Although many studies indicated a relation between attachment insecurity and bullying, there are few studies on how attachment might affect bullying behaviors and being victims of bullying. Is the link between attachment and bullying direct or mediated by other factors? A fascinating hypothesis in the literature is that the experience of growing up in a nurturing and responsive family environment, and consequently attachment security, may facilitate a prosocial value orientation—i.e., relatively stable individual preferences or desirable goals that reflect socialization and that serve as guiding principles during people’s lives (Schwartz, 1992; Bilsky and Schwartz, 1994; Knafo and Schwartz, 2003). According to this prosocial-security hypothesis, the internalization of a secure internal model of self and others may facilitate the development of “conservative social-oriented values”—i.e., being benevolent toward others, preferring security, being tied to traditions, behaving similarly to people considered to be friends and peers. These values can be measured on specific rating scales: universalism, benevolence, tradition, conformity, and security (van Lange et al., 1997). Individuals who scored higher on these values, in particular on benevolence, remembered warmer and less aggressive parents (Aluja et al., 2005). The choice of conservative social-oriented values, in turn, may protect teens from problem behaviors, and more specifically from externalizing behaviors (Barnea et al., 1992). Moreover, high levels of benevolence were related to socialized behaviors (Aluja-Fabregat et al., 1999). On the contrary, insecure adolescents may have learned to perceive interpersonal relationships as threatening, and this may increase “dynamic self-oriented values” (including power, achievement, hedonism, stimulation, and self-direction), as shown by van Lange et al. (1997; see also Sarracino et al., 2011; Sarracino and Innamorati, 2012). People who prefer these values, especially power, remember more neglecting and less warm parents (Meesters et al., 1995). Consequently, competitive and aggressive behaviors may be favored (Aluja-Fabregat and Torrubia-Beltri, 1998) and transmitted across generations (De Carli et al., 2016, 2017a,b; Wang et al., 2017).

A number of studies empirically supported the prosocial-security hypothesis through different phases of adolescence and early adulthood, showing a significant association between attachment security and prosocial values (van Lange et al., 1997; Mikulincer et al., 2003; Sarracino et al., 2011; Sarracino and Innamorati, 2012). Although we are far from having a comprehensive view, a plausible model of the relations among temperamental factors, attachment, social value orientation, and problem behaviors in adolescence can be suggested (Figure 1).

Bullying concerns both males and females. Although recent studies indicate that girls are more physically aggressive than in the past (see for example, Jamal et al., 2015), the psychological repercussions on girls and boys may differ (Besag, 2006). In fact, girls are usually involved in more indirect and relational forms of bullying than boys (Crick and Grotpeter, 1995; Caravita et al., 2009; Jimerson et al., 2009). These differences are in line with evidence that boys are more likely to have externalizing behaviors than girls—with the exception of hedonistic habits like smoking and drinking (for a meta-analysis, see Byrnes et al., 1999). Moreover, boys prioritize dynamic self-oriented values (i.e., achievement and power), while girls tend to favor conservative social-oriented values (i.e., benevolence and universalism). Thus, boys are believed to be more likely to experience problematic behaviors than girls, due to their attempts to achieve social power and authority with respect to their peers (see also Wilson and Daly, 1985; Liu et al., 2007). In these studies, boys also scored higher in sensation seeking than girls.

Moreover, several studies found gender differences in adolescents’ attachment. Adolescent girls report greater attachment security with their mothers than to their fathers; conversely, while boys experience greater attachment security with their fathers (Verschueren and Marcoen, 2005; Diener et al., 2008; Del Giudice, 2009; Sarracino et al., 2011). Williams and Kennedy (2012) observed that girls were more likely to be involved in bullying when they had higher levels of attachment avoidance and attachment anxiety with their mothers, and higher levels of attachment anxiety with their fathers. On the other hand, boys were more likely to bully when they experienced higher levels of attachment anxiety with their fathers. In light of these findings, Sarracino et al. (2011) hypothesized that gender not only has a direct effect on the single variables such as attachment, social value orientation, sensation seeking and bullying, but may influence the prosocial-security model. In other words, gender may affect the way attachment mediates the relationship between sensation seeking and problem behavior.

