The Contagious Effect of Deviant Behavior in Adolescence

Abstract

This article investigated how the development of deviant behavior in adolescence is influenced by the variability of deviant behavior in the peer group. Based on the social information-processing (SIP) model, we predicted that peer groups with a low variability of deviant behavior (providing normative information that is easy to process) should have a main effect on the development of adolescents’ deviant behavior over time, whereas peer groups in which deviant behavior is more variable (i.e., more difficult to process) should primarily impact the deviant behavior of initially nondeviant classroom members. These hypotheses were largely supported in a multilevel analysis using self-reports of deviant behavior in a sample of 16,891 adolescents in 1,308 classes assessed at two data waves about 1-year apart. The results demonstrate the advantages of studying cross-level interactions to clarify the impact of the peer environment on the development of deviant behavior in adolescence.

Keywords

multilevel analyses peer influences deviant behavior gender adolescence

Engaging in deviant behavior during adolescence is a predictor for a range of problematic outcomes, such as delinquency, risky sexual behavior, or substance abuse, well into adulthood (Brook, Whiteman, Finch, & Cohen, 1996; Lansford, Dodge, Fontaine, Bates, & Pettit, 2014). The term deviant behavior comprises a large number of different behaviors violating social or legal norms, such as skipping school, cheating on a test, or stealing (Snyder, Dishion, & Patterson, 1986). Therefore, understanding the factors that promote deviant behavior is a critical task. One established risk factor for deviant behavior is contact with deviant peers: Adolescents who socialize with deviant peers are more likely to show deviant behavior themselves, which is why the role of contact with deviant peers has attracted much research attention (for a review see Dishion & Tipsord, 2011). Most of this research has centered either on self-selected peer groups (Haynie, 2001) or on groups with a high presence of deviant peers (e.g., groups selected for intervention measures, for a review see Dodge, Dishion, & Lansford, 2006).

In contrast to these lines of research, the present study sought to investigate the long-term consequences of adolescents’ interactions with deviant peers whom they did not choose as friends and from whom they could not withdraw. These conditions are characteristic of classroom communities to which students are assigned without having a say. Experimental studies have shown that the communication patterns of deviant adolescents change when they interact with nondeviant peers (Mathys, Hyde, Shaw, & Born, 2013). Therefore, studies are needed that analyze how deviant peers in social groups that are not self-selected affect the behavior of nondeviant members. Such studies are obviously important, not only from an applied perspective but also from a theoretical perspective, to gain a better understanding of the multiple risk factors that predict both the deviant behavior itself and the affiliation with deviant peers (Patterson, Reid, & Dishion, 1992). Because the possibility of self-selection into deviant peer groups is minimized in classroom communities, studying deviant peer groups in the class context provides a way of disentangling peer-group effects from other risk factors for deviant behavior.

Peer Influences on Deviant Behavior

To explain the development of deviant behavior, influential theories emphasize the role of peers (Crick & Dodge, 1994; Huesmann, 1988), which is backed by ample empirical evidence (Vitaro, Boivin, & Tremblay, 2007). Theoretical considerations as well as empirical results suggest two different patterns by which members of a classroom may be affected by the deviant behavior of their peers. One pattern is by way of a main effect, assuming that all members of a class are affected by social interactions with deviant peers in the same way. The second pattern suggests that members of a class may be affected differentially by the deviant behavior of their classmates, assuming an interactive effect of peer behavior and individual characteristics.

Supporting the first pattern, Farrell, Henry, Mays, and Schoeny (2011) showed that deviant behavior was elevated in peer groups in which the collective normative acceptance of deviant behavior was high. Longitudinally, the overall level of reactive aggression in the classroom was shown to predict higher levels of individual reactive and proactive aggressions 6 months later (Frey, Higheagle Strong, & Onyewuenyi, 2016).

However, other studies demonstrated an interaction between classroom context variables and characteristics of individuals. Busching and Krahé (2015) found that in classes with a low acceptance of aggression, individual differences in aggressive behavior were stable across a time span of 3 years. By contrast, classes with a higher normative acceptance of aggression had a stronger impact on the initially nonaggressive students than on class members who were more aggressive from the start. A similar finding was reported by the study of Yarnell, Pasch, Brown, Perry, and Komro (2014).

A theoretical approach that may integrate these divergent empirical findings is the Social Information Processing (SIP) model (Crick & Dodge, 1994), which explains the influence of the social context by cognitive processes based on learning mechanisms. The processing of social information depends on individual characteristics, such as the individual’s level of deviant behavior. Individuals who show a high level of deviant behavior often have limited social information processing abilities (Hill, 2002). This limitation means that only those peer behaviors that are shown consistently and are therefore easy to discern will influence more deviant individuals in a class. If there is greater variability of deviant behavior in a group, it requires more information-processing efforts and should therefore have a stronger effect on classmates with a comparatively low level of deviant behavior. This theorizing suggests that the social information processing capabilities required to process information about peer behavior may act as a moderator of the degree to which deviant behavior in a classroom promotes deviant behavior in individual classroom members. Consistent with this reasoning, Rohlf, Krahé, and Busching (2016) found that the classroom level of physical aggression, a pattern of highly visible social behavior that is easy to recognize, affected everybody, whereas class-level relational aggression, a social behavior which requires more cognitive processing, had an effect only on the initially less aggressive class members.

Gender Differences in the Influence of Deviant Peers on Adolescents’ Deviant Behavior

A large body of research suggests that males show a higher level of deviant behavior compared to females (Chun & Mobley, 2010; Farrell, Kung, White, & Valois, 2000). However, evidence on the potential moderating role of gender on the relationship between contact with deviant peers and deviant behavior is inconsistent. Studies of groups in which adolescents could choose their interaction partners often found a stronger relationship between contacts with deviant peers and later deviant behavior in males compared to females (e.g., Bowman, Prelow, & Weaver, 2007; Crosnoe, Erickson, & Dornbusch, 2002; Heinze, Toro, & Urberg, 2004; Kellam, Ling, Merisca, Brown, & Ialongo, 1998). However, studies conducted in class communities have reported a stronger relationship between problematic behavior at the classroom level and individual behavior in girls compared to boys, for instance with respect to bullying (Isaacs, Voeten, & Salmivalli, 2013; Salmivalli & Voeten, 2004) and aggressive behavior (Rohlf, Krahé, & Busching, 2016). Given the lack of consistent evidence on gender differences in the susceptibility to deviant peer behavior, we included the gender as a moderator at the individual and the classroom levels in our analyses.

