Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Jan 1.
Published in final edited form as: Soc Sci Med. 2014 May 24;125:173–181. doi: 10.1016/j.socscimed.2014.05.023

Multiplex congruity: Friendship networks and perceived popularity as correlates of adolescent alcohol use

Kayo Fujimoto a,*, Thomas W Valente b
PMCID: PMC4242802  NIHMSID: NIHMS601518  PMID: 24913275

Abstract

Adolescents interact with their peers in multiple social settings and form various types of peer relationships that affect drinking behavior. Friendship and popularity perceptions constitute critical relationships during adolescence. These two relations are commonly measured by asking students to name their friends, and this network is used to construct drinking exposure and peer status variables. This study takes a multiplex network approach by examining the congruity between friendships and popularity as correlates of adolescent drinking. Using data on friendship and popularity nominations among high school adolescents in Los Angeles, California (N = 1707; five schools), we examined the associations between an adolescent's drinking and drinking by (a) their friends only; (b) multiplexed friendships, friends also perceived as popular; and (c) congruent, multiplexed-friends, close friends perceived as popular. Logistic regression results indicated that friend-only drinking, but not multiplexed-friend drinking, was significantly associated with self-drinking (AOR = 3.51, p < 0.05). However, congruent, multiplexed-friend drinking also was associated with self-drinking (AOR = 3.10, p < 0.05). This study provides insight into how adolescent health behavior is predicated on the multiplexed nature of peer relationships. The results have implications for the design of health promotion interventions for adolescent drinking.

Keywords: Social network analysis, Adolescent alcohol use, Peer status, Friendship network, Perceived popularity, Network multiplexity, Decomposed network exposure model

1. Introduction

Adolescents engage with peers in multiple social settings, which provides them with opportunities to have a variety of types of relationships (Brown, 1990). Among these relationships, those based on friendships most strongly organize the world of adolescents. Peer relationships are also organized around network status systems, for which certain individuals are the leaders (Coleman,1961), and peer status is determined by popularity among peers (Schwartz and Gorman, 2011). Peer networks guided by friendship and popularity play an integral part in the ways that risk-taking behaviors are engaged in among network members (Osgood et al., 2013; Schaefer et al., 2012). Among the substances that adolescents use, alcohol is the most prevalent (National Institute on Drug Abuse, 2011). Friendships provide adolescents not only with a means to gain or maintain peer status but also with exposure to drinking (Moody et al., 2011; Osgood et al., 2013).

Social network analysis has been widely used as a methodological tool to measure the structural aspects of peer networks in terms of adolescent drinking (Ennett et al., 2006; Valente et al., 2004). A number of network studies have documented the integral role that friends' drinking plays in an adolescent's drinking, specifically in terms of the similarity in drinking behaviors and the amount of alcohol consumed (Cleveland and Wiebe, 2003; Crosnoe et al., 2004; Fujimoto and Valente, 2012a; Urberg et al., 1997). Network measures of peer status have shown that greater popularity among peers leads to higher alcohol use (Ennett et al., 2006; Kobus and Henry, 2010; Mayeux et al., 2008; Osgood et al., 2013; Schwartz and Gorman, 2011).

An adolescent's peer status, particularly network centrality, also moderates the association between friends' drinking behavior and his or her own drinking behavior (Crosnoe et al., 2004). The peer status of one's friends also moderates the association between friends' drinking behavior and the adolescent's own drinking. One study found that, among seventh and eighth graders, the popularity of an adolescent's best friends was positively associated with substance use (Peters et al., 2010). Overall, the research on peer networks and adolescent alcohol use indicates that friendship and peer status are the main structural constituents of the peer context that are associated with adolescent drinking.

Most network research, however, takes a unitary view of peer relationships and focuses only on friendship, which is then used as a basis for popularity measurements. For example, one measure of popularity in a friendship network is “sociometric popularity,” which is a quantification of the extent to which an adolescent is liked by his or her peers, e.g., “name individuals who you like the most” (Coie et al., 1982), i.e., a “liking relationship” (Hartup, 1996). Although “sociometric popularity” may capture notions of system-level social status (Moody et al., 2011), it is still based on personal preference and affection. This is distinct from reputation-based “perceived popularity,” e.g., “name individuals who are popular” (Cillessen and Marks, 2011; Parkhurst and Hopmeyer, 1998), which is determined through the explicit naming of individuals who are considered popular by network members and which indicates visibility, prestige, and social dominance (Dijkstra et al., 2012; Mathys et al., 2013). Further, empirical studies have found that adolescents do not necessarily like the same person whom they perceive as being popular (reputation). As related to alcohol use, “perceived popularity” was reported to be predictive of increased use among tenth graders, while “sociometric popularity” was not (Mayeux et al., 2008). This finding suggests that friendship and “perceived popularity” are different types of peer relationships.

To take into account these different types of relationships, the current study uses a multiplex network approach to measuring peer context to examine the relationship between friendship, reputation-based “perceived popularity,” and adolescent drinking. Adolescents also make friends with those who are different from the peers whom they perceive as being popular (“popular friends”), and these peer relationships may or may not overlap. Nonetheless, we expect that the degree of the relationship between friends' drinking and the adolescent's will differ in the case of friends who are friends only (a uniplex relationship) versus that of a friendship with peers who are perceived to be popular (a multiplex relationship). Additionally, it is reasonable to postulate that, as these two types of relationships become more congruent and, thus, reinforce each other, the similarity of drinking behaviors between friends perceived to be popular and the adolescent should increase.

2. Theoretical framework

Social network analysis is a powerful methodological tool used to test social psychological theories that explain delinquent–peer association. Differential association theory (Sutherland, 1947; Sutherland and Cressey, 1978) considers the social context of peer networks through which interaction among an adolescent and his or her friends occurs (Krohn, 1986). The theory posits that adolescents learn deviance through the norms, attitudes, techniques, rationalizations, and motives of delinquent behavior. Social learning theory (Bandura, 1977; Burgess and Akers, 1966) posits that adolescents learn deviant behavior by observing, modeling, or imitating the behaviors of intimate others and through those behaviors' subsequent social reinforcement.

Both theories assume that personal peer networks based on friendship are key to transmitting deviant behavior among adolescents. The majority of research on differential association or social learning (Akers et al., 1979; Akers and Lee, 1996) focuses on peer delinquency based on direct friendships (Payne and Cornwell, 2007). For both theories, a peer context in which adolescents interact with their friends is an a priori condition. To explain the association between friends' delinquency and self-delinquency, research focuses on the process of how individuals learn to behave in a deviant manner or on the social mechanism in the process of learning deviancy. Neither differential association nor social learning theory clearly addresses how to operationalize the friendship network context. Thus, the current study focuses on the structure of these networks.

