Elsevier

Social Networks

Volume 34, Issue 3, July 2012, Pages 359-369
Social Networks

Actor-based analysis of peer influence in A Stop Smoking In Schools Trial (ASSIST)

https://doi.org/10.1016/j.socnet.2010.07.001Get rights and content

Abstract

As shown by the success of network intervention studies that exploit the occurrence of peer influence in their target group, the reliable assessment of peer influence processes can be important for informing public health policy and practice. A recently developed tool for assessing peer influence in longitudinal social network data is stochastic actor-based modeling. The body of the literature in which this method is applied is growing, but how reliable are the results? In this paper, we identify two shortcomings in this literature: the questionable assumption of temporal homogeneity, and the potential dependence of results on the inclusion of nuisance parameters in the model specification. These issues are resolved by analyzing the data of three schools selected from ASSIST, a large UK-based trial of a school-based smoking prevention intervention. Results show that the co-evolution of friendship and smoking is a time heterogeneous process, and that results are sensitive to specification details. However, the peer influence parameter is not affected by either, but emerges as surprisingly stable over time and robust to model variation. This establishes confidence in the method and encourages detailed future investigations of peer influence in ASSIST.

Introduction

For interventions aimed at the reduction of smoking to be a success, it is pivotal to know in detail the processes by which smoking habits start and stabilize. An adolescent's school friends and their smoking habits can play a decisive role in these processes, enhancing or suppressing individual tendencies to take up smoking through processes of peer influence (Alexander et al., 2001, Valente et al., 2003). The notion of peer influence refers in general to any social process by which a behavior or attitude of a focal individual is affected by the behavior and attitudes that are present among the peers that act as reference points for the individual. In the literature on adolescent development, peer influence has been studied as a process of contagion, by which adolescents gradually adopt their friends’ behavior, i.e., become more similar over time. Such similarity of friends is found on a host of behavioral and attitudinal dimensions that are salient in adolescence, for example, substance use patterns, delinquency, school performance, church attendance, participation in sports activities, or sexual behavior (Billy and Udry, 1985, Cohen, 1977, Kandel, 1978, McPherson et al., 2001). In the case of tobacco use, it has long been known that adolescent smokers tend to have more friends that smoke, while their non-smoking counterparts tend to have more non-smoking friends (e.g., Salber et al., 1963, Lanese et al., 1972). This alignment of similar smoking habits with friendship, however, is not an unequivocal indicator for the occurrence of peer influence. As discussed elsewhere (Cohen, 1977, Kandel, 1978, Ennett and Bauman, 1994, Mercken et al., 2009, Mercken et al., 2010a, Mercken et al., 2010b, Steglich et al., 2010), similarity of friends can also result from other processes, in particular from smoking homophily, the selection of friends based on already similar smoking habits. In the methodological literature on peer effects, problems associated with confounding peer influence with peer selection is known as endogeneity (Manski, 2000, Moffitt, 2001). Any analytical method that claims to reliably assess peer influence must control for the endogeneity of peer selection. When studying the effects of peer influence on smoking behavior, this issue gains further importance, considering that success or failure of public health policy and practice might crucially depend on the quality of analytical research results.

Recent advances in the modeling of peer influence processes in a dynamic social network context (Snijders et al., 2007, Snijders et al., 2010) enable the quantitative assessment of such peer effects on smoking, controlling for the endogeneity of partner choices (Moffitt, 2001). The literature making use of this method, called stochastic actor-based modeling of network-behavior co-evolution (henceforth abbreviated as SAB modeling), still is limited and, in general, focuses on diagnosing whether or not peer influence occurs at all (Burk et al., 2007, Knecht et al., forthcoming, Knecht et al., 2010), and to what degree it can account for behavioral similarity of friends (Mercken et al., 2009, Mercken et al., 2010a, Mercken et al., 2010b, Steglich et al., 2010).