The Current Study

The goal of the current study was to investigate the impact of the classroom level of deviant peer behavior on the development of individual deviant behavior in a longitudinal design. We argue that in addition to the level of deviant behavior or normative beliefs, it is critical to consider the variability of these behaviors or beliefs in a classroom, as they define the salience and clarity of deviant behavior patterns to which individual members of the classroom are exposed. The two-wave longitudinal design of our study enabled us to examine the following predictions.

Main effect of the classroom level of deviant behavior

We expected that students would show higher deviant behavior at Time 2 (T2) if they were in a classroom where the level of collective deviant behavior at Time 1 (T1) was high than if they were in a classroom with a low level of deviant behavior at T1 (Hypothesis 1).

Variability as a class-level moderator

The impact of the level of deviant behavior in a classroom was expected to be stronger if the variability at the classroom level was low, that is, if all peers contributed to a similar degree to the overall classroom level (Hypothesis 2). That is, high levels of deviant behavior should have a greater impact on the individual class members if variability is low (shown in a relatively similar fashion by many students in a class) rather than high (shown only by a few highly deviant students in the class).

Cross-level interactions between class level and individual characteristics

The SIP model predicts that individuals high in deviant behavior are less able to process social cues and are therefore less likely to be influenced by their classmates than individuals with a lower level of deviant behavior. We predicted that being surrounded by deviant peers as reflected in a high classroom level of deviant behavior would have a stronger impact on initially nondeviant than on initially more deviant class members (Hypothesis 3).

Based on the SIP model, we further predicted that in classes where the variability of deviant behavior was comparatively high, because they contained a few highly deviant members who raised the mean level of deviance, classmates would be exposed to a more ambiguous peer context that required cognitive effort for disambiguating. We predicted that adolescents with an initially low level of deviant behavior who were in a classroom with a high but more variable degree of deviant behavior would show a greater increase in deviant behavior over time than adolescents with an initially high level of deviant behavior in the same classroom context (Hypothesis 4).

Gender differences

As noted above, while there is ample evidence that the level of deviant behavior is higher in males than in females, it is as yet unclear whether there is also a gender difference in the susceptibility to peer group influences in the domain of deviant behavior. Therefore, we included a comparison between boys and girls in our analysis, proposing that the relationships outlined in the four hypotheses would hold for both gender groups.

Covariates

Since deviant behavior is caused by numerous factors, it is important to control for other potential risk factors to reduce the possibility that significant effects emerge as a result of third-variable correlations. One such variable is academic achievement, as studies have shown that low-achieving adolescents are at higher risk for deviant behavior (Jung, Krahé, Bondü, Esser, & Wyschkon, 2016). Another potential variable is age. There is an increase in deviant behavior from early adolescence up to the middle adolescence, which is followed again by a decrease (Lacourse, Nagin, Tremblay, Vitaro, & Claes, 2003). Furthermore, migration background may be linked to differences in deviant behavior (Wallner & Stemmler, 2014).

At the school level, there is variation due to the fact that the secondary school system in Germany is tracked, with schools leading to university entrance qualifications being more academically oriented and attracting a different student clientele than schools leading to vocational qualifications. Students in these tracks differ in many respects (e.g., socioeconomic status, cognitive abilities, migration background, Baumert, Stanat, & Watermann, 2006) so that selection effects might operate at the school-track level. Based on these considerations, academic achievement, age, migration background, and school track were selected as covariates for inclusion in the analyses.

Method

Participants

The data for our study were collected by the StEG-Konsortium (2011) to investigate the development of all-day schools in Germany and was made available to us for the purposes of the present analysis by the Research Data Centre (FDZ) at the Institute for Educational Quality Improvement (IQB). The final sample consisted of 1,308 classes from 252 schools distributed across the whole of Germany, with a total of 16,891 students (48% female). The mean number of students per class was 12.9. The mean age of participants at T1 was 14.00 years (SD = 1.2).¹ The interval between the two measurements was approximately 1 year.

Around 29% of the participants indicated that either they or one of their parents had not been born in Germany. These were coded as participants with a migration background. Participants came from all levels of the German multi-tier secondary school system. Based on whether they led to a university entrance qualification or a vocational qualification, schools were categorized as either academically or vocationally oriented.

Measures and Procedure

Deviant behavior

To assess deviant behavior, participants were presented with a list of 10 different items describing deviant behavior (e.g., skip school, take something away from someone, and hit somebody) and were asked how often they had shown each behavior during the last 12 months. Responses were made on a 5-point scale ranging from 1 (never) to 5 (almost every day). This questionnaire was adapted from Tillmann, Holler-Nowitzki, Holtappels, Meier, and Popp (1999). The internal consistency was good for both data waves (α_T1 = .77, α_T2 = .79).

Academic achievement

To assess academic achievement, participants were asked to indicate their grades on their latest report card in four different core subjects (Mathematics, German, Geography, and their first foreign language). In Germany, grades go from 1 (very good) to 6 (insufficient). For ease of interpretation, we reversed the coding so that higher numbers indicate better performance. The reliability of the overall score aggregated across the four subjects was good (α_T1 = .77, α_T2 = .79).

Procedure

After obtaining active parental consent, participants completed the paper-and-pencil questionnaires during class time. The questionnaires were administered by trained research assistants. A complete description of the design, procedure, the amount of missing data, and the reasons for missing data can be found in Furthmüller (2015). Ethics approval was obtained from an independent scientific advisory board overseeing the original data collection and from the school authorities of the participating federal states.

Data preparation and analyses

To test our hypotheses, three different scores were computed for each participant: (1) the overall classroom mean, (2) the individual deviation from the classroom mean, and (3) the coefficient of variation of the classroom-level score, which represents the variability of deviant behavior in the class. The coefficient of variation is the standard deviation divided by the mean and results in a lower correlation between the mean level and the measure of variability compared to using only the standard deviation (Everitt & Skrondal, 2010).

The models were estimated using lme4, version 1.1-12, in combination with parametric bootstrapping, which results in robust confidence intervals (CIs). Because this estimation approach does not use standard errors, exact p values are not available. Additionally, since there are no measures of effect size for multilevel models including cross-level interaction and random components, we additionally report beta weights, which provide an indication of the size of the effects and can be used in a meta-analysis (Peterson & Brown, 2005). To test for multicollinearity, we calculated the variance inflation factor (VIF), which should be lower than 10 (Myers, 1990), and the κ, which should be below 30 (Cohen, 2003). The highest VIF was 3.17 for the four-way interaction between class-level deviant behavior, class-level variability, individual-level deviant behavior, and gender, and the maximum κ was 10.60. Thus, the scores fell well below the critical threshold for collinearity. A complete list of all VIFs and the κ can be found in the Online Supplemental Material.