2.1. Network perspective of friendship and risk behavior

The social network perspective focuses on peer context and provides a method to measure structural aspects and personal attributes of peer relationships relevant to adolescent substance use (Ennett et al., 2006). The network perspective integrates the network contexts with differential association by identifying structural characteristics of friendship networks, such as density (i.e., cohesiveness), centrality (i.e., connectedness), popularity, and social proximity, as measured by the degrees of separation (i.e., social distance; Fujimoto and Valente, 2012b; Payne and Cornwell, 2007), which may moderate the association between friends' behavior and self-behavior (Crosnoe et al., 2004; Ennett et al., 2006; Haynie, 2001). As conditional on these network properties, some adolescents are more susceptible to being influenced by their friendship networks, and some friendship networks are more effective in directly controlling the risk behavior of their network members (Haynie, 2001).

Most network contexts measured by social network analysis, however, are limited to one type of peer relationship, i.e., friendship, and any structural conditions under which adolescents interact with each other are assumed to be grounded in the friendship network. This network is then assumed to be related to drinking behavior. Friendship entails personal preference and affection, and any network measures that are computed using friendship networks connote personal liking. For instance, popularity based on a friendship network refers to how much individuals are liked, which may be distinct from popularity based on the hierarchical nature of peer relations, which concerns visibility or prestige that is acquired among peers, regardless of one's liking. This is another type of peer network and one that also accounts for adolescent drinking behavior.

2.2. Perceived popularity and risk behavior

Popular adolescents are more likely than other group members to exert influence on group norms and, through visual or other cues, influence the acceptability of substance use (Sandstrom, 2011; Schwartz and Gorman, 2011). Further, influential peers are those who either occupy high status positions in a reputation-based peer hierarchy (perceived popularity) or who are popular in friendship network (sociometric popularity). A hierarchical peer structure may reflect peer consensus in regard to who is perceived as popular as well as affect peer norms.

According to popularity-socialization theory, higher levels of popularity are associated with being more strongly socialized by the peer group as well as an increase in the level of deviance over time (Allen et al., 2005). Popular adolescents are particularly likely to experience increased exposure to social pressures and influences (Schwartz and Gorman, 2011). Thus, they are likely to adopt behavior consistent with group norms as a means to establish their social identity and to reinforce their dominant position in the peer hierarchy (Michell and Amos, 1997). However, although conforming to the level of peers' alcohol use enhances popularity, exceeding the level, as set by group norms, leads to social rejection (Becker and Luthar, 2007).

In this study, we separate perceived popularity and friendships and simultaneously model their associations with adolescent drinking behavior. Most network studies measure popularity based on friendship nominations, yet this merely reflects a different aspect of the same type of relationship, namely, “friendship.” Perceived popularity, in comparison, is a different type of relationship, which may or may not include liking. It may simply be a reflection of peer agreement with regard to who is regarded as having high status or popularity. Thus, treating perceived popularity differently from friendship choice in the theoretical construction of a “multiplex network” provides a more nuanced understanding of the influence of these peer relationships on an adolescent's drinking behavior.

2.3. Network multiplexity

The network perspective offers a way to conceptualize a peer context that comprises multiple types of peer relations as well as the relevant methodology to measure the interdependency of the different types of peer relations. Multiplex relationships are ones in which individuals are connected through more than one relationship, and such relationships may foster greater behavioral similarity (Krohn, 1986; Krohn et al., 1988). Social network researchers view multiplexity as a coincidence of different types of relationships that have multiple contents (Skvoretz and Agneessens, 2007). Such multiplex relationships are expected to be stronger than uniplex relationships because they contain more than one basis for interaction (Skvoretz and Agneessens, 2007). Thus, they also are expected to have stronger consequences for interpersonal processes. These multiplex relationships reflect not only the simultaneous presence of multiplex ties but also contribute to the development of a local network structure that involves multiple types of ties, with interdependence among ties within dyadic and triadic network structures (Koehly and Pattison, 2005; Lazega and Pattison, 1999) or higher level configurations in combined one-mode and two-mode networks (Snijders et al., 2013; Wang et al., 2013).

When applying the concept of network multiplexity to our study, we assume that friendship and peer popularity constitute different types of peer relations and, taken together, form a multiplex peer network. These different types of relationships are assumed to overlap, which has consequences for peer interactions. Specifically, the overlap of friendship and popularity networks (multiplex networks) is stronger than friendship alone (uniplex network), as the former contains both a greater number of and stronger peer norms with regard to drinking than does the latter. This overlap is expected to contribute to greater similarity in drinking behaviors between friends and the adolescent.

2.4. Multiplex congruity

One limitation of network multiplexity is that social network researchers conceptualize it as mere co-occurrence of the different types of ties, which are then used to identify local network configurations. Thus, all overlapped ties are treated as equally important. This leaves out more detailed information on multiple ties and the relative importance of one type of relationship nested within another in the assessment of the similarity of drinking behavior between popular friends and the adolescent. This limitation stems mainly from the standard network approach to operationalizing uniplex/multiplex networks through binary networks (simultaneous existence of multiple types of ties). Our study addresses this limitation by introducing the concept of “multiplex congruity,” which encompasses various levels of congruity between different types of ties. Congruity is measured by the nomination rank of one type of relationship nested within another type. In this study, the nomination rank of popularity choices is nested within the list of friendship ones.

Analyzing a binary network of friendship nomination but ignoring the rank information can result in misleading statistical inferences (Hoff et al., 2013). To address this concern, this study extends the order of friendship nomination to multiplex relationships and takes into account the degree to which one type of relationship is congruent with another type of relationship. We expect that the association between friends' drinking and an adolescent's drinking will be stronger when the friendships are more congruent with popularity relationships than when they are less congruent. We hypothesize that an adolescent's drinking will be more strongly associated with the drinking of a first-nominated friend whom the adolescent perceives as popular than with the last-nominated friend who is also perceived as popular by the adolescent.

This study uses the concept of “multiplex congruity” and a corresponding network measure that combines both friendship and popularity relationships to measure the level of congruity between these relationships, based on the degree to which adolescents are exposed to their friends' drinking. This level of congruity is computed by weighting the relative importance of popularity nomination among friends (measured by the rank of the friendship nomination).

For the uniplex friendship case, multiplex congruity is zero (no congruence/no interdependence of friendship and popularity and, hence, only friendship), and, for the multiplex ties (i.e., simultaneous existence of friendship and popularity), multiplex congruity is non-zero. In the latter case, greater values indicate that the popularity nominations are ranked highly (sooner) in the friendship nominations, and lower values indicate that the popularity nominations are nominated later, which indicates a weaker friendship for each actor pair. The concept of “multiplex congruity” is distinguished from the social network concept of a “multiplex network,” which does not consider the relative importance of one type of tie (popularity nomination) nested within the other type of tie (friendship nomination list); that is, it considers only the simultaneous existence of different types of ties.

