We used a systems science perspective to examine adolescents' personal networks, school networks, and neighborhoods as a system through which emotional support and peer influence flow, and we sought to determine whether these flows affected past-month smoking at 2 time points, 1994–1995 and 1996. To test relationships, we employed structural equation modeling and used public-use data from the National Longitudinal Study of Adolescent Health (n = 6504). Personal network properties affected past-month smoking at both time points via the flow of emotional support. We observed a feedback loop from personal network properties to emotional support and then to past-month smoking. Past-month smoking at time 1 fed back to positively affect in-degree centrality (i.e., popularity). Findings suggest that networks and neighborhoods in this system positively affected past-month smoking via flows of emotional support.

Adolescent cigarette smoking remains a complex public health problem in the United States. Although lifetime smoking and current frequency of smoking among adolescents decreased between the late 1990s and 2003, prevalence remained unchanged from 2003 through 2005.1 Smoking prevalence among adolescents is currently estimated to be around 23%,1 posing ongoing challenges for tobacco-control efforts.

Several streams of literature suggest that adolescent smoking is inextricably connected to the social context in which it occurs. Literature shows that an adolescent's smoking behavior will tend to be similar to that of his or her peers.26 There is a longstanding debate over why this similarity occurs; some studies suggest that it is caused by peer influence on an individual adolescent's smoking,7,8 whereas others suggest that it is caused by the individual's selection of smoking peers,8 and still others attribute the cause to both influence and selection.2 Some of the literature implicating adolescents’ social contexts in their smoking behavior examines youths’ social networks of friends and peers from a structural perspective. Such studies focus on how structural and positional characteristics of these networks relate to adolescent smoking. Structural characteristics reflect information about linkages among individuals, and positional characteristics indicate the significance of occupying different network positions. In general, studies find that isolated youths are likely to smoke, although some studies have found that popular youths are likely to smoke.35,912 Implicit in each study is the notion that adolescents’ social context of friends and peers plays a critical role in their own smoking behavior.

Given the relevance of adolescents’ social context to their smoking behavior, it is important to understand how to conceptualize and measure this context. Previous research has used ecological models to inform the theoretical specification of the context of adolescent smoking and other substance-use behaviors.13 Ecological models allow this context to be theoretically partitioned into levels of influence. Although there are valuable insights yet to be gained from such models, more theoretically informed research is necessary to elaborate the complexity of the social context of adolescent smoking. Moreover, because various theories are often integrated at different levels in such models, it is difficult to ensure that conceptual coherence is achieved across and within levels, given the possibility that the theories applied at each level make incongruous assumptions. Furthermore, such models do not provide specific guidance about mechanisms through which levels of influence relate to outcomes such as adolescent smoking. There is a need for theoretical models that more specifically and holistically elaborate features of the social context of adolescent smoking and how they act in concert.

In this study, we incorporate valuable insights from ecological models that theoretically partition the environment into levels of influence to frame the social context of adolescent smoking as a complex social system. We employed a systems science14 approach in conceptualizing the social context of adolescent smoking, which emphasizes interdependence in complex relationships among people or organizations.15 The defining features of systems are (1) parts or components yielding a whole that is greater than the sum of the parts, (2) inputs and flows coursing through a system, and (3) feedback loops linking parts of a system. Such system features hold promise for informing theoretical models that elaborate adolescents’ social networks, the dimensionality of their social relationships, and the interdependence among levels of influence within their social context.

We conceptualized 3 key structures in the social environment of adolescent smoking as structural components defining the system under study: (1) personal networks of friends, (2) school networks, and (3) neighborhoods. School networks were defined as whole networks of all students in a school and the social ties among them. These networks were constructed from adolescents’ nominations of up to 10 friends from a roster listing all names of adolescents in their own high school and a geographically proximal “sister” junior or senior high school. Adolescents’ personal networks were subsets of whole school networks and comprised friends nominated within their schools and their friends’ ties. The neighborhoods under study were the physical areas where the adolescents lived. We investigated how characteristics of these 3 structural components influenced adolescent smoking via flows of 2 social processes: emotional support and peer influence (Figure 1). We hypothesized that these social processes were mechanisms through which the structural components under study influenced smoking.

In this introduction, we focused on the paths of main theoretical interest depicted in this model; we refer to the other constructs in the Methods section only. Throughout this article, we present the model, findings, and discussion of the results in terms of the causal directionality assumed in our model, but only for ease of exposition and not for the purpose of making any causal claims.

For this study, we focused on properties of personal and school networks that are relevant to the flow of network resources throughout the larger system under study and that have been related to adolescent smoking. Network ties carry resources that flow through a network and are exchanged by network members, such as social influence and social support. At the personal network level, conceptualized at the level of the adolescent, we focused on (1) adolescents’ popularity, or their in-degree centrality; (2) whether their friendships are mutually reciprocated; (3) personal network density, or the extent to which those named in adolescents’ networks know one another; (4) the social distance between adolescents in their networks or the number of path lengths between them, which are relationship ties linking individuals in a network16; and (5) the number of people they nominated as friends outside of their schools.

At the school network level, we focused on the network properties of size, density, and the mutuality of ties, all of which may affect smoking behavior.17 Size refers to the number of students in a school, density is the extent to which students know one another in a school, and mutuality is the extent to which youths’ relationships are mutually reciprocated within the broader school context. School-level network characteristics may affect the structure and positional attributes of personal networks as larger-scale versions of these constructs.

The theoretical intuition underlying each of the aforementioned network properties is what makes them relevant to adolescent smoking from a systems perspective. Popular youths are directly connected to many others and can quickly and disproportionately transmit and receive network resources, such as support. Being central or popular in networks is positively related to smoking among youths.11,12 Having reciprocated friendships may facilitate the flow of network resources throughout a network, as resources are likely exchanged in mutual friendships, both in personal networks and in school networks. One study found that adolescents with reciprocated ties with a best friend were less likely to smoke at age 11 and 13 years than those without such ties.4

Personal network density reflects the extent to which those in an adolescent's personal network know one another. The density of ties, either in personal networks or whole networks, likely plays a role in the flow of resources throughout a network by binding people together and strengthening adherence to pervasive norms and beliefs. Dense ties can also limit the inflow of resources from outside a network. Adolescents aged 15 years who had dense local networks were found to have lower odds of recent smoking.4 If adolescents are close to one another in a network—that is, if path length is low—then social influences and network resources may be easily transmitted because the probability of transmission decreases over longer path lengths.18 Previous research has found that adolescents who were socially proximal to a smoker had increased odds of smoking.4 Lastly, the number of people youths nominated as “friends” outside of their schools captures relationships that either reinforce or attenuate the effect of social influences from in-school friends, depending on the types of influences exerted by each of these friends. There is evidence that having ties to friends from outside one's school is positively related to adolescent smoking.4

Although previous research offers insight into how each of these network properties relates to adolescent smoking, smoking behavior may affect the structure and position of individuals in a network via network processes such as influence, social support, and selection. We explored this notion in our study by positing feedback pathways from smoking to the network characteristics under study.