To interpret the interaction effects, we used an approach outlined by Bauer and Curran (2005): In a first step, we drew Johnson–Neyman (J-N) plots, which were used to identify the different interaction patterns. In a second step, we plotted these patterns using conventional interaction plots.

Results

Descriptive Results and Intraclass Correlations

The mean scores of deviant behavior in the sample as a whole was 1.49 (SD = .73) at T1 and 1.56 (SD = .69) at T2. Given a scale range of 1–5, both the means and the SDs indicate that the overall level of deviant behavior was at the low end of the scale, which is an expected finding in a community sample of students attending mainstream schools. The descriptive statistics and correlations at the individual and the classroom level are presented in Table 1. Across both time points, higher individual deviant behavior was associated with lower academic performance, higher age, and a greater likelihood of being male and having a migration background. Additionally, deviant behavior was significantly correlated across both time points. The high correlation between the combined deviant behavior score and the individual-level score indicates that most of the variance occurs at the individual level. Additionally, it was tested whether deviant behavior differed between the two school tracks. At T1 as well as at T2, the participants in vocationally oriented schools reported higher deviant behavior (T1: academically oriented schools: M = 1.28, vocationally oriented schools, M = 1.55, T2: academically oriented schools: M = 1.44, vocationally oriented schools, M = 1.60) compared to participants in academically oriented schools.

Table 1.
Means and Correlations of Individual- and Classroom-Level Variables.

Mean (SD) 1 2 3 4 5 6 7 9 10 11

Individual Level

1. T1 Deviant behavior raw score (comprising both individual- and class-level information) (1–5) 1.49 (.73)

2. T2 Deviant behavior raw score (comprising both individual- and class-level information) (1–5) 1.56 (.69) .39*

3. T1 Individual-level deviant behavior^a 0.00 (.66) .91* .34*

4. T2 Individual-level deviant behavior^a 0.00 (.65) .34* .93* .36*

5. T1 Age (years) 12.17 (1.28) .20* .12* .05* .02

6. T1 Academic performance (1–6) 3.09 (.76) −.25* −.17* −.16* −.10* −.28*

7. Migration background (% yes) 29% .09* .08* .05* .06* .03* −.13*

8. Gender (% male; female = −1; male = 1) 52% −.18* −.20* −.17* −.19* −.05* .09* .01

Classroom Level

9. T1 Classroom-level deviant behavior^b 1.51 (.32)

10. T2 Classroom-level deviant behavior^b 1.58 (.27) .45*

11. T1 Classroom-level variability of deviant behavior 0.39 (.16) .63* .26*

12. T2 Classroom-level variability of deviant behavior 0.38 (.13) .24* .57* .26*

^aIndividual deviation from the classroom mean.

^bOverall classroom mean.

p < .01. ***p < .001.

Table 2 presents the intraclass correlation coefficients (ICCs), indicating the degree to which variance in individual deviant behavior can be traced back to differences between individuals, between classrooms, and between schools. ICCs are presented for the total samples as well as separately for boys and girls. At T1, 8% of the variance of deviant behavior in the total sample could be traced back to the class level and 6% of the variance could be traced back to the school level. At T2, 6% of the variance was associated with the class level, and 2% of the variance was associated with the school level. Compared to ICCs typically found in school settings, the variance at the individual level is quite high in our sample (Hedges & Hedberg, 2007). However, because all CIs of the ICC at the class level are above .05, considerable variance is still found at the class level. With one exception, the school-level CI did not include zero, indicating that there is variance at the school level. Ignoring this variance would bias the result at the class level; therefore, it is included in all analyses.

Table 2.
Intraclass Correlation Coefficients of Deviant Behavior.

T1 T2

ICC 95% CI ICC 95% CI

Total sample

Individual level .87 [.85, .88] .92 [.91, .93]

Class level .08 [.06, .09] .06 [.05, .07]

School level .06 [.04, .07] .02 [.02, .03]

Boys

Individual level .86 [.84, .88] .91 [.90, .93]

Class level .09 [.07, .10] .08 [.06, .09]

School level .05 [.04, .07] .01 [.00, .02]

Girls

Individual level .85 [.83, .87] .88 [.86, .90]

Class level .09 [.08, .11] .08 [.06, .09]

School level .06 [.04, .08] .04 [.03, .06]

Note. ICC = intraclass correlation coefficients; CI = confidence interval.

Multilevel Analyses

To examine the hypotheses, 3 three-level models were estimated. The first level was the individual level, the second level was the class level, and the third level was the school level. Because differences in deviant behavior at the school level are shaped more by general school rules and sociodemographic composition of the student body than by social interactions between the students, this level is included only as a control variable. In Model 1, the main effects and interactions on the same level were included. In Model 2, cross-level interactions between individual deviant behavior and the classroom-level constructs were added, and in Model 3, the interactions with gender were included. The coefficients for the three models are presented in Table 3.

Table 3.
Multilevel Models Predicting Deviant Behavior at T2.

Model 1 Model 2 Model 3

est(b) 95% CI β est(b) 95% CI β est(b) 95% CI β

Fixed Effects

(Intercept) 1.44* [1.39, 1.49] 0.04 1.44* [1.4, 1.49] 0.04 1.44* [1.39, 1.48] 0.04

Gender^a 0.08* [0.08, 0.09] 0.12 0.08* [0.07, 0.09] 0.12 0.09* [0.08, 0.1] 0.13

Age 0.02* [0.01, 0.03] 0.03 0.02* [0.01, 0.03] 0.03 0.02* [0.01, 0.03] 0.03

Migration background^b −0.03* [−0.04, −0.02] −0.04 −0.03* [−0.04, −0.02] −0.04 −0.03* [−0.04, −0.02] −0.04

Academic performance T1 0.06* [0.04, 0.07] 0.06 0.06* [0.04, 0.07] 0.06 0.06* [0.04, 0.07] 0.06

School track 0.01 [−0.01, 0.03] 0.02 0.01 [−0.01, 0.03] 0.02 0.01 [−0.01, 0.03] 0.02

Individual deviant behavior T1 0.41* [0.38, 0.43] 0.37 0.49* [0.46, 0.52] 0.46 0.48* [0.45, 0.51] 0.45

Class-level deviant behavior T1 0.40* [0.34, 0.47] 0.18 0.41* [0.35, 0.47] 0.19 0.41* [0.35, 0.47] 0.19

