Despite the large body of research documenting the role of peer influence in adolescent delinquency, research on the role of delinquent peers has been limited in three important ways. First, past research has used a less than precise definition of the friendship group in which normative influence is believed to occur. Most studies in the criminology literature examining the effect of peer influence on delinquency have simply asked adolescents to think about their friends in general and to report whether their friends have participated in a particular illegal behavior or set of illegal behaviors. As a result of this strategy, it is unclear who was included in adolescents’ definition of “friends.” For instance, the number of friends considered is unknown. In addition, no information on prosocial individuals (i.e., friends who abstain from crime/delinquency) has been collected.
Second, problematic measures of peer influence have been used. For the most part, past research has relied on adolescents’ perceptions of friends’ behavior. Therefore, the standard approach to measuring peer delinquency contains a same-source bias that substantially inflates similarity in behavior between peers. In almost all criminological studies, information about friends comes from adolescents’ descriptions of the behavior of their friends instead of from those friends’ reports of their own behavior. Such measures inflate the similarity in behavior between adolescents and their peers, because people tend to project their own attitudes and behavior onto their friends, a phenomenon social psychologists refer to as assumed similarity or projection. Although such findings have led several scholars to caution against the use of adolescents’ reports about peers, there has been limited recognition of this problem in research on crime and deviance. Such findings show that there is some truth in Gottfredson and Hirschi’s (1990) argument that adolescents’ reports of their peers’ delinquency “may merely be another measure of self-reported delinquency” (p. 157).
Third, prior research has neglected to consider the role of the structural properties of friendship relations. By overlooking the structure of friendship networks, past research has assumed that everyone in the friendship network is affected by friends’ behavior similarly. This is an oversimplification of network processes because it overlooks the adolescent’s position within the network (e.g., central vs. peripheral), the cohesiveness of the network (i.e., the interconnections among network members), and the adolescent’s prestige (e.g., popularity) within the network. These structural characteristics shape the degree to which adolescents are influenced by group dynamics.
Fortunately, recent work on social networks and network analyses has begun to make its way into the work of researchers interested in understanding peer processes as they relate to adolescent crime and delinquency. Much of this recent work has been spurred by the availability of a new novel data set, The Longitudinal Study of Adolescent Health (hereafter, Add Health), which allows researchers to overcome the limitations just described (for use of the Add Health data, see, e.g.,, Haynie, 2002). The advantage of these data is that they can be used to incorporate a social network perspective to elaborate on the normative influence process believed to generate peer similarity among friends. Specifically, a network perspective is guided by the assumption that the behaviors exhibited by network members, as well as the structure of the network, have important consequences for understanding subsequent behavior. In the context of delinquency, this suggests that exposure to pro- or anti-delinquent behaviors will depend upon the structure of the network, the adolescent’s position within the network, and the behaviors exhibited in the network.
In addition, and in contrast to past measurement strategies, a network perspective offers a more desirable measurement strategy whereby the friendship network is carefully mapped out, responses about behaviors come directly from the friends’ perspectives, and network homogeneity and structure are considered. The beginning point of network studies involves asking adolescents both to describe their own behavior and to identify their friends. The second step involves locating and interviewing the friends, with the friends describing their own behavior and then identifying their friends, and so on. In a best-case scenario, all adolescents and friends in the population of adolescents provide this information. This allows for the links among friends to be established for the purposes of constructing analytical friendship networks with identifiable structural properties and allows researchers to measure friends’ behavior based on the actual responses of friends themselves.
A. Add Health Data
Part of the reason the effects of friendship networks on adolescents’ delinquency has received less attention than it deserves is that the necessary data have not been available. Understanding social networks’ influence on adolescent delinquency requires detailed data on the structure of friendship networks within a school, for many different schools. Until recently, the only data that approached these stringent requirements came from Coleman’s (1961) landmark study of social relationship among high school students in the 1960s. Fortunately, more recent data are now available.
Add Health is a nationally representative sample of adolescents in Grades 7 through 12 located within randomly selected schools in the United States in 1995–1996. The innovative design of this sample, in particular its emphasis on the effects of multiple contexts of adolescents’ lives, allows for an examination of the causes of adolescent health and health behavior (including delinquency) that goes considerably beyond prior research.
Adolescents were included in the Add Health study on the basis of a sampling design that stratified schools by region, urbanicity, school type, ethnic mix, and size. This is important because, when used properly, these data allow findings to be generalized to all adolescents enrolled and attending middle and senior high schools in the United States. In addition, the data are longitudinal and currently consist of three waves of data: an initial in-school questionnaire followed by three in-home surveys conducted in 1995, 1996, and 2002.
Information collected in the in-school questionnaire is the critical component of the study for network analyses, because this is where the friendship networks of schoolage adolescents are measured. In the initial in-school survey administered in 1994–1995, all students attending school on the day of the self-administered questionnaire in each of 132 high schools and middle schools were surveyed. This sample is the basis for the construction of the measures of friendship network characteristics. To tie all of the students together in the schools, researchers asked each student who filled out the in-school questionnaire to nominate up to 5 of his or her closest female and 5 of his or her closest male friends (for a maximum of 10 friends). They identified their friends by name from school rosters and entered a corresponding identification number. Because each student in the school was interviewed, global networks (i.e., school networks connecting all students in the school) were re-created. The behaviors of friends nominated by the adolescent, as well as those friends who nominated the adolescent, were matched to the adolescent’s record, allowing a unique opportunity to assess the actual effect of friends’ behaviors.
Friendship networks can be defined in various ways using the Add Health data. For instance, it is possible to define the network as consisting of those adolescents who reciprocate the friendship nomination (i.e., the friendship network contains only adolescents whose friendship ties sent to others are reciprocated), as containing only those nominations sent to others (i.e., including only those friends that the adolescent nominates), as containing only those nominations received from others (i.e., including as friends those adolescents in the school who nominated the adolescent as a friend), or as including both ties sent and received (i.e., defining the friendship network as including all of those friends the adolescent nominated as well as those adolescents in the school who nominated the adolescent in the school). In addition to examining characteristics of the adolescent’s friendship network (including behavior, demographics, and structure), these data make it possible to measure characteristics of the overall school network in which the adolescent’s friendship network is located.
Following the in-school questionnaire, in-home surveys were administered to a smaller sample of adolescents selected from school rosters and involved a longer series of questions, including items concerning more serious delinquency involvement. By the time of the third wave of data collection, the sample was approaching young adulthood (i.e., between the ages of 18 and 26). Unfortunately, network information for all students was available only during the initial in-school questionnaire. However, for a small number of schools, network data are available for two points in time. Because of Add Health’s interest in social networks, there were 12 schools from which all enrolled students were selected for the in-home interviews (instead of a random sample). The 12 schools (2 very large schools and 10 small schools) have various characteristics, including location in rural and urban areas, designation as public and private schools, and differing degrees of ethnic heterogeneity. In this saturated sample, all adolescents in these schools were interviewed in depth in their homes. In addition to answering a series of questions relating to involvement in serious delinquency, students in these schools also nominated their closest friends at two points in time (during the first and second in-home interviews). These data, therefore, provide a unique opportunity to study the effect of peer influence processes over time. More information on the Add Health data design can be found at the following Web site: http://www.cpc.unc.edu/projects/addhealth/design.