While the different studies focused on specific aspects of the problem, to date few models consider and integrate the links between temperamental, psychological and psychosocial variables, which may be involved in adolescent bullying. The contribution of the present study consists in the attempt to extend the prosocial-security hypothesis to the study of bullying. We have included sensation seeking in this model because, in a previous study, two of the authors of this paper found sensation seeking to be strongly tied to at risk behavior in pre-adolescence through the mediation of attachment (Sarracino et al., 2011), and decided to verify if the same factor is a predictor of bullying.

Specifically, our study explored the relationships between sensation seeking (as a temperamental predisposition to risk-taking), attachment security (related to family environment and parenting style), and social value orientation (connected to family and the socio-cultural context), and how this network of innate and cultural-driven influences can affect bullying-related behaviors (bullying or being bullied) in early adolescence. In this regard, we explored the possibility that a social value orientation can mediate the relationship between attachment security and both bullying and being victimized by bullying. In line with Sarracino et al.’s (2011) findings, we expected that more secure adolescents would have more “conservative social-oriented values,” less sensation seeking, and fewer bullying-related behaviors. Conversely, we hypothesized that insecure attachment, dynamic self-oriented values, and sensation seeking would be associated with more bullying-related behaviors.

Another aim of the study was to explore the role of gender in our model. Specifically, we expected different patterns for boys and girls, in which attachment moderates the relationship between sensation seeking and bullying.

Research assistants recruited 375 Italian early adolescents from two middle schools in Northern Italy. Average ages of participants ranged from 11 to 16 years (M = 12.58, SD = 1.08). Forty-seven percent of participants were 11–12 years old, 49% were 13–14 years old, and 4% were 15–16 years old. Fifty-three percent of participants were female. Seventy-six percent of participants reported being born in Northern Italy, 7% in Central or Southern Italy, 4% in other European countries, and 13% in non-European countries. With regard to parent nationality, 77% of parents were both Italians, 19% were both foreigners, and 4% were mixed. Eighty-one percent of participants reported living with both parents, 15% with mothers only, 2% only with their fathers, 1% with other relatives, and 1% lived in welfare structures. Twelve percent of participants reported living in a family with a low or medium-low income, 53% reported belonging to a family with a medium income, and 35% declared a medium-high or high income. Non-native Italian participants spoke and understood Italian adequately as attested by their teachers, and thus the school or parents deemed no help was necessary.

The present study is part of a multidisciplinary research project concerning the role of psychodynamic factors in problem behaviors in different stages of life (Innamorati et al., 2010; Dellantonio et al., 2012; Lo Verde et al., 2012; Sarracino et al., 2014; Sassaroli et al., 2015). The review board of the University of Rome “Tor Vergata” approved the study protocol, in accordance with the Helsinki Declaration. After the school principals’ approval of the study, research assistants described it to the classes, answered questions, and provided them with the informed consent form to be signed by parents, along with a letter, signed by the study authors and the school principals, explaining the importance of collaboration between schools and the universities. Participation rate was unexpectedly high (98%), which may be explained by the fact that data collection occurred during school hours, replacing the usual lessons. At a given date, research assistants administered during class information forms (containing questions on gender, age, birthplace, parents’ nationality, family situation, and income), and a set of pencil-and-paper questionnaires. The questionnaire took approximately 1 h to complete. A research assistant informed participants that the questionnaire was anonymous and that they would not be judged by their answers. Subsequently, she explained the aims of the study and answered any further questions.

We performed all the analyses with R software, using the package for structural equation modeling “lavaan” (Rosseel, 2012). In order to handle missing values (89.6% of the respondents had no missing values, with a maximum of 2.9% per variable), we imputed data using the EM algorithm (Schafer and Graham, 2002). Then we adopted a more conservative approach, and repeated all the analyses performing multiple imputations with the “Amelia” package (Honaker et al., 2011). We reported only single imputation analyses since results did not differ for multiple imputed data.

Table 2 presents the descriptive analyses of the variables of the study divided by gender. Boys showed higher scores for sensation seeking, dynamic self-oriented values (PVQ), and bullying behaviors. Girls had higher scores on conservative social-oriented values.

