urgent summary

IdentifyingChildrensPeerSocialNetworksinSchoolClassrooms-LinksBetweenPeerReportsandObservedInteractions2222.pdf

Home >Education homework help >urgent summary

Identifying Children’s Peer Social Networks in School Classrooms: Links Between Peer Reports and Observed Interactions Scott D. Gest, Pennsylvania State University, Thomas W. Farmer, University of North Carolina at Chapel Hill, Beverley D. Cairns, University of North Carolina at Chapel Hill, and Hongling Xie, University of North Carolina at Chapel Hill

Abstract

Links between peer reports of social cluster membership and observed classroom interactions were examined in a sample of 72 children in 4th grade and 7th grade. All participating children in each classroom identified as many social clusters in the class- room as they could recall. Using the social-cognitive map (SCM) procedure, these individual reports were aggregated to summarize the number of times a given child was nominated as being in the same social cluster as each of his or her classmates (i.e., a co-nomination profile) and to identify the classmates in each child’s social cluster. Extensive classroom observations allowed for a parallel summary of the number of times a given child was observed to interact with each of his or her class- mates (i.e., an interaction profile). Results indicated that correlations between co- nomination profiles and interaction profiles were positive and statistically reliable. Children were observed to interact with members of their SCM-identified social cluster at a rate four times higher than with other same-sex classmates. These effects did not vary reliably by grade, sex or aggressive risk status.

Keywords: peers; social networks; behavior observations

Children’s peer social networks provide distinctive and important contexts for indi- vidual development (Cairns & Cairns, 1994; Kindermann, 1993). Research on chil- dren’s peer affiliations has often been synonymous with research on children’s dyadic friendships (Hartup, 1996). More recently researchers have begun to study children’s informal peer groups, often called social cliques or social clusters (Cairns, Cairns, Neckerman, Gest, & Gariepy, 1988; Cairns, Xie, & Leung, 1998). Empirical research on peer social clusters has been limited, in part, by a lack of convenient and valid methods to identify these larger social groupings (Bagwell, Coie, Terry, & Lochman, 2000; Rubin, Bukowski, & Parker, 1998). Most researchers have identified informal

Correspondence should be addressed to Scott D. Gest, Human Development & Family Studies, Pennsylvania State University, S110B Henderson South Building, University Park, PA 16802, U.S.A. Email: [email protected]

514 Scott D. Gest et al.

peer groups by direct observation (Ladd, 1983; Strayer & Santos, 1996) or by quali- tative description (Evans & Eder, 1993; Kinney, 1993). The Social Cognitive Map (SCM) method, which is based on peer reports of cluster membership, has proven to be a valuable alternative approach (Cairns, Perrin, & Cairns, 1985; Cairns et al., 1988). The goal of the present study is to examine the strength and generalizability of the links between patterns of peer nominations to social clusters obtained with the SCM method and patterns of directly observed classroom interactions.

Methods for Identifying Children’s Peer Affiliation Patterns

Direct Observations

Researchers have long relied on direct observations to study the social organization of preschool classrooms. Much of the early work in this area was inspired by etho- logical theory and verified the existence of relatively stable dominance hierarchies that could be observed in patterns of children’s agonistic exchanges (Strayer & Strayer, 1976) or visual attention (Vaughn & Waters, 1981). Only a small number of studies, however, have relied on direct observations to identify children’s peer affiliation pat- terns (as opposed to their dominance relations). For example, Strayer and Santos (1996) used direct observations to describe age-related changes in the structure of chil- dren’s social clusters; and Ladd (1983) used playground observations to describe dif- ferences in the social networks of children with varying sociometric status. Despite the face-validity of directly observing children’s affiliation patterns, the high expense of direct observations has precluded their use in most studies.

Researchers’ limited access to important settings in which older children and ado- lescents interact (e.g., hallways, restrooms, buses, locker rooms) also limits the utility of direct observations. Perhaps as a result, observational studies of peer affiliations in middle school and beyond have relied on participant-observer methodologies in which access to some of these settings is gained (Evans & Eder, 1993; Kinney, 1993). Such qualitative studies are rich in descriptive power but have never been used to identify the peer affiliation patterns of large numbers of children.

Reciprocated Friendships

Children’s peer affiliation patterns are most commonly studied in the form of dyadic friendships. Research in this area highlights the importance of establishing whether or not a child’s self-reported friendships are reciprocal, mutual-liking relationships (Hartup, 1996). Reciprocated friendships differ from unreciprocated friendships in several important ways: for example, reciprocated friendships are characterized by more equitable patterns of conflict resolution and higher levels of mutual positive affect (Hartup, 1996). Reciprocated friendships are also characterized by higher rates of observed interaction: for example, preschoolers were observed to interact more fre- quently with their reciprocated friends than with their unreciprocated friends (Vaughn, Colvin, Azria, Caya, & Krzysik, 2001). This suggests that children’s reports of their friendships, when they are ‘verified’ by the specific classmates identified as friends, can provide valid indicators of actual peer interaction patterns. Analyses of recipro- cated friendships, however, do not help to characterize children’s larger, informal peer group affiliations.

Peer-Reported Social Clusters and Observed Interactions 515

Peer Reports of Informal Peer Groups: The Social-Cognitive Map (SCM) Method

The need for an efficient means of identifying children’s peer social networks moti- vated Cairns to develop the peer-report SCM method (Cairns et al., 1985; Cairns et al., 1988). In the SCM method, informants are asked to identify as many social clusters in the social setting (typically a school or classroom) as they can. Informants are not restricted to naming only social clusters of which they are a member; they are not provided with lists of children in the network; and they are not required to cate- gorize every member of the network into a social cluster.

Cairns’ SCM method reflected two methodological insights (Cairns et al., 1985; Cairns et al., 1988). The first insight was that children are expert observers of the entire peer social network who share reasonably convergent views on the composition of social clusters in the network. With a relatively simple prompt, most children can describe the composition of several social clusters, not just the one in which they par- ticipate. The second insight was that these individual social-cognitive maps can be aggregated into a composite social-cognitive map that provides a valid approximation of actual peer interaction patterns. The composite social-cognitive map is a co- nomination matrix in which each cell summarizes the number of times a given pair of children were named by peers as being in the same cluster. The co-nomination matrix is analogous to the ‘interaction matrix’ that is generated in observational studies of social networks (in which each cell summarizes the number of times two children were observed to interact). In the SCM approach, the co-nomination matrix is thought to provide an approximation of actual peer interaction patterns and underlies all sub- sequent steps in the process of describing peer-reported affiliation patterns. Various algorithms can be applied to the co-nomination matrix to identify discrete social clus- ters or to identify a given child’s significant peer relationships (Cairns et al., 1985; Cairns et al., 1988; Cairns, Gariepy, & Kindermann, 1991a; Kindermann, 1996).