3. Data and methods

3.1. Data

The current study is nested within a larger study of social networks and networking as influences on adolescent substance use behavior (Valente et al., 2013b). The data for this study come from a cross-sectional sample of 1707 students who were interviewed in October 2010 in five schools in one school district, El Monte Union High School District (EMUHSD), in Los Angeles County. The students at EMUHSD, a predominantly Hispanic/Latino district, are a potentially high-risk population, based on their socioeconomic status. We obtained 2016 valid parental consent forms (88.0%) out of the 2290 tenth grade students, and 1823 agreed to participate in the study, while 28 did not provide student assent, which reduced the eligible pool to 1795 students, of whom 1707 completed the survey questionnaire (a 74.5% overall participation rate).

For the collection of network data, respondents were provided a grade roster that contained the students' school photos with an ID number unique for each student within the school and were asked to write the roster ID numbers in certain places on the survey questionnaire. The study used friendship information, for which respondents were asked to nominate up to seven best friends in the same grade. The nomination question also allowed respondents to name up to 12 additional friends through the inclusion of, “Are there other people in the tenth grade whom you consider a close friend?” The study included all friendship nominations for a maximum of 19, and we created friendship networks based only on sending ties (outdegrees). We also collected popularity information by asking respondents to write the roster ID numbers of seven students from the entire tenth grade whom they think are the most popular. This information was based on the respondents' subjective assessment of perceived popularity. Using the above-noted information, we constructed two one-mode networks based on friendship and perceived popularity, and each was used for the computation of network measurements. Survey data also included substance use behavior, drinking alcohol, as well as standard demographic information, such as gender, ethnicity, academic grades, and socioeconomic status.

3.2. Measures

Drinking behavior was measured by respondents' answers to questions that inform lifetime drinking, such as, “How many days have you had at least one drink of alcohol during your life?” We created a dichotomous variable of “ever drank alcohol” (coded as 1) or “not” (coded as 0). The same variable was used to measure the alter's behavior in regard to lifetime drinking in the computation of network exposure terms described below.

3.3. Network measures

Our study specified two one-mode adjacency matrices, Xij and Pij, to create the network measures. The adjacency matrix Xij represents friendship, with Xij = 1 if actor i (ego) nominated actor j (alter) as friend, and Xij = 0 otherwise. The adjacency matrix Pij represents popularity with Pij = 1 if actor i (ego) nominated actor j (alter) as being popular, and Pij = 0 otherwise. Using these matrices, various network statistics were computed.

3.3.1. Network exposure to friends' drinkers

We used the network exposure model (Valente, 1995, 2005), using friendship network Xij, and computed the level of overall exposure to friends' drinkers, Ei,(f),which is defined as follows:

Ei,(f)=∑jXijYj∑jXiji,j=1,…,N,i≠j (1)

where Xij is an adjacency matrix based on friendship, and Yj is alter's drinking status (0 or 1). The level of an ego's exposure to friends' drinkers is measured as the proportion of friends who ever drunk alcohol in an ego's network. The level of an ego's exposure to popular peer drinkers is measured in a similar way by replacing the friendship network Xij with the popularity matrix Pij.

3.3.2. Segregation of network exposure

We computed the decomposed network exposure model (DNEM; Fujimoto, 2012) using two matrices of Xij and Pij. A DNEM is an extended version of the network exposure model that can decompose portions of friendship by using perceived popularity. This study decomposed exposure to friend drinkers into those perceived as (i) popular, Di(fp), and (ii) not popular, (friend only), Di,(fp̄). Note that (i) represents exposure based on a binary multiplex network, or congruence, and (ii) represents exposure based on non-congruence. The DNEM of (i) for a given actor i is defined as follows:

Di,(fp)=∑j[(Xij·Pij)Yj]∑j(Xij·Pij),i,j=1,…,N,i≠j (2)

where Xij and Yj are defined above, Pij is an adjacency matrix based on popularity nominations. (i) was computed by the element-wise product of the matrix Xij by a matrix Pij (friend nominations that overlap with popular ones) and then row-normalized and matrix-multiplied by the alter's behavior Yj. To compute (ii), we subtracted the adjacency matrix Pij from a unit matrix (where all off-diagonal elements are ones), and everything identical to the computation of (i), which is defined in the following formula:

Di,(fp¯)=∑j{[Xij·(1−Pij)]Yj}∑j[Xij·(1−Pij)],i,j=1,⋯,N,i≠j (3)

Essentially, the friendship matrix Xij was partitioned by the popularity network Pij. The resulting values range from 0 to 1. Decomposed exposure variables have many cases with 0 values: Their friends never drink alcohol and/or none of their nominations overlapped, e.g., friendship and popularity are unique, which exhibits the zero-spiked distribution. To address this issue, we applied a parameterization of the logit model (Robertson et al., 1994), which treats exposed and non-exposed cases separately by creating separate terms, one indicating non-zero exposures versus zero and the other indicating the actual exposure values (Hosmer and Lemeshow, 2000) that were then mean-centered to minimize potential collinearity.

3.3.3. Multiplex congruity measure and exposure to drinkers

We developed the measure of multiplex congruity (MC) to quantify the extent to which the nomination rank of friendship is congruous with the nomination rank of the popularity relationships. MC ranges from 0 to 1, with 0 indicating no overlap between both relationships and unity indicating that the first nominated friend also was nominated in the popularity relationship. Hybrid cases, which fall between these two extremes, are assigned an intermediate fractional value. Thus, MC can serve as the valued (non-binary) tie strength of the combined multiplexed relationship that reflects the “distance” in the nomination order. For example, consider an ego whose best friend is also perceived as popular. In this scenario, the congruence is total and is assigned a value of 1. As a second example, consider an ego whose friend is not perceived as popular. In this scenario, there is no congruence, and it is assigned a value of 0. As a third (hybrid) example, consider an ego whose second-best friend is perceived as popular. In this scenario, because there is another friend who is considered more important (or closer), even though the congruence is less than perfect, there is still some congruence. This case is assigned a non-zero fractional value based on the friend's relative ranking among the friend and popular nominations. The pairwise measure of multiplex congruity (MCij) converts this ranking into a weight that diminishes as the relative rank increases. The mathematical expressions and derivations of multiplex congruity can be found in the Online Supplement.

Multiplex congruity exposure (MCEi) uses MCij values to construct a valued adjacency matrix and then combines this matrix with the standard network exposure model (see Equation (1)) to form:

MCEi=∑jMCijYi∑jMCij,i,j=1,⋯,N,i≠j (4)

where MCij and Yj were defined previously. Our MCEi measure takes into account the level of congruity between friendship and popularity relations when computing network exposure to drinkers by taking the weighted average of the drinking attribute.

