The magazine publisher is the copyright holder of this article and it is reproduced with permission. Further reproduction of this article in violation of the copyright is prohibited.

ABSTRACT
Natural behavior contingencies were examined as a mechanism by which peers can influence children's school motivation in classroom interactions. Sequential observations in a fifth grade classroom identified contingencies that children experienced from peer group members, nonmembers, and the teacher as consequences of their behavior; peer groups were identified with a Composite Social Map procedure. The more students were motivated, the more likely they were to receive approval from peer group members following their active on-task behaviors. The less students were motivated, the more they received disapproval from nonmembers following their disruptive off-task behaviors. These contingency patterns constitute learning conditions that can be seen as a mechanism through which a child's peer group members can influence that child's school motivation.
The study of group influences on individuals' behavior has enjoyed a long and rich history, predominantly in the fields of social and experimental psychology. The effects of group processes have been shown to be powerful and robust, to generalize across many experimental and naturalistic conditions, and to exert their influences on various target variables (e.g. Asch, 1955; Sherif, Harvey, White, Hood, & Sherif, 1961). Developmentalists also have shown a longstanding interest in children's peer groups (for reviews, see Hartup, 1983; Rubin, Bukowski, & Parker, 1998). Theorists (e.g., Piaget, Vygotski) have argued that influences from friends and peer affiliates play a central role in children's development.
Experimental studies are usually considered to provide the strongest evidence for group influence effects. In these studies, peer influences are typically exerted by randomly assigned individuals who do not share a pre-established relationship and a prior history. However, for developmentalists, more interesting questions about peer influence processes during childhood do not concern effects from random strangers, but effects from children's natural friends and peer affiliates. Thus, the focus of developmental research is more on influences from natural group contexts and on questions about the extent to which change in individuals across time is a result of the characteristics of their peer contexts.
Studies that have experimentally identified influences from children's natural peer contexts are rare. One example is a study by Berndt, Laychak, and Park (1990) in which the role of discussions among adolescent friends were examined for changes in their self-reported behavior. Pairs of friends who discussed academically related dilemmas with their friends (e.g., whether to do school work or attend a rock concert) converged more in their decisions than did pairs who discussed nonacademic topics. This study shows that influences from friends can lead to increasing similarity in behavior, and that discussions can produce such convergence.
In order to examine outcomes of natural socialization processes across longer time frames, researchers necessarily have to rely on correlational studies that examine similarities between the characteristics of a child and the characteristics of his or her peer affiliates. During recent years, a considerable number of such studies has accumulated. Researchers have examined friendship dyads (e.g., Berndt & Keefe, 1996; Dishion, Spracklen, Andrews, & Patterson, 1996; Hallinan & Williams, 1990; Kandel, 1978), groups of friends (e.g., Ennett & Bauman, 1994; Mounts & Steinberg, 1995; Urberg, 1992), and peer group networks (e.g., Cairns, Cairns, & Neckerman, 1989; Kindermann, 1993). Overall, the findings are consistent with the hypothesis that influences can occur from natural peer contexts on children's attitudes, beliefs, and behaviors, and that these influences can be important for children's further development.
However, correlational studies on the influences from natural peer contexts are faced with one drawback. Although findings of significant relations between characteristics of peer affiliations and children's own characteristics can indicate influence processes, alternative explanations also are possible. The relations also can be results of pre-existing conditions or concurrent external factors, without actual involvement of direct influences from peers.
One such pre-existing factor is peer selection. Childhood peer contexts have the characteristic that children are able to select for themselves (from a pool of candidates) those other children with whom they want to affiliate. These self-selected affiliations are based on mutual liking, shared interests, or shared activities. Thus, high levels of similarity on a variety of variables are the basis from which groups of affiliates are created (e.g., groups of friends or peers who merely "hang out" together). Overall, selection processes in natural groups appear to be as powerful as socialization influences in creating person-to-group similarity (e.g., Kandel, 1978).
Peer selection also can explain increasing group convergence across time, because peers may have joined who were already on similar developmental pathways beforehand. For example, if children select as friends other children who are low on parental monitoring (e.g., Dishion, Patterson, Stoolmiller, & Skinner, 1991), then it may be that the low monitoring, and not peer socialization, increases the probability of future deviant behavior. Thus, part of what appears to be peer influence may be change within peers who have been on specific pathways themselves, and influence processes would only add to the enactment of this potential.
Finally, influences outside the context of the peer group, such as socialization forces from teachers or parents, also can cause individuals within a group to become more similar and groups to become more different from one another. With regard to school contexts, for example, teachers have been found to interact differently with students who enter a classroom motivationally "rich" than with students who enter the classroom motivationally "poor" (Skinner & Belmont, 1993), and this has effects on how students change across time. Differential teacher treatment may extend to entire groups of students who are motivationally similar (Brophy, 1985). Teachers may treat students alike whom they perceive to be similar, and the students may change in a similar way even when they were not influencing each other at all.