Finally, we took adolescents’ neighborhoods into account, utilizing insights from social disorganization theory.19,20 This ecological theory posits that the key structural characteristics of economic disadvantage, racial/ethnic heterogeneity, and residential instability lead to an overall milieu of social disorganization within certain neighborhoods, giving rise to higher levels of delinquency. In such neighborhoods, this disorganization limits residents’ ability to provide the informal social control that would otherwise reduce adolescents’ delinquent behavior. Although the bulk of research using this theory has focused on the generation of criminal forms of delinquency,2123 recent research has tested whether these structural characteristics also increase the delinquent behavior of cigarette smoking.2426

To study how the structural components of interest exert influence on adolescent smoking, we hypothesized that 2 social processes—emotional support and peer influence—may be mechanisms through which properties of these structural components influence adolescent smoking. Emotional support relates to health through both direct effects and buffering effects.27 Social support may be a pathway linking social networks and health indicators; studies have related structural network characteristics to health via emotional support and social influence processes.28 The findings of previous research highlight the need for future work to focus on how social influence mechanisms may mediate the relationship between social networks and health,2830 particularly the association between certain types of supportive networks and health-compromising behavior.31 Therefore, we focused on emotional support—which generally extends to feelings of closeness, encouragement, and belonging32—and peer influence, narrowly defined as the influence exerted by adolescents’ friends who smoke, as mechanisms linking social network ties and adolescent smoking. Although more inclusive conceptualizations of peer influence exist, we focused only on the influence derived from adolescents’ friends who smoke because they may be particularly proximal to and relevant for adolescents’ smoking behavior.

Studies have found that emotional support is positively associated with smoking.33 Perhaps the closeness generated from an emotionally supportive tie reinforces social bonding as friends smoke together. An analogous finding comes from the injection drug use literature, which has shown that the provision of emotional support in relationships mediated the association between closeness of ties and needle sharing.34 Among injection drug users, needle sharing behavior was likely a symbolic act of solidarity, exclusivity, and bonding. For adolescents, smoking behavior may carry a similar symbolic meaning related to social bonding, especially in the context of emotionally supportive and valued friendships. Moreover, in some friendships receipt of emotional support may be contingent upon engaging in a delinquent behavior because failure to smoke and being different from a friend on this dimension might be perceived as a strike against the friendship. It is not surprising that emotional support has potential to compromise health because previous research has indicated that social support can reinforce delinquent behaviors among youths and their network members via modeling processes.35,36

Beyond the effect of emotional support on smoking behavior, there is reason to expect that the network characteristics under study increase emotional support. It is likely that more central individuals are in a position to provide support to others or may be connected to others who could also provide support.37 Likewise, mutually reciprocated ties, as well as the density of ties,38 likely increase emotional support through increased closeness. Also, shorter path lengths in networks likely increase emotional support because being proximal to others may lead to the provision or receipt of more emotional support.

The second social process hypothesized to potentially mediate the effect that properties of structural components have on smoking is peer influence. In this study, peer influence was conceptualized as the influence exerted by an adolescent's friends’ smoking behavior on the adolescent's own smoking behavior. The positive relationship between peer influence and smoking is well-documented.39,40 Peer influence processes have been measured in relation to smoking in numerous ways, including the number of friends who smoke.41,42 Furthermore, studies have found that network characteristics affect the number of friends who smoke. For example, studies have found that reciprocity43 and density of network ties are associated with greater peer influence, arguably through tightly binding people together in a network and amplifying social influence.44,45 It is also likely that peer influence is positively related to being central in networks, because highly central individuals can quickly receive and transmit influence, and to path length in networks, because social influences may be quickly transmitted if individuals are close to one another in a network.

Although there is likely an influence effect of friends’ smoking behavior on adolescent smoking, there is also likely a selection effect, in which persons who smoke are more likely to choose friends who also smoke. To the extent that having friends who smoke then affects one's position in the network, we suggest that there will be a feedback loop from personal network properties, to friends’ smoking behavior, to smoking at wave 1 (Figure 1).

In sum, we examined direct, mediated, and feedback pathways through which these structural components influenced smoking at 2 time points approximately 1 year apart via flows of emotional support and peer influence. Given the systems nature of the study, we examined a number of pathways in 2 categories: (1) direct pathways indicating how properties of personal and school networks and neighborhoods influenced smoking and (2) indirect pathways through which properties of networks and neighborhoods influenced smoking via emotional support and peer influence exerted by friends’ smoking behavior. Last, we investigated 2 feedback loops: (1) a loop from network properties to emotional support, to smoking, and back to the network variables and (2) a loop from personal network characteristics to past-month smoking via friends’ smoking.

The data for this study were taken from the first wave of the National Longitudinal Study of Adolescent Health (i.e., Add Health), a school-based longitudinal study consisting primarily of in-school and at-home surveys. We also included a measure of smoking from the second wave of Add Health, which occurred 1 year after the first wave. The Add Health wave 1 surveys were administered to a nationally representative sample of students in grades 7 through 12 (and their parents) from September 1994 through December 1995; wave 2 surveys were administered from April 1996 to August 1996. We used Add Health's public-use data, which comprise a random sample of 6504 individuals from the full study. Social network measures are based on a network elicitation item asking a respondent to nominate up to 5 male friends and 5 female friends; respondents could name persons outside of their school. We also used contextual data from the 1990 US Census based on the 2407 block groups of residence in the sample (block groups had an average population of about 1100 persons in 1990). The design sampled first on schools and then on students within schools; the neighborhoods under study simply arose as a byproduct of this sample design because they were defined by where adolescents in the sample lived.


The measures used in the analyses are described in Table 1, along with their summary statistics and the levels at which they were measured. To aid in identifying the model (described next), we also included 4 measures of smoking risk (measured at the individual level) that likely affected smoking behavior but not network characteristics. We included demographic characteristics that were likely related to our endogenous variables. We measured racial/ethnic heterogeneity on the basis of a dispersion formula:

where K is the number of groups, N2 is the number of persons squared, and fk is the frequency of group k (D ranges from 0 to 1).