3.3.4. Popularity (indegree)

Numerous studies have found a positive association between adolescents' popularity and their substance use (Alexander et al., 2001; Mayeux et al., 2008; Pearson et al., 2006; Schwartz and Gorman, 2011; Valente et al., 2013a, 2005). Depending on how popularity is measured, its meaning becomes noticeably different such that reputation-based popularity (perceived popularity) indicates prestige or social dominance, while popularity based on friendship ties (sociometric popularity) indicates preference or affection. This study computed both types of popularity measures by counting the total number of nominations received, i.e., indegree, based on friendship matrix Xij (“sociometric popularity”) and popularity matrix Pij (“perceived popularity”). Both measures were transformed by taking their square root to approximate a normal distribution. We also computed the total number of valid nominations made, i.e., outdegree, for each matrix Xij of and, Pij respectively.

3.3.5. Socio-demographic and other control variables

We controlled for variables that have been shown to be related to adolescent substance use in other studies, such as “academic grades” (Crosnoe, 2006) and “parental drinking” (Latendresse et al., 2008). We controlled for the demographic and socioeconomic variables of age (in years), gender, ethnicity (a dummy variable for Hispanics/Latinos versus others), whether respondents received a free or reduced-cost lunch, and the ratio of the number of rooms in the house to the number of people in the household. Academic grades were self-reported and ranged from 1 (“mostly Ds or Fs”) to 4 (“mostly As”). Parental drinking was measured by a dummy variable that indicated whether at least one parent drinks alcohol at least once per week.

3.4. Statistical analyses

We conducted logistic regression analysis to test individual lifetime drinking as a function of various types of network measures, controlling for demographic variables, and we fit a school-level fixed effect model by entering four dummy-school identifiers. We specified three models:

  • Model 1 includes the network exposure to friends' drinking and to popular peers' drinking (Equation (1)), plus the control variables.

  • Model 2 is Model 1 plus decomposed network exposure to drinkers (Equations (2) and (3)).

  • Model 3 is Model 2 plus the multiplex congruity exposure (Equation (4)).

Our statistical analysis allows an individual's error term to be correlated with an alter's behavior, as we use an outcome variable (Y) on both the left side (as an ego's drinking) and right side (as an alter's behavior, to compute a percentage of drinkers among alters) of the equation, which violates the assumption that the observations would be i.i.d. (independent and identically distributed). To address this limitation, an alternate approach is to use an estimation procedure for the network autocorrelation model (Doreian, 1989; Dow, 1984) that assumes that an endogenous network effect variable is correlated with the error term and to use a network autocorrelation model that accommodates categorical outcome variables, using a two-stage conditional maximum likelihood estimation (Dow, 2008) or Bayesian estimation (Zhang et al., 2013). However, the validity of these statistical methods when dealing with complex nested data structures with the missing values that one commonly encounters in health behavioral research has yet to be proven. We also considered the use of multiple regression quadratic assignment procedures (MRQAP; Dekker et al., 2007) or two-way clustered standard errors (Cameron et al., 2011; Lindgren, 2010) to address network autocorrelation. However, because these methods are designed for modeling network using dyadic data and not for modeling individuals' behavior, we ultimately chose the parametric approach to fitting a fixed-effect logistic regression model.

As our regression approach weakly addresses potential network autocorrelation with the error term, there is a concern about model consistency and, thus, our model may over-identify the parameters (Lyons, 2011). This problem becomes magnified when the relationship is mutual (VanderWeele et al., 2012). Although our exposure term represents a summary function of the outcome variable (as opposed to the outcome variable itself) based on outgoing ties (as opposed to mutually nominated ties), we are aware of the potential that this technique generates invalid standard errors for performing meaningful hypothesis testing. To address the potential over-identification in our models, we conducted stepwise confirmation to check whether the unexplained variance is reduced by adding a main exposure term using a likelihood ratio test and Akaike information criterion (AIC). However, this may not be an ideal way to address the over-identification concern, and, thus, the estimated results should be interpreted with this limitation in mind.

Our statistical models assumed relational stability among the close friendship ties (such as the first nomination), as such close ties have been known to have a significant influence in changing an individual's behavior. We imputed any missing values for at least one of the socio-demographic control variables (required for approximately 24% of our sample)by using switching regression, an iterative multivariable regression technique of multiple imputations by chained equations (Royston, 2004) implemented in Stata 12.

4. Results

4.1. Descriptive statistics

Table 1 presents the descriptive statistics for each variable used for our statistical analysis. As indicated, 52% of adolescents reported having ever drunk alcohol. The average age was 15 years old, and the majority of adolescents were Hispanic (74%). For the “average” adolescent, 58% of their friends have drunk alcohol, 18% of their popular friends have drunk alcohol, and 56% of their non-multiplex friends (non-overlapped with popular nomination) have drunk alcohol.

Table 1.

Descriptive statistics: percentages or means with standard deviations and minimum and maximum values for outcome, control variables, and network exposure measures (N = 1707).

Variable Percentage or mean
(SD; min, Max)
Lifetime drinking 52%
Age 15.07 (0.43; 14, 18)
Female 48%
Hispanic/Latino 74%
Academic grades 2.60 (0.93; 1, 4)
Receipt of free or reduced-cost lunch 82%
Ratio of rooms to people in a household 0.93 (0.64; 0.17, 6)
Parental drinking 41%
Number of friendship nominations 5.45 (4.08; 0, 19)
Number of popularity nominations 1.35 (1.92; 0, 7)
Popularity
  Indegree based on friendship 5.45 (3.72; 0, 23)
  Indegree based on popularity 1.35 (2.87; 0, 35)
Network exposure
  Friends' drinking 0.58 (0.34; 0, 1)
  Popular friends' drinking 0.18 (0.37; 0, 1)
  Non-popular friends' drinking 0.56 (0.36; 0, 1)
  Multiplex congruity
  (popular friends rank order)		 0.18 (0.29; 0, 1)
Open in a new tab

Note. The variable of “Hispanic/Latino” includes Central American, Chicano or Chicana, Hispanic, Latino or Latina, Mexican, Mexican-American, or South American. Academic grades range from 1 for “mostly D's or F's” to 4 for “mostly A's.” Indegrees and outdegrees were computed based on valid nominations. % of lifetime drinking includes 12.6% of missing values.

4.2. Descriptive statistics for multiplexed relationship

With regard to frequency of friendship ties overlapping with the popularity ties, 1237 adolescents (73%) had no overlapped popular friends, while 232 adolescents (14%) named one popular friend,108 adolescents (6%) named two popular friends, 50 adolescents (3%) named three popular friends, 40 adolescents (2%) named four popular friends, and 40 adolescents (2%) nominated five or more popular friends. In sum, 470 adolescents (27%) named at least one friend whom they perceived as popular. This result indicates that approximately two-thirds of adolescents did not have friends with peers whom they perceived as popular (see summary table in the Online Supplement).