MECHANISMS OF PEER INFLUENCE
When selection processes, external influences, and socialization processes occur together, it is difficult to disentangle their relative contributions to individuals' change. Although there are theoretical expectations and experimental indications that peer socialization processes exist, correlational findings of similarity in outcomes are open to rival interpretations. In this paper, we present an additional route to studying peer influences, namely, direct examinations of the mechanisms of peer influence that can be expected to operate in natural peer contexts.
Most correlational studies on peer influences do not focus on specific mechanisms. Rather, findings that antecedent peer characteristics are related to individuals' change across time are taken as signs that one or many different peer group mechanisms have produced this change. Noted exceptions are studies in which groups of (natural) friends were instructed to discuss controversial topics or to work together on problem-solving tasks (Berndt et al., 1990; Dishion et al., 1996). These studies have shown that specific mechanisms of influence can be simulated in natural peer contexts. Hence, they suggest that direct peer influences can lead to individual change, and cast doubts on arguments that selection and/or external factors would be the primary processes that are responsible for group homogeneity in naturalistic studies. In essence, a focus on natural mechanisms of peer influences requires the adoption of a research strategy of "convergent operations" (Baer, 1973; Baltes, 1996) for the study of peer group influences.
Two kinds of influence mechanisms have been suggested. Studies in controlled conditions have focused on social cognitive mechanisms, suggesting that change in groups comes about in discussions, through joint work on tasks, via modeling, or internalization processes (e.g., Berndt & Keefe, 1996; Schunk & Zimmerman, 1996; Tudge, 1992). Studies based on a social learning model suggest an alternative mechanism, in which group members' social contingencies in interactions represent learning conditions that organize how individuals change (Dishion et al., 1996; Gottman, Gonso, & Rasmussen, 1975; Masters & Furman, 1981).
In the current study, we intend to extend the learning theoretical approach. The focus is on mechanisms by which behaviors of an individual can be shaped by members of that individual's group of peers. There are strong theoretical expectations and ample empirical evidence that such mechanisms can induce change in controlled environments (e.g., Hartup, 1983), and the goal of the current study is to determine whether such mechanisms also occur among natural affiliates. If it can be shown that such mechanisms of influence exist in children's natural interactions, this would support expectations that they also have operated in the existing longitudinal studies.
In specific, the present study examines social contingencies as one such mechanism in school settings. The focus is on children's developing school motivation or engagement in the classroom (Wellborn, 1991). In school settings, student engagement is highly valued by teachers and parents, children's peer groups have been found to be organized around this variable, and experimental and correlational studies indicate that peer influences on academic behaviors are possible (e.g., Berndt & Keefe, 1996; Kandel, 1978; Kindermann, 1993). A study that shows a specific mechanism of peer influence exists in natural classroom interactions provides information about whether peer influences, by themselves, can support or undermine engagement, independently of contributions of selection processes or external factors.
Rather than examining influences based on analyses of peer group characteristics and (subsequent) individual change, the targets were learning mechanisms themselves by which peer influences can be expected to occur. Because social interactions among classmates are presumed to play a key role in these processes, the goal was to observe, in the natural environment, the socialization mechanisms in everyday classroom interactions that could be relevant for children's developing school motivation. To show that these mechanisms actually produce individual change is beyond the scope of the current study. However, because correlational and experimental evidence suggests that peer influences can occur with regard to school motivation and classroom behavior, we have strong expectations that mechanisms of transmission can be found. Hence, a study of these mechanisms may be small in scope. If these mechanisms exist, they should be identifiable in almost any classroom.

CONTEXT IDENTIFICATION
Who among a child's classmates should be most important in providing learning conditions that shape his or her school motivation? Evidence points to closest friends (Berndt & Keefe, 1996), to groups of friends (Ladd, Kochenderfer, & Coleman, 1997; Wentzel & Caldwell, 1997), as well as to broader networks of peers (Kindermann, 1993). All of these contexts have been shown to be influential for school motivation. However, for a given child, these contexts seem to consist of (partly) different partners (Cairns, Leung, Buchanan, & Cairns, 1995; Kindermann, 1996). Children's friends are classmates with whom they share intimate and long-lasting emotional bonds. Peer networks are conceived of as those classmates with whom a given child is publicly known to affiliate, spend time, and share activities (Cairns, Perrin, & Cairns, 1985).
Because there seem to be differences in the specific partners who make up a child's friends or peer group network, one also expects differences in the kinds of influences that these contexts may have. Children's friends may be more influential for their identity and socioemotional development (Newcomb & Bagwell, 1995), whereas public peer network affiliations may be more influential for their public behavior (Kindermann, 1996). With regard to development in school, children's peer networks may be more influential for the extent of engaged behavior that classmates (or the teacher) can observe in the classroom, but children's friends, because of the more intimate nature of the relationship, may be more influential for private perceptions and beliefs. Because our main interest in this study is on peer contingencies for academically-related classroom behavior, we focused on children's peer group networks.

SOCIAL NETWORKS AND SOCIAL LEARNING MECHANISMS
From a learning theoretical perspective, one would expect that the natural contingencies children experience in their everyday classroom interactions can be considered to constitute learning conditions that are influential for their further development. Thus, the target variables of interest were the positive and negative contingencies that children experienced as consequences of their own behaviors in the classroom.
We had several expectations with regard to the structure of children's peer groups and their everyday interactions. First, because previous studies have shown that academic variables and motivation can be criteria around which peer groups are organized (Cairns et al., 1989; Kindermann, 1993), we also expected that peer networks are homogeneous in their motivational composition, so that a child's engagement would be more similar to the engagement level of the members of that child's peer network than to nonpeer network members.
Second, it was expected that in everyday interactions children experience different contingency patterns from the members of their peer groups than from nonmembers or the teacher. We assumed that the members of a child's peer group are this child's strongest supporters in the classroom. However, we also attended to teacher contingencies as a contrast, because research on student's motivation has been focused traditionally on teachers (e.g., Ames & Ames, 1985), and because teacher influences can be alternative explanations for peer-group homogeneity. Hence, a child's peer group members (as well as the teacher) were expected to show contingent support for his or her engaged (ontask) behavior, and peer group members (but not the teacher) were expected to show support for his or her off-task behavior.
Third, we expected that the positive and negative contingencies that children experience from their social partners are related to their own level of engagement in the classroom. Highly motivated students would experience more positive contingencies from members of their peer groups as consequences of their engaged classroom behaviors, whereas students with lower motivation would receive more positive peer-group contingencies following their disruptive (off-task) behaviors. In order to show that peer contingencies can operate as independent mechanisms of influence, we hoped to find that they made unique contributions in predicting children's engagement, over and above the contributions of those contingencies which children experienced in interactions with the teacher.

METHOD

SETTING AND PARTICIPANTS
Observational, questionnaire, and interview data were collected 1 month after the beginning of the school year in a fifth grade classroom of a suburban elementary school. From a total of 25 students, 22 students (all were from middle-class families, 18 were of European descent, 10 were boys) and the male teacher agreed to participate.