TABLE 1 Descriptions and Summary Statistics of Study Variables: National Longitudinal Study of Adolescent Health, 1994–1996

TABLE 1 Descriptions and Summary Statistics of Study Variables: National Longitudinal Study of Adolescent Health, 1994–1996

VariableHow Variable Was MeasuredRange of ValuesMean (SD) or %
Smoking and support (individual level)
    Past-month smoking (days, logged), wave 1Number of days the respondent smoked cigarettes during the previous month (log transformed)0 = no days to 30 = 30 days4.921 (10.082)
    Past-month smoking (days, logged), wave 2Number of days the respondent smoked cigarettes during the previous month (log transformed)0 = no days to 30 = 30 days5.182 (10.299)
    Emotional support (proportion)Proportion of friends in the respondent's personal network with whom they discussed a problem in the previous 7 days0–10.347 (0.327)
    Friends’ smoking behaviorRespondent's perception of how many of their 3 best friends smoked0 = no friends to 3 = 3 friends0.817 (1.066)
Personal network measures (individual level)
    In-degree centralityThe number of persons in the network who nominated the respondent as a friend1–31 people5.551 (3.692)
    Personal network densityThe number of existing ties in a respondent's network divided by the total possible number of tiesTheoretically from 0 to 1; actually from 0.09 to 10.412 (0.203)
    Mean distance to reachable peopleThis is computed by (1) determining all the people the respondent could reach in the network either directly or indirectly, (2) computing the minimum number of path lengths to reach each person, and (3) computing the mean of those distances1–21.395.284 (1.620)
    Ties outside the schoolNumber of people nominated as friends who were not members of the respondent's school0–101.406 (2.144)
    Best male friend reciprocatesWhether the respondent's best male friend reciprocated their tie choice0 = did not reciprocate, 1 = reciprocate54.4
    Best female friend reciprocatesWhether the respondent's best female friend reciprocated their tie choice0 = did not reciprocate, 1 = reciprocated62.7
School network measures (school level)
    School network densityThe proportion of existing ties to the number of possible ties in a schoolTheoretically from 0 to 1; actually from nearly 0 to 0.350.017 (0.037)
    Size of school networkNumber of persons in the school network30–2559 students671.5 (488.5)
    Mutuality indexThe tendency for ties to be reciprocated relative to expectations on the basis of chance; higher values indicate more mutualityTheoretically from 0 to 1; actually from 0.23 to 0.530.377 (0.052)
Neighborhood measures (block group level)
    Median home valueMedian value of homes in block group, 1990$15 000–$300 00095 407 (62 950)
    Racial/ethnic heterogeneityBased on a dispersion formula0–10.340 (0.294)
    Residential stabilityThe proportion of residents who moved into their housing unit between 1985 and 1990, categorized into 3 groups, with 1 standard deviation above and below the mean as the cutoffsLow, medium, high1.996 (0.562)
Demographic characteristics and smoking risk variables (individual level)
    AgeAge of the respondent at the time of the survey10–19 y14.871 (1.729)
    Mother's educationHighest level of mother's educational achievement1 = eighth grade or less, 2 = ninth to 12th grade, 3 = high school graduate or GED, 4 = vocational school, 5 = some college, 6 = graduated from college, 7 = professional or graduate training5.275 (2.344)
    Parent smokingNumber of parents who smoked0 = none, 1 = 1 parent smokes, 2 = both parents smoke1.050 (0.781)
    Wear seatbeltsFrequency of wearing seatbelts when riding in a car0 = never, 1 = rarely, 2 = sometimes, 3 = most of the time, 4 = always3.071 (1.190)
    Motorcycle ridingFrequency of riding a motorcycle in the previous 12 months0 = never, 1 = once or twice, 2 = about once a month, 3 = about once a week, 4 = almost every day0.363 (0.863)
    Cigarettes in homeWhether cigarettes are easily available in the home0 = no, 1 = yes30.5
    African AmericanSelf-reported race/ethnicity1 = African American, 0 = not24.6
    AsianSelf-reported race/ethnicity1 = Asian, 0 = not4.4
    LatinoSelf-reported race/ethnicity1 = Latino, 0 = not13.5
    Other raceSelf-reported race/ethnicity1 = other race, 0 = not3.3
    White (Ref)Self-reported race/ethnicity1 = White, 0 = not54.2
    FemaleSelf-reported gender1 = Female, 0 = male38.2

Note. Data are from the public-use version of the National Longitudinal Study of Adolescent Health (n = 6504).

Structural Equation Modeling

We specified the 10 equations in our system as a series of simultaneous equations using structural equation modeling. Structural equation modeling is ideal for our approach because it allows the equations to be estimated simultaneously with a maximum likelihood estimator, and it allows reciprocal effects and feedback loops to be specified while appropriately accounting for possible endogeneity. We accounted for clustering within schools by computing robust standard errors. Although hierarchical linear models handle clustering, they cannot handle reciprocal relationships and feedback loops, given current software constraints. We estimated the model displayed in Figure 1. Note that the exogenous variables are depicted on the left-hand side of the figure and are not predicted by any other variables. The other variables displayed in this figure are endogenous variables in our system (i.e., each is predicted by some other variable or variables). Each of these endogenous variables is represented by an equation based on our theoretical discussion above. For instance, in-degree centrality is a function of the following equation:

where β1 shows the effect of friends’ smoking behavior on the in-degree centrality of the respondent, β2 shows the effect of the respondent's past-month smoking on their own popularity, Γ1 is a vector capturing the effects of the neighborhood variables on respondent popularity, Γ2 is a vector capturing the effects of the respondent's demographics on his or her own popularity, Γ3 is a vector capturing the effects of the school network variables on respondent popularity, and ζ1 is an error term. Analogous equations can be written for the other endogenous variables.

We allowed for correlated errors among the personal network variables and between emotional support and friends’ smoking behavior. We used various techniques to confirm that our model was identified (i.e., there are unique values for the parameters), including estimating models with varying starting values.42 We also used the Sargan test to determine whether our instrumental variables were appropriate, and the test's nonsignificant results confirmed that these instrumental variables were indeed independent of the error term in these equations, as hypothesized. These instrumental variables explained a reasonable amount of the partial variance in the first-stage equation, which is an important indicator of their strength.

