihood of drinking decreased by 53% for drinking onset (100(1�exp(�0.758))); 67% for getting drunk (100(1�exp(�1.122))); and 58% for binge drinking (100(1�exp(�0.874))) compared to the reference category of being the lone non-drinker in the friendship group.

Overall, our results suggest non-drinking friends in the behavioral minority can decrease the odds of a non-drinker’s decision to consume any alcohol even when facing an in-network drinking majority. A single non-drinking friend in the behavioral minority significantly decreases the odds of getting drunk and

binge drinking while two non-drinking friends does so for all outcomes.

Next, being in a friendship group with all non-drinking friends results in a significantly lower likelihood of drinking onset compared to the reference category (�2.14, p< 0.01). This finding is consistent for the outcomes of getting drunk (�3.69, p < 0.01) and binge drinking (�2.88, p < 0.01). This pattern holds for being in a friendship group with an even number of non-drinking and drinking friends (�1.07, p < 0.01 for drinking; �1.34, p < 0.01 for getting drunk; �1.08, p < 0.01 for binge drinking) and also for having a majority of non-drinking friends along with a minority of friends who drink (�1.12, p < 0.01 for drinking; �1.50, p < 0.01 for getting drunk; �1.19, p < 0.01 for binge drinking).

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 41

In contrast, the Poisson portion of the hurdle model in Table 2 indicates conditional alcohol onset, there are no significant dif- ferences in the amount a lone non-drinker consumes alcohol compared other social scenarios. This is consistent across all outcomes. That is, given a non-drinking respondent has experi- enced the onset of alcohol use or abuse, their risk of engaging in more of these behaviors is not significantly different from those with any number of non-drinking friends. For adolescents who initially identify as non-drinkers, being in the minority position does not seem to affect level of drinking once drinking has begun.

Table 3 presents results from Adjusted Wald tests of coefficient differences for all social scenarios within the logit model. First, as expected, being in an undisputed majority of non-drinkers significantly decreases the odds of alcohol use and abuse compared to all other social scenarios. Next, having an equal number of drinking and non-drinking friends lowers the odds of drinking onset by half (exp(�0.68)) compared to being in the non- drinking minority with one friend; this is not found for getting drunk or binge drinking or for other social scenarios. Next, being in the minority with a single non-drinking friend increases the odds of drinking onset (0.37, p < 0.05) compared to having at least 2 non-drinking friends. Similarly, having a minority of drinking friends by increases the odds of onset by a factor of 2.03 (exp(0.71), p < 0.001). Lastly, adolescents with at least 2 friends in the non-drinking minority are 1.40 times (exp(0.34), p < 0.01) as likely to experience drinking onset compared to those with a majority of non-drinking friends and a minority of drinking friends.

Drinkers: minority influence on the majority

Table 4 presents the relationship between the six minority- majority positions faced by non-drinking adolescents across the three alcohol use and abuse outcomes. Again, the logit models estimate having drank before versus. any level of drinking while the Poisson models estimate drinking levels from 1 to 6. The reference social scenario here is being in the unanimous majority of drinkers. We also include a dummy variable that signals that the respondent is the only drinker in their network. Lastly, these models also control for the level of respondent time I drinking.

The logit results are consistent with our second hypothesis. The odds of getting drunk are decreased by 38% (100(1�exp(�0.485)), p < 0.001) and binge drinking by 36% (100(1�exp(�0.454)), p < 0.001) if a drinking respondent is in a majority of drinkers but with one non-drinking friend compared to being in a unanimous majority of drinkers. However, a single non-drinking friend in a minority position does not decrease the odds of the decision to participate in drinking (exp(�0.249)¼ 0.78, p> 0.05). Respondents in the drinking majority and having two or more non-drinking friend decreases the odds of drinking (exp(�0.39) ¼ 0.67, p< 0.01); getting drunk (exp(�0.725)¼ 0.48, p< 0.001); and binge drinking (exp(�0.435) ¼ 0.65, p < 0.001).

When examining lone drinkers, being the only drinker in a network decreases the odds of drinking by 68% (100(1�exp(�1.128, p < 0.001))), getting drunk by 80% (100(1�exp(�1.646, p < 0.001))), and binge drinking by 54% (100(1�exp(�0.789, p < 0.001))). Friendship groups with an even number of non- drinking and drinking friends decrease the odds of drinking (�0.898, p < 0.001); getting drunk (�0.896, p < 0.001); and binge drinking (�0.763, p < 0.001). Similarly, having a majority of non- drinking friends along with a minority of friends who drink decrease the likelihood of drinking for all outcomes (�1.02,

p < 0.001 for drinking; �1.328, p < 0.01 for getting drunk; �1.214, p < 0.01 for binge drinking).