MEASURES AND DESIGN
Individual engagement. Student engagement was assessed using teacher reports in a 24- item questionnaire on his perceptions of each participating student's behavioral engagement and orientation to school (e.g., "In my class this student just acts like he/she is working," Wellborn, 1991). Ratings are on a 4-point scale; note that the original measure also contains a scale on emotional engagement that was not used in this study. Engagement ratings were shown previously to have high internal consistency (alpha = .95; Wellborn, 1991) and high stability across the school year (r = .72, p < .001, n = 144; Skinner & Belmont, 1993).
Peer group identification. Cairns and colleagues' (1985) Composite Social Map Interview was used. Informants were asked to nominate, from free recall, whom they knew to "hang out together" in the classroom. They were encouraged to nominate as many groups as they wanted (including dyads) and were free to nominate classmates as belonging to more than one group. Depending on students' responses to the question, further probes were used in addition. For example, if participants named only girls, they were asked about any groups of boys; if they did not include themselves, they were asked whether they had a group of their own. At the end, students were asked about people who did not belong to a group or preferred to be alone.
Observational design. The observational strategies were adapted from Baltes (1996). Observations commenced after observers were trained in the observation system for about 4 months up to a high level of reliability (consistent 90% agreement), after students and observers had become acquainted with one another for several days, and after observers knew the students' names reliably. Observations were balanced across a wide range of lesson structures from lecture-type lessons to situations in which students worked on problems together in self-selected groups. Across 10 days, 5 trained observers (uninformed about the hypotheses and peer networks) observed focal students for 3 min each, in random sequences according to pre-arranged lists. During each day of observation, two observers were present so that each student was observed twice per day. At the beginning and end of the observation days, the two observers met in order to code one (randomly selected) student simultaneously.
Observations were coded using audio-cassette recorders and headset microphones. A pilot study had shown that students could not easily detect when and what was coded, that natural routines were not disturbed, and that many different social partners' responses could be coded in a fast pace. Beginning with the focal student, observers coded his or her own behaviors and any responses by any social partner as they occurred in the natural sequence of events (focal students and partner(s) were identified by name). Three changes were made by observers to the natural stream of events. A focal student's behavior was recoded if it lasted longer than 10 s without change or a response from a partner, and when many partners responded simultaneously (thus, the different partners' behaviors were registered as if they each had followed the focal student's behavior). Finally social partners' behaviors were included only if they occurred in direct response to a focal student (e.g., teacher lecturing was not coded when the focal student was listening).
Observation system. Observation categories developed by Charlesworth and Hartup (1967), Horn, Conners, and Wells (1986), and Kerr, Zignmond, Schaeffer, and Brown (1986) were adapted for the current study. The coding system consists of 12 mutually exclusive and exhaustive categories, 4 for focal children, 6 for social partner(s), and two could be used for either interactant. Four categories are of interest for the current study. With regard to focal children, we were interested in active on-task and off-task behaviors because these are behaviors most likely involved in interactive exchanges. Active on-task behavior was defined as making a class contribution. Examples include initiating (or participating in) class-related activities by asking questions, commenting on class-related topics, working on the blackboard, raising a hand, smiling or laughing in response to on-task discussions, or showing one's own on-task work to another person. Active off-task behavior was defined as a disruption of others' on-task activity. Examples include interfering with others' on-task work, making remarks unrelated to the class topic (e.g., jokes), and laughing in response to others' off-task behavior.
With regard to social partners, we were most interested in their responses indicating social approval and disapproval. Although these behaviors are not the most frequent responses of children's classmates (because students are supposed to work on class materials and are usually not coaches for each other), they are the behaviors that can indicate social learning mechanisms most clearly. Partner approval was defined as a display of direct approval of the focal student's behavior (usually accompanied by a positive emotion). Examples are praising ("That's great"), laughing, or smiling. Partner disapproval was defined as a display of direct disapproval of a focal student's behavior (usually accompanied by a negative emotion). Examples include ridiculing, critiquing, and abruptly changing a topic of discussion.
The other behavior categories not of interest for the current study were children's passive on-task and off-task behaviors (silent work and nondisruptive off-task behaviors), as well as partners' cooperation, factual disagreement, ignoring, and prompting. Two additional codes, leaving and "other," were used for focal children as well as their interaction partners.
Reliability of observations. Aggregated across all days and observers, interobserver reliability was sufficient (kappa = .71; agreement percentages of the categories of interest ranged from 73% for off-task-active behavior to 50% for disapproval; note that all errors in the coding of disapproval were observer omissions). Reliability indices were obtained for 14 sessions (on 2 days, one observer was ill and not present in the classroom; on 2 other days, class periods ended early so that agreement was only checked at the beginning of observations). Reliability ranged from one kappa score of zero (in a session in which observers agreed perfectly but coded only one behavior category), to two instances of perfect agreement (1.0). There were no indications for changes in reliability over time or for systematic differences across observers.

RESULTS
The results are presented in three parts. First, children's social networks are identified and examined with regard to their members' motivational composition. Second, the results are outlined on the relative observation frequencies of the four key observational categories (children's active on- and off-task behavior; approval and disapproval from peer group members and nonmembers, as well as from the teacher), in order to show that these categories captured motivationally relevant behavior. Last, and at the core of the study, contingency patterns are examined that were observed in children's classroom interactions. Analyses focus on social partners' contingencies for children's on- and off-task behavior.

SOCIAL NETWORKS
The classroom's peer network structure was determined by aggregating the informants' group nominations into a co-occurrence matrix that contained the frequencies with which each child was nominated to be in the same group as any other child (see Table 1 for the matrix of girls in this classroom; a computer program Networks can be made available upon request). Binomial z tests were used to determine, for any given child, the probability with which he or she was found to be connected to other children.(FN1) Based on these tests, a composite cognitive social map was formed using a 1% significance level for the interconnections (see Figure 1; note that three European American boys are included who did not directly participate in the study). Across reporters, group nominations were consistent with the composite map (kappa = .73); there were no gender differences in reliability. On average, a student had 3.6 other students in his or her peer network, and network size ranged from dyads to one network of eight students. There was no overlap between boys' and girls' peer networks.

MOTIVATIONAL COMPOSITION OF PEER NETWORKS
On average, children were quite engaged (3.0); individual scores ranged from 2.06 to 3.84 on the 4-point scale. In order to form peer context scores, the engagement scores of those other children who were connected with each child were averaged (see Figure 1). For example, FAY's peer network score was the average of AMY's, CAM's, DEE's, EVE's, and GIN's engagement scores. Note that scores of the three children for whom we did not have permission to participate were estimated as the averages of the participating other children of the same gender; this made it possible to include the three boys who had peer group averages but missing individual values.
Overall, students tended to be somewhat similar in their engagement to the members of their peer networks, but they were different from their other classmates. There was a low correlation between students' own engagement and the engagement profile of their peer group members, r = .28, n = 25, p < .10, and a negative correlation between individuals' own engagement and the averages of their nonpeer group members, r = -.56, n = 25, p < .01.