Note that we cannot estimate the reciprocal effect of the selection-influence relationship between friends’ smoking behavior and adolescent's past-month smoking because we do not have any plausible instrumental variables to estimate both of these paths. Thus, having friends who smoke likely increases an adolescent's past-month smoking, but those who smoke are also more likely to associate with friends who also smoke. Rather than simply assuming that the degree of association between these 2 constructs is entirely attributable to an influence effect, we adopted a novel technique to test the robustness of our system, assuming various values for the relative proportion of this relationship attributable to the selection effect.46 We estimated models in which we fixed the selection effect at various values: (1) zero selection effect, (2) selection effect one third the size of the influence effect, (3) equal selection and influence effects, and (4) selection effect 3 times the size of the influence effect. This technique has only occasionally been employed.46

Another advantage of structural equation modeling is that it allows us to test the overall fit of the model. Structural equation modeling allows the specification of a causal model, and it can then test how well the model represents the observed data as a test of causality, as has been described elsewhere.4749 This tests the similarity of the model's implied covariance matrix to the sample covariance matrix. Although a good model fit would be consistent with our theorized model, it is possible that other models that might be specified could fit these data equally well. Therefore, the causal conclusions must necessarily be tempered, despite the fact that the specified model is inherently causal. The comparative fit index of 0.998, the Tucker-Lewis index of 0.991, and the root mean square error approximation of 0.008 suggest an excellent fit for our model.49

The summary statistics for the variables used in the analyses are shown in Table 1. At wave 1, 70.5% of the entire sample had not smoked, whereas this figure was 67% by wave 2. The mean days of smoking per month for the entire sample increased from 4.9 to 5.2 over the 2 waves. We performed bivariate analyses on our key outcome measures and found that Whites smoked significantly more, on average, than did other groups (5.65 days per month), whereas Blacks and Latinos smoked significantly less (1.81 and 3.97 days per month, respectively; for all P values in this paragraph, P < .01). We also found that females smoked fewer days per month than did males (4.16 vs 5.39), as did those whose mothers had higher levels of education (3.11). The pattern is similar for friends’ smoking behavior: Whites had more friends who smoked (0.87 friends who smoked), whereas Blacks (0.53), females (0.74), and those with more educated mothers (0.62) had fewer friends who smoked. For emotional support, we found that Whites and females had more (38% and 44% of their friends provided emotional support, respectively), whereas Latinos (30%) and Blacks (31.5%) had less. Finally, Whites had higher in-degree centrality (named by 5.95 ties), as did females (5.73 vs 5.35 for males), whereas Blacks (5.06) and Latinos (4.77) had fewer social ties.

The results for our full simultaneous equations model are displayed in Table 2. Given the complexity of the results, we will now focus on key findings based on our theoretical discussion, beginning with the system component of school network measures. School networks with greater density and mutuality increased the likelihood that the respondents’ personal network would have greater density and reciprocity (P < .01; Table 2 equations 1 through 3). Respondents in school networks with more density (b = 0.289; P < .01) and mutuality (b = 0.714; P < .10) had greater in-degree centrality. Results for equation 6 in Table 2 show that adolescents in the largest and smallest schools received the fewest nominations. When we took the first derivative and set it equal to zero, we found that students in schools with about 1580 students had the largest in-degree centrality (size = –[0.766/(–0.243*2)] = 1.576), whereas students in smaller and larger schools had fewer nominations. Although there was no evidence that these school network measures directly affected adolescents’ past-month smoking (on the basis of ancillary models not presented, in which we added these variables to Table 2 equations 9 and 10), adolescents in larger school networks did have fewer friends who smoked (b = −0.251; P < .01; Table 2 equation 7).


TABLE 2 Simultaneous Equations Model Testing Pathways Among Characteristics of Social Networks and Neighborhoods, Emotional Support, Peer Influence, and Smoking: National Longitudinal Study of Adolescent Health, 1994–1996

TABLE 2 Simultaneous Equations Model Testing Pathways Among Characteristics of Social Networks and Neighborhoods, Emotional Support, Peer Influence, and Smoking: National Longitudinal Study of Adolescent Health, 1994–1996

Equation 1: Personal Network Density, b (SE)Equation 2: Best Female Friend Reciprocates, b (SE)Equation 3: Best Male Friend Reciprocates, b (SE)Equation 4: Distance to Reachable Others, b (SE)Equation 5: Ties Outside the School, b (SE)Equation 6: In-Degree Centrality, b (SE)Equation 7: Friends’ Smoking Behavior, b (SE)Equation 8: Emotional Support, Proportion, b (SE)Equation 9: Past-Month Smoking, Time 1, b (SE)Equation 10: Past-Month Smoking, Time 2, b (SE)
Past-month smoking0.005 (0.005)0.002 (0.007)0.007 (0.008)0.034 (0.021)0.009 (0.021)0.023*** (0.007)0.085 (NA)0.553*** (0.018)
Emotional support (proportion)0.381** (0.155)0.660*** (0.169)
Friends’ smoking behavior–0.047 (0.039)0.012 (0.052)–0.106** (0.054)–0.169 (0.158)0.139 (0.171)–0.153*** (0.057)0.765*** (0.050)0.323*** (0.045)
Personal network measures
    In-degree centrality0.233** (0.118)0.087*** (0.010)
    Ties outside the school–0.019 (0.035)0.032*** (0.003)
    Mean distance to reachable people0.046 (0.066)0.001 (0.004)
    Best male friend reciprocates0.188 (0.225)0.022 (0.015)
    Best female friend reciprocates–0.324 (0.232)0.050*** (0.018)
    Personal network density0.083 (0.199)–0.052*** (0.015)
School network measures
    School network density0.117*** (0.025)0.058*** (0.022)0.051** (0.023)–0.899*** (0.153)0.019 (0.161)0.289*** (0.064)–0.137* (0.080)–0.009 (0.015)
    Size of school network0.024 (0.036)0.040 (0.038)0.035 (0.038)0.138 (0.184)0.096 (0.247)0.766*** (0.208)–0.251*** (0.071)0.019 (0.018)
    Size of school network squared–0.243*** (0.063)
    Mutuality index0.467* (0.244)0.582** (0.238)0.947*** (0.232)0.365 (1.682)1.542 (1.437)0.714** (0.354)0.164 (0.511)0.229* (0.123)
Neighborhood (block group) measures
    Median home value0.106*** (0.031)0.029 (0.026)0.015 (0.025)0.399*** (0.132)0.217** (0.102)–0.005 (0.036)–0.125** (0.055)–0.010 (0.011)–0.002 (0.125)–0.009 (0.096)
    Median home value squared–0.035* (0.019)0.080*** (0.028)
    Racial/ethnic heterogeneity0.063** (0.031)0.077** (0.036)–0.035 (0.051)0.138 (0.137)–0.098 (0.207)–0.008 (0.057)0.014 (0.060)0.040** (0.020)0.099 (0.126)–0.208 (0.165)
    Residential stability0.001 (0.016)–0.032* (0.018)–0.002 (0.021)–0.069 (0.060)0.016 (0.067)–0.080*** (0.023)0.019 (0.025)–0.008 (0.011)0.120** (0.055)–0.111* (0.061)
Demographic characteristics
    African American–0.060** (0.024)–0.115*** (0.034)–0.091** (0.036)0.023 (0.108)0.489*** (0.153)–0.110*** (0.038)–0.218*** (0.056)–0.062*** (0.013)–0.967*** (0.115)–0.790*** (0.119)
    Asian0.000 (0.033)0.028 (0.049)–0.004 (0.054)0.327* (0.174)–0.132 (0.226)–0.094 (0.064)–0.044 (0.067)–0.036 (0.026)–0.438** (0.198)–0.184 (0.242)
    Latino0.031 (0.032)–0.074** (0.036)–0.075* (0.044)0.088 (0.082)–0.026 (0.126)–0.104** (0.042)–0.036 (0.058)–0.052*** (0.013)–0.460** (0.190)–0.305 (0.203)
    Other race–0.035 (0.030)–0.070* (0.038)–0.063 (0.040)0.039 (0.063)0.188 (0.123)–0.101*** (0.036)0.031 (0.053)0.008 (0.013)0.072 (0.134)–0.362*** (0.138)
    Age0.026*** (0.006)0.025*** (0.007)0.029*** (0.008)0.056*** (0.021)0.098*** (0.032)0.001 (0.010)0.087*** (0.013)0.023*** (0.003)0.213*** (0.029)0.019 (0.029)
    Female0.048*** (0.015)0.315*** (0.020)–0.156*** (0.021)–0.013 (0.078)0.495*** (0.068)0.111*** (0.025)0.068 (0.082)0.164*** (0.011)–0.131 (0.090)–0.058 (0.076)
    Mother's education–0.028*** (0.007)–0.027 (0.019)–0.029 (0.021)
    Parent smoking0.101*** (0.015)0.270*** (0.041)0.209*** (0.043)
    Wear seatbelts–0.092*** (0.013)–0.259*** (0.037)–0.091** (0.037)
    Motorcycle riding0.094*** (0.012)0.323*** (0.050)0.062 (0.041)
    Cigarettes in home0.207*** (0.037)0.729*** (0.096)–0.032 (0.101)