Class-level variability T1 −0.11 [−0.22, 0.01] 0.02 −0.14* [−0.25, −0.03] −0.02 −0.15* [−0.27, −0.04] −0.02

Class-level deviant beh. × Class variability −0.56* [−0.87, −0.25] −0.04 −0.62* [−0.93, −0.3] −0.04 −0.61* [−0.93, −0.28] −0.04

Class level × Individual deviant beh. −0.15* [−0.24, −0.06] −0.04 −0.14* [−0.23, −0.05] −0.04

Class variability × Individual deviant beh. −0.93* [−1.12, −0.75] −0.13 −0.96* [−1.16, −0.77] −0.14

Class level × Class variability × Individual deviant beh. 0.29 [−0.26, 0.84] 0.01 0.39 [−0.16, 0.97] −0.02

Individual deviant beh. × Gender 0.02 [0, 0.04] 0.02

Class level × Gender 0.00 [−0.05, 0.04] 0.00

Class variability × Gender 0.04 [−0.05, 0.12] −0.01

Class level × Individual deviant beh. × Gender −0.02 [−0.09, 0.04] 0.02

Class variability × Individual deviant beh. × Gender 0.01 [−0.15, 0.16] 0.00

Class level × Class variability × Gender −0.23* [−0.47, −0.01] 0.02

Class level × Class variability × Individual deviant beh. × Gender −0.04 [−0.47, 0.39] 0.00

Random Parameters

Level 1—Individual level

Intercept 0.57* [0.57, 0.58] 0.84 0.57* [0.57, 0.58] 0.84 0.57* [0.57, 0.58] 0.84

Level 2—Class level

Intercept 0.08* [0.07, 0.09] 0.07 0.06* [0.05, 0.07] 0.05 0.06* [0.05, 0.07] 0.05

Deviant behavior 0.02* [0.02, 0.03] 0.04 0.02* [0.02, 0.03] 0.04 0.02* [0.02, 0.03] 0.04

Level 3—School level

Intercept 0.00* [0, 0.01] 0.01 0.00* [0, 0.01] 0.01 0.00* [0, 0.01] 0.01

^aGender coded: female = −1, male = 1.

^bMigration background coded −1 = migration background, 1 = no migration background.

*p < .05, coefficients in bold are plotted.

The predictors at the individual level were the individual’s deviation from the mean and his or her gender, with age, migration background, and academic performance included as covariates. Predictors at the class level were the mean and the variability of deviant behavior at T1. School track (academic vs. vocational) was included as covariate at the school level.² In each model, the dependent variable was the individual score of deviant behavior at T2. It was impossible to separate the class mean from the individual’s deviation from the class mean on the outcome side, as this would have led to a misestimation of all standard errors. The results for all models without the covariate can be found in the Online Supplemental Materials.

In addition to the random intercepts for each higher level, a random slope for the score of individual deviant behavior was estimated. These random components encompass all classroom-level influences on T2 deviant behavior that are due to variables not included in the model.

Within-Level Effects

The main effects at the individual level are significant in all three models. Participants who scored higher on deviant behavior at T1 also reported more deviant behavior 1 year later at T2. Of the covariates, being male, being older, having a migration background, and doing less well academically were prospective predictors of deviant behavior at T2, controlling for deviant behavior at T1.

At the classroom level, both the main effect of deviant behavior and the interaction with variability in deviant behavior were significant. The significant main effect indicates that individuals showed more deviant behavior at T2 if they were surrounded by classmates with a high level of deviant behavior than if they were surrounded by nondeviant peers. This finding is consistent with Hypothesis 1. However, the main effect was qualified by an interaction with classroom-level variability, as displayed in Figure 1. In support of Hypothesis 2, being in a classroom with a high overall level of deviant behavior was more closely associated with later individual class members’ deviant behavior if the peers were less variable in their deviant behavior. By contrast, no significant effect of class-level variability was found for classrooms with a low level of deviant behavior (b _{variability at 25th percentile of class level deviant behavior} = .07; 95% confidence interval [CI] [−.05, .18], b _{variability at the 75th percentile of class level deviant behavior} = −.28; 95% CI [−.42, −.16]). This interaction effect within the classroom level was independent of the cross-level interaction between the classroom and the individual level discussed later, as it remained significant even after cross-level interactions were introduced. Although the size of this effect was small, the pattern was consistent with the prediction that the impact of deviant peers increases the less variable (i.e., the more homogeneous) they are in terms of their deviant behavior.

Figure 1.
Interaction between T1 classroom-level deviant behavior and variability of deviant behavior predicting individual-level deviant behavior at T2. For clarity of presentation, we plot within-level interactions as bar diagrams and cross-level interactions as line diagrams.

Cross-Level Interactions

In Model 2, interactions between individual deviant behavior and the two classroom-level measures as well as the respective three-way interaction were included. The interaction between individual deviant behavior and classroom-level deviant behavior was significant. As shown in Figure 2, initially nondeviant participants showed more deviant behavior at T2 and the higher the overall level of deviant behavior in their class had been at T1 (b _{class-level at 25th percentile of individual deviant behavior} = .47; 95% CI [.41, .53]). By contrast, individuals with a high level of deviant behavior a T1 were largely unaffected at T2 by the level of deviant behavior among their classmates (b _{class-level at 85th percentile of individual deviant behavior} = .33; 95% CI [.27, .41]). This result is in line with our prediction in Hypothesis 3 that deviant peer groups have a more pronounced effect on initially nondeviant individuals.

Figure 2.
Interaction between T1 individual-level and classroom-level deviant behavior predicting T2 deviant behavior.

A second significant cross-level interaction was found between individual deviant behavior and the variability of deviant behavior in the classroom. However, as plotted in Figure 3, the effect ran counter to the predicted pattern. In classrooms with a low variability of deviant behavior, individual differences in deviant behavior at T1 did not play a smaller role (b _{individual level at 25th percentile of variability} = .39; 95% CI [.37, .42]), compared to classrooms with a high variability in deviant behavior. Rather, individual differences were magnified, with initially deviant individuals showing significantly higher T2 scores than initially nondeviant individuals (b _{individual level at 75th percentile of variability} = .62; 95% CI [.58, .67]). The three-way interaction between classroom variability, classroom-level deviant behavior, and individual deviant behavior was nonsignificant, which fails to support Hypothesis 4.

Figure 3.
Interaction between T1 classroom-level variability of deviant behavior and individual-level deviant behavior predicting deviant behavior at T2.