OBSERVATION FREQUENCIES
In total, 12,043 behaviors were coded for focal students and social partners across the 487 (3 min) observation sessions. With regard to children's own behaviors, we examined active on- and off-task behaviors. Children were considerably more often observed to be active on-task (43%) than off-task (9%), t(24) = 14.38, p < .001. With regard to social partners, approval (10%) was more frequent than disapproval (2%). A repeated measures analysis of variance on the partners' (relative) behavior frequencies with the factors partner (members, nonmembers of one's peer group, teacher), and type of response (approval vs. disapproval) showed an interaction of both factors, F(2, 23) = 13.19, p < .001. As expected, different social partners showed different behaviors in interactions with focal students.
Follow-up analyses showed that the differences were due mainly to the teacher. In contrast to our expectations, children's peer group members did not show more approval than did nongroup members, but both partners showed more approval than did the teacher, t(24) = 3.75, p < .001 and t(24) = 6.03, p < .001, respectively. Peer group members, however, showed less frequent disapproving responses than did nonmembers, t(24) = 2.20, p < .05, and nonmembers showed even more disapproval than did the teacher, t(24) = 2.99, p < .01.
To examine whether children's profiles of their own behavior and of the social responses they received in interactions were related to their engagement in the classroom, a multiple regression was conducted (controlling for children's network size and gender), including the frequencies of individuals' on- and off-task behaviors and the frequencies of approving and disapproving partner behaviors from peer group members, nonmembers, and the teacher. Absolute frequencies were used because of linear dependencies in relative frequencies. Only two behaviors showed significant relations. The more engaged children were, the less frequently they showed off-task behavior, beta = -1.33, t(24) = -2.85, p < .05, and the more likely they were to experience approval from peer group members, beta = .81, t(24) = 3.08. p < .01.

SOCIAL CONTINGENCIES AND CHILDREN'S CLASSROOM ENGAGEMENT
Sequential analyses were used to examine contingency patterns in children's interactions with peer group members, nongroup members, and the teacher. The analyses determined whether specific partner responses were more likely to occur than could be expected by chance. Deviations from chance are denoted by adjusted residual scores that are larger than 1.96 (equivalent to Allison & Liker's adjusted z scores; Bakeman & Quera, 1995). "Lumped" analyses of lag one ("offset" analyses with lags from the end of chains of identical events) were used because classroom routines often involved long sequences of uninterrupted student behaviors. Structural zeroes were included for those behavior codes which could not follow each other (so that expected frequencies for a partner behavior to follow another partner behavior were set to zero).(FN2)
In a first step, a repeated measures analysis of variance was conducted to examine contingency differences across social partners using the adjusted residual contingency scores from the lag analyses, with the factors partner (3), on- versus off-task antecedent, and approval versus disapproval contingency. There was an interaction of all three factors, F (2, 23) = 8.29, p < .01. As expected, social partners differed in their approval and disapproval contingencies following children's on- and off-task behaviors. However, this was due mostly to the teacher and to consequences of children's off-task behavior. Following students' off-task behavior, the teacher showed less approval contingencies than did both group members, t (24) = 5.86, p < .001, and nonmembers, t (24) = 5.86, p < .001, but also less disapproval contingencies than did nongroup members, t(24) = 2.28, p < .05. There were no differences in on-task contingencies.(FN3)
The second step of the analyses was to determine whether the social contingencies that children experienced as consequences of their on-and off-task behaviors in interactions with peer group members, non-members, and the teacher (adjusted residuals for approval and disapproval) were related to their own level of engagement. A multiple regression was used to identify unique contributions of specific contingencies, controlling for children's gender and network size. For illustration, graphs also are presented of (separate) pooled sequential analyses on categories of high versus low engaged students (median-split).
Two peer contingencies were significantly related to students' engagement in the classroom; there were no relations with teacher contingencies. Following their active on-task behaviors, highly motivated students were more likely to receive approval from members of their peer groups, beta = .63, t (24) = 2.24, p < .05. As Figure 2 shows, only highly motivated students received contingent approval from group members at all. Group member approval was almost inhibited as a consequence for low-motivated students who had only the teacher to rely on.
Following their active off-task behaviors, lower-motivated students were more likely to experience disapproval from classmates who were not members of their peer networks, beta = -.88, t (24) = -2.54, p < .05. As can be seen in Figure 3, disapproval was a contingent response of nonpeer group members for both high- as well as low-engaged students. Less engaged students, however, experienced this response with relatively higher levels of contingency.
Contrary to our expectations, approval contingencies from peer group members that children experienced as consequences of their off-task behavior were not related to their own engagement. As Figure 4 shows, approval contingencies were highly significant from peer group as well as nongroup members, approval contingencies from members were not different from contingencies from nonmembers, and there were no differences between high-and low-engaged children. It can be noted, though, that for low-motivated students, peer group approval was only a significant consequence of off-task behavior; the probability of such approval (.16) was eleven times higher than the probability of group member approval for on-task behavior (.014). In contrast, highly engaged students could expect peer group approval for on-task as well as off-task behavior. Although we had no expectations with regard to disapproval responses following children's on-task behaviors, Figure 5 shows these results for reasons of completeness. There were no significant disapproval contingencies following on-task behavior.