Bagwell and colleagues (2000) recently described a peer report method that has important similarities and differences with the SCM method. They asked each partici- pant to identify classmates he or she ‘hangs around with’ by circling names on a class roster. These self-reports were summarized in a voter (rows) by votee (columns) matrix that was factor-analyzed to identify social clusters. The procedure is similar to the NEGOPY computer program for analyzing social networks (Richards & Rice, 1981). The SCM method and Bagwell and colleagues’ method share three important features: both methods include reports from all possible peer reporters in a given network; both methods aggregate data across peer informants into a summary matrix (although the summary matrices take somewhat different form); and both methods derive social clusters based on children’s similarity in patterns of nominations. The two methods also differ in three important respects. First, in the SCM method, participants are asked to describe all of the social clusters in the classroom; whereas Bagwell and colleagues ask participants to describe only the social cluster to which they belong. Second, in the SCM method, similarity in patterns of co-nomination is analyzed via matrix manipulations; whereas Bagwell and colleagues analyze similarity patterns with factor-analysis. Third, in the SCM method, network centrality (not a focus of the present discussion) is defined in terms of a child’s total number of nominations to social clusters; whereas Bagwell and colleagues define network centrality in terms of children’s factor loadings on their factor-defined social clusters.

The present study does not address the relative merits of different approaches to describing children’s peer affiliations (i.e., direct observations, reciprocated

516 Scott D. Gest et al.

friendships, peer report methods). Rather, the current focus is on the validity of the assumptions underlying the one particular approach: the SCM peer report method. Specifically, the present study focuses on the generalizability of the link between observed patterns of interaction in the classroom and peer nominations of social cluster membership generated with the SCM method.

Research on the Validity of the SCM Method

Cairns’ initial report describing the SCM method provided critical evidence that it could yield valid estimates of actual peer interaction patterns (Cairns et al., 1985). Cairns and colleagues employed the SCM method to identify discrete social clusters in one 7th grade classroom. Five girls and five boys from the classroom were also observed extensively in the classroom and in the lunchroom. The key finding was that children were observed to interact substantially and reliably more frequently with members of their SCM-derived social cluster (M = 4.35 interactions per cluster mate per hour) than with other same-sex classmates (M = 1.12 interactions per other same- sex classmate per hour). This remains the only study examining the extent to which the peer-reported social clusters upon which the SCM method rests are associated with observed patterns of peer interaction. Other evidence for the validity of the SCM method comes from research on the structure, stability and functions of SCM-derived social clusters.

Research on the structure of SCM-derived social clusters indicates that dyadic friendships are often embedded within these clusters (Cairns, Leung, Buchanan, & Cairns, 1995; Gest & Fletcher, 1996). Self-reports of cluster membership are gener- ally concordant with SCM-derived clusters; discrepancies are consistent with a self- enhancement bias in that self-reports tend to exclude low-status cluster members and (much less often) include higher-status non-cluster members (Gest & Fletcher, 1996; Leung, 1996). There is moderate stability in social cluster membership across a three- week interval (Cairns et al., 1995) and modest stability over a one-year interval when there is continuity in classroom membership (Neckerman, 1996).

Research on SCM-derived social clusters has also addressed substantive questions about possible peer influences on individual adjustment. The SCM method has been used to demonstrate that: (a) aggressive children tend to belong to social clusters with other aggressive children (Cairns et al., 1988; Xie, Cairns, & Cairns, 1999); (b) bullies often belong to social clusters with peers who reinforce their bullying behavior (Salmivalli, Huttunen, & Lagerspetz, 1997); (c) peers inside and outside a child’s peer group respond differently to children’s achievement behaviors (Sage & Kindermann, 1999), and the average school motivation of a child’s peers predicts change over time in the child’s own school motivation (Kindermann, 1993); (d) children with emotional and behavioral disorders are typically well integrated in classroom social clusters (Farmer & Hollowell, 1994); and (e) children who affiliate with aggressive peers in middle school are at increased risk for subsequent school dropout (Cairns et al., 1988), criminal activity (Mahoney, 2000) and early parenthood (Xie, Cairns, & Cairns, 2001).

The Present Study

The research summarized above provides important evidence for the reliability and external validity of social clusters derived from the SCM method. The initial report by Cairns and colleagues (1985), however, remains the only study that directly exam-

Peer-Reported Social Clusters and Observed Interactions 517

ines the link between SCM nominations and observed peer interactions. Although that study established such a link, it was based on 10 students attending a single 7th grade classroom. Given the potentially broad applicability of the SCM method, it is impor- tant to replicate the link between SCM nominations and observed interactions in a larger and more diverse sample.

The aim of the present study is to replicate and generalize the association between peer-reported social clusters and observed patterns of peer interaction in a larger sample of 4th grade and 7th grade children. We replicated the findings reported by Cairns and colleagues (1985) in two ways. First, we compared patterns of observed peer interactions to patterns of peer nominations of social cluster membership. Specifi- cally, we constructed two ‘profiles’ for each student: a ‘co-nomination profile’ sum- marizing the number of times a student was named as being in the same social cluster as each member of his or her classroom; and an ‘interaction profile’ summarizing how many social interactions a child was observed to have with each member of his or her classroom. For each student, the correlation between these two profiles provided a direct measure of the extent to which patterns of peer-nominated social cluster affilia- tions were associated with patterns of observed peer interaction. We expected that, on average, these correlations would be positive and statistically reliable. Second, we compared students’ rates of interaction with members of their SCM-identified social clusters to their rates of interaction with other same-sex classmates. We expected that students’ rates of interaction would be substantially higher with members of their SCM-identified clusters.