The Poisson section of the hurdle model measures the level of alcohol consumption contingent on the decision to continue to participate in drinking. Having at least one non-drinking friend decreases the rate at which respondents drink by 10% (100(1�exp(�0.103))). A majority of non-drinking friends with a drinking minority decreases the rate at which respondents drink (�0.271, p < 0.001). Two non-drinking friends in the majority decrease the rate by a factor of 0.83. An even split of drinkers and non-drinkers also decreases the rate at which drinkers continue to consume alcohol (�0.180, p < 0.01). However, drinkers with a majority of non-drinking friends but a minority of drinking friends is the only social scenario which decreases the rate at which respondents get drunk (�0.319, p < 0.01). This social scenario has a similar affect for binge drinking (�0.269, p < 0.01) along with results showing at least two or more friends in the non-drinking minority decrease the level of binge drinking (�0.169, p < 0.01).

Next, Table 5 presents results from Adjusted Wald tests of coefficient differences for all social scenarios within time I drinker hurdle modeldboth logit and Poisson. Here, we focus on discussing the differences of particular theoretical importance for the paperdnamely a “majority of friends have drank, 1 has not” and a “majority of friends have drank, 2 þ have not”dthough all results are presented in the table. We begin with the results from the logit models. First, lone drinkers have lower odds of continuing to drink at time II than individuals with a majority of friends that have drank, though 1 has not (�0.88). This result is similar for getting drunk, but not binge drinking which is non- significant. Lone drinkers have lower odds of continuing to drink at time II than individuals with a majority of friends that have drank, though 2 or more have not (�0.74). Again this holds for getting drunk, but not binge drinking. Individuals with an even number of drinkers and non-drinkers in a network see decrease the likelihood of drinking (�0.65, p < 0.001) and getting drunk (�0.41, p < 0.05) but not binge drinking compared to having a majority of friends who drink but one non-drinking friend. Increasing that minority to two or more friends also shows a decrease in the likelihood of drinking, but shows no significant differences for getting drunk or binge drinking. Drinkers have a higher likelihood of alcohol use and abuse if they have a majority of drinking friends but one non-drinking friend compared to a majority of drinkers but minority non-drinkers (0.77, p < 0.001 for drinking; 0.84, p < 0.001 for getting drunk; 0.76, p < 0.001 for binge drinking). Similar results are found for having two or more non-drinking friends in the minority (0.63, p < 0.001 for drinking; 0.60, p < 0.001 for getting drunk; 0.78, p < 0.001 for binge drinking).

The Poisson model in Table 5 indicates, conditional upon the decision to participate in drinking, lone drinkers in non-drinking networks have an increased drinking consumption levels compared to individuals in networks where there is one non- drinker in the minority (exp(0.31) ¼ 1.36, p < 0.05) or two or more non-drinkers in theminority (exp(0.38)¼ 1.46, p< 0.05). This effect is also seen for getting drunk, but not binge drinking. We have seen no significant differences from individuals in an even drinking network (50e50 split) and individuals who are in a network that consists of a majority of friends who have drank, though 1 has not or similarly, where two people have not drank; this is across all outcomes. Lastly, drinkers with a minority of one non-drinking friend show higher levels of alcohol use and abuse across all outcomes (0.17, p < 0.05 for drinking; 0.22, p < 0.05 for getting drunk; 0.19, p < 0.05 for binge drinking). The remaining comparisons outcomes show no significant differences.

Table 5 Social situation coefficient differences for time I drinkers.

Drinking Getting drunk

Binge drinking

B- difference

B- difference

B- difference

Logit Lone drinker in network versus: 50% friends have drank, 50% have not

�0.23 �0.75* �0.03

Majority friends have drank, 1 has not

�0.88** �1.61** �0.33

Majority friends have drank, 2þ have not

�0.74** �0.92* �0.35

Majority friends non-drinkers, minority have drank

�0.11 �0.32 0.425

50% friends have drank, 50% have not versus: Majority friends have drank, 1 has not

�0.65*** �0.41* �0.31

Majority friends have drank, 2þ have not

�0.51** �0.17 �0.33

Majority friends non-drinkers, minority have drank

0.12 0.43* 0.45*

Majority friends have drank, 1 has not versus: Majority friends have drank, 2þ have not

0.14 0.24 �0.02

Majority friends non-drinkers, minority have drank

0.77*** 0.84*** 0.76***

Majority friends have drank, 2þ have not versus: Majority friends non-drinkers, minority have drank

0.63*** 0.60*** 0.78***

Poisson Lone drinker in network versus: 50% friends have drank, 50% have not

0.38* 0.75* 0.11

Majority friends have drank, 1 has not

0.31* 0.62* 0.08

Majority friends have drank, 2þ have not

0.38** 0.63* 0.17

Majority friends non-drinkers, minority have drank

0.47** 0.84** 0.27

50% friends have drank, 50% have not versus: Majority friends have drank, 1 has not