In order to test the prosocial security hypothesis, we performed a path analysis, which estimated the effects of sensation seeking, attachment security, and social value orientation on the bully/victim behaviors. The developmental onset of the predictors determined the direction of the effects of the path models. Thus, since sensation seeking is a personality trait (e.g., Zuckerman, 2007), we hypothesized that it affects attachment security, which depends on environmental factors and develops later in childhood (e.g., Pasco Fearon et al., 2006). In a similar way, our path model relied on the assumption that attachment security matures prior to social value orientation and therefore influences the latter. Finally, we assumed that attachment security, social value orientation and sensation seeking affect bullying at a behavioral level. We allowed for the two attachment security factors to co-vary with each other, as well as the two social value orientation factors and the two bullying outcome variables (bullying and victim of bullying).

We used the following indices to evaluate the goodness of fit of the model: The chi-square statistic, which should be non-significant; the comparative fit index (CFI) and the Tucker-Lewis index (TLI), which should approach 0.95 to indicate optimal fit (Brown, 2015); the root mean squared error of approximation (RMSEA) and the root mean square residual (SRMR), which should approach 0.05 and 0.08, respectively, to indicate a good fit (Kline, 2015); non-significant probability values associated with the Cfit of RMSEA. The sample size seems adequate for the requirements of structural equation models presented by Kline (2015).

Figure 2 illustrates the path analysis of the hypothesized model. The overall fit measures indicated that the fit of the model was adequate (χ² = 5.20, p = 0.07, CFI = 0.99, TLI = 0.98, RMSEA = 0.06, p = 0.26, 95% CI [0.00, 0.14], SRMR = 0.005). Significant paths showed the effect of sensation seeking in predicting BO and BV both directly and, only for BO, through the mediating role of conservative social-oriented values (mediated path: b = 0.04 [β = 0.11], p < 0.001, 95% CI [0.019, 0.059], total effect: b = 0.09 [β = 0.25], p < 0.001, 95% CI [0.053, 0.120]). The results also showed direct significant paths from attachment to the outcome variables.

In order to test the mediation effect of social orientation in the association between attachment to mother and bullying, we computed the indirect effect. Results show a significant mediation effect (b = -0.029, [β = -0.05], z = -2.93, p = 0.003, 95% CI [-0.048, -0.010]) and a significant total effect (b = -0.128, [β = -0.22], z = -3.88, p < 0.001, 95% CI [-0.192, -0.063].

In order to study whether the same pattern of associations was found both in males and females, we tested the previous model in a multiple group analysis. The sample size can support only an exploratory approach, so the findings have to be considered only preliminary and needs further replications. The fit indices can be considered adequate (χ² = 7.06, p = 0.13, CFI = 0.99, TLI = 0.98, RMSEA = 0.06, p = 0.31, 95% CI [0.00, 0.14], SRMR = 0.005). The significant paths are presented in Figure 3.

We also repeated the test of the indirect effect of social orientation in the association between attachment to mother and bullying to explore whether it is present both for males and females. Results show a significant mediation effect of social orientation in males (b = -0.045, [β = -0.06], z = -2.07, p = 0.04, 95% CI [-0.09, -0.002]) but not in females (b = -0.01, [β = -0.04], z = -1.71, p = 0.09, 95% CI [-0.03, 0.002]).

Our study has a number of limitations. Firstly, the study only employed self-report measures. The reliance on a single informant may have caused shared-method variance (i.e., the relation between variables may be in part due to similar measurement methods). Furthermore, self-report measures may cause systematic distortions when investigating problem behavior in adolescents. In fact, even though we assured anonymity to participants, their answers may not be entirely reliable, since we asked them to report if they had participated in deviant and illegal activities at times. Therefore, future research should integrate reports from caregivers and/or observational data relative to caregiver–adolescent interactions.

Second, this study has to be considered as explorative and thus causal interpretations should be made with caution since our data are cross-sectional, and the variables of the study have not been extensively researched in the early adolescence age-group. Longitudinal studies are therefore required in order to explore the direction of effects, and to understand how attachment security and other social experiences interact, and influence children’s social development in general, and bullying more specifically.

Third, our sample was somewhat homogenous concerning social class, family structure, and ethnicity. Therefore, future research should be extended to adolescents from other ethnic backgrounds, family structures (e.g., both single and dual-earners), social contexts (e.g., at-risk populations), and different nationalities.