We examined the generalizability of these findings with respect to students’ grade, sex and aggressive risk status. We did not have a priori reasons to expect the SCM method to vary in validity across levels of these factors, but we wanted to document explicitly its validity for different groups of students. We examined variation by grade level (4th and 7th) because one might expect that children become more competent observers of classroom social networks as they develop, in which case the SCM method would demonstrate more robust associations with classroom behavior among older children. We examined variation by aggressive risk status (Aggressive and Nonaggressive) because there is broad interest in the affiliation patterns of aggressive children, so it is important to verify the validity of the SCM method for aggressive children. We examined variation by sex because many researchers are interested in sex differences in the form and function of peer relations (e.g., patterns of indirect and relationship-oriented aggression).

We further explored the generalizability of the SCM method by examining whether there is a minimum number of peer social cluster nominations a child must receive to arrive at a valid estimate of that child’s peer interaction patterns. Recall that with the SCM method, children are not provided with a list of children in the social network and they are not required to name every child in the network to a social cluster. As a result, there are typically some children who are named to social clusters by only a handful of their classmates and others who are named by virtually all of their class- mates. We expected that the SCM method would provide a less robust approximation of actual interaction patterns for students who received relatively few social cluster nominations. This can be regarded as a statistical phenomenon (i.e., less variability in nomination profiles would tend to produce weaker external correlates); but given that variability in nomination frequency is an inherent aspect of the SCM method, it is important to examine whether the validity of SCM method varies by frequency of nomination to social clusters.

518 Scott D. Gest et al.

Methods

Sample

Overall Sample. Participants were drawn from the Carolina Longitudinal Study (CLS; Cairns & Cairns, 1984, 1994), which consists of two cohorts of children who were first seen in either 4th Grade (Cohort I; N = 220) or 7th Grade (Cohort II; N = 475). Participants in Cohort I were recruited from four elementary schools (M = 10.2 years, SD = .57). Participants in Cohort II were recruited from three middle schools (M = 13.4 years, SD = .58). The seven schools were located in two counties, one of which was classified as a suburban metropolitan area and the other as a rural area according to the 1980 U.S. Census. Family socioeconomic status ranged from unemployed or unskilled labor to professional positions (doctors, lawyers); on the Duncan scale, the mean SES was 30.2 (SD = 17.1) in Cohort I and 31.6 (SD = 17.8) in Cohort II. Twenty- five percent of the participants were minority status, primarily African-American. In the schools from which the cohorts were recruited, no systematic differences were observed between participants and non-participants in terms of ethnic status, race, and probability of being nominated as highly aggressive.

Observational Sub-sample. For each cohort, school personnel were asked to identify children whom they considered to be highly aggressive. Ten boys and ten girls in each cohort who were nominated as highly aggressive by two or more school personnel were considered to be at Aggressive Risk. For each of these 40 students, a classmate was identified who matched the Risk student in terms of sex, age, and race, but who did not receive any school personnel nominations as being highly aggressive. Teacher ratings of aggression were much higher for students identified as being at Aggressive Risk (Cairns, Cairns, Neckerman, Ferguson, & Gariepy, 1989); and Aggressive Risk students experienced substantially higher rates of problem behavior in adolescence (Cairns & Cairns, 1994). Classroom behavioral observations were conducted for 78 of these 80 students (observations could not be arranged with the teacher in one class- room that contained two students). Peer reports of social cluster membership were obtained from all participating students in the classrooms attended by these 78 stu- dents. Six of the 78 students were not named to a social cluster by any of their class- mates and were therefore excluded from analyses (i.e., these students had no variation in their ‘nomination profiles,’ making it statistically impossible to examine covari- ation between their nomination profiles and their observed interaction profiles). All analyses in the present paper include the 72 students for whom both behavioral obser- vations and peer-reported social cluster membership were available. These 72 students varied by grade level (36 4th grade, 36 7th grade), sex (36 girls, 36 boys), Aggressive Risk Status (38 non-risk, 34 risk) and Race (56 White, 15 African-American, 1 Asian- American).

Social-Cognitive Map (SCM) Procedures

Details of the social-cognitive Map (SCM) method are described in previous reports (Cairns et al., 1985; Cairns et al., 1988). Here we summarize three key steps in the procedure.

Individual Social-Cognitive Maps. Students provided reports of social clusters during individual interviews that were audiotaped and subsequently transcribed. The specific

Peer-Reported Social Clusters and Observed Interactions 519

question regarding peer social clusters was, ‘Are there some kids here at school who hang around together a lot?’ Interviewers used general prompts to encourage children to name as many clusters as they could recall. Children who named only same-sex groups were prompted to consider whether there were any other-sex groups. (This prompt was not used if, in response to the more general prompts, children named some mixed-sex groups or some groups of boys and some groups of girls.)

Composite Social-Cognitive Maps. These individual social-cognitive maps were aggregated across informants to yield a ‘co-nomination matrix’ (sometimes called a co-occurrence matrix) that is symmetric around the diagonal. This co-nomination matrix represented a composite social-cognitive map of the classroom. Each cell in the co-nomination matrix summarized the total number of times a given pair of chil- dren was named by peers as being in the same social cluster. (Each cell along the diag- onal contained the total number of times a given child was named to a social cluster.) There was substantial variation in the total number of times children were named as being in a social cluster: some children were named by virtually all of the children who were interviewed, but others were named by very few (range from 1 to 21, M = 7.67, SD = 4.43). Regardless of this variation, each column (or row) in the matrix can be regarded as a co-nomination profile that summarized the frequency with which a given child was reported by peers to be in the same social cluster as each of his or her classmates.

Identification of Discrete Social Clusters. Discrete, non-overlapping social clusters were derived from the social-cognitive maps by grouping together children who shared similar co-nomination profiles. To ensure continuity with previous reports from this project (Cairns et al., 1988, 1989; Cairns & Cairns, 1994; Mahoney, 2000), we use the social clusters established by Cairns and colleagues (1988), who describe the fol- lowing procedures in detail. First, the columns of the co-nomination matrix were inter- correlated. Each cell in the resulting correlation matrix summarized the degree of similarity between the co-nomination profiles of a given pair of students. Second, pre- liminary social clusters were defined by grouping together those students whose co- nomination profiles were reliably intercorrelated (p < .05). Transitivity in patterns of reliable intercorrelations was common but not universal (e.g., if child A was reliably correlated with B and C, then B and C were typically reliably correlated); as a result, the preliminary grouping process sometimes led to partially overlapping social clus- ters. Third, to arrive at non-overlapping social clusters (which was an a priori goal of the process), several alternative groupings were compared with confirmatory factor analysis to arrive at the best-fitting model of non-overlapping social clusters. It is important to note that although there are several alternative approaches to analyzing the composite social map to arrive at final decisions regarding social cluster bounda- ries (e.g., contingency analyses, multi-dimensional scaling, cluster analysis), the various approaches yield highly similar solutions (Cairns et al., 1991a). In the overall sample of 695 students, 115 social clusters were identified; 64% of those clusters had between 4 and 7 members (M = 5.68, SD = 2.33; median and mode = 6, range from 2 to 12). The average size of social clusters did not differ reliably across grade (4th

grade M = 5.31, SD = 2.53; 7th grade M = 5.87, SD = 2.31).