�0.08 �0.13 �0.03

Majority friends have drank, 2þ have not 0.00 �0.12 0.06 Majority friends non-drinkers, minority have drank

0.09 0.09 0.16

Majority friends have drank, 1 has not versus: Majority friends have drank, 2þ have not

0.08 0.01 0.09

Majority friends non-drinkers, minority have drank

0.17* 0.22* 0.19*

Majority friends have drank, 2þ have not versus: Majority friends non-drinkers, minority have drank

�0.08 0.22 0.1

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e4542

Discussion

Alcohol use among adolescents has serious health conse- quences, therefore understanding the correlates of adolescent drinking is amajor health concern in the United States (Ali & Dwyer, 2010). Previous research demonstrates that alcohol use is devel- opmentally normative for adolescents and profoundly influenced by peers (Chassin, Pitts, & Prost, 2002; Clark & Lohéac, 2007). Our study expands upon socio-psychology experiments on conformity to examine real-world adolescent friendship networks and the potential positive impact of peers on an adolescent’s ability to abstain from alcohol use and abuse. We find that in the face of a group majority that promotes drinking, other in-group peers may facilitate individuals’ abstinence from alcohol use or decrease use of alcohol.

First, individuals in the minority alone who face a majority of drinkers are significantly more likely to drink than individuals in other minority-majority positions. Effectively, we see the classic effects of peer influence here. With little power to exert their in- fluence or opinion, minority position non-drinkers who stand alone are more likely to begin drinking, and engage in varying levels of drunkenness and binge drinking. However, we also see ev idence that people can resist the onset of alcohol use and problem alcohol use when they are in the minority position and have a friend or two who are also resisting alcohol use. This finding is consistent with lab-based experiments on social influence (Asch, 1951, 1952, 1955, 1956; Moscovici, 1976, 1980). In these situa- tions, the non-drinking adolescent and friends are the behavioral deviants of the group. Deviating from the norm is not limited to the delinquent adolescent intent on pushing back against con- ventional society.

The non-conformity displayed by the adolescents in our study is the type of behavior encouraged by parents and schools. This ability is not necessarily surprising given adolescence is a time of identity development and establishing autonomy from parents (Bagwell & Schmidt, 2011; Brechwald & Prinstein, 2011). When establishing an independent but social self, adolescents select friends for a va- riety of reasons (McPherson et al., 2001). This selection results in a mix of drinking and non-drinking friends. For a non-drinking adolescent, support from other non-drinking friends appears to be enough to significantly dampen the effect of the drinking ma- jority compared to adolescents who are the lone non-drinker in a network.

Minority influence is also impacts adolescents that do drink. Drinkers in the majority at time I have a lower likelihood of continued drinking, getting drunk or binge drinking if their in- group includes a minority of non-drinkers. This effect persisted when examining levels of drinking behavior: adolescents with minority groups in their network (i.e., people who did not drink) have lower levels of drinking behavior (not getting drunk or binge drinking) than adolescents with other social situations. Drinking adolescents in the majority have non-drinkers as friends, thus allowing the minority to significantly dampen the probability of continuation and levels at which drinkers continue alcohol con- sumption. This is consistent with conversion theory’s proposition of the minority’s ability to influence those in the majority. Our study also provides evidence that behavioral consistency across minority members increased the probability of abstention or desistance. Also, increasing the number of non-drinkers in the minority, thereby increasing consistency in the minority message, generally increased the probability of drinking desistance and lowered levels of consumption.

These findings strongly parallel those of past experimental research on minority influence, but we acknowledge there are limitations to our study. We note that unpacking the black box of causal mechanisms is a difficult task within the non-experimental social sciences and must be pursued with caution (Cialdini & Goldstein, 2004). It is often unclear which one of a number of proposed mechanisms “is responsible for observed behavior, or even if the different mechanisms are analytically distinct” (Watts & Dodds, 2009: 4; see also Cialdini & Goldstein, 2004). While conversion theory is no exception, Moscovici (1980) offers a rationale for the mechanisms by which majority and minority influence operate. He concedes that compliance to the majority may simply be a product of inter-personal attraction. A sensible proposition given the majority is often seen as correct and in control of social rewards. However, the minority position is generally disliked, unattractive, and thought of as incompetent. Moscovici (1980) therefore reasons conversion mechanisms of minority influence cannot be confounded by attraction or liking.

N Mean SD Range

Dependent variable Respondent drinking time II 4008 0.29 0.80 0e6 Respondent drunkenness time II 4008 0.10 0.49 0e6 Respondent binge drinking time II 4008 0.10 0.51 0e6 Independent variables Fully united non-delinquent 4008 0.12 0.33 0e1

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 43

As such, the influence of the minority position cannot be explained away to the simply push and pull of attraction within friendships. Next, we take a wide, sociological approach towards defining friendship. Like many adolescent drinking and network studies, our definition of friendship is respondent driven (see Fujimoto & Valente, 2012a for a review). A strict psychological definition of friendship would suggest that both people recognize that the friendship exists for it to exert influence. While a concern, there is ample evidence suggesting any friendship, even those that are not reciprocated, exert influence on drinking behaviors (Fujimoto & Valente, 2012a).

