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

• Author: Baerveldt, Chris
• Position: Canada

1. Introduction

Most criminologists agree that delinquents prefer relationships with other delinquents (e.g., Aseltine 1995; Baerveldt and Snijders 1994; Baerveldt, Van Rossem, and Vermande 2003; Baron and Tindall 1993; Bender and Losel 1997; Dishion, Andrews, and Crosby 1995; Elliot, Huizinga, and Ageton 1985; Fisher and Bauman 1988; Fletcher, Darling, Steinberg, and Dornbusch 1995; Gilmore, Hawkins, Day, and Catalano 1992; Haynie 2001; Jussim and Osgood 1989; Marcus 1996; Ploeger 1997; Poulin, Dishion, and Haas 1999; Reed and Rose 1998; Thornberry, Lizotte, Krohn, Farnworth, and Jang 1994; Vitaro, Tremblay, Kerr, Pagani, and Bukowski 1997). However, while there is certainly more similarity among friends than among non-friends regarding delinquent behaviour, the nature of the underlying mechanism for this similarity remains an open question. Some authors argue that the similarity is primarily the result of selection processes (Hirschi 1969). According to this view, delinquents choose other delinquents as friends, while non-delinquents prefer other non-delinquents. Others argue that it is mainly the result of influence processes (e.g., Sutherland and Cressey 1974). This latter perspective implies that people adapt their behaviour to that of their friends. Someone who has more delinquent than non-delinquent friends is more likely to become delinquent him- or herself. Although there is some evidence (e.g., Matsueda and Anderson 1998) that selection and influence are simultaneous processes, most authors see selection and influence processes as mutually exclusive.

Until recently, most studies on selection and influence processes were hampered by three important limitations on measurement, study design, and methods of analysis. First, many studies measured peer delinquency indirectly; that is, respondents were asked to give information about their own behaviour as well as that of their friends (Brownfield and Sorenson 1993; Bruinsma 1992; Frauenglass, Routh, Pantin, and Mason 1997; Gardner and Shoemaker 1989; Gilmore et al. 1992; Hayes 1997; Heimer 1997; Hundleby and Mercer 1987; Keenan, Loeber, Zhang, and Stouthamer-Loeber 1995; McCarthy 1996; Mears, Ploeger, and Warr 1998; Ploeger 1997; Reed and Rose 1998; Warr 1993; Warr and Stafford 1990; White, Johnson, and Garrison 1985). This approach has been criticized (Aseltine 1995; Kandel 1996), as people tend to exaggerate the similarity in behaviour between themselves and their friends. As a result, the association between the delinquent behaviour of respondents and that of their friends may be overestimated (Weerman and Smeenk 2005). Second, studying selection or socialization processes requires longitudinal data, which are quite rare (Baerveldt et al. 2003; Haynie 2001; Knecht 2008). Cross-sectional data do not allow distinguishing between influence and selection processes. Third, until recently, one lacked the appropriate advanced techniques to analyse nested data such as students within classes or friends within networks of individuals. Multilevel techniques, which can account for interdependencies in the relationships of students nested in classes or schools, have only come into widespread use in the past decade (Snijders and Bosker 1999). In addition, the techniques necessary to analyse the processes of influence and selection simultaneously were developed just recently. Before that, most analyses of selection and influence accounted only for one side of the process and neglected the other. Fortunately, there have been important innovations recently (see Snijders 2001 and below, for a description of simulation investigation for empirical network analysis [SIENA]).

This article contributes to our understanding of selection and influence processes, while attempting to address the limitations mentioned above. The study on which it is based used direct measurements of friendship networks and of delinquency (measured at two points in time) and a newly developed technique for longitudinal network analysis (SIENA).

The remainder of the article is organized as follows. In section 2, we discuss some important studies on networks and delinquent behaviour. In section 3, we focus on the explanation for the similarity in delinquency levels among friends; that is, on selection and influence processes. Our data, our measurements, and the analytical model applied are outlined in section 4. Section 5 presents our results; and, in the last section, section 6, we conclude and discuss our findings.