In the final model, interactions with gender were added, as displayed in the last column of Table 3. The classroom-level interaction between variability and deviant behavior was qualified by an interaction with gender as an individual-level variable. The plot of this interaction in Figure 4 shows that for boys, this interaction was similar to the pattern for the sample as a whole. In classrooms where a high level of deviant behavior at T1 was combined with a low variability, boys reported more deviant behavior at T2, (b _{variability at 25th percentile of classroom level for boys} = 0.52; 95% CI [0.42; 0.63]) compared to classrooms with a higher variability of deviant behavior (b _{variability at 75th percentile of classroom level for boys} = 0.32; 95% CI [0.25; 0.39]). For girls, the level of deviant behavior in the classroom did not influence the relationship between class-level variability behavior and individual deviant behavior over time (b _{variability at 25th percentile of classroom level for girls} = 0.47; 95% CI [0.36; 0.58]; b _{variability at 75th percentile for girls} = 0.38; 95% CI [0.30; 0.46]).

Figure 4.
Interaction between T1 classroom-level deviant behavior, variability of deviant behavior, and gender predicting individual-level deviant behavior at T2.

Discussion

This aim of this study was to investigate the influence of deviant behavior in a class community on the development of individual class members’ deviant behavior over a 1-year period. The result showed a contagion effect in the sense that students with an initially low level of deviant behavior were particularly influenced by their deviant classmates, while adolescents who already showed high deviant behavior at T1 were less affected. This result is in line with the theoretical predictions of the SIP model and with other empirical findings showing that children and adolescent with a lower level of aggressive behavior are more influenced by the aggressive behavior among their peers (Rohlf et al., 2016). At the same time, our findings extend past evidence collected in homogeneous deviant peer groups, where it was found that adolescents learn deviant behavior from each other (Dodge et al., 2006). We argue that exposure to the deviant behavior of peers also provides learning experiences for initially nondeviant peers through modeling and reinforcement.

Going beyond previous research, our study examined not only the level of deviant behavior in a classroom but also the variability with which it was shown. The contagious effect of deviant behavior in a class was augmented when it was shown consistently (i.e., with low variability) by many students in the class rather than being due to a small number of highly deviant peers. This result is consistent with the theoretical consideration that social cues are easier to process if they are shown in a uniform fashion by many people in a group and should therefore affect all classroom members in a similar way (Crick & Dodge, 1994). The finding that no effect of variability was observed in classes with low levels of deviant behavior is also consistent with the SIP model, because the absence of cues does not need any processing resources. However, as indicated by the significant three-way interaction of deviant behavior, variability, and gender, this reasoning was only supported for boys. Socially shared normative beliefs often constrain the expression of aggressive behavior (Busching & Krahé, 2015). Since deviant behavior is more normatively acceptable for boys compared to girls, peer influences of deviant behavior may be stronger for boys than for girls.

It is noteworthy that no other interactions with gender emerged. Crick and Dodge (1994) argue that gender norms moderate the relationship between social information processing and deviant behavior. Girls should be especially affected by deviant peers who behave according to feminine gender norms, whereas boys should be more affected by deviant peers who act in accordance with masculine gender norms. Because our measure of deviant behavior did not include gender-specific forms, such a moderation could not be assessed. To investigate gender-specific effects, a more fine-grained, gender-sensitive measurement of deviant behavior is required for future research.

An unexpected result was that individual differences played a greater role as moderators of the impact of deviant class communities when the level of variability was higher, which contradicts the theoretical expectation that individual differences should be more influential moderators in classes with a low level of variability. One explanation could be that if there is a match between individual characteristics and the social environment (e.g., a deviant adolescent in a more deviant classroom and a nondeviant adolescent in a nondeviant classroom), individual behavior is socially reinforced. If there is a mismatch (e.g., a deviant adolescent in a nondeviant classroom), individuals are more likely to feel excluded and reach out to other peers outside their class who are more similar to them (Maner, DeWall, Baumeister, & Schaller, 2007).

Strengths and Limitations

This study has several strengths. It uses a large nationally representative sample to answer research questions about the development of deviant behavior in adolescence which have a high theoretical and applied relevance. It uses state-of the-art multilevel analysis to estimate the longitudinal impact of the collective deviant behavior as well as its variability at the classroom level. Moreover, it largely eliminates self-selection effects into like-minded peer groups by studying class communities to which individuals are assigned by third parties. A major limitation is that the variable of interest is based on self-reports. Since deviant behavior is undesirable, it is likely to be underreported. However, although mean levels of deviant behavior may be deflated, the influence of underreporting on the relationship between the variables, which is the focus of this article, is likely to be small. The low level of deviant behavior in this sample raises the problem of floor effects and the associated question of whether the findings could be an artifact of regression to the mean. Since floor effects are most pronounced among nondeviant peers in nondeviant classrooms, a regression-to-the-mean explanation would predict an increase in deviant behavior in this group. However, as shown in Figure 2, this group showed no increase in deviant behavior form T1 to T2, whereas nondeviant peers in deviant classrooms did. This pattern of results goes against the assumption of a regression-to-the-mean effect. Another weakness is that the processes involved in promoting the contagion of deviant behavior cannot be examined on the basis of the present data set because variables underlying the impact of deviant peers, such social status or normative influence, were not assessed.

Despite these limitations, our results support Huesmann’s (2012) proposition that aggression and deviant behavior can be seen as a contagious disease. Our study has shown that deviant behavior should not only be conceptualized as a developmental problem in the individual child or adolescent, but instead should be regarded as problem at the collective level of social environments, such as classroom communities. Our finding that classroom effects are stronger on individuals who start off with a low level of deviant behavior suggests that interventions need to start as early as possible, because deviant behavior spreads from deviant peers to the initially less deviant members of a class.