Adolescent social networks can change over time, and poten- tially, substance use and friendship ties change with it. While we have longitudinal data and control for days between waves, we see value in further developing conversion theory’s concept of persis- tence in minority behavior over time. Diffusion of innovations (Valente, 1996; Rogers, 2003) seems particularly relevant in terms of individual thresholds for adoption, social contexts and how adoption processes vary across time.

Add Health includes only self-reports of drinking behavior. While studies have demonstrated that self-reported use of alcohol is a valid measure (Campanelli, Dielman, & Shope, 1987; Del Boca & Darkes, 2003), others show concern over circumstance that could change the veracity of individuals’ reports (Del Boca & Darkes, 2003). Furthermore, our definition of the minority position is numerically based and consistent with most studies of minority influence. However, other definitions of minority status based on power, status, or even outrageousness can be used (see Crano, 1994 for review). Furthermore, numerically based classifications of mi- nority positions by definition lead to less exposure to levels of drinking because the minority positions include non-drinkers and thereby pull down the overall network average level of drinking. Sensitivity analyses presented in Tables 2b and 4b (see Appendix C) also control for the average level of drinking by friends who drink and these results parallel those found in Tables 2 and 4

Lastly, our data is observational and invites concern for un- observed heterogeneity in the sense of idiosyncratic groups reacting differently to the “treatment”; that is, after controlling for observed time constant or even time-varying variables other unobserved person specific attributes might affect the dependent variable. Our models employ the commonly used lagged dependent variable approach to deal with this issue (see Halaby, 2004; Morgan & Winship, 2007). Theoretically the lagged dependent variable serves as a proxy for any unobserved dif- ferences between person characteristics and in turn leads to less biased estimates of predictor effects. This is but one way to deal with this issue and we recognize that others have been proposed such as random effects and fixed effects panel models (Wooldridge, 2010).

Stands alone 4008 0.06 0.23 0e1 50/50 4008 0.10 0.30 0e1 In minority with 1 friend 4008 0.14 0.35 0e1 In minority with 2 þ friends 4008 0.17 0.38 0e1 In majority with friends 4008 0.40 0.49 0e1 Male (%) 4008 0.42 0.49 0e1 Black (%) 4008 0.21 0.41 0e1 Other Race (%) 4008 0.12 0.47 0e1 White (%) 4008 0.66 0.33 0e1 Age 4008 14.52 1.70 11e19 Two parents in home 4008 0.78 0.41 0e1 Parental education 4008 0.87 0.34 0e1 Family on public assistance 4008 0.07 0.26 0e1 Grade point average 4008 3.01 0.75 1e4 School attachment 4008 3.81 0.87 1e5 Parental attachment 4008 4.80 0.51 1e5 Friend attachment 4008 4.30 0.75 1e5 Network size e drinking 4008 5.81 2.07 3e10 Time between surveys (days) 4008 233.01 47.87 93e397

Conclusion

Our study highlights the potential for implementation problems in school-based alcohol use interventions. School-based in- terventions have often used peer leaders (i.e., opinion leaders) to help implement the delivery of substance-abuse based in- terventions (Valente, Gallaher, & Mouttapa, 2004). Peer leaders are crucial to the school-based interventions since they tend to be similar to their peers, influence norms within groups, and are early adopters of the intervention (Valente et al., 2004). Yet, as Valente et al. (2004) point out, network based studies can also inform implementation in ways beyond simply identifying peer leaders or socio-metric stars. Group behavior is far easier to change than in- dividual behavior, however, when groups have minorityemajority

contingents regarding the behavior interventions are aimed at affecting, using peers to aide implementation becomes complicated and potentially less effective. Research has yet to demonstratewhat the optimal group formation is for substance abuse interventions using peer leaders (Valente, 2012), however, our study demon- strates how in-group minorities can be influential for problem behavior. Because, as our results show, minority positions can both help people to continue to desist from alcohol use and lower con- sumption levels among group members who do drink, simply tar- geting peer leaderswithin schools but notwithin specific friendship groups may not be the most powerful way to use peers as an intervention component. A greater understanding of minoritye majority dynamics may facilitate better peer-based interventions.