Note. Selection effect of friends’ smoking on individual smoking is fixed at same size as influence effect on individual smoking. Models were estimated using maximum likelihood estimation, with standard errors corrected for clustering within schools. χ258 = 79.6; P = .03. Data are from the public-use version of the National Longitudinal Study of Adolescent Health (n = 6504).

*P < .05 (1-tail test); **P < .05 (2-tail test); ***P < .01 (2-tail test).

For the structural component of neighborhood properties, adolescents from block groups with more economic resources had more ties outside the school (Table 2 equation 5) and higher mean distance to reachable people (Table 2 equation 4). Neighborhood economic resources had a curvilinear relationship with the density of personal networks and friends’ smoking behavior: the peak density of personal networks occurred in neighborhoods with median home values of approximately $151 000, which were somewhat above average (value = –[0.106/(–0.035*2)] = 1.514). By contrast, the number of smokers in adolescents’ networks was lowest in middle-class neighborhoods. Residential stability had a direct negative effect on in-degree centrality (b = −0.081; P < .01) and increased past-month smoking (b = 0.120; P < .05; in Table 2 equation 9).

Regarding the role that emotional support plays in the system, results for Table 2 equation 8 show that several personal network characteristics affect the amount of emotional support. A 10% increase in in-degree centrality led to a 0.087 proportionate increase in persons providing emotional support. Likewise, those with more ties outside the school, and those with reciprocation from the best friend (especially females), received an increased amount of emotional support. Only personal network density showed a negative effect on emotional support.

There is also evidence that adolescents with more emotional support engaged in more past-month smoking. This effect was positive (0.381) in the equation predicting past-month smoking at the same time point (Table 2 equation 9), and in Table 2 equation 10 for past-month smoking at the next time point (0.66), even after control for past-month smoking at the previous time point. This implies that emotional support plays a long-term role in mediating the relationship between the various personal network measures and smoking the following year. For example, those with more social ties inside and outside the school, and those with reciprocated ties with their best friend, had more emotional support, which led to more smoking 1 year later.

There is little evidence that friends’ smoking behavior mediated the relationship between personal network measures and past-month smoking. On the one hand, friends’ smoking behavior increased past-month smoking (Table 2 equation 9) and past-month smoking at the next wave, after control for time 1 smoking (Table 2 equation 10). Each additional friend who smoked increased the number of days smoked per month by 77%. On the other hand, there is evidence in Table 2 equation 7 that those who were more popular (i.e., those with high in-degree centrality) had more friends who smoked. The other network measures had no effect on friends’ smoking behavior.

Although there is little evidence that these personal network measures increased friends’ smoking behavior, there is evidence in Table 2 equation 6 that the popularity of adolescents (in-degree centrality) was affected both by their own past-month smoking and by their friends’ smoking behavior. A 1% increase in past-month smoking increased in-degree centrality by 2.3% (b = 0.023: P < .01). However, there was a countervailing effect if one had friends who smoked: each additional friend who smoked reduced in-degree centrality by 15.3% (b = −0.153; P < .01). There is little evidence that past-month smoking or friends’ smoking behavior affected the other personal network measures (Table 2 equations 1 through 5).

We performed sensitivity tests of our model. As described already, we set the parameter for the effect of adolescents’ past-month smoking on friends’ smoking behavior to various values and assessed the robustness of the system. In short, the system appeared to be relatively robust, regardless of the ratio of influence to selection. For example, the effect of adolescent past-month smoking on in-degree centrality was 0.014 if we assumed no selection effect, 0.018 if 25% of the relationship was attributable to selection, 0.023 if 50% was attributable to selection, and 0.030 if 75% was attributable to selection (for all effects, P < .05). Likewise, the effect of friends’ smoking behavior on in-degree centrality ranged from −0.128 to −0.171 (for all effects, P < .05). The other parameters in the model showed even more stability over these various parameterizations, indicating that our results are robust regardless of the portion of the relationship that was attributable to selection and the portion attributable to influence.

We also assessed whether emotional support and friends’ smoking worked multiplicatively at the level of social ties. We tested this by constructing a measure that multiplied the emotional support score and the smoking behavior for each of an individual's friendship ties and then computing the average of these values among the ties of an individual. In ancillary models including personal and school network measures and demographics, we tested the effect of this variable on past-month smoking behavior at time 1 and time 2 and found no significant effects (results not presented; available upon request). Finally, we estimated a school-level fixed-effects model to control for unobserved differences across schools, and the substantive results were very similar to those presented in the text (results available upon request).