Fig. 1 illustrates the average rank/order of the popular nomination in the friendship list (1st to 19th) among adolescents who have at least one popular friend (N = 470). For approximately 19% of the adolescents, their popular friend(s) was either their first or, for those with more than one, on average, less than the second nomination; for 25%, the popular friend(s) was their second, or the average was less than their third nomination; and so on.

Fig. 1.

Fig. 1

Open in a new tab

Average rank of the popular nominations in the friendship list (1st to 19th) among Adolescents with at least one popular friend (N= 470).

The average rank across all nominated popular friends was 3.69 (SD = 2.59). These results indicate a wide variation in the order of friendship nomination list among popular friends. Therefore, we need to consider the situation where the ego may nominate popular peers, but these nominations are their distant friends, or the case where there is only one nomination, but it is their “best” friend. The multiplex congruity measure quantifies the extent to which the nomination rank of one type of relationship is congruous with the nomination rank of the other one, as described above.

4.3. Fixed-effect logistic regression results

Table 2 presents the estimated adjusted odds ratios (AOR) of lifetime drinking based on the fixed-effect logistic regression models.

Table 2.

Estimated adjusted odds ratios (AOR) of lifetime drinking using fixed effect logistic regression model (N = 1707).

Variable Model 1 Model 2 Model 3
Age 1.51** (0.22) 1.48** (0.22) 1.47* (0.22)
Female 1.36† (0.16) 1.31* (0.16) 1.29* (0.16)
Hispanic/Latino 2.41*** (0.38) 2.26*** (0.37) 2.28*** (0.37)
Academic grades 0.63*** (0.05) 0.64*** (0.05) 0.64*** (0.05)
Receipt of free
  or reduced-cost lunch		 1.19 (0.22) 1.20 (0.22) 1.21 (0.22)
Ratio of rooms to people
  in a household		 1.10 (0.11) 1.10 (0.11) 1.09 (0.11)
Parental drinking 2.13*** (0.26) 2.13*** (0.27) 2.13*** (0.27)
Number of friendship
  nominations		 0.96* (0.02) 0.99 (0.02) 0.99 (0.02)
Number of popularity
  nominations		 0.99 (0.03) 1.01 (0.04) 0.96 (0.04)
Popularity
  Indegree based
  on friendship		 1.15† (0.09) 1.16† (0.10) 1.16† (0.10)
  Indegree based
  on popularity		 1.37*** (0.11) 1.34*** (0.11) 1.35*** (0.11)
Network exposure
  Friends' drinking 1.88** (0.36) 1.07 (0.53) 0.91 (0.46)
  Popular friends'
  drinking		 – 2.23 (1.36) 2.55 (1.59)
  Non-popular friends'
  drinking		 – 3.51* (1.84) 3.89* (2.06)
  Multiplex congruity
  (popular friends
  rank order)		 – – 3.10* (1.44)
Open in a new tab

Note.

†

p < 0.1;

*

p < 0.05;

**

p < 0.01;

***

p < 0.001 for two-tailed test. Parentheses show standard errors. All models include four dummy school identifiers. A set of dummy variables indicating zero exposure terms was controlled for in the decomposed exposure terms (not included in the table).

Model 1 included the main effects of network exposures to friends' drinking and popular peers' drinking sequentially. The likelihood ratio test between the full and reduced models (with and without popularity main exposure effect) was non-significant, with a chi-square value of 1.62 (df = 1, p = 0.204). Comparison of the AIC values also indicated similar results (AIC for full model = 1.2079, and AIC for reduced model = 1.2082). Therefore, we excluded the main effect of the popularity exposure in specifying our Model 1. The results showed that friends' drinking was significantly associated with an adolescent's drinking (AOR = 1.88, p < 0.01). This result is consistent with those of previous network studies that have reported robust associations between drinking and having drinking friends. The result also indicated that exposure to popular peer's drinking was not associated with an adolescent's drinking.

Model 2 added two decomposed exposure terms that segregated network exposure into popular and non-popular friends' drinking and included the corresponding zero-dummy variables to Model 1. The results showed that an adolescent's drinking was significantly associated with decomposed exposure to non-popular friends' drinking (AOR = 3.51, p < 0.05) but not with popular friends' drinking at alpha = 0.05 level (two-tailed test). This result for drinking similarity between friends and an adolescent support differential associations or social learning theory and is consistent with previous network studies based on these theoretical frameworks. However, the non-significant association between popular friends' drinking and an adolescent's drinking is counter to the concept of network multiplexity, in which we expect this association to be stronger when friendships overlap.

There are a few additional points to note with regard to the results of Model 2. First, the association between friends' drinking and an adolescent's drinking (network exposure term in Model 1) became non-significant after inclusion of the decomposed exposure measures. This result indicates that the association between drinking and exposure to friends' drinking (regardless of whether an adolescent perceived them as popular) may be the perceived status of those peers. Because the decomposed exposure to friends' drinking is nested within exposure to all friends, these two exposure terms are inherently correlated (the standardized correlation coefficient for two parameters was −0.86). After removing the term for exposure to all friends from Model 2, the association between the decomposed exposure to friends' drinking and self-drinking became more significant (AOR = 3.72, SE = 1.09, p < 0.001).

Model 3 added the multiplex congruity exposure term to Model 2. The results indicated a significant association between multiplex contiguity exposure to drinkers and an adolescent's drinking (AOR = 3.10, p < 0.05). This result indicates that, as adolescents are exposed to peer drinkers who are closer friends and perceived to be popular, they are more likely to drink. The correlation of parameter estimates between the zero dummy of the decomposed exposure to popular friends' drinkers and multiplex congruity exposure was moderate (−0.77). Subsequent multicollinearity diagnostic analysis to measure the strength of interrelationships among all exposure terms in Model 3 indicated that none provided clear evidence of multicollinearity, as evaluated with a VIF greater than 10 or tolerance smaller than 0.1. These results should be interpreted with caution by taking these issues into account.

As for the effects of different measures of an adolescent's popularity, only indegree based on perceived popularity was significantly associated with an adolescent's drinking (AOR = 1.35, p < 0.001). In comparison, indegree based on friendship network was marginally associated with lifetime drinking (AOR = 1.16, p < 0.10).

5. Discussion

The literature shows that friendships and popularity can create risks for adolescent substance use behavior. Most existing network studies take a unitary view of peer context derived solely from the friendship network. Exposure and popularity are calculated from this network, and these measures are correlated with individual substance use behaviors. Friendship consists of personal likability (sociometric popularity), which may be different from popularity conceived of as status or prestige that does not necessarily consist of an affective relationship between peers (perceived popularity). This study took a multiplex view of the peer context by measuring both friendship and perceived popularity and decomposing friendship into those that overlapped and did not overlap with popularity. Each of these measures was then correlated with adolescent drinking. The findings supported social psychological theories of differential association and social learning: Interaction with intimate friends was associated with similarity in behavior.