In sum, non-drinking adolescents in the minority position are at risk for the onset of alcohol use as are drinkers with a majority of drinking friends. However, having one or more non-drinking friends occupying a minority position can lower the probability of alcohol use even when the in-group norm is pro-drinking. Impor- tantly, abstaining friends in the minority position can decrease a majority member’s drinking behavior as well as increasing the probability of abstention of a fellow minority member facing a drinking majority. Consequently, being in the minority position does not unilaterally put individuals at increased risk for alcohol onset and abuse; there are peer group circumstances that dampen this risk. Nor does being in the minority with a friend relegate their behavior to being non-influential. Pointedly, the minority position matters, and non-conformity to the group norm or majority posi- tion does not mean disengaging from the overall group (Packer & Miners, 2012). Adolescence is a dynamic developmental period and friends play an important role in guiding behaviors. Minority influence is not about ‘winning’ but about authentic in-group dissent that is beneficial for both the minority and majority (Nemeth, 2011). A continued view of social influence as merely conformity to a group norm as defined by the majority risks perpetuating an over-socialized version of the adolescent and fails to account for dissention as a healthy behavior.

Appendix A. Descriptive statistics for time I non-drinkers

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e4544

Appendix B. Descriptive statistics for time I drinkers

N Mean SD Range

Dependent variable Respondent drinking time II 4765 1.72 1.50 0e6 Respondent drunkenness time II 4765 0.99 1.36 0e6 Respondent binge drinking time II 4765 1.02 1.46 0e6 Independent variables Fully united non-delinquent 4765 0.22 0.41 0e1 Stands alone 4765 0.02 0.15 0e1 50/50 4765 0.08 0.27 0e1 In minority with 1 friend 4765 0.26 0.44 0e1 In minority with 2 þ friends 4765 0.24 0.43 0e1 In majority with friends 4765 0.18 0.38 0e1 Male (%) 4765 0.45 0.50 0e1 Black (%) 4765 0.18 0.38 0e1 Other race (%) 4765 0.76 0.43 0e1 White (%) 4765 0.07 0.25 0e1 Age 4765 15.34 1.55 11e19 Two parents in home 4765 0.74 0.44 0e1 Parental education 4765 0.85 0.35 0e1 Family on public assistance 4765 0.07 0.26 0e1 Grade point average 4765 2.75 0.77 1e4 School attachment 4765 3.57 0.96 1e5 Parental attachment 4765 4.65 0.67 1e5 Friend attachment 4765 4.33 0.72 1e5 Network size e drinking 4765 5.86 2.03 3e10 Time I respondent drinking 4765 2.11 1.33 3e10 Time between surveys (days) 4765 231.09 48.93 60e418

Appendix C. Supplementary data

Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.socscimed.2014.01.040

References

Ali, M. M., & Dwyer, D. S. (2010). Social network effects in alcohol consumption among adolescents. Addictive Behaviors, 35(4), 337e342.

Allen, J. P., Porter, M. R., McFarland, F. C., Marsh, P., & McElhaney, K. B. (2005). The two faces of adolescents’ success with peers: adolescent popularity, social adaptation, and deviant behavior. Child Development, 76(3), 747e760.

Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgments. In Groups, leadership, and men. S (pp. 222e236).

Asch, S. E. (1952). Social psychology. Englewood Cliffs, New Jersey: Prentice Hall. Asch, S. E. (1955). Opinions and social pressure. Asch, S. E. (1956). Studies of independence and conformity: I. A minority of one

against a unanimous majority. Psychological Monographs: General and Applied, 70(9), 1e70.

Bachanas, P. J., Morris, M. K., Lewis-Gess, J. K., Sarett-Cuasay, E. J., Sirl, K., Ries, J. K., et al. (2002). Predictors of risky sexual behavior in African American adolescent girls: implications for prevention interventions. Journal of Pediatric Psychology, 27(6), 519e530.

Bagwell, C. L., & Schmidt, M. E. (2011). Friendships in childhood and adolescence. Guilford Press.

Balsa, A. I., Homer, J. F., French, M. T., & Norton, E. C. (2011). Alcohol use and popularity: social payoffs from conforming to peers’ behavior. Journal of Research on Adolescence, 21(3), 559e568.

Bandyopadhyay, D., DeSantis, S. M., Korte, J. E., & Brady, K. T. (2011). Some con- siderations for excess zeroes in substance abuse research. American Journal of Drug and Alcohol Abuse, 37(5), 376e382.

Bauman, K. E., & Ennett, S. T. (1996). On the importance of peer influence for adolescent drug use: commonly neglected considerations. Addiction, 91(2), 185e198.

Berndt, T. J., & Murphy, L. M. (2003). Influences of friends and friendships: myths, truths, and research recommendations. Advances in Child Development and Behavior, 30, 275e310.

Bond, R., & Smith, P. B. (1996). Culture and conformity: a meta-analysis of studies using Asch’s (1952b, 1956) line judgement task. Psychological Bulletin, 119(1), 111e137.

Botticello, A. L. (2009). School contextual influences on the risk for adolescent alcohol misuse. American Journal of Community Psychology, 43(1e2), 85e97.