Our findings indicate that when a system of pathways between characteristics of personal networks, school networks, and neighborhoods is taken into account, together with its constituent flows of emotional support and the influence exerted through friends’ smoking behavior, important insights into the complexity of the social context of adolescent smoking can be gained. Personal network characteristics—being central in a network, having ties outside the school, having a best female friend reciprocate friendship, and the density of ties—influenced the flow of emotional support, which in turn influenced past-month smoking at both time points. We found that although the flow of influence from friends’ smoking behavior was not affected by personal or school-level network characteristics, it did affect past-month smoking. Findings pertaining to the school component indicate that school-level density and size affected personal network characteristics, density, reciprocation of ties from best female friend, mean distance to reachable people, in-degree centrality, and number of nominations outside of respondents’ schools. Findings related to the neighborhood component of the system indicate that median home value negatively affected past-month smoking at time 1. Neighborhood characteristics also affected network characteristics, including density, reciprocation, and ties outside the school. Findings indicate a feedback loop from personal network characteristics, to emotional support, to past-month smoking, and then back to personal network characteristics, including the provocative finding that past-month smoking affected in-degree centrality.

Our finding that personal network characteristics increased emotional support, which then increased past-month smoking, supports previous research identifying emotional support as a mechanism through which networks relate to health and to risk behavior.29,34 These findings were not surprising because being central in a network likely affords opportunities for giving and receiving emotional support. Second, having friendships outside of school may increase the number of friends an adolescent has, thus increasing the probability of receiving support from any one friend, and may also suggest that friends outside of school are close friends who provide emotional support. Third, having a female friend reciprocate friendship may increase the emotional support exchanged in a mutually reciprocated friendship tie, given that females may often be viewed as sources of emotional support. Last, the negative effect of density on emotional support may be explained by the numerous relationship obligations and constraints that densely connected ties can impose, constituting a great demand on one's personal resources. Moreover, if density of ties limits the resources entering from outside the network, this can further limit the amount and diversity of personal resources to expend as emotional support to others.

It is notable that the only school-level characteristic to increase emotional support was the mutuality index. Perhaps a whole network structure with a large proportion of mutually reciprocated ties increases the possibility that emotional support will be exchanged in any one of these close ties. We observed that neither school-level density nor size had an effect on the flow of emotional support; the former finding is consistent with the idea that density may limit support in a network. The latter finding may indicate that larger schools promote anonymity and consequently fewer support resources, leading to more diffuse networks and fewer close and supportive ties.

Our finding that emotional support influences past-month smoking at both time points is consistent with previous work showing that emotional support positively relates to smoking.33 Perhaps the effect between emotional support and smoking is more likely to occur between close friends in emotionally supportive relationships. The persistence of this relationship at the second time point may indicate its strength and stability over time.

Although we found modest effects for our neighborhood component, findings of note were the curvilinear relationships that neighborhood economic resources had with the density of personal networks and with friends’ smoking behavior. Adolescents living in middle-income neighborhoods had networks with the highest density, suggesting a relative cohesion among their personal ties. At the same time, adolescents in middle-income neighborhoods had the fewest smokers in their networks, suggesting a relatively low effect from influence of friends’ smoking behavior. It is notable that adolescents in both low- and high-income neighborhoods had networks with more smokers than adolescents in middle-income neighborhoods.

Friends’ smoking behavior was not affected by any of the network characteristics under study, but it did increase past-month smoking at the first time point. The lack of any effects of network characteristics on friends’ smoking behavior suggests that although these characteristics may be important for promoting smoking behavior among youths,11,12 they are not important for the smoking behavior of youths’ friends. This finding runs counter to the many studies indicating homogeneity in the smoking status of friends. The findings that friends’ smoking behavior reduced both in-degree centrality and best male friend reciprocation suggest that having friends who smoke actually decreases popularity and the reciprocation of friendship ties. Overall, such findings suggest that having friends who smoke was not well received among the greater social milieu of youths in our study.

Our findings suggest evidence of a feedback loop: personal network characteristics increased the emotional support received by adolescents, which then appears to have led to more smoking at both waves. In addition, adolescent smoking at time 1 flowed back in the other direction through the system by bringing about more friends who smoked (through a selection effect) and then leading to greater in-degree centrality. This greater in-degree centrality and greater distance to reachable people then led to more emotional support, and thus the loop begins again. Our findings are indicative of a feedback process that encompasses the amplifying effects of personal network characteristics on emotional support, the reinforcing effect of emotional support on smoking, and the effect of smoking on popularity and distance to reachable others. Such a “reinforcing” loop might suggest that smoking brings social gains in the way of emotional support and popularity in the social system under study.

Our findings also have implications for extant and future studies that employ the general strategy of examining relationships between network characteristics and smoking among youths. Previous research found a positive association between the popularity of students (as measured by in-degree centrality) and adolescent smoking, and it has been assumed conceptually that the direction proceeded from popularity to smoking,12 but we have specified a system that allows this directionality to proceed in either direction. As a consequence, we were able to detect more evidence that smoking behavior and peers who smoked affected one's popularity, rather than the reverse. This finding has the potential to inform future models used to investigate the relationship between in-degree centrality and smoking among youths. More broadly, this finding suggests that a social behavior—cigarette smoking—could alter an important positional attribute of a social network. It is notable that the individuals who showed a relatively high degree of autonomy—in that they smoked, but did not hang out with fellow smokers—were the most popular, based on in-degree centrality. Alternatively, smokers who affiliated with friends who smoked were generally no more popular than average adolescents.

Our findings suggest the need to examine how the pathways represented in the systems model under study might differ across gender and racial/ethnic groups, given the possible group differences. Also, future studies should examine how other types of social support, such as confidant support, might function in lieu of emotional support in our study model. Confidant support has been associated with positive health outcomes50,51and is relevant given the notable effects of reciprocated ties and emotional support, both likely characteristics of a confidant relationship, on adolescent smoking in this study.