Mixed-sex Social Clusters. One composite social map was created for each classroom; individual reports of mixed-sex social clusters were represented within this composite

520 Scott D. Gest et al.

social-cognitive map. In principle, this meant that mixed-sex social clusters could emerge from the SCM method; but in practice, this seldom occurred. Most children described social clusters comprised of either all boys or all girls. When discrete social clusters were derived from the composite social map using the procedures described above, mixed-sex social clusters emerged for only 3 of the 72 participants (all 3 were in 7th grade).

Out-of-classroom Nominations. Composite social-cognitive maps were constructed separately for each classroom because the vast majority of children reported clusters existing within their own classrooms. If a particular child outside the target classroom was reported as a member of a cluster in the target classroom on only one occasion, that report was discarded as idiosyncratic. If there were two or more such reports for the same child, however, that child was added to the composite social-cognitive map for the target classroom. That child was also included in the composite social- cognitive map of his or her ‘home’ classroom. Children invariably received more social cluster nominations from peers within their ‘home’ classroom than from peers in other classrooms. Consequently, all children’s final social cluster assignments were within their ‘home’ classroom.

Convergence Among Individual Social-cognitive Maps. The SCM practice of aggre- gating individual social-cognitive maps into a composite social-cognitive map pre- sumes there is at least moderate agreement among classmates regarding the social clusters existing in a classroom. To verify this assumption, we compared individual social-cognitive maps to the final SCM social clusters that were derived from the com- posite social-cognitive maps for a 10% sub-sample of the larger sample of 695 chil- dren who provided reports of social clusters. Because individuals were not required to name all of their classmates to social clusters, we only considered those classmates whom a child included in his or her individual social-cognitive map. For example, if an individual named 11 classmates to social clusters, there were 55 possible dyadic connections among those 11 classmates (i.e., [11*10]/2 = 55). Dyadic connections among those 11 classmates that were implied by both the individual social-cognitive map and the final SCM social clusters (or by neither individual social-cognitive map nor the final SCM social clusters) were considered agreements; connections that were implied by the individual social-cognitive map but not by the final SCM social clus- ters (or vice versa) were considered disagreements. Agreement was calculated sepa- rately for each individual using kappa. We selected six boys and six girls at random from each of three 4th grade and three 7th grade classrooms (total N = 72). To mini- mize the inherent part-whole relationship between individual social-cognitive maps and social clusters derived from the composite social-cognitive map, we selected the 72 children from classrooms in which a large number of students contributed to the composite social-cognitive map (range 24 to 28). Individuals in these classes con- tributed, on average, about 4% of the information to the composite social-cognitive map. Across these 72 individuals, the average number of dyads included in the kappa analysis was 56 (SD = 25.7, range 15 to 136). Average levels of agreement were sub- stantial: Mk = .65, median k = .61, range k = .18 to k = 1.00. Average percent agree- ment was 85%. The average kappa of .65 indicates that peers provided convergent reports of peer affiliation patterns and validates the practice of aggregating individual social-cognitive maps into composite social-cognitive maps.

Peer-Reported Social Clusters and Observed Interactions 521

Classroom Observations

The observation system was designed to provide a description of children’s social interactions and the classroom context in which they occurred (for details, see Cairns & Cairns, 1984, 1994; Cairns, Santoyo, Ferguson, & Cairns, 1991b). All observations took place during classroom instructional time. (Pilot testing had indicated that, ironi- cally, social interaction rates were typically higher during classroom instruction than during tightly regulated ‘transition times’ in the hallways.) The observation system consisted of three stages: observing classroom interactions; identifying social inter- action episodes; and creating interaction profiles.

Observing Classroom Interactions. Two observers worked simultaneously in the same classroom to record a target child’s behavior and the classroom social context in which it occurred. One observer provided a brief narrative summary of the target child’s behavior and social interactions in five-second time blocks. Observers did not use a priori categories to describe children’s behavior, thereby minimizing demands on observers to interpret the meaning of behaviors and maximizing the amount of descrip- tive information contained in the observation records. When the target child engaged in a social interaction, observers recorded the initiator of the social interaction, the identity of the social partner, and described the nature of the interaction in a very brief narrative note (e.g., ‘T [target child] tapped [peer’s name] w/pencil’). The second observer summarized the broader classroom context at the end of each one-minute interval (i.e., nature of instructional task, classroom atmosphere and any major dis- ruptions). The two observers switched roles every five minutes to reduce observer fatigue and to reduce the chances that students would identify the target of observa- tions. Observers gathered a median of 160 minutes of observations on each student (M = 169 minutes, SD = 40, range 80 to 285) across a median of eight observation days (M = 9.47 days, SD = 4.14, range 4 to 20).

Identifying Interaction Episodes. The two observers worked together to reconstruct a single record of the child’s social interaction episodes and the classroom context in which they occurred later in the same day on which classroom interactions were observed. By reconstructing this record of interaction episodes later in the same day, observers could more effectively interpret and characterize the interactions described in the ‘raw’ observation records. An interaction episode was defined as an action on the part of a child that produced, or should have produced, a response from another child (e.g., a verbal exchange; a tap on the shoulder whether answered or not). Some interaction episodes included a single bid-response sequence; others included multi- ple bids-responses. Interaction episodes were punctuated by at least one five-second block in which no bids or responses occurred. Each summary of an interaction episode included a brief description of the bid-response sequence, the identity of the peer with whom the child interacted, and the classroom social context at the time of the inter- action. There was substantial variation in the total number of observed interaction episodes (M = 88.67 episodes, SD = 42.41, range 21 to 188).

