Both human and animal studies of risk-taking behavior have proposed models that predict risky choice. Recent studies have provided two new perspectives: the decision-making model (e.g., Millstein and Halpern-Felsher, 2001) and the social-neuroscience model (e.g., Steinberg, 2008). Decision-making theorists posit that individuals’ beliefs about the consequences of their actions and perceptions of their vulnerability regarding those consequences play a key role in their behavior (Millstein and Halpern-Felsher, 2001). Researchers are interested in understanding why adolescents make the decisions they do and their competence in making these decisions. In conceptualizing and measuring perceptions of risk, decision-making theorists observe whether people recognize the benefits and risks inherent in a given situation in order to measure their sense of risk and vulnerability (Slovic, 1987; Siegrist et al., 2000). Recent studies have suggested that engagement in risky activities is determined by individual differences in cognitive appraisal (expected benefit minus expected risk) (Fromme et al., 1997, 1999). When a person overestimates the benefit of or underestimates the risk of certain events it increases the likelihood of their participation in a risky activity (Byrnes, 1998; Steinberg, 2004, 2005). In short, decision theorists generally believe that by adopting a risk-benefit analysis, one might obtain the net expectancy of the risky event, which might provide useful information in predicting an individual’s involvement in risk-taking behavior (Steinberg, 2005). Under the decision-making model, risk perceptions play a fundamental role in behavioral intervention programs that try to encourage adolescents to recognize and acknowledge their own vulnerability to negative outcomes.

School bullying involves repeated aggression toward those who are perceived as weaker and less dominant and it has negative consequences for both bullies and victims (Ttofi et al., 2012). Many factors play an important role in school aggression: individual differences on how one encodes and emotionally processes evocative situations, interpretation and response to the behaviors of others, and motivation toward obtaining reward and avoiding punishment (e.g., Crick and Dodge, 1996). Another view of risk-taking behavior highlights the role of affective intensity and sensitivity to reward. According to this perspective, risky behavior cannot be fully explained by a deficiency in comprehending the potential consequences of these actions (Reyna and Farley, 2006). Moreover, leading environmental cues may “win” over cognitive control in emotionally charged circumstances. This model is supported by epidemiological reports of heightened affective responsiveness and incentive-based behavior changes when cognitive-based measures are hampered. Steinberg (2010) examined the social-neuroscience model of risk taking using behavioral measures. Steinberg used a gambling task to capture the reward-related decision-making process of participants and concluded that a heightened reward sensitivity and impulsivity were seen consistently and clearly in high-risk takers.

In sum, while many studies on bullying and its associations with a variety of risk-taking behaviors have been performed, it is surprising that no study has examined and compared the risk-taking pattern among bullying groups (i.e., bully, victim, and bully victim). This study investigated the link between risk taking and bullying in adolescents in hopes to elucidate the potential relationship between bullying and risk-taking behavior by addressing the two models of risk taking. Within the framework of these two models, this study hopes to offer a holistic model for understanding and differentiating a risk-taking pattern among the three bullying groups. Particularly, this study adopted a risk-appraisal questionnaire to address a cognitively focused process and a computerized gambling task to measure a motivational, emotion-focused process. It was hypothesized that bullies and victims would exhibit different patterns of risk appraisal and propensity. Moreover, bully victims were expected to present co-morbid patterns of both bullies and victims.

The study sampled students from Hong Kong public secondary schools. A multi-stage sampling procedure was used to obtain the sample. Mass invitation has been sent to all public secondary schools in Hong Kong. Schools to be studied were then randomly selected such that the proportion of students in each selected district represented the number of students in that area. This selection procedure resulted in the selection of 4 schools respectively with one from Hong Kong Island, one from Kowloon Peninsula, and two from the New Territories. As schools were given the option to withdraw from participation, a further four schools were selected by the same procedure to act as a back-up. Schools that withdrew from participation were replaced by schools on the back-up list from within the same district. Of the originally selected schools, none withdrew from participation. From each school, six students were randomly selected from grade 7 to grade 12, such that a total of 36 students were selected per school for participation. With eight students dropped out in the middle of the study, 136 adolescents were recruited and tested. Participants ranged in age from 12 to 17 years (M = 14.45, SD = 1.614) and comprised of 65 males (M) (47.8%) and 71 females (F) (52.2%). Based on self-reported experience of school bullying and victimization in the last 4 months, the participants were categorized into four distinct groups: control group (n = 52, M = 19, F = 33), bully group (n = 27, M = 10, F = 17), victim group (n = 20, M = 10, F = 10), and bully victim group (n = 37, M = 26, F = 11). All participants had a clean medical history; spoke Cantonese as their first language; had normal intelligence; and had no suspected brain damage or neurological, sensory, or psychiatric problems.