Our study has some limitations. First, the network elicitation items were limited in the number of friendship nominations. Capping friendship nominations is a common strategy, though it is a potential drawback among studies utilizing network generator items. It remains unclear how social position and network structure would differ if the number of nominations were not capped at this level. Second, network data were not collected for the full national sample at wave 2; therefore, we could not account for network variables at time 2 in our models. It is unclear how the inclusion of these variables might have changed our results. Future studies should include network variables at multiple time points, to permit observation of the evolution of the system. Third, because we conducted a secondary analysis, we were restricted in the types of network variables, social processes, and outcomes available for study. Nevertheless, we investigated theoretically informed pathways composing a larger system of adolescent smoking. Lastly, what constitutes a friendship tie is of note here because it is unclear whether there was uniformity in the strength, duration, and frequency of contact in friendship ties.

Implications for Prevention

In spite of these limitations, our findings provide insight into the importance of the strength of reciprocated friendships and the emotional support they can transact to help adolescents support each other in remaining nonsmokers or in quitting smoking. These friendship pairs could be targeted for a school-based intervention, either to help both adolescents in a pair remain nonsmokers or so that they could help each other stop smoking. This could be done by teaching youths in these pairs how to use emotional support as reinforcement for helping one another remain nonsmokers (among nonsmoking pairs) and for considering quitting (among smoking pairs). Second, adolescents could learn self-regulatory techniques (e.g., journaling) to help one another identify cues in the social environment that elicit interest in smoking or smoking associations. All participating adolescent pairs could form task forces in schools and lead smoking awareness campaigns. Reciprocated relationship pairs would become a channel through which antismoking messages permeate personal and school networks.

Findings suggest the need to target adolescents who smoke, have nonsmoking friends, and are not yet popular. Research suggests that popular youths can set norms in a school context.12 A corollary is that if popular youths smoke, others will emulate them. Building on previous research12 suggesting that popular youths will need to adopt antismoking norms in order for programs to become effective, we suggest that interventions should target youths who smoke before they become popular. Perhaps these youths are not yet frequent smokers, given that they affiliate with nonsmoking friends, and thus may be tolerant of antismoking norms. Such adolescents could be educated about the risks of smoking with the hope that they would adopt antismoking norms, which their nonsmoking friends might reinforce. This intervention would be disseminated through adolescents’ personal networks and would have the potential to solidify antismoking norms over time, as these messages spread from personal to school networks.

This study demonstrates the merit of using a systems science approach to conceptualize complexity in the social context of adolescent smoking. We found evidence of direct pathways and feedback processes. Emotional support was a pathway linking personal network characteristics and past-month smoking, but the peer influence process of friends’ smoking behavior was not. We found evidence of a feedback process, as past-month smoking had a direct effect on the popularity of students (in-degree centrality). Overall, our findings suggest complexity in the social context of adolescent smoking and the need for theory to account for it.


This research uses data from Add Health, a program project directed by Kathleen Mullan Harris and designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill and funded by grant P01-HD31921 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design of the Add Health Study. Information on how to obtain the Add Health data files is available on the Add Health Web site (http://www.cpc.unc.edu/addhealth). No direct support was received from grant P01-HD31921 for this analysis.

Human Participant Protection

No protocol approval was necessary because data were obtained from publicly available secondary sources.