The findings did not, however, support the prevailing social network concept of network multiplexity in which multiplex relationships are conceived of as stronger and, hence, more likely to be associated behavioral similarity. Instead, we find that multiplex congruity matters in assessing the similarity in drinking between friends and adolescents themselves. Specifically, the more congruent friendship was with a popularity relationship, the more likely it was for an association between friends' drinking and an adolescent's drinking.

Differential patterns of association in behavioral similarity as dependent on different types of relationships could be explained by a tendency for adolescents to tap into different types of peer networks for different types of health outcomes. Bearman and Parigi (2004) introduced the idea of “role-topic dependency,” that is, the type of topics people discuss depends on with whom they talk. For example, the “important matters” network is often associated with the achievement of instrumental outcomes (Bearman and Parigi, 2004). Conversely, if the matters are not important, people talk about them with people who are not important. Based on this argument, diffusion of alcohol use might flow through certain types of networks but not others: popular friends who are actually considered friends and have close affective ties rather than popular friends who are not really friends. Perhaps different health outcomes, such as smoking cigarettes or using marijuana, may be associated with other types of networks.

Our findings also indicated that reputation-based popularity measured by a perception of who is popular was more strongly associated with self-drinking, as compared to a preference-based popularity based on friendship nominations. This result is consistent with the popularity-socialization theory, which states that the more popular an adolescent, the more likely he or she is to engage in delinquent behavior (Allen et al., 2005). It should be noted that peer status in this context is more applicable to reputation-based popularity (perceived popularity), as this type of popularity entails peer dominance.

On a methodological level, this study offered new insights into social network and health behavioral research in two respects. First, this study introduced the network concept of multiplex congruity for use when analyzing multiplex networks and demonstrated the importance of taking into account information about the relative importance of one type of relationship compared to another type of relationship. Second, the study demonstrated the utility of a DNEM that has the capability to separate friends' drinking into the effects of non-popular and popular friends' drinking. One of the advantages of this approach is that it can be used to model the multiplexity (or overlap) in any network. For example, the overlap of friendships and romance, the overlap of one-mode and two-mode networks (Fujimoto and Valente, 2013), or two different two-mode (affiliation) networks (Fujimoto et al., 2013) could be tested to assess behavioral similarity. Both multiplex relationships and congruity between multiplex networks have been understudied in both social network analysis and in health, and we believe that this study takes a step in that direction.

Our study has limitations that need to be taken into consideration. First, the study used cross-sectional, not longitudinal, data, so we are limited in our ability to understand the mechanisms of peer associations (both uniplex and multiplex relationships) in relation to substance use. Our finding of the significant association between individual drinking and drinking by non-popular friends or congruent friends could come from peer selection, the tendency for peers to select friends who engage in similar behaviors. A recent study has reported that reputation-based popularity moderated friendship selection based on alcohol use among tenth graders, with popular adolescents' being more likely than less-popular adolescents to select friends with frequent alcohol use (Mathys et al., 2013). Future research should take this issue into account. Second, our regression approach to statistically test an association between drinking by different types of peer associates and individual drinking does not take network dependencies into account. Perhaps friendships among non-popular friends are formed by having another non-popular friend in common. In turn, adolescent drinking behaviors may likely lead to a clustering of drinkers among connected people. In addition, these models may be biased due to potential over-identification of the parameters, and, thus, the results of our statistical testing need to be interpreted with some caution. Future research will need to verify our findings by taking a different non-parametric approach to statistically test network effects whose utilities have been demonstrated in prior research (Christakis and Fowler, 2013; Dekker et al., 2007).

Despite these limitations, our study offers insight into the utility of investigating the multiplex congruity of peer relationships in relation to adolescent drinking behavior. The findings may relate to the design of network interventions (Valente, 2010, 2012) and inform policy implications for school-based substance use prevention programs. Interventionists should consider friendship in combination with perceived popularity (and other networks). Indeed, an adolescent's peer context is composed of different dimensions of structural relationships, including friendship, popularity, and romance. Similarity in drinking between friends and adolescents is conditioned on how adolescents perceive their friends' popularity; placing their friends either outside or inside the popularity network will shift these friends in terms of their relative importance based on their friendship order. Additionally, interventions that recruit popular adolescents as opinion leaders to aid in program delivery based on status will be potentially more effective than those who recruit popular adolescents based on friendship. To assess the role of multiple networks, the research should examine how social relationships correspond to one another first and explore different combination possibilities.

Supplementary Material

01

Acknowledgment

We gratefully acknowledge Chris Robertson, Bin Zhang, Erik Lindsley, anonymous reviewers, and guest editors for their contributions in completing this article. This study was support by NIH/ NIAAA 4R00AA019699-03, NIH/NIAAA 1RC1AA019239-01, and 1R01MH100021.

Footnotes

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.socscimed.2014.05.023.