In this paper, we identify and investigate two previously unaddressed methodological concerns about the reliability of results obtained by this method. The first issue of concern is the assumption of time homogeneity of the co-evolution process in the earlier studies. It is known that during adolescence, social networks as well as risk behavior, including substance use, are highly dynamic—which suggests that time homogeneity is questionable. A more adequate approach, which we take here, is to allow for time heterogeneity in SAB modeling, and assess to what degree analytical results are affected by time. The second issue is the seemingly ad hoc choice of model specifications. Notably, the way in which friendship dynamics are controlled for differs considerably across earlier studies, and the degree to which conclusions drawn from SAB modeling, in particular the conclusions about peer influence, are sensitive to such differences in these details remains unclear. We will address this by comparing results from two parallel analyses obtained with very different specifications of friendship dynamics: one is exclusively formulated on the dyad level, reflecting the psychological tradition of research on interpersonal relations; the other includes effects of higher order network structure (triads, classrooms), i.e., pays much more attention to the modeling of friendship selection.

The empirical data with which we will address these concerns were collected in A Stop Smoking In Schools Trial (ASSIST; Audrey et al., 2004, Campbell et al., 2008, Starkey et al., 2005). This was a cluster randomized controlled trial of the effectiveness of a school-based, peer-led smoking prevention intervention conducted between 2001 and 2004 in England and Wales. ASSIST collected social network data from over 10,000 students in 59 secondary schools on three annual occasions when the majority of students were aged 12–15. We confine the analyses in this paper to a subsample of 629 students in three schools. Considering the rather technical nature of questions addressed, a larger sample would not necessarily be of added value for our study. Future analyses of ASSIST have the potential to exploit the size of the dataset much better, and substantially deepen our insights into the workings of peer influence in adolescent smoking, provided the reliability issues identified above have been satisfactorily resolved.

The paper is organized as follows. In the following section, after a brief recall of the main methodological challenges encountered in peer influence research, a sketch of SAB modeling is given, followed by a brief summary of the adolescence literature making use of this methodology. Based on this we elaborate our two methodological research questions. As technical background, we sketch the use of goodness of fit criteria that allow testing our hypotheses, as well as the commonly applied meta-analytical procedures for aggregation of results from several independent studies. In the empirical section, we introduce our three schools subsample from ASSIST. We estimate SAB models for these three schools, and answer our research questions. In the discussion, we address implications of our results, and identify several issues on which future analyses of ASSIST will focus.

Section snippets

Stochastic actor-based modeling of peer influence

In early comprehensive reviews of the various theoretical accounts for homogeneity among friends, Billy and Udry (1985) and Fisher and Bauman (1988) broadly distinguish between theories of influence processes, according to which, for example, friendship begets similarity (assimilation to friends hypothesis), and theories of selection processes, according to which, for example, similarity facilitates friendship (homophilic selection hypothesis). A separation of both processes based on empirical

Goodness of fit and meta-analysis

A few technical hurdles need to be overcome before addressing the research questions in an empirical analysis of the three ASSIST schools. We need to know how results from different time periods can be compared, how models with differing specification can be compared, and how results for the three schools can be combined into one comprehensive set of results. To make a comparison of results obtained in different analyses – be it of different time periods, or according to different model

Empirical analysis

The two research topics identified above are (1) the potential occurrence of time heterogeneity in the co-evolution of friendship and smoking and (2) the sensitivity of SAB model results to potentially distracting details in the model specification. While the focal interest is on peer influence, the analyses offer the opportunity to study the heterogeneity and sensitivity issues also in relation to other dynamic processes. We accordingly follow an exploratory rather than a hypothesis-driven

Results

For each of the three schools, the two nested models, ‘interpersonal’ and ‘context’, are fitted separately to the data of the two observation periods, which gives a total of 12 sets of parameter estimates. In Table 2, Table 3, they are summarized per time period and per model specification. For each parameter we report its true mean over the three schools as calculated according to Cochran's procedure, its standard error, the results of Fisher's combined one-sided test in the direction of this

Discussion

This study uses stochastic actor-based models for network-behavior co-evolution (SAB models, Snijders et al., 2010) to assess the strength of peer influence effects in three control schools of ASSIST, a school-based, peer-led intervention study to prevent adolescent smoking (Audrey et al., 2004, Campbell et al., 2008, Starkey et al., 2005). The analyses were conducted for the purpose of settling open methodological issues related to the application of SAB models, before applying them to the

References (47)

  • S. Audrey et al.