Brechwald, W. A., & Prinstein, M. J. (2011). Beyond homophily: a decade of advances in understanding peer influence processes. Journal of Research on Adolescence, 21(1), 166e179.

Brown, B. B., Bakken, J. P., Ameringer, S. W., & Mahon, S. D. (2008). A comprehensive conceptualization of the peer influence process in adolescence. In Understanding peer influence in children and adolescents (pp. 17e44).

Burt, C., & Rees, C. I. (2014). Behavioral heterogeneity in adolescent friendship networks. Justice Quarterly (Advance online publication). http://dx.doi.org/10. 1080/07418825.2013.856932.

Cameron, A. C., & Trivedi, P. K. (1998). Regression analysis of count data. Econometric society monograph No. 30. Cambridge: Cambridge University Press.

Campanelli, P. C., Dielman, T., & Shope, J. T. (1987). Validity of adolescents’ self- reports of alcohol use and misuse using a bogus pipeline procedure. Adoles- cence, 22(85), 7e22.

Chantala, K., & Tabor, J. (2010). Strategies to perform a design-based analysis using the add health data. http://www.cpc.unc.edu/projects/addhealth/data/guides/ weight1.pdf.

Chassin, L., Pitts, S. C., & Prost, J. (2002). Binge drinking trajectories from adolescence to emerging adulthood in a high-risk sample: predictors and substance abuse outcomes. Journal of Consulting and Clinical Psychology, 70(1), 67e78.

Chatterji, P., & DeSimone, J. (2005). Adolescent drinking and high school dropout. National Bureau of Economic Research.

Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: compliance and conformity. Annual Review of Psychology, 55, 591e621.

Clark, A. E., & Lohéac, Y. (2007). “It wasn’t me, it was them!” Social influence in risky behavior by adolescents. Journal of Health Economics, 26(4), 763e784.

Crano, W. (1994). Context, comparison, and change: methodological and theoretical contributions to a theory of minority (and majority) influence. In S. Moscovici, A. Mucchi-Faina, & A. Maass (Eds.), Minority influence (pp. 17e46). Chicago: Nelson Hall Publishers.

Crano, W. (2012). The rules of influence: Winning when you’re in the minority. New York: St. Martin’s Press.

Crosnoe, R., Muller, C., & Frank, K. (2004). Peer context and the consequences of adolescent drinking. Social Problems, 51(2), 288e304.

Del Boca, F. K., & Darkes, J. (2003). The validity of self-reports of alcohol consumption: state of the science and challenges for research. Addiction, 98(s2), 1e12.

DeWit, D. J., Adlaf, E. M., Offord, D. R., & Ogborne, A. C. (2000). Age at first alcohol use: a risk factor for the development of alcohol disorders. American Journal of Psychiatry, 157(5), 745e750.

Diego, M. A., Field, T. M., & Sanders, C. E. (2003). Academic performance, popularity, and depression predict adolescent substance use. Adolescence, 38(149), 35e42.

Donovan, J. E. (2004). Adolescent alcohol initiation: a review of psychosocial risk factors. Journal of Adolescent Health, 35(6), 529.

Donovan, J. E., & Jessor, R. (1985). Structure of problem behavior in adolescence and young adulthood. Journal of Consulting and Clinical Psychology, 53(6), 890.

Elliott, D. S., Huizinga, D., & Ageton, S. S. (1985). Explaining delinquency and drug use. Beverly Hills, CA: Sage Publications.

Enders, C. K. (2010). Applied missing data analysis. New York: Guilford Press. Farrow, J. A., & Brissing, P. (1990). Risk for DWI: a new look at gender differences in

drinking and driving influences, experiences, and attitudes among new adolescent drivers. Health Education & Behavior, 17(2), 213e221.

Friedkin, N. E., & Johnsen, E. C. (2011). Social influence network theory: A sociological examination of small group dynamics (Vol. 33). Cambridge University Press.

Fujimoto, K., & Valente, T. W. (2012a). Decomposing the components of friendship and friends’ influence on adolescent drinking and smoking. Journal of Adoles- cent Health, 51(2), 136e143.

Fujimoto, K., & Valente, T. W. (2012b). Social network influences on adolescent substance use: disentangling structural equivalence from cohesion. Social Sci- ence & Medicine.

Gardikiotis, A. (2011). Minority influence. Social and Personality Psychology Compass, 5(9), 679e693.

Halaby, C. N. (2004). Panel models in sociological research: theory into practice. Annual Review of Sociology, 507e544.

Hawkins, J. D., Catalano, R. F., & Miller, J. Y. (1992). Risk and protective factors for alcohol and other drug problems in adolescence and early adulthood: impli- cations for substance abuse prevention. Psychological Bulletin, 112(1), 64.

Haynie, D. L. (2001). Delinquent peers revisited: does network structure matter? American Journal of Sociology, 106(4), 1013e1057.