References

  1. Akers RL, Krohn MD, Lanza-Kaduce L, Radosevich M. Social learning and deviant behavior: a specific test of a general theory. Am. Sociol. Rev. 1979;44(4):636–655. [PubMed] [Google Scholar]
  2. Akers RL, Lee G. A longitudinal test of social learning theory: adolescent smoking. J. Drag. Issues. 1996;26:317–343. [Google Scholar]
  3. Alexander C, Piazza M, Mekos D, Valente TW. Peers, schools, and adolescent cigarette smoking. J. Adolesc. Health. 2001;29(1):22–30. doi: 10.1016/s1054-139x(01)00210-5. [DOI] [PubMed] [Google Scholar]
  4. Allen JP, Porter MR, McFarland CF, Marsh PA, McElhaney KB. The two faces of adolescents' success with peers: adolescent popularity, social adaptation, and deviant behavior. Child. Dev. 2005;76:747–760. doi: 10.1111/j.1467-8624.2005.00875.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bandura A. Social Learning Theory. Prentice-Hall; Englewood Cliffs, NJ: 1977. [Google Scholar]
  6. Bearman P, Parigi P. Cloning headless frogs and other important matters: conversation topics and network structure. Soc. Forces. 2004;83(2):535–557. [Google Scholar]
  7. Becker BE, Luthar SS. Peer-perceived admiration and social preference: contextual correlates of positive peer regard among suburban and urban adolescents. J. Res. Adolesc. 2007;17(1):117–144. doi: 10.1111/j.1532-7795.2007.00514.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Brown BB. Peer groups and peer cultures. In: Feldman SS, Elliott GR, editors. At the Threshold: the Developing Adolescent. Harvard University Press; Cambridge, MA: 1990. pp. 171–196. [Google Scholar]
  9. Burgess RL, Akers RL. A differential association-reinforcement theory of criminal behavior. Soc. Probl. 1966;14:128–147. [Google Scholar]
  10. Cameron AC, Gelbach JB, Miller DL. Robust inference with multiway clustering. J. Bus. Econ. Stat. 2011;29(2):238–249. [Google Scholar]
  11. Christakis NA, Fowler JH. Social contagion theory: examining dynamic social networks and human behavior. Stat. Med. 2013;32:556–577. doi: 10.1002/sim.5408. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cillessen AHN, Marks PEL. Conceptualizing and measuring popularity. In: Cillessen AHN, Schwartz D, Mayeux L, editors. Popularity in the Peer System. The Guilford Press; New York, NY: 2011. pp. 25–56. [Google Scholar]
  13. Cleveland HH, Wiebe RP. The moderation of adolescent-to-peer similarity in tobacco and alcohol use by school levels of substance use. Child. Dev. 2003;74(1):279–291. doi: 10.1111/1467-8624.00535. [DOI] [PubMed] [Google Scholar]
  14. Coie JD, Dodge KA, Coppotelli H. Dimensions and types of social status: a cross-age perspective. Dev. Psychol. 1982;18:121–132. [Google Scholar]
  15. Coleman J. The Adolescent Society. The Free Press of Glencoe; New York, NY: 1961. [Google Scholar]
  16. Crosnoe R. The connection between academic failure and adolescent drinking in secondary school. Sociol. Educ. 2006;79(1):44–60. doi: 10.1177/003804070607900103. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Crosnoe R, Muller C, Frank KA. Peer context and the consequences of adolescent drinking. Soc. Probl. 2004;51(2):288–304. [Google Scholar]
  18. Dekker D, Krackhardt D, Snijders TAB. Sensitivity of MRQAP tests to collinearity and autocorrelation conditions. Psychometrika. 2007;72(4):563–581. doi: 10.1007/s11336-007-9016-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Dijkstra JK, Cillessen AHN, Borch C. Popularity and adolescent friendship networks: selection and influence dynamics. Dev. Psychol. 2012;49:1242–1252. doi: 10.1037/a0030098. [DOI] [PubMed] [Google Scholar]
  20. Doreian P. In: Network autocorrelation models: problems and prospects. Griffith DA, editor. Spatial Statistics: Past, Present, Future. Michigan Document Services; Ann Arbor, MI: 1989. [Google Scholar]
  21. Dow MM. A bi-parametric approach to network autocorrelation: Galton's problem. Sociol. Methods Res. 1984;13:201–217. [Google Scholar]
  22. Dow MM. Network autocorrelation regression with binary and ordinal dependent variables. Cross-Cultural Res. 2008;42(4):394–419. [Google Scholar]
  23. Ennett ST, Bauman KE, Hussong A, Faris R, Foshee VA, Cai L, DuRant RH. The peer context of adolescent substance use: findings from social network analysis. J. Res. Adolesc. 2006;16(2):159–186. [Google Scholar]
  24. Fujimoto K. Using mixed-mode networks to disentangle multiple sources of social influence. In: Yang SJ, Greenberg AM, Endsley M, editors. Lecture Notes in Computer Science (LCNS) Vol. 7227. Springer, Verlag; Germany: 2012. pp. 214–221. [Google Scholar]
  25. Fujimoto K, Valente TW. Decomposing the components of friendship and friends' influence on adolescent drinking and smoking. J. Adolesc. Health. 2012a;51:136–143. doi: 10.1016/j.jadohealth.2011.11.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Fujimoto K, Valente TW. Network influence on adolescent substance use: disentangling structural equivalence from cohesion. Soc. Sci. Med. 2012b;74(12):1952–1960. doi: 10.1016/j.socscimed.2012.02.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Fujimoto K, Valente TW. Alcohol peer influence from participating in organized school activities among U.S. adolescents: a network approach. Health Psychol. 2013;32(10):1084–1092. doi: 10.1037/a0029466. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Fujimoto K, Wang P, Valente TW. The decomposed affiliation exposure model: a network approach to segregating peer influences from crowds and organized sports. Netw. Sci. 2013;1(2):154–169. doi: 10.1017/nws.2013.7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Hartup WW. Friendships and their developmental significance. Child. Dev. 1996;67:1–13. [PubMed] [Google Scholar]
  30. Haynie DL. Delinquent peers revisited: does network structure matter? Am. J. Sociol. 2001;106:1013–1057. [Google Scholar]
  31. Hoff P, Fosdick B, Volfovsky A, Stovel K. Likelihoods for fixed rank nomination networks. Netw. Sci. 2013;1(3):253–277. doi: 10.1017/nws.2013.17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Hosmer DW, Lemeshow S. Applied Logistic Regression. Wiley; New York, NY: 2000. [Google Scholar]
  33. Kobus K, Henry DB. Interplay of network position and peer substance use in early adolescent cigarette, alcohol, and marijuana. J. Early Adolesc. 2010;30(2):225–245. [Google Scholar]
  34. Koehly LM, Pattison P. Random graph models for social networks: multiple relations or multiple raters. In: Carrington PJ, Scott J, Wasserman S, editors. Models and Methods in Social Network Analysis. Cambridge University Press; New York, NY: 2005. pp. 162–214. [Google Scholar]
  35. Krohn MD. The web of conformity: a network approach to the explanation of delinquent behavior. Soc. Probl. 1986;33(6):S81–S93. [Google Scholar]
  36. Krohn MD, Massey JL, Zielinski M. Role overlap, network multiplexity, and adolescent deviant behavior. Soc. Psychol. Q. 1988;51(4):346–356. [Google Scholar]