Creating Interaction Profiles. Each student’s interaction episodes were tallied into an interaction profile that summarized the total number of interaction episodes that the student engaged in with each of his or her classmates.1

522 Scott D. Gest et al.

Reliability of the Observation System. Data on the reliability of the procedures for observing classroom interactions and identifying interaction episodes come from the study of a separate sample described by Cairns and colleagues (1985). That study is relevant to the present study because: (a) both studies were directed by the developer of the observation system; (b) the same individual (the third author) served as the primary observer in both studies; and (c) the studies were conducted in consecutive years in similar self-contained classroom settings. Ten 7th grade students were each observed for a total of 30 minutes by two observers who independently and simulta- neously recorded the target child’s behavior in five-second time blocks (Cairns et al., 1985). The reliability of the procedures for observing classroom interactions was high: the two observers reached 98% agreement regarding whether an interaction occurred or did not occur in a particular five-second time block; and achieved over 98% agreement regarding the identity of the peer with whom a child interacted. The reliability of the procedure for identifying interaction episodes was also high: each of the two observers independently coded her own observation protocol to identify inter- action episodes as defined above; there was strong agreement regarding the rank- ordering of the target students in terms of their total number of interaction episodes, r (10) = .97.

Results

Overall Links Between Peer-Nominated Affiliations and Observed Interactions

Correlations Between Co-nomination Profiles and Interaction Profiles. Co- nomination profiles were correlated with interaction profiles separately for each student. These Pearson correlations were averaged across the 72 students. (Each cor- relation was transformed using Fisher’s r-to-z transformation before computing the average; the resulting average was converted back from z-to-r for display here.) The average correlation across all 72 students was substantially and reliably greater than zero, Mr = .55, t(71) = 10.67, p < .001; median r = .51, minimum r = -.36, maximum r = .96. This indicates that, on average, there was a moderate association between pat- terns of nominations that underlie the SCM method and patterns of observed peer interaction in the classroom. It is important to note that these correlations did not depend on decisions about membership in a particular social cluster; whereas the next analyses address the validity of decisions regarding discrete social clusters.

Rates of Interaction with SCM-derived Cluster Members and Other Same-sex Class- mates. Rates of interactions with SCM-derived cluster members were compared with rates of interaction with other same-sex classmates. The comparison was made with other same-sex classmates because: (a) for 69 of 72 participants, social clusters were entirely same-sex groups; and (b) rates of interaction were higher for same-sex class- mates than for other-sex classmates; so that (c) comparing interaction rates with cluster members to interaction rates with all other classmates would have confounded dif- ferences in interaction rate due to cluster membership status with differences in inter- action rate due to same-sex versus other-sex status. Rates of interaction were evaluated in a repeated measures analysis of variance with one within-subjects factor (Cluster Membership: Member vs. Other Same-Sex Classmate). Results indicated a significant main effect for Cluster Status, F(1,71) = 31.97, p < .001. On average, students inter- acted about four times more frequently with members of their own social cluster (M = 3.18 interaction episodes per cluster member per hour, SD = 3.58) than with

Peer-Reported Social Clusters and Observed Interactions 523

other same-sex classmates (M = .80 interaction episodes per other same-sex classmate per hour, SD = .47).

Variations by Grade, Sex, Aggressive Risk Status, and Number of Social Cluster Nominations

In the following analyses, the total number of social cluster nominations each child received from his or her classmates was treated as a continuous covariate because it was positively correlated with the strength of the association between co-nomination profiles and interaction profiles, r = .35, p < .01. This variable was examined in some detail because it has practical implications for the use of the SCM methodology to estimate the interaction patterns of children who are seldom named to social clusters by their classmates. In contrast, the total number of observed interaction episodes was not treated as a covariate because it was not reliably correlated with the strength of the association between co-nomination profiles and interaction profiles and because it was an outcome of the quantity of observations collected.

Correlations Between Co-nomination Profiles and Interaction Profiles. We tested a 2 ¥ 2 ¥ 2 ANCOVA model in which co-nomination/interaction correlations were the dependent variable (transformed from r-to-z), with two levels of Sex (Male/Female), two levels of Grade (4th/7th) and two levels of Aggressive Risk Status (Aggressive/ Non-aggressive). Cell sizes ranged from 8 to 10. The total number of nominations to social clusters was treated as a continuous covariate. Results indicated no reliable main effects or interactions involving Sex, Grade or Aggressive Risk Status.

The total number of social cluster nominations was a significant covariate, F(1,63) = 6.89, p < .01. To better understand this effect, co-nomination/interaction correla- tions were summarized separately for students with differing numbers of social cluster nominations: 1 to 3; 4 to 6; 7 to 9; 10 or more (see Table 1). These ranges resulted in roughly equal-N groupings of sufficient size to conduct separate statistical tests. Cor- relations were strongest for students with 10 or more nominations, Mr = .67, t(20) = 8.39, p < .001; and were similarly robust for students with 7 to 9 nominations, Mr = .62, t(18) = 6.49, p < .001. Correlations were modest but remained reliable for stu- dents with 4 to 6 nominations, Mr = .43, t(16) = 4.56, p < .001; and for students with 1 to 3 nominations, Mr = .38, t(14) = 2.85, p < .05. The less robust correlations for students with fewer nominations to social clusters can be seen as a statistical phe- nomenon (i.e., their co-nomination profiles contain less variability than those of more frequently nominated classmates). Nonetheless, these findings indicate that even a small number of SCM nominations can yield a co-nomination profile that is a valid approximation of actual classroom interaction patterns.

Rates of Interaction with Cluster Members Versus other Same-sex Classmates. We tested a 2 ¥ 2 ¥ 2 repeated-measures MANCOVA model that included one within- subjects factor (Cluster Membership: Member vs. Other Same-Sex Classmate), three between-subjects factors (Grade, Sex, Aggressive Risk Status) and one covariate (number of social cluster nominations). The effect of Cluster Membership was not reliably moderated by Grade, Sex or Aggressive Risk Status. These patterns are con- sistent with the analyses of correlations above in suggesting that the validity of the SCM method did not vary significantly for 4th and 7th grade boys and girls with varying aggressive risk status.

524 Scott D. Gest et al.

There was, however, a marginally reliable interaction effect between Cluster Mem- bership and number of social cluster nominations, F(1,67) = 2.83, p < .10. To clarify this effect, Table 1 summarizes interaction rates with cluster members and other same- sex classmates for students at each of the four levels of social cluster nominations. Univariate F-tests revealed reliable effects of Cluster Membership at each level of social cluster nominations. As with the results for correlations, Cluster Membership effects were most robust for students named to social clusters 10 or more times, F(1,20) = 11.15, p < .01; and for students named to social clusters between 7 and 9 times, F(1,18) = 27.47, p < .001. Nonetheless, even among children who received between 1 and 3 social cluster nominations, rates of observed interaction with cluster members were reliably greater than rates of interaction with other same-sex class- mates, F(1,14) = 5.96, p < .05.