DISCUSSION
The goal of this study was to identify natural contingency patterns in children's everyday classroom interactions as mechanisms by which social influences could be exerted by children's peer networks on the individual members. The results are consistent with notions that a child's peer group members can be influential socialization agents for his or her developing school motivation. Peer group members and nongroup members appeared to provide different learning conditions in everyday classroom interactions (cf., Hartup, 1983, 1993). Students who were highly motivated for classroom activities were likely members of peer groups that, on average, were also slightly more motivated, and these students were likely to experience approval contingencies for their on-task efforts from the members of their groups. Conversely, students who were less motivated were found to be with slightly less motivated peer groups, and the only source of approval for their on-task behavior was the teacher.
The results were less clear with regard to learning conditions for off-task behaviors. In contrast to our expectations, children's peer group members were not more supportive of off-task behavior than were nonpeer group members, and there was no relation between these contingencies and students' own motivation. Apparently, all students in the classroom, regardless of their peer group affiliations, enjoyed their classmates' off-task behaviors (to some extent), and approved of these behaviors when they occured. The only differences were found for nongroup members' disapproval contingencies following off-task behavior. The strength of these contingencies was related to children's own level of engagement, indicating that children who were less motivated (and were more frequently off-task) were more likely to receive contingent disapproval from classmates who were not members of their groups.
We regard these findings as consistent with assumptions of a learning theoretical peer influence mechanism, and expect that they indicate rein-forcement patterns that can lead to behavior change. Across time, existing correlational data indicate that children who are found with highly motivated peer groups tend to increase in their motivation, whereas children who are with peer groups who are less motivated remain at this low level or even decrease (e.g., Kindermann, 1993). The current findings suggest that peer group member approval for on-task behavior and nonmember disapproval for off-task behavior can be a mechanism that has the potential to facilitate positive change in highly motivated students, and behavior maintenance in lower motivated students. However, because the data are not longitudinal, there is no information about whether this mechanism actually produces this pattern of change.

STRENGTHS AND LIMITATIONS OF THE STUDY
Although the study provides some support for expectations that the members and nonmembers of a child's peer network can provide different kinds of learning conditions for that child's behavior in the classroom, the results need to be regarded with some caution. A first limitation of the study appears to be its generalizability. Peer group socialization mechanisms were examined in a single classroom. Although the study has high generalizability across a variety of situations that normally occur in everyday classroom interactions (including structured lecture-type lessons, class discussions, and work-group settings), replications are needed across different classrooms and teachers. Nevertheless, the literature strongly supports expectations that influence mechanisms exist in classroom settings, and evidence of such a mechanism was found in a small sample, with relatively low power of the analyses.
A second question concerns the correlational and descriptive nature of the study. What was shown (at one point of measurement) is that a peer socialization mechanism can exist, over and above teacher mechanisms. A strength of the study's approach is that this interpretation is not jeopardized by peer selection processes. Selection for peer group members may be based on engagement, as we have assumed, or it may be based on the specific contingencies that peers provide; the contingencies would nevertheless influence focal children in the same way. However, not examined was whether this mechanism actually leads to specific outcomes. Longitudinal research is necessary to show that the mechanism, in the natural environment, can produce individual changes that are also consistent with the existing correlational findings. A key question will be whether the contingency patterns identified should be regarded as mechanisms for behavior maintenance, or whether they would actually represent learning mechanisms that can lead to intraindividual change.
A third question concerns the extent to which the structure of peer networks in this classroom is typical for fifth graders' affiliation patterns. We found surprisingly low group homogeneity in terms of engagement, which could have diluted contingency patterns. Although there was a high negative correlation between children's own motivation and the average scores of their nonpeer group members, this low homogeneity in academic characteristics is in contrast to earlier studies with peer groups (Cairns et al., 1995; Kindermann, 1993) as well as friendship groups (Urberg, Degirmencioglu, Tolson, & Halliday-Scher, 1995).
One reason for this could be timing. In the current study, children's social networks may not have been as well established as in previous studies. Although the early time window (the second month of the school year) was the same as in some of the earlier studies, the overall consistency between individual students' reports about peer groups and the composite map of peer group structures was .73, whereas consistencies had been above .80 in prior studies.
However, it may also be that the teacher had a critical role in this classroom. Along with nationwide trends to organize classroom activities as more group oriented, this teacher paid more attention to organizing work groups than did teachers in our earlier studies. Although the observed groups were freely chosen by children themselves, the teacher's support for group work may have affected the networks' homogeneity. Groups that work on class projects may have specific goals, and peer selection processes may take different routes than when children associate for other reasons. Still, we regard it as an encouraging sign that mechanisms of peer influence were shown in peer networks that were less homogeneous than the ones found in earlier studies.
A fourth limitation concerns the sampling strategy. The observed classroom can be considered quite typical for fifth graders' everyday interactions in school, but there was much variability in contingency patterns, even when differences in students' motivational levels were taken into consideration. Although this is common in observational studies, there were children in the current study who received only a small amount of approval or disapproval from any of their partners. These students were not isolated, but rather seemed to prefer to work by themselves (which the teacher supported), and showed little active on-task behavior. We did not omit these children from analyses, but instead tolerated their low interaction frequencies. Although more than 10,000 observations were collected, more extensive observations seem necessary to capture the contingency patterns of children who tend to work independently.
A final limitation concerns the observation system and the observational design. We aimed to develop a rather "clean" category of off-task behavior, taking the risk that this category would make up only a small proportion of observed behaviors. Thus, only behaviors were included that represented a clear disruption of ongoing classroom activities (in a pilot study, we encountered reliability problems with broader category definitions). However, the category still included students' jokes and funny remarks, and these may have elicited overall positive responses from all classmates. Other kinds of off-task behavior may not have received such uniform approval. For example, students who are with disaffected peer groups might not receive approval from nonmembers for their (rare) outright "obnoxious" off-task behavior.
A similar point can be made with regard to social partners' disapproval. There were reliability problems with this category (although errors were errors of omission only), and it made up only a small portion of observations. In further studies, a refined observational system, different sampling techniques, or simulations designed to increase the rates of off-task and disapproval behaviors may need to be used.