2. Social networks and delinquency

   The importance of relationships for delinquent behaviour has long been recognized in the criminological literature. As early as the 1930s, Shaw and McKay (1931) observed that more than 80% of delinquent youths who were caught by the police acted, not alone, but together with others. Additionally, early criminological theories acknowledged the importance of the environment as opposed to personality characteristics for explaining delinquent behaviour (Sutherland 1947). Some early criminological theories also acknowledged that social relations were more important than personality characteristics for explaining delinquency (e.g., Sutherland 1947). Despite these early studies stressing the importance of social relations, many studies focused either on the personality of the offender or on contextual characteristics, such as neighbourhood or school composition, while the effects of social relationships were not considered systematically (Birbeck and LaFree 1993). Over the last two decades, the number of studies examining the effects of social relations on delinquent behaviour has grown substantially (Haynie 2001). Haynie's research showed that the association between a respondent's delinquent behaviour and that of his or her peers is stronger than that of most other factors investigated (Birbeck and LaFree 1993: 1014). Further, the effects of socio-demographic characteristics decrease substantially when one controls for friendship relations (Sampson and Groves 1989).

3. Two explanations: Selection and influence

   As mentioned earlier, two possible causes for similarities in the delinquent behaviour of friends suggested in the criminological literature are influence and selection. (2)

   The idea that people prefer similar others as friends--or, to put it differently, that homophily is a fundamental principle of social structure--can be traced at least back to the work of Lazarsfeld and Merton (1954), who discuss value- and status homophily as principles for the selection of others. According to Homans (1954/1974) people select similar others because those who are similar in socio-demographic background characteristics, in attitudes, and in behaviour understand each other better. This makes relationships with similar others relatively more rewarding and stable. Later, this argument was augmented by a supply-side argument (Blau 1978; Feld 1981; Verbrugge 1977)--the selection of friends is contingent upon the social composition of the pool of available others. The composition of such a pool is anything but random and it structures choices among possible friends. Within a school, the others available to become network members tend to have much more homogeneous background characteristics (socio-economic status, ethnicity, religion, locality, and so forth) than in society at large. As schools tend to be more homogenous in composition, the chances of meeting are considerably higher for two students with similar background characteristics than for two students with distinct backgrounds. Feld (1981) has noted that shared foci (social settings that structure a person's actions and interactions) increase the likelihood of similarity in behaviour (for an overview of the literature, see Feld 1981; McPherson, Smith-Lovin, and Cook 2001). Hence, homophily with respect to delinquency may be a result not only of preference but also of the opportunity to meet--the "supply" of similar other persons. In fact, for many types of relationships and forms of behaviour, assortative pairing is corroborated by social research (Kalmijn and Flap 2001).

   In criminology, Glueck and Glueck (1950) were among the first to introduce the principle of selection as a basis for (delinquent) associations. However, Glueck and Glueck never explicitly state that delinquency itself is the selection criterion. Reasons for selection might be the popularity or the status of the other person or some other feature rather than his or her delinquent behaviour. Studies of organized and gang-related crime introduced delinquency as a selection criterion in its own right: Individuals involve themselves in relationships to commit delinquent acts together (Venkatesh 1997). Social control theory (Hirschi 1969) also predicts that delinquents will preferentially select other delinquents for friends. Their poor social integration and general social disability prevents them from establishing strong and stable relationships, and delinquents associate with one another because of external threats or for lack of better alternatives. As a consequence, their relationships are weak, emotionally lacking, and superficial; or, as described by Hirschi (1969), "cold and brittle" (141; see also Hansell and Wiatrowski 1981, on the social ability model).

   The contrasting idea that behaviour is influenced by social relationships can be traced back to Durkheim's (1897/1951) argument that all types of behaviour are influenced by social norms. Norm conformity is enforced through membership and integration in social groups. The more someone is integrated, the more she or he complies with the existing norm of the group in question. Hence, the degree of integration in, for example, church, family, schools, or voluntary organizations is crucial for explaining an individual's behaviour. Later, the arguments of Durkheim were extended by the assertion that norms can differ among intermediate groups and that, in general, all behaviour is...