This study consisted of a screening phase and an assessment phase. During the screening phase, participants were administered a standard Raven Progressive Matrices (RPM) as a proxy of intelligence in order to rule out low intellectual functioning. Only students with normal intelligence were invited to the assessment phase. During the assessment phase, the Computerized Cambridge Gambling Task (CGT) was administered individually to each participant. Participants then completed a demographic questionnaire including questions about age, educational level, family income, and the Cognitive Appraisal of Risky Events Questionnaire (CARE). The assessment phrase lasted approximately 1 h. All assessments were conducted by well-trained research assistants with undergraduate degrees in psychology who had been trained for approximately 2 h in test administration. All parents and adolescents provided informed consent to participate in this study. Ethical approval for the research was gained through the Human Research Ethics Committee at the Education University of Hong Kong.

Table 1 summarizes the descriptive statistics for each group. Close to a marginally significant level of group differences were found in general intellectual ability [F(3,125) = 2.58, p = 0.06, η_p² = 0.06]. Post hoc pairwise comparisons revealed that the control group had significantly outperformed the victim group (p < 0.05). No significant difference was found between the control and the bully victim group (p = 0.68). The four groups did not differ in age, education level (self, father, and mother), and average family income (all ps > 0.10). Total bully scores and victim scores captured from the school bullying questionnaire are also summarized in Table 1 with higher scores representing more frequent bullying or victimization behavior. The bully victim group showed significant higher bullying frequency than the bully group (p < 0.001); however, there was no difference in victim score between the victim and bully victim groups (p > 0.10). For sex, a chi-square analysis was applied and a statistical significance was found [see Table 2, χ²(3, N = 136) = 9.72, p = 0.02].

To examine whether bullying identity (control, bully, victim, bully victim) was significantly associated with the three outcome expectancies measured by CARE, a series of 2 (bully vs. not bully) × 2 (victim vs. not victim) between subject ANCOVAs with general intellectual ability as covariate were computed and summarized in Table 3. A post hoc Fisher’s Least Significant Difference (LSD) for group comparison within each sub-questionnaire was also conducted. Based on the framework of the decision-making model, a hierarchical regression analysis was conducted to examine the expected net (CARE_EN) (CARE_EN = CARE_EB – CARE_ER) in predicting CARE_EI (see Table 4). Results showed that the aggregate effect of expected benefit minus expected risk significantly predicted the expected involvement of risky events (p < 0.001).

A hierarchical regression analysis was conducted to examine what CARE variables best predicted the severity of bullying and victimization. Table 5 shows that CARE accounted for an additional 16.9% of the variance in Bullying severity, and CARE_EI was the only significant predictor of Bullying severity (β = 0.40, t = 4.30, p < 0.001). Regarding the severity of victimization, Table 6 shows that CARE accounted for only an additional 2.5% of the variance in victimization severity, and none of the CARE variables significantly predicted the frequency of being bullied.

In order to capture behavior differences in adolescents’ reward-related sensitivity and risk-taking propensity, the six CGT factors previously listed were analyzed with general intellectual ability as a covariate using a 2 (bully vs. not bully) × 2 (victim vs. not victim) between-subjects ANCOVA (see Table 7). A one-way ANCOVA followed by a post hoc LSD were conducted to examine significant differences between groups within each factor. Correlations between the CGT factors are shown in Table 8.

A hierarchical regression analysis was conducted to examine what CGT variables best predicted the severity of bullying and victimization. Table 9 shows that CGT accounted for an additional 10.2% of the variance in bullying severity and risk adjustment was the only significant predictor of bullying severity (β = 0.25, t = 2.77, p < 0.01). Regarding the severity of victimization, Table 10 shows that CGT only accounted for an additional 3.5% of the variance in the frequency of victimization and none of the CGT variables significantly predicted the severity of victimization.

