Esther Mertens

100 | Chapter 5 as reinforcement or as correction. Reactions were coded as reinforcement when the reaction indicated a positive evaluation of the other peer’s behavior (e.g., “Indeed”, laughing). Reactions were coded as deviant reinforcement when deviant behavior of the peer was reinforced and as prosocial reinforcement when prosocial behavior of the peer was reinforced. Corrections were coded when the reaction indicated a negative evaluation of the other peer’s behavior (e.g., “No”, shaking head). Proportions of deviant and prosocial reinforcement were calculated relative to all coded reaction codes (i.e., reinforcements and corrections). Interrater reliability of the three independent coders was good concerning deviant and prosocial reinforcement (ICC deviant  = .82, ICC prosocial  = .87) based on 22 observations coded over time. Dyadic mutuality. Dyadic mutuality was assessed at T1, based on the video- observations, by a combination of coding systems. Each member of the dyad was coded on three items: Responsiveness (i.e., the extent to which the student responded verbally and nonverbally to his or her peer), self-centeredness (i.e., the extent to which the student redirected the conversational flow to focus on personal ideas and experiences), and communicative efficiency (i.e., the appropriateness and competence of the messages send during the discussion; Piehler & Dishion, 2004a; Whalen, Henker, Collins, McAuliffe, & Vaux, 1979). Additionally, each dyad as a whole was coded on three items: Reciprocity (i.e., verbal reciprocity such as engaging in a conversation-like interaction, and behavioral reciprocity such as eye-contact and posture orientation), shared attitudes and values (i.e., similar beliefs and attitudes about the discussed ideas) and affective valence (i.e., the emotional tone of the discussion and nonverbal behavior such as gestures, facial expression, and tone of voice (Piehler & Dishion, 2004a). All items were rated on a 6-point Likert type scale (1 =  rarely or never , 6 = always or throughout ) and coded for the session as a whole. The item self-centeredness was reversed coded, so high values representing low self-centeredness. Subsequently, the 9 items (i.e., two times three individual items and three items of the dyad) were averaged to form a score on dyadic mutuality per dyad. Interrater reliability of the three independent coders was good (ICC = .73) based on 22 observations coded over time. Analyses First, we tested whether changes in students’ modeling and reinforcement mediated the relation between R&W and students’ perceived classroom peer context in M plus 8.2. We modeled multilevel mediation models which allowed us to analyze changes in modeling, reinforcement, and perceived classroom peer context at classroom level. In other words, we examined to what extent changes in modeling and reinforcement in the classroom mediated the relation between condition and the classroom peer context (Preacher, Zyphur, & Zhang, 2010). The mediators deviant and prosocial modeling were analyzed as parallel mediators in one model per outcome measure.