THE CLASSROOM ECOLOGY AND SOCIAL CONTINGENCY PATTERNS
Although questions remain with regard to the generalizability of the study, the results suggest that peer group influence mechanisms can exist in the form of social contingencies. Specifically, the results indicate that children's peer groups can be supportive contexts for on-task behavior, that both members as well as nonmembers of peer networks can be supportive contexts for off-task behavior, and that nongroup members can be social partners who set limits to the proliferation of off-task behaviors. Overall, the role of peer groups appears more positive than negative. Although this may be partly due to our observation system and design, there are also indications in the literature that classroom peer influences often can have a more positive nature. Some reports point out that peer affiliates generally encourage positive classroom behavior and represent a support system for adjustment to school (Berndt & Keefe, 1996; Brown, Clasen, & Eicher, 1986; Ladd, 1990). Others (e.g., Juvonen, 1996) argue that students themselves know what is important in the classroom and use "self-representational tactics" that have a high likelihood of eliciting approval from teachers and peers. Perhaps, it is safe to assume that in middle-class elementary school settings (and if the focus is not just on deviant populations), peer groups are more a source of support than a hindrance for classroom engagement.
It was nevertheless against our expectations that peer group approval contingencies for off-task behavior were not found to be part of the learning theoretical mechanism. This contingency was not uniquely related to students' motivation and there were no differences between peer group members and nonmembers. This is in contrast to (at least) one of the existing observational studies on peer influences. Dishion and colleagues (1996) showed that the rates of reinforcement that adolescent boys' best friends provided as consequences of socially adjusted and nonadjusted behaviors in laboratory interactions were related to increases in self-reported delinquent behavior and even to subsequent arrest patterns within a 2-year time frame. Based on this finding, peer group approval contingencies following students off-task behavior also could be part of the mechanism of peer influence in the current study.
Because of the current study's inherent limitations, it is possible that this aspect of the peer influence mechanism was missed. In some respects, our data suggest that for low-engaged students (who likely have disaffected peer group members), peer group contingencies also could support changes in negative directions. For these students, approval rates for off-task behavior were higher than those for on-task behavior and off-task behavior was more likely to elicit peer approval than disapproval. In fact, off-task behavior was their only route to secure peer group approval. However, approval from group members and nonmembers occurred at about the same rate, and the rate was similar for low and highly engaged students. In addition, both low and highly motivated students experienced peer group approval more often than disapproval following off-task behavior. Thus, approval contingencies for off-task behavior should represent learning mechanisms in interactions with overall classmates, not just with peer group members. Hence, the conservative conclusion seems to be that these contingencies, which occur naturally in a fifth grade classroom, do not seem to be part of the peer group influence mechanism on school motivation.
Teacher contingencies. Although the focus of this study was on peer group interactions, the findings cannot be viewed as entirely independent from student-teacher interactions that take place at the same time. Because children's classroom interactions are highly organized by teachers' instructional efforts and classroom management techniques, it was important to include student-teacher interactions in the observations. Most important to the present study was to determine whether teacher contingencies are a viable alternative explanation to peer contingencies in shaping the engagement of differentially motivated students.
In many respects, the teacher provided learning conditions that differed from those of peer group members and nonmembers. Frequencies of teacher approval and disapproval were low (which casts some doubts on their robustness), but the low levels were in marked contrast to the more frequent peer contingencies. Following highly motivated children's off-task behavior, the teacher showed no disapproval at all, and lower approval contingencies than did either members or nonmembers of children's peer groups. It is unlikely that these differences were just an outcome of the observation schedule or the amount of attention that the teacher could give to a focal student. Although there was only one teacher who could respond to an average child (but more than three peer group members), the teacher's contingencies following highly motivated children's on-task behaviors were not different from those of their group members.
Hence, it appears that, in this classroom, the teacher played a different role than did children's peer group members or nonmembers. The teacher was especially supportive in interactions with low-engaged students (in terms of behavior frequencies as well as in terms of contingencies). For low-motivated students, the teacher's approval contingencies were their sole source of support for on-task behaviors. We also re-examined the data and did not find that the teacher treated children differentially depending on the peer groups to which they belonged. There were no relations between the teacher contingencies that children experienced themselves and the (average) teacher contingencies that their group members experienced (approval following on-task behavior: r = .04, disapproval following off-task behavior: r = -.12). In other words, in the observations of this classroom, it is unlikely that the teacher's contingency patterns denote behavioral manifestations of biases toward groups of students (cf., Brophy, 1985).
However, absence of teacher biases and of unique teacher contributions to students' motivation (at one point in time) does not imply that teacher influences on students' motivation would be negligible. The study focused on teacher contingencies only to an extent that provided a contrast to peer contingencies. What the results indicate is that in this particular classroom, as soon as peer contingencies are included in the observations, two peer contingencies can play a unique role for students' engagement, over and above teacher contingencies. Thus, with regard to a mechanism that can explain the existing correlational indications of peer influence effects, the alternative of a teacher mechanism can be ruled out.
Gender versus peer group influence mechanisms. For the age range examined in this study, gender segregation of peer affiliations is an often replicated finding (e.g., Hartup, 1983). However, because the groups are gender segregated (see Figure 1), it is possible that contingency patterns indicating a peer group influence mechanism could be just a gender-specific pattern. For example, it may be just other girls (who were girls' only peer group members) who approved of girl's on-task behavior, and it also may be girls (namely, boys' nonpeer group members) who showed disapproval following boys' off-task behavior.
It was possible to rule out this interpretation. Regression analyses included children's gender as a control. Post hoc comparisons, based on sequential analyses that included the sex of interaction partners, showed that boys and girls experienced similar contingency patterns in interactions with peer partners of the same sex as they did in interactions with partners of the opposite sex. Hence, peer influence mechanisms appeared to be the same for boys and girls.