Post-Hoc Analyses: Validity of SCM Procedure for Racial Minority Students

To the extent that social clusters are segregated by race, one might expect the peer affiliations of racial minority students to be described with less validity by racial majority classmates. This possibility was not examined in the ANCOVA and MANCOVA models because race was not a planned factor in the study design and its inclusion would have resulted in unacceptably small cell sizes. Instead, the key analy- ses were repeated including only racial minority students (N = 16). Results indicated that the correlation between co-nomination profiles and interaction profiles was sta- tistically reliable, Mr = .55, t(15) = 5.19, p < .001; and that rates of interaction with SCM-derived social cluster members were higher (M = 4.56 interactions per SCM © Blackwell Publishing Ltd. 2003 Social Development, 12, 4, 2003

Table 1. SCM Nominations and Observed Interactions

Observed Interaction Episodes

Total Number of Average r: SCM Other Peer Nominations to Co-Nomination & Cluster Same-Sex Social Clusters Interaction Profiles Members Classmates F-value

Overall (N = 72) .55***a 3.18 (3.58) .80 (.47) 31.97***b 1, 2 or 3 (N = 15)c .38* 2.70 (.78) .78 (.43) 5.96* 4, 5, or 6 (N = 17) .43*** 2.26 (2.03) 1.00 (.62) 5.95* 7, 8 or 9 (N = 19) .62*** 2.81 (1.68) .76 (.40) 27.47*** 10 or more (N = 21) .67*** 4.60 (5.54) .70 (.38) 11.15**

Note: Figures in parentheses are standard deviations. * p < .05; ** p < .01; *** p < .001. a Average Pearson r between co-nomination profiles (based on SCM peer nominations) and interaction profiles (based on observed classroom interactions), N = 72. Correlations were trans- formed from r-to-z for statistical tests and were transformed back from z-to-r for display here. b F-Test for difference in rates of observed interaction between classmates in a child’s SCM- defined social cluster and other same-sex classmates, averaged across all 72 children in the study. c Row entries are interpreted the same as for the first row, but apply only to the N = 15 students who were nominated to a social cluster by peers a total of 1, 2 or 3 times.

Peer-Reported Social Clusters and Observed Interactions 525

member per hour, SD = 6.12) than with other same-sex classmates (M = .84 interac- tions per classmate per hour, SD = .41), F(1,15) = 6.27, p < .05. These effects are con- sistent with the results for the overall sample.

Discussion

Children’s reports of their classmates’ peer affiliations obtained with the social- cognitive map (SCM) procedure were substantially and consistently associated with observed patterns of peer interactions in the classroom. This was demonstrated by the positive association between the number of times two children were named by peers as being in the same social cluster and the number of times those two children were observed to interact in the classroom; and by the finding that students were observed to interact with members of their SCM-derived social clusters four times more often than with other same-sex classmates. These two findings held true for 4th grade and 7th grade students; for boys and girls; for aggressive and non-aggressive children; and for children in the racial minority. These effects were most robust when students were named frequently to social groups, but they remained reliable among students who were seldom named to social clusters.

These findings validate two key premises of the SCM methodology. First, children are expert observers of the peer social networks in their classrooms who provide rea- sonably convergent information about these peer networks. This premise underlies the SCM practice of aggregating children’s individual peer reports (individual social- cognitive maps) into a single co-nomination matrix (composite social-cognitive map) in which each cell summarizes the number of times a given pair of children were named as being in the same social cluster. The fact that children’s individual social- cognitive maps tended to agree with the social clusters derived from the composite social-cognitive map (median kappa = .65) supports this premise. It is important to note, however, that the power of the SCM method comes not from the reliability of individual reports, but from the aggregation of many such reports into the co- nomination matrix.

Second, the co-nomination matrix (composite social-cognitive map) provides a valid estimate of actual classroom peer interaction patterns. The positive correlations between co-nomination profiles and observed interaction profiles directly support this premise. These correlations existed despite numerous constraints on children’s inter- action patterns during observation sessions (e.g., classroom instructional activities; seating arrangements; occasional absences of classmates). The co-nomination matrix was also used to derive a set of non-overlapping social clusters within each classroom by grouping together children whose co-nomination profiles were relatively highly intercorrelated (Cairns et al., 1988). The validity of this approach was supported by the finding that children were observed to interact with members of their SCM-derived social cluster four times more often than with other same-sex classmates. In fact, the ratio of within-cluster to outside-of-cluster interaction rates in the present study (3.18/.80 = 3.98) is strikingly similar to the ratio that Cairns and colleagues (1985) observed in their initial study (4.35/1.12 = 3.88). In sum, the co-nomination matrix produced by the SCM method provides a valid summary of overall classroom interaction patterns that can be used to identify the key social partners of individual children.

These findings support the validity of other methods for identifying peer affiliation patterns that are anchored in the co-nomination matrix. Kindermann (1993), for

526 Scott D. Gest et al.

example, uses the co-nomination matrix as the basis for a dyadic linkage analysis that results in the identification of each child’s peer social network. The present results do not address the validity of dyadic linkage analysis, but early work on the SCM method indicated that several alternative approaches to analyzing the co-nomination matrix (e.g., correlations; multidimensional scaling; dyadic linkage) yielded highly conver- gent solutions (Cairns et al., 1991b).

The present findings extend the earlier work of Cairns and colleagues (1985) by demonstrating the generalizability of the SCM methodology. Given that the validity of the SCM method did not differ reliably for 4th grade and 7th grade students, it seems reasonable to hypothesize that the method may have validity at grade levels outside of that range. The SCM method has been used among children in 1st to 3rd grades (Estell et al., 2002; Gest et al., 2001), but these studies have not validated SCM nom- inations with respect to observed interaction patterns. The less robust validity of the SCM method for individuals receiving relatively few nominations may be relevant to the validity of the SCM method among younger children. If younger children provide less complete reports of classroom social clusters (e.g., because of less complete knowledge of peers’ affiliation patterns; or greater difficulty in retrieving such know- ledge in response to open-ended prompts), then individual co-nomination profiles would have less variability (being based on fewer total reports of social clusters) and would therefore be somewhat less robust. At higher grade levels, it is not clear to what extent the SCM method would be valid in middle schools and high schools in which there are no self-contained classrooms and students at a given grade level may not be familiar with each other’s affiliation patterns. In sum, the SCM method is clearly valid in relation to observed interaction patterns among children in 4th grade to 7th grade; validation in relation to observed interaction patterns at lower and higher grade levels awaits further study.