  37. Latendresse S, Rose RJ, Viken RJ, Pulkkinen L, Kaprio J, Dick DM. Parenting mechanisms in links between parents' and adolescents' alcohol use behaviors. Alcohol. Clin. Exp. Res. 2008;32(2):322–330. doi: 10.1111/j.1530-0277.2007.00583.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Lazega E, Pattison P. Multiplexity, generalized exchange and cooperation in organizations. Soc. Netw. 1999;21:67–90. [Google Scholar]
  39. Lindgren KO. Dyadic regression in the presence of heteroscedasticity–an assessment of alternative approaches. Soc. Netw. 2010;32:279–289. [Google Scholar]
  40. Lyons R. The spread of evidence-poor medicine via flawed social-network analysis. Stat. Polit. Policy. 2011;2(1):1–26. [Google Scholar]
  41. Mathys C, Burk WJ, Cillessen AHN. Popularity as a moderator of peer selection and socialization of adolescent alcohol, marijuana, and tobacco use. J. Res. Adolesc. 2013;23(3):513–523. [Google Scholar]
  42. Mayeux L, Sandstrom MJ, Cillessen AHN. Is being popular a risky proposition? J. Res. Adolesc. 2008;18:49–74. [Google Scholar]
  43. Michell L, Amos A. Girls, pecking order and smoking. Soc. Sci. Med. 1997;44(12):1861–1869. doi: 10.1016/s0277-9536(96)00295-x. [DOI] [PubMed] [Google Scholar]
  44. Moody J, Brynildsen WD, Osgood DW, Feinberg ME, Gest S. Popularity trajectories and substance use in early adolescence. Soc. Netw. 2011;33(2):101–112. doi: 10.1016/j.socnet.2010.10.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. National Institute on Drug Abuse. Monitoring the Future: National Results on Adolescent Drug Use. 2011 Retrieved from http://monitoringthefuture.org/pubs/monographs/mtf-overview2011.pdf.
  46. Osgood DW, Ragan DT, Wallace L, Gest SD, Feinberg ME, Moody J. Peers and the emergence of alcohol use: influence and selection processes in adolescent friendship networks. J. Res. Adolesc. 2013;23(3):500–512. doi: 10.1111/jora.12059. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Parkhurst JT, Hopmeyer A. Sociometric popularity and peer-perceived popularity: two distinct dimensions of peer status. J. Early Adolesc. 1998;18:125–144. [Google Scholar]
  48. Payne DC, Cornwell B. Reconsidering peer influences on delinquency: do less proximate contacts matter? J. Quant. Criminol. 2007;23:127–149. [Google Scholar]
  49. Pearson M, Sweeting H, West P, Young R, Gordon J, Turner K. Adolescent substance use in different social and peer contexts: a social network analysis. Drugs Educ. Prev. Policy. 2006;13(6):519–536. [Google Scholar]
  50. Peters E, Cillessen AHN, Riksen-Walraven JM, Haselager GJT. Best friends' preference and popularity: associations with aggression and prosocial behavior. Int. J. Behav. Dev. 2010;34(5):398–405. [Google Scholar]
  51. Robertson C, Boyle P, Hsieh C, MacFarlane GJ, Maisonneuve P. Some statistical considerations in the analysis of case-control studies when the exposure variables are continuous measurements. Epidemiology. 1994;5(2):164–170. doi: 10.1097/00001648-199403000-00006. [DOI] [PubMed] [Google Scholar]
  52. Royston P. Multiple imputation of missing values. Stata J. 2004;4(3):227–241. [Google Scholar]
  53. Sandstrom MJ. The power of popularity: influence processes in childhood and adolescence. In: Cillessen AHN, Schwartz D, Mayeux L, editors. Popularity in the Peer System. The Guilford Press; New York, NY: 2011. pp. 219–244. [Google Scholar]
  54. Schaefer DR, Haas SA, Bishop NJ. A dynamic model of US adolescents' smoking and friendship networks. Am. J. Public Health. 2012;102(6):E12–E18. doi: 10.2105/AJPH.2012.300705. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Schwartz D, Gorman AH. The high price of high status: popularity as a mechanism of risk. In: Cillessen AHN, Schwartz D, Mayeux L, editors. Popularity in the Peer System. The Guilford Press; New York, NY: 2011. pp. 245–273. [Google Scholar]
  56. Skvoretz J, Agneessens F. Reciprocity, multiplexity, and exchange: measures. Qual. Quantity. 2007;41:341–357. [Google Scholar]
  57. Snijders TAB, Lomi A, Torlo VJ. A model for the multiplex dynamics of two-mode and one-mode networks, with an application to employment preference, friendship, and advice. Soc. Netw. 2013;35:265–276. doi: 10.1016/j.socnet.2012.05.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Sutherland EH. Principles of Criminology. fourth ed. J. B. Lippincott; Philadelphia, PA: 1947. [Google Scholar]
  59. Sutherland EH, Cressey DR. Criminology. J. B. Lippincott; Philadelphia, PA: 1978. [Google Scholar]
  60. Urberg KA, Değirmencioğlu SM, Pilgrim C. Close friend and group influence on adolescent cigarette smoking and alcohol use. Dev. Psychol. 1997;33(5):834–844. doi: 10.1037//0012-1649.33.5.834. [DOI] [PubMed] [Google Scholar]
  61. Valente TW. Network models and methods for studying the diffusion of innovations. In: Carrington PJ, Scott J, Wasserman S, editors. Models and Methods in Social Network Analysis: Structural Analysis in the Social Sciences. Cambridge University Press; New York: 2005. pp. 98–116. [Google Scholar]
  62. Valente TW. Social Networks and Health: Models, Methods, and Applications. Oxford University Press; New York, NY: 2010. [Google Scholar]
  63. Valente TW. Network interventions. Science. 2012;337(6):49–53. doi: 10.1126/science.1217330. [DOI] [PubMed] [Google Scholar]
  64. Valente TW, Fujimoto K, Soto D, Ritt-Olson A, Unger JB. A comparison of peer influence measures as predictors of smoking among predominately Hispanic/Latino high school adolescents. J. Adolesc. Health. 2013a;52(3):358–364. doi: 10.1016/j.jadohealth.2012.06.014. http://dx.doi.org/10.1016/j.jadohealth.2012.06.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Valente TW, Fujimoto K, Unger JB, Soto D, Meeker D. Variations in network boundary and type: a study of adolescent peer influences. Soc. Netw. 2013b;35(3):309–316. [Google Scholar]
  66. Valente TW, Gallaher P, Mouttapa M. Using social networks to understand and prevent substance use: a transdisciplinary perspective. Subst. Use Misuse. 2004;39:1685–1712. doi: 10.1081/ja-200033210. [DOI] [PubMed] [Google Scholar]
  67. Valente TW, Unger JB, Johnson CA. Do popular students smoke? The association between popularity and smoking among middle school students. J. Adolesc. Health. 2005;37(4):323–329. doi: 10.1016/j.jadohealth.2004.10.016. [DOI] [PubMed] [Google Scholar]
  68. VanderWeele TJ, Ogburn EL, Tchetgen EJT. Why and when "flawed" social network analyses still yield valid tests of no contagion. Stat. Polit. Policy. 2012;3:1–11. doi: 10.1515/2151-7509.1050. [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Wang P, Robins G, Pattisona P, Lazega E. Exponential random graph models for multilevel networks. Soc. Netw. 2013;35(1):96–115. http://dx.doi.org/10.1016/j.socnet.2013.01.004. [Google Scholar]
  70. Zhang B, Thomas AC, Doreian P, Krackhardt D, Krishnan R. Contrasting multiple social network autocorrelations for binary outcomes, with applications to technology adoption. ACM Trans. Manag. Inf. Syst. 2013;3(4):18:11–18:21. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

01
NIHMS601518-supplement-01.doc (892KB, doc)

RESOURCES