Haynie, D. L. (2002). Friendship networks and delinquency: the relative nature of peer delinquency. Journal of Quantitative Criminology, 18(2), 99e134.

Hilbe, J. M. (2011). Negative binomial regression. New York: Cambridge University Press.

Hirschi, T. (1969). Causes of delinquency. Berkeley: University of California Press. Jessor, R., & Jessor, S. L. (1977). Problem behavior and psychosocial development: A

longitudinal study of youth. Academic Press New York. Jetten, J., & Hornsey, M. J. (2011). The many faces of rebels. In Rebels in groups: Dissent,

deviance, difference, and defiance (pp. 1e13). Keefe, K. (1994). Perceptions of normative social pressure and attitudes toward

alcohol use: changes during adolescence. Journal of Studies on Alcohol and Drugs, 55(1), 46.

Knecht, A. B., Burk, W. J., Weesie, J., & Steglich, C. (2011). Friendship and alcohol use in early adolescence: a multilevel social network approach. Journal of Research on Adolescence, 21(2), 475e487.

C. Rees, D. Wallace / Social Science & Medicine 108 (2014) 34e45 45

Kobus, K., & Henry, D. B. (2010). Interplay of network position and peer substance use in early adolescent cigarette, alcohol, and marijuana use. The Journal of Early Adolescence, 30(2), 225e245.

Kroger, J. (2004). Identity in adolescence. New York: Routledge. Kulbok, P. A., & Cox, C. L. (2002). Dimensions of adolescent health behavior. Journal

of Adolescent Health, 31(5), 394e400. Lang, S. W., Waller, P. F., & Shope, J. T. (1997). Adolescent driving: characteristics

associated with single-vehicle and injury crashes. Journal of Safety Research, 27(4), 241e257.

Laursen, B., Popp, D., Burk, W. J., Kerr, M., Stattin, H., & Card, N. (2008). Incorporating interdependence into developmental research: examples from the study of homo- phily and homogeneity. In Modeling dyadic and interdependent data in the developmental and behavioral Sciences (pp. 11e37).

Levine, J. M. (1999). Solomon Asch’s legacy for group research. Personality and Social Psychology Review, 3(4), 358e364.

Loeys, T., Moerkerke, B., De Smet, O., & Buysse, A. (2012). The analysis of zero- inflated count data: beyond zero-inflated Poisson regression. The British Jour- nal of Mathematical and Statistical Psychology, 65(1), 163e180.

Maass, A., & Clark, R. D. (1984). Hidden impact of minorities: fifteen years of mi- nority influence research. Psychological Bulletin, 95(3), 428e450.

Maguin, E., & Loeber, R. (1996). Academic performance and delinquency. Crime and Justice, 145e264.

Martin, R., & Hewstone, M. (2010). Introduction: Theory and research on minority influence. In Minority influence and innovation: antecedents, processes, and consequences.

McDowell, A. (2003). From the help desk: hurdle models. The Stata Journal, 3, 178e 184.

McPherson, M., Smith-lovin, L., & Cook, J. M. (2001). Birds of a feather : homophily in social networks. Annual Review of Sociology, 27, 415e444.

Meropol, S. B., Moscati, R. M., Lillis, K. A., Ballow, S., & Janicke, D. M. (1995). Alcohol- related injuries among adolescents in the emergency department. Annals of Emergency Medicine, 26(2), 180e186.

Miller, M., Lippmann, S. J., Azrael, D., & Hemenway, D. (2007). Household firearm ownership and rates of suicide across the 50 United States. The Journal of Trauma and Acute Care Surgery, 62(4), 1029e1035.

Morgan, S. L., & Winship, C. (2007). Counterfactuals and causal inference: Methods and principles for social research. New York: Cambridge University Press.

Moscovici, S. (1976). Social influence and social change (Vol. 10). London: Academic Press.

Moscovici, S. (1980). Toward a theory of conversion behavior. Advances in Experi- mental Social Psychology, 13, 209e239.

Moscovici, S. (1985). Innovation and minority influence. In (3rd ed), Handbook of social psychology (Vol. 2); (pp. 347e412) New York: Random House.

Moscovici, S., & Faucheax, C. (1972). Social influence, conformity bias, and the study of active minorities. Advances in Experimental Social Psychology, 6, 149e202.

Moscovici, S., & Lage, E. (1976). Studies in social influence III: majority versus mi- nority influence in a group. European Journal of Social Psychology, 6(2), 149e174.

Moscovici, S., Mugny, G., & van Avermaet, E. (1985). Perspectives on minority influ- ence (Vol. 9). Cambridge University Press.

Mugny, G., & Pérez, J. A. (1991). The social psychology of minority influence. Cam- bridge University Press.

Nagin, D., & Paternoster, R. (2000). Population heterogeneity and state dependence: state of the evidence and directions for future research. Journal of Quantitative Criminology, 16(2), 117e144.