The fact that the SCM method was equally valid for aggressive and non-aggressive children suggests it can be a valuable tool in current efforts to understand the role of peers in the emergence and maintenance of aggressive behavior in middle childhood. As described by Cairns and colleagues (Cairns et al., 1988), aggressive children in the present study were as likely as their non-aggressive peers to be members of social clusters; but they tended to belong to clusters with other aggressive children. The present findings underscore that patterns of peer-reported affiliations are as strongly anchored in observable interactions for aggressive children as they are for non- aggressive children.

Perhaps surprisingly, the SCM methodology also proved valid among children who were named to social clusters only one to three times. Even among these children there was a reliable correlation between co-nomination profiles and interaction profiles; and rates of interaction with SCM-derived social cluster members were three times higher than with other same-sex classmates. These findings diminish potential concerns about the utility of the SCM procedure for children who are seldom named to social clusters.

These findings also highlight the importance of distinguishing between using the SCM methodology to measure social affiliations (i.e., social clusters) and using the SCM methodology to describe individual differences in social network centrality (i.e., the total number of times a child is named to a social cluster). There is an emerging literature in which SCM methodology is employed to examine the meaning of social network centrality (e.g., Farmer & Rodkin, 1996; Gest et al., 2001; Rodkin, Farmer, Pearl, & Van Acker, 2000). The present findings that social clusters can be identified

Peer-Reported Social Clusters and Observed Interactions 527

with validity for individuals with varying social network centrality indicates that, to some degree, it is possible to study children’s affiliation patterns independently of their centrality within the network.

In choosing among the various methods available for identifying children’s peer affiliations, the SCM methodology has several attractive features. Like widely used peer-report measures of social behavior (Masten, Morison, & Pellegrini, 1985), the SCM method takes advantage of peers’ access to the diverse settings in which peer interactions take place (e.g., hallways, playgrounds, buses); and of the efficiency of obtaining peer reports in either brief individual interviews or in classroom question- naires. One unique advantage of the SCM method is that a child’s affiliation patterns can be identified even when the child does not provide a self-report of affiliations. This advantage derives from the fact that with the SCM method, a child’s affiliation pat- terns are determined based on reports from all of the peers in the social network, not solely the reports of the child and his or her friends.

Several features of this study limit the generalizations that can be made. First, this study was limited to 4th and 7th grade students attending schools with relatively self- contained classrooms. As noted above, explicit validation of the SCM procedure with respect to observed interaction patterns at lower and higher grades awaits further study. Also, it is unclear whether the present findings would have been different if observa- tions had occurred in school settings other than classrooms (e.g., lunchroom, play- ground, P.E. class). Cairns and colleagues (1985) found highly similar patterns of affiliation in the classroom and in P.E. class, but the current observations were limited to classrooms. Finally, although the present results indicated that, overall, the SCM method was valid among the racial minority students in this study, sample size limi- tations precluded a more detailed examination of potential interactions between racial minority status and other factors (e.g., grade, sex, aggressive risk status).

Future studies could extend these findings by examining the links between SCM nominations and observed interactions among younger students and among older stu- dents attending schools without self-contained classrooms. It will also be important to understand the extent to which peer reports of social clusters are responsive to short- term changes in peer interaction patterns. For example, we know that other aspects of children’s peer reputation can be quite stable and resistant to change (e.g. Hymel, Wagner, & Butler, 1990), which has implications both for children seeking to con- solidate behavioral changes and for researchers seeking to measure short-term behav- ioral change. There is currently very little data regarding such issues with regard to peer reports of social clusters (Cairns et al., 1995).

In sum, this study validates two key premises of the SCM methodology: peers can provide convergent reports of classroom affiliation patterns; and these reports can be aggregated into a matrix that provides a valid approximation of actual peer interac- tion patterns. The SCM methodology relies on the same power of aggregated peer reports that has proven successful in studies of individual differences in social status (Asher & Coie, 1990) and social reputation (Masten et al., 1985). It is distinct, however, in eliciting and aggregating peer reports of peer affiliation patterns (Cairns et al., 1998). Given the increasing interest in the role of peer affiliations as a source of influence on social behavior, sparked in part by the early work of Cairns (Cairns et al., 1985; Cairns et al., 1988), the present validation of the SCM methodology provides impetus for further studies of the role of peer affiliations in children’s development.

528 Scott D. Gest et al.

References

Asher, S. R. & Coie, J. D. (1990). Peer rejection in childhood. New York: Cambridge Univer- sity Press.

Bagwell, C. L., Coie, J. D., Terry, R. A., & Lochman, J. E. (2000). Peer clique participation and social status in preadolescence. Merrill-Palmer Quarterly, 46, 280–305.

Cairns, R. B. & Cairns, B. D. (1984). Predicting aggressive patterns in girls and boys: A devel- opmental study. Aggressive Behavior, 10, 227–242.

Cairns, R. B. & Cairns, B. D. (1994). Lifelines and risks. Cambridge: Cambridge University Press.

Cairns, R. B., Gariepy, J-L., & Kindermann, T. (1991a). Identifying social clusters in natural settings. Unpublished manuscript. University of North Carolina at Chapel Hill.

Cairns, R. B., Perrin, J. E., & Cairns, B. D. (1985). Social structure and social cognition in early adolescence: Affiliative patterns. Journal of Early Adolescence, 5, 339–355.

Cairns, R. B., Xie, H., & Leung, M-C. (1998). The popularity of friendship and the neglect of social networks. New Directions for Child Development, 80, 25–53.

Cairns, R. B., Leung, M-C., Buchanan, L., & Cairns, B. D. (1995). Friendships and social net- works in childhood and adolescence: Fluidity, reliability, and interrelations. Child Develop- ment, 66, 1330–1345.

Cairns, R. B., Santoyo, V. C., Ferguson, L., & Cairns, B. D. (1991b). Integrating interactional and contextual information: The synchronized observation procedure. Mexican Journal of Behavior Analysis, 17, 105–120.