CHILDREN'S PEER RELATIONSHIPS AND MECHANISMS OF GROUP INFLUENCE
Beyond the specific results of this study, one of the main implications is that the direct examination of peer influence mechanisms may open up new avenues for understanding the role of peer relationships in children's development. the study focused on social learning conditions as one specific mechanism through which peer influences can occur. However, peer group contingency patterns are not the only pathway by which peer contexts can influence individuals.
First, alternative mechanisms also can be considered. Alternatives include social-cognitive and modeling mechanisms (e.g., Berndt & Keefe, 1996; Schunk & Zimmerman, 1996), as well as identification and norm internalization pathways (Ennett & Bauman, 1994; Harris, 1995). All of these are viable candidates for mechanisms to explain how influences from peer groups on individuals can come about. Second, because the current study is focused on classroom engagement (a behavior that can be observed by the teacher as well as classmates), the peer contexts of interest were children's publicly known affiliations with classmates. Other studies (e.g., Berndt & Keefe, 1996; Dishion et al., 1996) have shown that mechanisms of influence also exist among friends. Children's peer group networks seem to consist of friends and nonfriend agemates (Kindermann, 1996).
Third, the current study is focused on a mechanism of peer influence that is observable in the classroom. Although children spend a considerable amount of their lives in school, they usually also have friends and peer affiliates outside, and these are also likely to be influential for children's adjustment to school and their engagement in the classroom. In sum, there are other mechanisms through which influences can occur from peers to individuals, and there are other peer agents from whom those influences can emanate.
On the one hand, different influence mechanisms may not be just alternatives; they may combine with social learning mechanisms (at least at some ages, for some target variables, and under some conditions), and thus have the potential of working synergistically. For example, peer group contingencies in school settings can work in the same direction as processes of identification with popular or academically excellent students. Both mechanisms together may support adjustment and school motivation more than either one by itself. On the other hand, the different kinds of influences could also work antagonistically. In adolescence, for example, peer contingencies in disaffected peer groups may provide learning conditions that foster continued (or increased) disaffection in focal students, even though simultaneous identification with academically successful students or friendships with well-adjusted partners outside of school may be able to counterbalance these influences.
To determine which mechanisms of influence exist in different peer contexts and how they combine or interact with one another can be a promising goal for research on peer influences. Our study's findings, that one such mechanism of transmission is identified that may theoretically produce changes in individuals across time, can be seen as a first step in that direction. The identification of such a mechanism provides support for interpreting correlational findings on individual change in peer systems as evidence of causal influences. However, this also suggests a further step, to examine whether the mechanism, in natural environments, also leads to differential outcomes. Based on the current study, and from a learning theoretical viewpoint, one would expect that children who experience supportive contingencies for on-task behavior from peer group members increase in engagement across time, if their peer groups remain stable. Conversely, children with disaffected groups should not increase, and perhaps even decrease in motivation, unless they manage to join more engaged peer groups.
Necessarily, pursuing such questions will involve longitudinal studies. If patterns of social contingencies can be replicated, showing that a specific mechanism of peer transmission operates in children's natural ecology, and if this mechanism is shown to be related to actual change in individuals across time, there would be a solid base to assume that existing correlational studies do indeed contain substantial estimates of peer group influences.
ADDED MATERIAL
Nicole A. Sage and Thomas A. Kindermann
Portland State University
Nicole A. Sage and Thomas A. Kindermann, Department of Psychology.
The study was supported by an Academic Research Enhancement Award from NICHD to the second author (1R15HD31687-01). We thank Robert B. Cairns for his advice on social network analyses, Ellen A. Skinner for her insightful critique of an earlier version of this paper, Wendy DeCourcey and Rebecca Sanders for help in designing the observation system and conducting interviews along with Janine Carroll, and Todd Colvin for help with sequential analyses. Thanks also to Juleen Meyers for help with data formatting, and to the observers Angela Albrecht, Marla Day, Danica Davis, Kristen Hillier, and Trevor Lockwood. Finally, we thank Mr. Robert Shotola from Greenway Elementary School in Beaverton, Oregon, as well as the children who participated. The study was the first author's masters thesis.
Correspondence may be sent to Nicole A. Sage, Department of Psychology, P.O. Box 751, Portland State University, Portland Oregon, 97207-0751. Electronic mail may be sent via Internet to sagen@ch1.ch.pdx.edu.
Table 1. Co-Occurence Matrix of Girls' Network Nominations in the Classroom

                                                                                         Nomi-
Student   AMY   BEV    DEE    CAM   EVE   INA    HEA   JOY    LYN    GIN    FAY   KIM    nations
AMY       --    15     15      13     8     2      3     1      1      3      7     1       19
BEV       15    --     14      12     7     1      2     1      1      2      5     1       17
DEE       15    14     --      11     7     2      2     1      1      2      6     1       16
CAM       13    12     11      --    11     1      2     1      1      2      6     1       16
EVE        8     7      7      11    --     1      3     1      1      4      8     1       17
INA        2     1      2       1     1    --      7     4      4      1      2     3       10
HEA        3     2      2       2     3     7     --     4      4      8      1     5       16
JOY        1     1      1       1     1     4      4    --     13      2      0     5       14
LYN        1     1      1       1     1     4      4    13     --      2      0     4       13
GIN        3     2      2       2     4     1      8     2      2     --      5     4       12
FAY        7     5      6       6     8     2      1     0      0      5     --     1       12
KIM        1     1      1       1     1     3      5     4      4      4      1    --       10

Note. The matrix shows the number of times each given individual was nominated as being in the same group as any other individual. In the classroom, 21 respondents generated a total of 107 groups.
Figure 1. Social networks in a fifth grade classroom. Depicted are significant interconnections among individuals (p < .01). Note that individuals' positions are arbitrary and based on drawing convenience only.
Figure 2. Social partner's approval contingencies following students' active on-task behaviors.
Figure 3. Social partners' disapproval contingencies following students' active off-task behaviors.
Figure 4. Social partners' approval contingencies following students' active off-task behaviors.
Figure 5. Social partners' disapproval contingencies following students' active on-task behaviors.

FOOTNOTES
1 In cases of low expected frequencies, Fisher's exact test was used as an alternative.
2 In these analyses, we were interested in children's actual experiences in the classroom, and not in whether a child's behavior actually influenced a social partner's response. Thus, we did not want to control for auto-contingencies in partners' behaviors. In other words, if the teacher was disgruntled with the behavior of a student over some time, we did not want to control for this teacher tendency, but regard his negative reactions as the actual contingency experience of that child.
3 The estimation of missing values did not substantially change the results: The person-to-group correlation for engagement increased slightly, analyses of variance on group differences were slightly more significant (because of the increased degrees of freedom), but the regression analyses on behavior frequencies and contingencies were more conservative using the estimated values.