1. Centers for Disease Control and Prevention. Cigarette use among high school students–United States, 1991–2005. MMWR Morb Mortal Wkly Rep. 2006;55(26):724726. MedlineGoogle Scholar
2. Ennett ST, Bauman KE. The contribution of influence and selection to adolescent peer group homogeneity: the case of adolescent cigarette smoking. J Pers Soc Psychol. 1994;67(4):653663. Crossref, MedlineGoogle Scholar
3. Ennett ST, Bauman KE. Peer group structure and adolescent cigarette smoking: a social network analysis. J Health Soc Behav. 1993;34(3):226236. Crossref, MedlineGoogle Scholar
4. Ennett ST, Bauman KE, Hussong A, et al.. The peer context of adolescent substance use: findings from social network analysis. J Res Adolesc. 2006;16(2):159186. CrossrefGoogle Scholar
5. Pearson M, West P. Drifting smoke rings: social network analysis and Markov processes in a longitudinal study of friendship groups and risk-taking. Connections. 2003;25:5976. Google Scholar
6. Kirke DM. Chain reactions in adolescents’ cigarette, alcohol and drug use: similarity through peer influence or the patterning of ties in peer networks? Soc Networks. 2004;26(1):328. CrossrefGoogle Scholar
7. Flay BR, Hu FB, Siddiqui O, et al.. Differential influence of parental smoking and friends’ smoking on adolescent initiation and escalation and smoking. J Health Soc Behav. 1994;35(3):248265. Crossref, MedlineGoogle Scholar
8. Hoffman BR, Monge PR, Chou CP, Valente TW. Perceived peer influence and peer selection on adolescent smoking. Addict Behav. 2007;32(8):15461554. Crossref, MedlineGoogle Scholar
9. Abel G, Plumridge L, Graham P. Peers, networks or relationships: strategies for understanding social dynamics as determinants of smoking behaviour. Drugs Educ Prev Policy. 2002;9(4):325338. CrossrefGoogle Scholar
10. Fang X, Li X, Stanton B, Dong Q. Social network positions and smoking: experimentation among Chinese adolescents. Am J Health Behav. 2003;27(3):257267. Crossref, MedlineGoogle Scholar
11. Alexander C, Piazza M, Mekos D, Valente T. Peers, schools, and adolescent cigarette smoking. J Adolesc Health. 2001;29(1):2230. Crossref, MedlineGoogle Scholar
12. 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):323329. Crossref, MedlineGoogle Scholar
13. Bronfenbrenner U. Toward an experimental ecology of human development. Am Psychol. 1977;32(7):513531. CrossrefGoogle Scholar
14. Midgley G. Systems Thinking. Vol 1–4. Thousand Oaks, CA: Sage; 2003. CrossrefGoogle Scholar
15. Leischow SJ, Milstein B. Systems thinking and modeling for public health practice. Am J Public Health. 2006;96(3):403405. LinkGoogle Scholar
16. Wasserman S, Faust K. Social Network Analysis: Methods and Applications. New York, NY: Cambridge University Press; 1994. CrossrefGoogle Scholar
17. Ennett ST, Faris R, Hipp JR, et al.. Peer smoking, other peer attributes, and adolescent cigarette smoking: a social network analysis. Prev Sci. 2008;9(2):8898. Crossref, MedlineGoogle Scholar
18. Moody J. The importance of relationship timing for diffusion. Soc Forces. 2002;81(1):2556. CrossrefGoogle Scholar
19. Shaw C, McKay HD. Juvenile Delinquency and Urban Areas. Chicago, IL: University of Chicago Press; 1942. Google Scholar
20. Sampson RJ, Groves WB. Community structure and crime: testing social-disorganization theory. Am J Sociol. 1989;94(4):774802. CrossrefGoogle Scholar
21. Osgood DW, Anderson AL. Unstructured socializing and rates of delinquency. Criminol. 2004;42(3):519550. CrossrefGoogle Scholar
22. Sampson RJ. Family management and child development: insights from social disorganization theory. In: , McCord J, ed. Facts, Frameworks, Forecasts: Advances in Criminological Theory. New Brunswick, NJ: Transaction; 1992:6393. Google Scholar
23. Gottfredson DC, Mcneil RJ, Gottfredson GD. Social area influences on delinquency: a multilevel analysis. J Res Crime Delinq. 1991;28(2):197226. CrossrefGoogle Scholar
24. Kandel DB, Kiros G-E, Schaffran C, Hu M- C. Racial/ethnic differences in cigarette smoking initiation and progression to daily smoking: a multilevel analysis. Am J Public Health. 2004;94(1):128135. LinkGoogle Scholar
25. Powell LM, Tauras JA, Ross H. The importance of peer effects, cigarette prices and tobacco control policies for youth smoking behavior. J Health Econ. 2005;24(5):950968. Crossref, MedlineGoogle Scholar
26. Xue Y, Zimmerman MA, Caldwell CH. Neighborhood residence and cigarette smoking among urban youths: the protective role of prosocial activities. Am J Public Health. 2007;97(10):18651872. LinkGoogle Scholar
27. House JS, Kahn RL. Measures and concepts of social support. In: , Cohen S, Syme SL, eds. Social Support and Health. Orlando, FL: Academic Press; 1985:83105. Google Scholar
28. House JS, Umberson D, Landis KR. Structures and processes of social support. Annu Rev Sociol. 1988;14:293318. CrossrefGoogle Scholar
29. House JS, Landis KR, Umberson D. Social relationships and health. Science. 1988;241(4865):540545. Crossref, MedlineGoogle Scholar
30. Berkman LF, Glass T. Social integration, social networks, social support, and health. In: , Berkman LF, Kawachi I, eds. Social Epidemiology. Oxford, England: Oxford University Press; 2000:137173. Google Scholar
31. Berkman LF, Glass T, Brissette I, Seeman TE. From social integration to health: Durkheim in the new millennium. Soc Sci Med. 2000;51(6):843857. Crossref, MedlineGoogle Scholar
32. Schaefer C, Coyne JC, Lazarus RS. The health-related functions of social support. J Behav Med. 1981;4(4):381406. Crossref, MedlineGoogle Scholar
33. Romano PS, Bloom J, Syme SL. Smoking, social support, and hassles in an urban African-American community. Am J Public Health. 1991;81(11):14151422. LinkGoogle Scholar
34. Lakon CM, Ennett ST, Norton EC. Mechanisms through which drug, sex partner, and friendship network characteristics relate to risky needle use among high risk youth and young adults. Soc Sci Med. 2006;63(9):24892499. Crossref, MedlineGoogle Scholar
35. Goehl L, Nunes E, Quitkin F, Hilton I. Social networks and methadone treatment outcome: the costs and benefits of social ties. Am J Drug Alcohol Abuse. 1993;19(3):251262. Crossref, MedlineGoogle Scholar
36. Power R, Jones S, Kearns G, Ward J. Drug user networks, coping strategies, and HIV prevention in the community. J Drug Issues. 1995;25(3):565581. CrossrefGoogle Scholar
37. Walker ME, Wasserman S, Wellman B. Statistical models for social support networks. Sociol Methods Res. 1993;22(1):7198. CrossrefGoogle Scholar
38. Walker KN, MacBride A, Vachon M. Social support networks and the crisis of bereavement. Soc Sci Med. 1977;11(1):3541. Crossref, MedlineGoogle Scholar
39. Conrad KM, Flay BR, Hill D. Why children start smoking cigarettes: predictors of onset. Br J Addict. 1992;87(12):17111724. Crossref, MedlineGoogle Scholar
40. Kobus K. Peers and adolescent smoking. Addiction. 2003;98(suppl 1):3755. Crossref, MedlineGoogle Scholar
41. Ary DV, Biglan A. Longitudinal changes in adolescent cigarette smoking behaviour: onset and cessation. J Behav Med. 1988;11(4):361382. Crossref, MedlineGoogle Scholar
42. Collins LM, Sussman S, Rauch JM, et al.. Psychosocial predictors of young adolescent cigarette smoking: a sixteen-month, three wave longitudinal study. J Appl Soc Psychol. 1987;17(6):554573. CrossrefGoogle Scholar
43. Mercken L, Candel M, Willems P, de Vries HH. Disentangling social selection and social influence effects on adolescent smoking: the importance of reciprocity in friendships. Addiction. 2007;102(9):14831492. Crossref, MedlineGoogle Scholar
44. Laumann EO. Bonds of Pluralism: The Form and Substance of Urban Social Networks. New York, NY: John Wiley & Sons; 1973. Google Scholar
45. Krohn M. The web of conformity: a network approach to explanation of delinquent behavior. Soc Probl. 1986;33(6):S81S93. CrossrefGoogle Scholar
46. Beyerlein K, Hipp JRA. A two-stage model for a two-stage process: how biographical availability matters for social movement mobilization. Mobilization. 2006;11(3):299320. CrossrefGoogle Scholar
47. Pearl J. Causality: Models, Reasoning, and Inference. New York, NY: Cambridge University Press; 2000. Google Scholar
48. Pearl J. Direct and indirect effects. In: Proceedings of the 17th Conference on Uncertainty in Artificial Intelligence. San Francisco, CA: Morgan Kaufmann; 2001:411420. Google Scholar
49. Bollen KA. Structural Equations with Latent Variables. New York, NY: John Wiley & Sons; 1989. CrossrefGoogle Scholar
50. Gottlieb B. Social Networks and Social Support. Beverly Hills, CA: Sage Publications; 1981. Google Scholar
51. Miller PM, Ingham JG. Friends, confidants, and symptoms. Soc Psychiatry Psychiatr Epidemiol. 1976;11(2):5158. Google Scholar


No related items




Cynthia M. Lakon, PhD, MPH, John R. Hipp, PhD, and David S. Timberlake, PhDCynthia M. Lakon and David S. Timberlake are with the Department of Population Health and Disease Prevention, Program in Public Health, University of California, Irvine. John R. Hipp is with the Department of Criminology, Law and Society and the Department of Sociology, University of California, Irvine. “The Social Context of Adolescent Smoking: A Systems Perspective”, American Journal of Public Health 100, no. 7 (July 1, 2010): pp. 1218-1228.


PMID: 20466966