Cairns, R. B., Cairns, B. D., Neckerman, H. J., Ferguson, L. L., & Gariepy, J-L. (1989). Growth and aggression: I. Childhood to early adolescence. Developmental Psychology, 25, 325–330.

Cairns, R. B., Cairns, B. D., Neckerman, H. J., Gest, S. D., & Gariepy, J-L. (1988). Social net- works and aggressive behavior: Peer support or peer rejection? Developmental Psychology, 24, 815–823.

Estell, D. B., Cairns, R. B., Farmer, T. W., & Cairns, B. D. (2002). Aggression in inner-city early elementary classrooms: Individual and peer-group configurations. Merrill-Palmer Quarterly, 48, 52–76.

Evans, C. & Eder, D. (1993). ‘No exit’: Processes of social isolation in the middle school. Journal of Contemporary Ethnography, 22, 139–170.

Farmer, T. W. & Hollowell, J. H. (1994). Social networks in mainstream classrooms: Social affiliations and behavioral characteristics of students with emotional and behavioral disor- ders. Journal of Emotional and Behavioral Disorders, 2, 143–155.

Farmer, T. W. & Rodkin, P. C. (1996). Antisocial and prosocial correlates of classroom social positions: The social network centrality perspective. Social Development, 5, 174–188.

Gest, S. D. & Fletcher, A. C. (1996). Concordance between self-reported and consensus-based peer groups as a function of social network status in early adolescence. Poster presented at the Conference of the Society for Research on Adolescence, Boston, March.

Gest, S. D., Graham-Bermann, S. A., & Hartup, W. W. (2001). Peer experience: Common and unique features of number of friendships, social network centrality, and sociometric status. Social Development, 10, 23–40.

Hartup, W. W. (1996). The company they keep: Friendships and their developmental signifi- cance. Child Development, 67, 1–13.

Hymel, S., Wagner, E., & Butler, L. (1990). Reputational bias: View from the peer group. In S. R. Asher & J. Coie (Eds.), Peer rejection in childhood (pp. 156–186). Cambridge, England: Cambridge University Press.

Kindermann, T. A. (1993). Natural peer groups as contexts for individual development: The case of children’s motivation in school. Developmental Psychology, 29, 970–977.

Kindermann, T. A. (1996). Strategies for the study of individual development within naturally- existing peer groups. Social Development, 5, 158–173.

Kinney, D. A. (1993). From nerds to normals: The recovery of identity among adolescents from middle school to high school. Sociology of Education, 66, 21–40.

Ladd, G. W. (1983). Social networks of popular, average, and rejected children in school set- tings. Merrill-Palmer Quarterly, 29, 3, 283–307.

Leung, M-C. (1996). Social networks and self-enhancement in Chinese children: A compari- son of self reports and peer reports of group membership. Social Development, 5, 146–157.

Peer-Reported Social Clusters and Observed Interactions 529

Mahoney, J. (2000). School extracurricular activity participation as a moderator in the devel- opment of antisocial patterns. Child Development, 71, 502–516.

Masten, A. S., Morison, P., & Pellegrini, D. S. (1985). A revised class play method of peer assessment. Developmental Psychology, 21, 523–533.

Neckerman, H. J. (1996). The stability of social groups in childhood and adolescence. Social Development, 5, 131–145.

Richards, W. D. & Rice, R. E. (1981). The NEGOPY network analysis program. Social Net- works, 3, 215–223.

Rodkin, P. C., Farmer, T. W., Pearl, R., & Van Acker, R. (2000). The heterogeneity of popular- ity in boys: Antisocial and prosocial configurations. Developmental Psychology, 36, 14–24.

Rubin, K. H., Bukowski, W., & Parker, J. G. (1998). Peer interactions, relationships and groups. In W. Damon (Series Ed.) and N. Eisenberg (Vol. Ed.), Handbook of child psychology, Vol. 3: Social, emotional and personality development (pp. 619–700). New York: John Wiley & Sons, Inc.

Sage, N. & Kindermann, T. A. (1999). Peer networks, behavior contingencies, and children’s engagement in the classroom. Merrill-Palmer Quarterly, 45, 143–171.

Salmivalli, C., Huttunen, A., & Lagerspetz, K. (1997). Peer networks and bullying in schools. Scandinavian Journal of Psychology, 38, 305–312.

Strayer, F. F. & Santos, A. J. (1996). Affiliative structures in preschool peer groups. Social Development, 5, 117–130.

Strayer, F. F. & Strayer, J. (1976). An ethological analysis of social agonism and dominance relations among preschool children. Child Development, 47, 980–989.

Vaughn, B. E., Colvin, T. N., Azria, M. R., Caya, L., & Krzysik, L. (2001). Dyadic analyses of friendship in a sample of preschool-age children attending Head Start: Correspondence between measures and implications for social competence. Child Development, 72, 862–878.

Vaughn, B. E. & Waters, E. (1981). Attention structure, sociometric status, and dominance: Interrelations, behavioral correlates, and relationships to social competence. Developmental Psychology, 17, 275–288.

Xie, H., Cairns, R. B., & Cairns, B. D. (1999). Social networks and configurations in inner- city schools: Aggression, popularity, and implications for students with EBD. Journal of Emotional and Behavior Disorders, 7, 147–155.

Xie, H., Cairns, B. D., & Cairns, R. B. (2001). Predicting teen motherhood and teen father- hood: Individual characteristics and peer affiliations. Social Development, 10, 488–511.

Note 1. Preliminary analyses addressed the potential utility of separate interaction profiles summarizing ‘nega-

tive’ and ‘non-negative’ interaction episodes. Interaction episodes that clearly involved an attempt to cause harm or discomfort to the interaction partner (e.g., teasing, insulting, hitting, pushing, throwing an object) were coded as negative; all other interaction episodes were coded as non-negative. Inter-coder agreement for the negative/non-negative distinction was high, kappa = .86, based on the complete observation pro- tocols of two participants (N = 170 interaction episodes). Separate negative and non-negative interaction profiles were constructed for all 72 individuals in the sample. These profiles were highly correlated, median r(72) = .78, indicating that students showed a general tendency to interact more with some classmates than with others, and that distinct negative and non-negative interaction profiles were unlikely to reveal impor- tant differences. Consequently, the distinction between negative and non-negative interaction episodes was dropped and a single, overall interaction profile was constructed for each student.