REFERENCES
AMES, C., & AMES, R. (Eds.). (1985). Research on motivation in education: The classroom milieu (Vol. 2). New York: Academic.
ASCH, S. E. (1955). Opinions and social pressure. Scientific American, 193, 31-55.
BAER, D. J. (1973). The control of developmental processes: Why wait? In J. R. Nesselroade & H. W. Reese (Eds.), Life-span developmental psychology: Methodological issues. New York: Academic.
BAKEMAN, R., & QUERA, V. (1995). Analyzing interaction: Sequential analysis with SDIS and GSEQ. New York: Cambridge University Press.
BALTES, M. M. (1996). The many faces of dependency in old age. New York: Cambridge University Press.
BERNDT, T. J., & KEEFE, K. (1996). Friends' influence on school adjustment: A motivational analysis. In J. Juvonen & K. R. Wentzel (Eds.), Social motivation: Understanding children's school adjustment. New York: Cambridge University Press.
BERNDT, T. J., LAYCHAK, A.E., & PARK, K. (1990). Friends' influence on adolescents' academic achievement motivation: An experimental study. Journal of Educational Psychology, 82, 664-670.
BROPHY, J. (1985). Teacher's expectations, motives, and goals for working with problem students. In C. Ames & R. E. Ames (Eds.), Research on motivation in education: The classroom milieu (Vol. 2). Orlando, FL: Academic.
BROWN, B. B., CLASEN, D. R., & EICHER, S. A. (1986). Perceptions of peer pressure, peer conformity dispositions, and self-reported behavior among adolescents. Developmental Psychology, 22, 521-530.
CAIRNS, R. B., CAIRNS, B. D., & NECKERMAN, H. J. (1989). Early school dropout: Configurations and determinants. Child Development, 60, 1437-1452.
CAIRNS, R. B., LEUNG, M., BUCHANAN, L., & CAIRNS, B. D. (1995). Friendships and social networks in childhood and adolescence: Fluidity, reliability, and interrelations. Child Development, 66, 1330-1345.
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.
CHARLESWORTH, R., & HARTUP, W. W. (1967). Positive social reinforcement in the nursery school peer group. Child Development, 38, 993-1002.
DISHION, T. J., PATTERSON, G. R., STOOLMILLER, M., & SKINNER, M. L. (1991). Family, school, and behavioral antecedents to early adolescent involvement with antisocial peers. Developmental Psychology, 27, 172-180.
DISHION, T. J., SPRACKLEN, K. M., ANDREWS, D. W., & PATTERSON, G. R. (1996). Deviancy training in male adolescent friendships. Behavior Therapy, 27, 373-390.
ENNETT, S. T., & BAUMAN, K. E. (1994). The contribution of influence and selection to adolescent peer group homogeneity: The case of adolescent cigarette smoking. Journal of Personality and Social Psychology, 67, 653-663.
GOTTMAN, J. M., GONSO, J., & RASMUSSEN, B. (1975). Social interaction, social competence, and friendship in children. Child Development, 46, 709-718.
HALLINAN, M. T., & WILLIAMS, R. A. (1990). Students' characteristics and the peer-influence process. Sociology of Education, 63, 122-132.
HARRIS, J. (1995). Where is the child's environment? A group socialization theory of development. Psychological Review, 102, 458-489.
HARTUP, W. W. (1983). Peer groups. In P. H. Mussen (Series Ed.) & E. M. Hetherington (Vol. Ed.), Handbook of child psychology: Vol. 4. Socialization, personality, and social development (4th ed.). New York: Wiley.
HARTUP, W. W. (1993). Adolescents and their friends. In B. Laursen (Ed.), Close friendships in adolescence. San Francisco: Jossey-Bass.
HORN, W. F., CONNERS, C. K., & WELLS, K.C. (1986). Use of the Abikoff classroom observation coding system on a children's inpatient psychiatric unit. Journal of Psychopathology and Behavioral Assessment, 8, 9-23.
JUVONEN, J. (1996). Self-presentation tactics promoting teacher and peer approval: The function of excuses and other clever explanations. In J. Juvonen & K. R. Wentzel (Eds.), Social motivation: Understanding children's school adjustment. New York: Cambridge University Press.
KANDEL, D. B. (1978). Homophily, selection, and socialization in adolescent friendships. American Journal of Sociology, 84, 427-436.
KERR, M. M., ZIGNMOND, N., SCHAEFFER, A. L., & BROWN, G. M. (1986). An observational follow-up study of successful and unsuccessful high school students. The High School Journal, 70, 20-24.
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.
LADD, G. W. (1990). Having friends, keeping friends, making friends, and being liked by peers in the classroom: Predictors of children's early school adjustment? Child Development, 61, 1081-1100.
LADD, G. W., KOCHENDERFER, B. J., & COLEMAN, C. C. (1997). Classroom peer acceptance, friendship, and victimization: Distinct relational systems that contribute uniquely to children's school adjustment? Child Development, 68, 1181-1197.
MASTERS, J. C., & FURMAN, W. (1981). Popularity, individual friendship, and specific peer interaction among children. Developmental Psychology, 17, 344-350.
MOUNTS, N. S., & STEINBERG, L. (1995). An ecological analysis of peer influence on adolescent grade point average and drug use. Developmental Psychology, 31, 915-922.
NEWCOMB, A. F., & BAGWELL, C. L. (1995). Children's friendship relations: A meta-analysis. Psychological Bulletin 117, 306-347.
RUBIN, K. H., BUKOWSKI, W., & PARKER, J. G. (1998). Peer interactions, relationships, and groups. In W. Damon (Series Ed.) & R. M. Lerner (Vol. Ed.), Handbook of child psychology: Vol 1. Theoretical models of human development (5th ed.). New York: Wiley.
SCHUNK, D. H., & ZIMMERMAN, B. J. (1996). Modeling and self-efficacy influences on children's development and self-regulation. In J. Juvonen & K. R. Wentzel (Eds.), Social motivation: Understanding children's school adjustment. New York: Cambridge University Press.
SHERIF, M., HARVEY, O. J., WHITE, B. J., HOOD, W. E., & SHERIF, C. W. (1961). Intergroup conflict and cooperation: The Robbers Cave experiment. Norman: University of Oklahoma Book Exchange.
SKINNER, E. A., & BELMONT, J. B. (1993). Motivation in the classroom: Reciprocal effects of teacher behavior and student engagement across the school year. Journal of Educational Psychology, 85, 571-581.
TUDGE, J. R. H. (1992). Processes and consequences of peer collaboration: A Vygotskian analysis. Child Development, 63, 1364-1379.
URBERG, K. A. (1992). Locus of peer influence: Social crowd and best friends. Journal of Youth and Adolescence, 21, 439-450.
URBERG, K. A., DEGIRMENCIOGLU, S. M., TOLSON, J. M., & HALLIDAYSCHER, K. (1995). The structure of adolescent peer networks. Developmental Psychology, 31, 540-547.
WELLBORN, J. G. (1991). Engaged versus disaffected action: Conceptualization and measurement of motivation in the academic domain. Unpublished doctoral dissertation, Graduate School of Human Development and Education, University of Rochester, Rochester, NY.
WENTZEL, K. R., & CALDWELL, K. (1997). Friendships, peer acceptance, and group membership: Relations to academic achievement in middle school. Child Development, 68, 1198-1209.