Supplemental Material

Supplemental Material, SPPS725151_suppl_mat - The Contagious Effect of Deviant Behavior in Adolescence: A Longitudinal Multilevel Study

Supplemental Material, SPPS725151_suppl_mat for The Contagious Effect of Deviant Behavior in Adolescence: A Longitudinal Multilevel Study by Robert Busching, and Barbara Krahé in Social Psychological and Personality Science

	T1	T2
Total sample
Individual level	.87	[.85, .88]	.92	[.91, .93]
Class level	.08	[.06, .09]	.06	[.05, .07]
School level	.06	[.04, .07]	.02	[.02, .03]
Boys
Individual level	.86	[.84, .88]	.91	[.90, .93]
Class level	.09	[.07, .10]	.08	[.06, .09]
School level	.05	[.04, .07]	.01	[.00, .02]
Girls
Individual level	.85	[.83, .87]	.88	[.86, .90]
Class level	.09	[.08, .11]	.08	[.06, .09]
School level	.06	[.04, .08]	.04	[.03, .06]

	Model 1	Model 2	Model 3
	Fixed Effects
(Intercept)	1.44*	[1.39, 1.49]	0.04	1.44*	[1.4, 1.49]	0.04	1.44*	[1.39, 1.48]	0.04
Gender^a	0.08*	[0.08, 0.09]	0.12	0.08*	[0.07, 0.09]	0.12	0.09*	[0.08, 0.1]	0.13
Age	0.02*	[0.01, 0.03]	0.03	0.02*	[0.01, 0.03]	0.03	0.02*	[0.01, 0.03]	0.03
Migration background^b	−0.03*	[−0.04, −0.02]	−0.04	−0.03*	[−0.04, −0.02]	−0.04	−0.03*	[−0.04, −0.02]	−0.04
Academic performance T1	0.06*	[0.04, 0.07]	0.06	0.06*	[0.04, 0.07]	0.06	0.06*	[0.04, 0.07]	0.06
School track	0.01	[−0.01, 0.03]	0.02	0.01	[−0.01, 0.03]	0.02	0.01	[−0.01, 0.03]	0.02
Individual deviant behavior T1	0.41*	[0.38, 0.43]	0.37	0.49*	[0.46, 0.52]	0.46	0.48*	[0.45, 0.51]	0.45
Class-level deviant behavior T1	0.40*	[0.34, 0.47]	0.18	0.41*	[0.35, 0.47]	0.19	0.41*	[0.35, 0.47]	0.19
Class-level variability T1	−0.11	[−0.22, 0.01]	0.02	−0.14*	[−0.25, −0.03]	−0.02	−0.15*	[−0.27, −0.04]	−0.02
Class-level deviant beh. × Class variability	−0.56*	[−0.87, −0.25]	−0.04	−0.62*	[−0.93, −0.3]	−0.04	−0.61*	[−0.93, −0.28]	−0.04
Class level × Individual deviant beh.				−0.15*	[−0.24, −0.06]	−0.04	−0.14*	[−0.23, −0.05]	−0.04
Class variability × Individual deviant beh.				−0.93*	[−1.12, −0.75]	−0.13	−0.96*	[−1.16, −0.77]	−0.14
Class level × Class variability × Individual deviant beh.				0.29	[−0.26, 0.84]	0.01	0.39	[−0.16, 0.97]	−0.02
Individual deviant beh. × Gender							0.02	[0, 0.04]	0.02
Class level × Gender							0.00	[−0.05, 0.04]	0.00
Class variability × Gender							0.04	[−0.05, 0.12]	−0.01
Class level × Individual deviant beh. × Gender							−0.02	[−0.09, 0.04]	0.02
Class variability × Individual deviant beh. × Gender							0.01	[−0.15, 0.16]	0.00
Class level × Class variability × Gender							−0.23*	[−0.47, −0.01]	0.02
Class level × Class variability × Individual deviant beh. × Gender							−0.04	[−0.47, 0.39]	0.00
	Random Parameters
Level 1—Individual level
Intercept	0.57*	[0.57, 0.58]	0.84	0.57*	[0.57, 0.58]	0.84	0.57*	[0.57, 0.58]	0.84
Level 2—Class level
Intercept	0.08*	[0.07, 0.09]	0.07	0.06*	[0.05, 0.07]	0.05	0.06*	[0.05, 0.07]	0.05
Deviant behavior	0.02*	[0.02, 0.03]	0.04	0.02*	[0.02, 0.03]	0.04	0.02*	[0.02, 0.03]	0.04
Level 3—School level
Intercept	0.00*	[0, 0.01]	0.01	0.00*	[0, 0.01]	0.01	0.00*	[0, 0.01]	0.01

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflict of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

The supplemental material is available in the online version of the article.

Notes

References

Bauer

Curran

(2005). Probing interactions in fixed and multilevel regression: Inferential and graphical techniques. Multivariate Behavioral Research, 40, 373–400. doi:10.1207/s15327906mbr4003_5

Baumert

Stanat

Watermann

(2006). Schulstruktur und die Entstehung differenzieller Lern- und Entwicklungsmilieus. In Baumert

Stanat

Watermann

(Eds.), Herkunftsbedingte Disparitäten im Bildungswesen: differenzielle Bildungsprozesse und Probleme der Verteilungsgerechtigkeit. Vertiefende Analysen im Rahmen von PISA 2000 ( ed., pp. 95–188). Wiesbaden, Germany: VS Verl. für Sozialwiss.

Bowman

M. A.

Prelow

H. M.

Weaver

S. R.

(2007). Parenting behaviors, association with deviant peers, and delinquency in African American adolescents: A mediated-moderation model. Journal of Youth and Adolescence, 36, 517–527. doi:10.1007/s10964-006-9117-7

Brook

J. S.

Whiteman

Finch

S. J.

Cohen

(1996). Young adult drug use and delinquency: Childhood antecedents and adolescent mediators. Journal of the American Academy of Child and Adolescent Psychiatry, 35, 1584–1592. doi:10.1097/00004583-199612000-00009

Busching

Krahé

(2015). The girls set the tone: Gendered classroom norms and the development of aggression in adolescence. Personality and Social Psychology Bulletin, 41, 659–676. doi:10.1177/0146167215573212

Chun

Mobley

(2010). Gender and grade-level comparisons in the structure of problem behaviors among adolescents. Journal of Adolescence, 33, 197–207. doi:10.1016/j.adolescence.2009.03.010

Crick

N. R.

Dodge

K. A.

(1994). A review and reformulation of social information-processing mechanisms in children’s social adjustment. Psychological Bulletin, 115, 74–101. doi:10.1037/0033-2909.115.1.74

Crosnoe

Erickson

K. G.

Dornbusch

S. M.

(2002). Protective functions of family relationships and school factors on the deviant behavior of adolescent boys and girls: Reducing the impact of risky friendships. Youth & Society, 33, 515–544. doi:10.1177/0044118X02033004002

Dishion

T. J.

Tipsord

J. M.

(2011). Peer contagion in child and adolescent social and emotional development. Annual Review of Psychology, 62, 189–214. doi:10.1146/annurev.psych.093008.100412

10.

Dodge

K. A.

Dishion

T. J.

Lansford

J. E.

(2006). Deviant peer influences in intervention and public policy for youth. Social policy report. Society for Research in Child Development, 20, 3–19.

