Esther Mertens

| 99 Classmate Influence in Intervention Deviant and prosocial modeling and reinforcement. Observations of peer interactions took place at the students’ school during school hours and were videotaped by trained research assistants. The research assistant explained the procedure of the observation task and kept track of time outside the observation room. The research assistant was not present in the observation room during the discussions to enable the dyad to talk freely. The observation task was based on the Peer Interaction Task (e.g., Dishion et al., 1996). The interaction consisted of four vignettes which students each discussed for five minutes. The first vignette was planning an activity together as a warm-up. The other three vignettes, systematically counterbalanced, concerned daily school situations involving: Student at work in the class, student with new clothes, and sitting together with classmates. For example, “classmate A is in the classroom working on an assignment in his book. Classmate B is doing nothing. Classmate B is annoying and throws pieces of paper towards classmate A.” Two different versions of all vignettes were used for the baseline and post measurements. Participants were instructed to read the vignettes in turn aloud and discuss the situation together for five minutes. After five minutes, the research assistant re-entered the observation room to end the discussion and provided the next vignette. Students were also given three questions they could use in order to help them discuss the situation for the full five minutes: 1) What do you think of the situation? Could this happen at your school? 2) Imagine you are classmate A. What would you do? 3) How could this end? The Conversation topic code (Piehler & Dishion, 2004b) was used to assess frequencies of deviant and prosocial verbal and nonverbal modeling. Verbal modeling was coded based on verbatim transcription of the discussion. Deviant modeling was all utterances that violated community or societal rules or were not appropriate to the setting or task (e.g., “I would hit him in his face.”). Prosocial modeling was all utterances referring to positive or prosocial values, principles, or actions (e.g., “I wouldn’t bully him.”). Neutral modeling was all utterances that did not fit in the deviant or prosocial categories (e.g., “This situation happens all the time.”). Nonverbal modeling was coded, while watching the videotaped observation, when participants used gestures to support their utterance or only used gestures. Depending on the content of the gesture it was coded as deviant (e.g., making a punch movement, making weird faces) or prosocial (e.g., waving their hand as a greeting). Proportions of (verbal and nonverbal) deviant and prosocial modeling were calculated over all verbal (i.e., prosocial, deviant, and neutral utterances) and nonverbal behaviors during the interaction, representing the proportion of prosocial and deviant modeling relative to all coded modeling. Interrater reliability of the three independent coders was good concerning deviant and prosocial modeling (ICC deviant  = .96, ICC prosocial  = .96) based on 22 observations coded over time. In addition to deviant and prosocial modeling, verbal (e.g., “Indeed”, “True”, “No”) and nonverbal (e.g., laughing, giving thumbs up, shaking head) reactions were coded (Piehler & Dision, 2007; Van de Bongardt et al., 2017). Reactions were coded 5