Bullying and risk-taking behavior are two common problems among adolescents worldwide and can cause serious impact to adolescents’ physical, psychological, social, and educational functioning when both coexist in the same youth (Currie et al., 2012). Despite the associations between bullying and risk-taking behaviors, no study thus far has examined and compared the risk-taking pattern among bullying groups (i.e., bully, victim and bully victim). This study elucidated the risk-taking pattern in these three bullying groups by addressing two risk-taking models (i.e., the decision-making model and the social-neuroscience model) and clarifying how adolescents’ characteristics in risk taking may associate with outcomes in bullying or victimization.

These findings suggest that being a bully or victim is associated with distinct results on the two approaches, and these results contribute several novel findings to the current literature. Under the decision-making framework, bullying was associated with higher negative expectancy on risky activities and more expected involvement in them. In other words, being a bully meant more anticipated negative outcomes in risky activities. Interestingly, despite the higher anticipation of negative outcomes, bullies expected themselves to have higher involvement in these activities compared to the non-bully groups. In CGT, bullying was associated with positive risk-taking adjustment, an increased amount of gambling their points when the odds were in their favor, and vice versa. These results indicated that being a bully was vigilant in situational deliberation and risk modulation, and contradicted the general agreement that there was a positive link between bullying and impulsivity (Olweus, 1995; Schwartz et al., 2001; O’Brennan et al., 2009). Perhaps bullying is a unique, complex form of interpersonal aggression or group phenomenon (e.g., Olweus, 2001; Salmivalli, 2001; Rodkin and Hodges, 2003) and bullies tend to be hyper-vigilant to social cues and attribute negative intentions to others (Vaillancourt et al., 2003). The paradox of aggression is that it is both adaptive and maladaptive. Aggressive adolescents are at risk for a host of possible negative consequences (e.g., disciplinary punishment, school suspension, etc.) (Dodge et al., 2006). However, aggression is often a successful means in changing other’s behavior and can be used to acquire resources and maintain group boundaries. Moreover, researchers generally agree that bullies tend to exhibit high levels of social intelligence and the ability to manipulate peers (e.g., Peeters et al., 2010). In other words, bullies may be a group of adolescents who are sensitive and respond quickly to external cues through peer manipulation, resulting in potential risk for themselves and victims.

The results of the CARE questionnaire suggested that victims were associated with an overestimation of benefits, an underestimation of risks, and higher expected involvement in risky events. According to the decision-making model, these patterns suggested the lowest sense of risk and the highest vulnerability to it (Hunter and Boyle, 2004; Hunter et al., 2004; Wachs et al., 2012). Motivationally, victims were associated with less deliberation time and more delay aversion in CGT. In CGT, even though delay response does not increase the available information for decision making, shorter deliberation time may indicate impulsive decision-making. This assumption was further supported by their poor performance in delay aversion and that victims were more likely to bet larger amounts when bets were displayed in a descending rather than an ascending way. It is well established that impulsivity is particularly relevant to peer victimization. For example, children with attention-deficit/hyperactivity disorder who share common features of impulsivity and high risk-taking propensity, show marked impairment in peer relationships and are significantly bullied by peers (e.g., Hoza, 2007). Researchers argued that individuals who are low on self-control or high on impulsivity are unable to see the consequences of their actions and more likely to place themselves in risky situations without regard for their long-term outcome (e.g., Higgins et al., 2009).

Consistent with our hypothesis, bully victims presented the combined pattern of the two pure groups. Cognitively, bully victims showed similar, but not identical patterns of the two pure groups. Bully victims overestimated the benefits of risk as well as benefit and projected the highest possibility of engaging in risky behavior. Motivationally, this unique group shared identical patterns with bullies in terms of positive risk adjustment. In other words, bully victims exhibited similar risk tendencies to victims in terms of cognitive-focused processes and to bullies on emotion-focused processes. Importantly, compared with bullies, bully victims had significantly higher bullying scores, showing that they engaged in a wider range of more frequent bullying activities. In fact, previous studies that focused on the frequency and forms of bullying also reported that bully victims used wider and more aggressive strategies than the pure groups did (Olweus, 1993; Schwartz, 1999; Olafsen and Viemerö, 2000). Additional evidence has suggested that bully victims may exhibit the highest aggression level of all three groups as well as the poorest emotion regulation (e.g., Nansel et al., 2004). Olweus (1978) suggested that bullies were generally more functional, more likely to use proactive aggression, and more likely to have an extensive social network than bully victims, who were more likely to react aggressively and show troubling risk patterns across virtually all adjustment indicators.