11.

Everitt

Skrondal

(2010). The Cambridge dictionary of statistics (4th ed.). Cambridge, NY: Cambridge University Press.

12.

Farrell

A. D.

Henry

D. B.

Mays

S. A.

Schoeny

M. E.

(2011). Parents as moderators of the impact of school norms and peer influences on aggression in middle school students. Child Development, 82, 146–161. doi:10.1111/j.1467-8624.2010.01546.x

13.

Farrell

A. D.

Kung

E. M.

White

K. S.

Valois

R. F.

(2000). The structure of self-reported aggression, drug use, and delinquent behaviors during early adolescence. Journal of Clinical Child Psychology, 29, 282–292. doi:10.1207/S15374424jccp2902_13

14.

Frey

K. S.

Higheagle Strong

Onyewuenyi

A. C.

(2016). Individual and class norms differentially predict proactive and reactive aggression: A functional analysis. Journal of Educational Psychology. doi:10.1037/edu0000118

15.

Furthmüller

(2015). Methodenbericht Dokumentation der Scientific-Use-Files: Erhebungsverfahren und Datensätze von 2005 bis 2009. München, Germany: Deutsches Jugendinstitut e.V.

16.

Haynie

D. L.

(2001). Delinquent peers revisited: Does network structure matter? American Journal of Sociology, 106, 1013–1057. doi:10.1086/320298

17.

Hedges

L. V.

Hedberg

E. C.

(2007). Intraclass correlation values for planning group-randomized trials in education. Educational Evaluation and Policy Analysis, 29, 60–87. doi:10.2307/30128045

18.

Heinze

H. J.

Toro

P. A.

Urberg

K. A.

(2004). Antisocial behavior and affiliation with deviant peers. Journal of Clinical Child and Adolescent Psychology, 33, 336–346. doi:10.1207/s15374424jccp3302_15

19.

Hill

(2002). Biological, psychological and social processes in the conduct disorders. Journal of Child Psychology and Psychiatry, 43, 133–164. doi:10.1111/1469-7610.00007

20.

Huesmann

L. R.

(1988). An information processing model for the development of aggression. Aggressive Behavior, 14, 13–24. doi:10.1002/1098-2337(1988)14:1<13:: AID-AB2480140104>3.0.CO;2-J

21.

Huesmann

L. R.

(2012). The contagion of violence: The extent, the processes, and the outcomes. In Patel

D. M.

& Taylor

R. M.

(Eds), Social and economic costs of violence: Workshop summary (pp. 63–69). Washington, DC: National Academies Press.

22.

Isaacs

Voeten

Salmivalli

(2013). Gender-specific or common classroom norms? Examining the contextual moderators of the risk for victimization. Social Development, 22, 555–579. doi:10.1111/j.1467-9507.2012.00655.x

23.

Jung

Krahé

Bondü

Esser

Wyschkon

(2016). Dynamic progression of antisocial behavior in childhood and adolescence: A three-wave longitudinal study from Germany. Applied Developmental Science, 22, 1–15. doi:10.1080/10888691.2016.1219228

24.

Kellam

Ling

Merisca

Brown

Ialongo

(1998). The effect of the level of aggression in the first grade classroom on the course and malleability of aggressive behavior into middle school. Development and Psychopathology, 10, 165–185. doi:10.1017/S0954579498001564

25.

Lacourse

Nagin

D. S.

Tremblay

R. E.

Vitaro

Claes

(2003). Developmental trajectories of boys’ delinquent group membership and facilitation of violent behaviors during adolescence. Development and Psychopathology, 15, 183–197. doi:10.1017/S0954579403000105

26.

Lansford

J. E.

Dodge

K. A.

Fontaine

R. G.

Bates

J. E.

Pettit

G. S.

(2014). Peer rejection, affiliation with deviant peers, delinquency, and risky sexual behavior. Journal of Youth and Adolescence, 43, 1742–1751. doi:10.1007/s10964-014-0175-y

27.

Maner

J. K.

DeWall

C. N.

Baumeister

R. F.

Schaller

(2007). Does social exclusion motivate interpersonal reconnection? Resolving the “porcupine problem”. Journal of Personality and Social Psychology, 92, 42–55. doi:10.1037/0022-3514.92.1.42

28.

Mathys

Hyde

L. W.

Shaw

D. S.

Born

(2013). Deviancy and normative training processes in experimental groups of delinquent and nondelinquent male adolescents. Aggressive Behavior, 39, 30–44. doi:10.1002/ab.21456

29.

Myers

R. H.

(1990). Classical and modern regression with applications (2nd ed.). Duxbury classic series . Pacific Grove, CA: Duxbury/Thomson Learning.

30.

Patterson

G. R.

Reid

J. B.

Dishion

T. J.

(1992). A social learning approach: IV. Antisocial boys. Eugene, OR: Castalia Pub Co.

31.

Peterson

R. A.

Brown

S. P.

(2005). On the use of beta coefficients in meta-analysis. Journal of Applied Psychology, 90, 175–181. doi:10.1037/0021-9010.90.1.175

32.

Rohlf

Krahé

Busching

(2016). The socializing effect of classroom aggression on the development of aggression and social rejection: A two-wave multilevel analysis. Journal of School Psychology, 57–72. doi:10.1016/j.jsp.2016.05.002

33.

Salmivalli

Voeten

(2004). Connections between attitudes, group norms, and behaviour in bullying situations. International Journal of Behavioral Development, 28, 246–258. doi:10.1080/01650250344000488

34.

Snyder

Dishion

T. J.

Patterson

G. R.

(1986). Determinants and consequences of associating with deviant peers during preadolescence and adolescence. The Journal of Early Adolescence, 6, 29–43. doi:10.1177/0272431686061003

35.

StEG-Konsortium. (2011). Studie zur Entwicklung von Ganztagsschulen (StEG). Datensatz der Längsschnitterhebungen von 2005, 2007 und 2009 (Version: 1). IQB - Institut zur Qualitätsentwicklung im Bildungswesen. doi:10.5159/IQB_StEG_v1.

36.

Tillmann

K.-J.

Holler-Nowitzki

Holtappels

H. G.

Meier

Popp

(1999). Schülergewalt als Schulproblem. Verursachende Bedingungen, Erscheinungsformen und pädagogische Handlungsperspektiven. Weinheim/München, Germany: Juventa.

37.

Vitaro

Boivin

Tremblay

R. E.

(2007). Peers and violence: A two sided developmental perspective. In Flannery

D. J.

Vazsonyi

A. T.

Waldman

I. D.

(Eds.), The Cambridge handbook of violent behavior and aggression (pp. 361–387). Cambridge, England: Cambridge University Press.