    The development and implementation of a peer-led intervention to prevent smoking among secondary school students using their established social networks

    Health Education Journal

    (2004)
  • C. Baerveldt et al.

    Revisiting selection and influence: an inquiry into the friendship networks of high school students and their association with delinquency

    Canadian Journal of Criminology and Criminal Justice

    (2008)
  • R. Berardo et al.

    Self-Organizing Policy Networks: Risk, Partner Selection and Cooperation in Estuaries

    American Journal of Political Science

    (2010)
  • T.J. Berndt

    Children's friendships: shifts over a half-century in perspectives on their development and their effects

    Merrill-Palmer Quarterly

    (2004)
  • J.O.G. Billy et al.

    Patterns of adolescent friendship and effects on sexual behavior

    Social Psychology Quarterly

    (1985)
  • W.J. Burk et al.

    Beyond dyadic interdependence: actor-oriented models for co-evolving social networks and individual behaviors

    International Journal of Behavioral Development

    (2007)
  • William G. Cochran

    The combination of estimates from different experiments

    Biometrics

    (1954)
  • J.M. Cohen

    Sources of peer group homogeneity

    Sociology of Education

    (1977)
  • R. de Cuyper et al.

    De co-evolutie van vriendschapsrelaties en delinquent gedrag onder Nederlandse jongeren

    Mens en Maatschappij

    (2009)
  • J.K. Dijkstra et al.

    Influence and selection processes in weapon carrying during adolescence: The roles of status, aggression, and vulnerability

    Criminology

    (2010)
  • S.T. Ennett et al.

    The contribution of influence and selection to adolescent peer group homogeneity: the case of adolescent cigarette smoking

    Journal of Personality and Social Psychology

    (1994)
  • L.A. Fisher et al.

    Influence and selection in the friend-adolescent relationship: findings from studies of adolescent smoking and drinking

    Journal of Applied Social Psychology

    (1988)
  • H. Goldstein

    Multilevel Statistical Models

    (1995)
  • Cited by (53)

    • Limits to inferring status from friendship relations

      2019, Social Networks
      Citation Excerpt :

      In the growing field of statistical network modelling using ERGMs (Exponential Random Graph Models; Lusher et al., 2012) or SAOMs (Stochastic Actor-Oriented Models or Siena-models; Snijders et al., 2010), the original ideas by Davis and Leinhardt (1967) were picked up, and triadic network configurations (such as the transitive or cyclic triad) were explicitly modelled in analyses under the assumption that asymmetric relations of positive affect represent status differences. Estimated tendencies towards transitive and against cyclic closure in friendship are consistently interpreted in light of these status differences (e.g. Snijders, et al. 2010; Steglich, et al. 2012; McFarland, et al. 2014). With increasing interest in networks outside of sociology, the status interpretation of friendship was adopted by scholars from other fields.

    • Friends in locked places: An investigation of prison inmate network structure

      2017, Social Networks
      Citation Excerpt :

      Beyond the contextual factors noted above, the composition of prisons is unique. Prison inmates more often possess attributes strongly correlated with crime and arrest, such as low socioeconomic status, substance abuse, low self-control, racial minority status, and poor mental health – all of which have been shown to affect relationship processes (Schaefer, 2016; Schaefer et al., 2011; Steglich et al., 2011). Indeed, there is a long-standing debate over potential differences in the quality of ties for delinquents and criminals in comparison to their non-offending peers.

    • Network formation in the interbank money market: An application of the actor-oriented model

      2017, Social Networks
      Citation Excerpt :

      Stochastic actor-based models have been applied to a variety of research questions in different fields. They have been used to investigate the early evolution of the network of friendships among pre-school children (Schaefer et al., 2010), the influence of friendships and acquaintanceships on television use and smoking habits among adolescents (Mercken et al., 2010; Steglich et al., 2012), and the influence of networks of acquaintanceships on the formation of norms of individuals in different professional environments (e.g., de Klepper et al., 2010; Lazega et al., 2012). One advantage of an actor-based approach against other statistical approaches to network analysis lies in the possibility to distinguish between social selection and social influence.

    View all citing articles on Scopus

    The work reported in this paper is part of the project “Social Network Analysis of Peers and Smoking in Adolescence (SNAPS)” funded by the Medical Research Council of the UK.

    View full text