Nemeth, C. J. (1986). Differential contributions of majority and minority influence. Psychological Review, 93(1), 23.

Nemeth, C. (2011). Minority influence theory. In P. Van Lange, A. Kruglanski, & E. Higgins (Eds.), Handbook of theories of social psychology (pp. 362e379). London: SAGE Publications Ltd.

Packer, D. J., & Miners, C. T. (2012). At the first sign of trouble or through thick and thin? When nonconformity is and is not disengagement from a group. Journal of Experimental Social Psychology, 48(1), 316e322.

Papastamou, S., & Mungy, G. (1990). Synchronic consistency and psychologization in minority influence. European Journal of Social Psychology, 20(2), 85e98.

Paton, S. M., & Kandel, D. B. (1978). Psychological factors and adolescent illicit drug use: ethnicity and sex differences. Adolescence, 13(50), 187e200.

Payne, D. C., & Cornwell, B. (2007). Reconsidering peer influences on delinquency: do less proximate contacts matter? Journal of Quantitative Criminology, 23(2), 127e149.

Rees, C., & Pogarsky, G. (2011). One bad apple may not spoil the whole bunch: best friends and adolescent delinquency. Journal of Quantitative Criminology, 27(2), 197e223.

Reiter, J. P., Raghunathan, T. E., & Kinney, S. K. (2006). The importance of modeling the sampling design in multiple imputation for missing data. Survey Method- ology, 32(2), 143e149.

Rogers, E. (2003). Diffusion of innovations. New York: Free Press. Rossi, P. H., & Berk, R. A. (1985). Varieties of normative consensus. American So-

ciological Review, 333e347. Sampson, R. J., & Sharkey, P. (2008). Neighborhood selection and the social repro-

duction of concentrated racial inequality. Demography, 45(1), 1e29. Snyder, C. R., & Fromkin, H. L. (1980). Uniqueness: The human pursuit of difference.

New York: Plenum. Stangor, C. (2004). Social groups in action and interaction. New York: Psychology

Press. Urberg, K. A. (1992). Locus of peer influence: social crowd and best friend. Journal of

Youth and Adolescence, 21(4), 439e450. Urberg, K. A., Degirmencioglu, S. M., & Pilgrim, C. (1997). Close friend and group

influence on adolescent cigarette smoking and alcohol use. Developmental Psychology, 33(5), 834e844.

Valente, T. W. (1996). Social network thresholds in the diffusion of innovations. Social Networks, 18(1), 69e89.

Valente, T. W. (2012). Network interventions. Science, 337(6090), 49e53. Valente, T. W., Gallaher, P., & Mouttapa, M. (2004). Using social networks to un-

derstand and prevent substance use: a transdisciplinary perspective. Substance Use & Misuse, 39(10e12), 1685e1712.

Valois, R. F., Oeltmann, J. E., Waller, J., & Hussey, J. R. (1999). Relationship between number of sexual intercourse partners and selected health risk behaviors among public high school adolescents. Journal of Adolescent Health, 25(5), 328e 335.

Vitaro, F., Brendgen, M., & Tremblay, R. E. (2000). Influence of deviant friends on delinquency: searching for moderator variables. Journal of Abnormal Child Psychology, 28(4), 313e325.

Von Hippel, P. (2007). Regression with missing Ys: an improved strategy for analyzing multiply imputed data. Sociological Methodology, 37(1), 87e113.

Warr, M. (2002). Companions in crime: The social aspects of criminal conduct. Cam- bridge University Press.

Watts, D. J., & Dodds, P. (2009). Threshold models of social influence. In P. Hedström, & P. Bearman (Eds.), The Oxford handbook of analytical sociology. Oxford Uni- versity Press.

Weerman, F., & Smeenk, W. H. (2005). Peer similarity in delinquency for different types of friends: a comparison using two measurement methods. Criminology, 43(2), 499e524.

Williams, K. D., Forgas, J. P., von Hippel, W., & Zadro, L. (2005). The social outcast: an overview. In K. Williams, J. Forgas, & W. von Hippel (Eds.), The social outcast: Ostracism, social exclusion, rejection, and bullying. Psychology Press.

Windle, M., Spear, L. P., Fuligni, A. J., Angold, A., Brown, J. D., Pine, D., et al. (2008). Transitions into underage and problem drinking: developmental processes and mechanisms between 10 and 15 years of age. Pediatrics, 121(Suppl. 4), S273e S289.

Wolf, S. (1987). Majority and minority influence: a social impact analysis. In Paper presented at the social influence: The Ontario symposium.

Wood, W., Lundgren, S., Ouellette, J. A., Busceme, S., & Blackstone, T. (1994). Mi- nority influence: a meta-analytic review of social influence processes. Psycho- logical Bulletin, 115(3), 323e345.

Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. Cambridge, MA: The MIT Press.