Concern about the frequency and effects of adolescent bullying was reflected in the increase in research aimed at understanding its causes and consequences in order to develop appropriate policy and intervention strategies. Nonetheless, the success of intervention programs to prevent or mitigate bullying in adolescence has been limited (e.g., Olweus, 1999; Merrell et al., 2008). Even when programs have an impact, the improvement appears to be in changing adolescents’ knowledge and perceptions on bullying, but not the behavior. This study extended the existing literature on bullying and victimization by addressing the co-occurrence of bullying and victimization and identifying a range of risk-taking patterns associated with the three groups. This study suggested that there were significant group differences on how bullies, victims, and bully victims appraised risky events and emotionally processed evocative risky situations. Moreover, this study identified different predictors of bullying and victimization. In accordance with the decision-making model, this study provided further support that participants’ perceived benefits and risk may play a key role in predicting their expected involvement in risky behaviors. Given that expected involvement in risky events is a significant predictor to the severity of bullying, future intervention programs should include cognitive training in risk evaluation (Steinberg, 2007). For instance, presenting strategies for effective risk and benefit evaluation will encourage less risky and healthier choices. Moreover, given the high vigilance of bullies to external cues, future interventions should target the enhancement of social skills and coping strategies on peer conflicts. Victims with marked impulsivity, on the other hand, may be targeted to deliver interventions on self-control that strengthen their ability to fit in with their peers and reduce the likelihood of rejection and victimization, or to provide supportive interpersonal relationships to reduce their isolation (Andreou et al., 2005). Effective and comprehensive evaluation of risk is another intervention focus for victims to help increase their awareness on the consequences of their risky choices.

This study has several limitations that should be considered when interpreting our findings. First, the sample size was relatively small. Future studies should include a larger sample size and a power calculation to determine what sample size would be needed to detect differences between groups. Second, this study was cross-sectional in nature. Therefore, causation cannot be determined. A temporal relationship between bullying and a risk-taking pattern could not be inferred. Questions of causation can only be answered in a longitudinal research design, which should be implemented in the future. Third, this study was localized to Chinese adolescents. Even though assessment tools were standardized, a comprehensive assessment tool on risk appraisal may be needed to elicit cultural differences on risk perception. In China, school bullying is often regarded as a collective act (Cheng et al., 2011; Huang et al., 2013; Chui and Chan, 2015), and whether this is a form of “collective bullying” that may cause an impact to the risk-taking pattern is yet to be determined. Fourth, although this study examined emotional aroused stimulation (i.e., the gambling task), it might not present an authentic context where bullying is most likely to occur. Lastly, this study was limited by its use of self-report measures (e.g., the bully or the victim). Biases such as social desirability and retrospective recall issues may occur. Moreover, the likelihood of under-reporting adolescents’ school bullying behavior was also possible. Future research should include other bullying behavioral assessments such as peer and teacher nominations and behavioral observations as supplementary measures to validate self-reported findings (e.g., Espelage and Holt, 2001; Espelage and Swearer, 2003).

With growing recognition that bullying is a complex phenomenon that is influenced by multiple factors, past research has been examined within a sociological framework. This study enhanced the understanding of this social phenomenon from a psychological perspective by examining the risk-taking pattern associated with bullies, victims, and bully victims. This analysis will aid in not only comprehending the mechanism of how these groups respond to risk taking, but also lead to improved practice for prevention and intervention. Overall, bullying behavior is the result of the interaction of multiple causes and factors, and it is too early to believe that the identification of various risk-taking patterns or specific intervention that target their unique risk-taking processes would be a cure-all solution for the elimination of bullying or victimization in these groups. However, this study provides a basis for future studies to adopt a holistic approach in investigating the link between bullying and risk-taking behavior.