38.

Wallner

Stemmler

(2014). Jugendliche Gewaltdelinquenz, psychosoziale Merkmale und Migrationsstatus. Forensische Psychiatrie, Psychologie, Kriminologie, 8, 84–95. doi:10.1007/s11757-014-0266-3

39.

Yarnell

L. M.

Pasch

K. E.

Brown

H. S.

Perry

C. L.

Komro

K. A.

(2014). Cross-gender social normative effects for violence in middle school: Do girls carry a social multiplier effect for at-risk boys? Journal of Youth and Adolescence, 43, 1465–1485. doi:10.1007/s10964-014-0104-0

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.57 MB

	Mean (SD)	1	2	3	4	5	6	7	9	10	11
Individual Level
1. T1 Deviant behavior raw score (comprising both individual- and class-level information) (1–5)	1.49 (.73)
2. T2 Deviant behavior raw score (comprising both individual- and class-level information) (1–5)	1.56 (.69)	.39***
3. T1 Individual-level deviant behavior^a	0.00 (.66)	.91***	.34***
4. T2 Individual-level deviant behavior^a	0.00 (.65)	.34***	.93***	.36***
5. T1 Age (years)	12.17 (1.28)	.20***	.12***	.05***	.02**
6. T1 Academic performance (1–6)	3.09 (.76)	−.25***	−.17***	−.16***	−.10***	−.28***
7. Migration background (% yes)	29%	.09***	.08***	.05***	.06***	.03***	−.13***
8. Gender (% male; female = −1; male = 1)	52%	−.18***	−.20***	−.17***	−.19***	−.05***	.09***	.01
Classroom Level
9. T1 Classroom-level deviant behavior^b	1.51 (.32)
10. T2 Classroom-level deviant behavior^b	1.58 (.27)								.45***
11. T1 Classroom-level variability of deviant behavior	0.39 (.16)								.63***	.26***
12. T2 Classroom-level variability of deviant behavior	0.38 (.13)								.24***	.57***	.26***

	T1		T2
	ICC	95% CI	ICC	95% CI
Total sample
Individual level	.87	[.85, .88]	.92	[.91, .93]
Class level	.08	[.06, .09]	.06	[.05, .07]
School level	.06	[.04, .07]	.02	[.02, .03]
Boys
Individual level	.86	[.84, .88]	.91	[.90, .93]
Class level	.09	[.07, .10]	.08	[.06, .09]
School level	.05	[.04, .07]	.01	[.00, .02]
Girls
Individual level	.85	[.83, .87]	.88	[.86, .90]
Class level	.09	[.08, .11]	.08	[.06, .09]
School level	.06	[.04, .08]	.04	[.03, .06]

	Model 1			Model 2			Model 3
	est(b)	95% CI	β	est(b)	95% CI	β	est(b)	95% CI	β
	Fixed Effects
(Intercept)	1.44*	[1.39, 1.49]	0.04	1.44*	[1.4, 1.49]	0.04	1.44*	[1.39, 1.48]	0.04
Gender^a	0.08*	[0.08, 0.09]	0.12	0.08*	[0.07, 0.09]	0.12	0.09*	[0.08, 0.1]	0.13
Age	0.02*	[0.01, 0.03]	0.03	0.02*	[0.01, 0.03]	0.03	0.02*	[0.01, 0.03]	0.03
Migration background^b	−0.03*	[−0.04, −0.02]	−0.04	−0.03*	[−0.04, −0.02]	−0.04	−0.03*	[−0.04, −0.02]	−0.04
Academic performance T1	0.06*	[0.04, 0.07]	0.06	0.06*	[0.04, 0.07]	0.06	0.06*	[0.04, 0.07]	0.06
School track	0.01	[−0.01, 0.03]	0.02	0.01	[−0.01, 0.03]	0.02	0.01	[−0.01, 0.03]	0.02
Individual deviant behavior T1	0.41*	[0.38, 0.43]	0.37	0.49*	[0.46, 0.52]	0.46	0.48*	[0.45, 0.51]	0.45
Class-level deviant behavior T1	0.40*	[0.34, 0.47]	0.18	0.41*	[0.35, 0.47]	0.19	0.41*	[0.35, 0.47]	0.19
Class-level variability T1	−0.11	[−0.22, 0.01]	0.02	−0.14*	[−0.25, −0.03]	−0.02	−0.15*	[−0.27, −0.04]	−0.02
Class-level deviant beh. × Class variability	−0.56*	[−0.87, −0.25]	−0.04	−0.62*	[−0.93, −0.3]	−0.04	−0.61*	[−0.93, −0.28]	−0.04
Class level × Individual deviant beh.				−0.15*	[−0.24, −0.06]	−0.04	−0.14*	[−0.23, −0.05]	−0.04
Class variability × Individual deviant beh.				−0.93*	[−1.12, −0.75]	−0.13	−0.96*	[−1.16, −0.77]	−0.14
Class level × Class variability × Individual deviant beh.				0.29	[−0.26, 0.84]	0.01	0.39	[−0.16, 0.97]	−0.02
Individual deviant beh. × Gender							0.02	[0, 0.04]	0.02
Class level × Gender							0.00	[−0.05, 0.04]	0.00
Class variability × Gender							0.04	[−0.05, 0.12]	−0.01
Class level × Individual deviant beh. × Gender							−0.02	[−0.09, 0.04]	0.02
Class variability × Individual deviant beh. × Gender							0.01	[−0.15, 0.16]	0.00
Class level × Class variability × Gender							−0.23*	[−0.47, −0.01]	0.02
Class level × Class variability × Individual deviant beh. × Gender							−0.04	[−0.47, 0.39]	0.00
	Random Parameters
Level 1—Individual level
Intercept	0.57*	[0.57, 0.58]	0.84	0.57*	[0.57, 0.58]	0.84	0.57*	[0.57, 0.58]	0.84
Level 2—Class level
Intercept	0.08*	[0.07, 0.09]	0.07	0.06*	[0.05, 0.07]	0.05	0.06*	[0.05, 0.07]	0.05
Deviant behavior	0.02*	[0.02, 0.03]	0.04	0.02*	[0.02, 0.03]	0.04	0.02*	[0.02, 0.03]	0.04
Level 3—School level
Intercept	0.00*	[0, 0.01]	0.01	0.00*	[0, 0.01]	0.01	0.00*	[0, 0.01]	0.01