Mar 16 2020. https://doi.org/10.1080/15374416.2020.1731820
ABSTRACT
Objective: Bullying affects approximately a quarter of schoolchildren and is associated with numerous adverse outcomes. Although distinct risk factors for bullying and victimization have been identified, few studies have investigated the genetic and environmental underpinnings of bullying and victimization. The aims of this study were twofold: first, to examine the contributions of genetic and environmental factors to bullying and victimization, and second, to analyze whether the KiVa antibullying program moderated the magnitude of these contributions by comparing estimates derived from the KiVa versus control groups.
Method: The sample comprised students from schools that participated in the evaluation of the KiVa antibullying program in Finland during 2007–2009. Bullying and victimization were measured using peer nominations by classmates. The sample for the twin analyses comprised of 447 twins (107 monozygotic and 340 dizygotic twins) aged 7–15.
Results: Genetic contributions accounted for 62% and 77% of the variance in bullying and in victimization at pre-intervention, respectively. There was a post-intervention difference in the overall role of genetic and environmental contributions between the intervention and the control group for bullying and victimization, with non-shared environmental effects playing a lesser role (and genes a larger role) in the intervention than in the control group context.
Conclusions: This study replicates previous findings on the genetic underpinnings of both bullying and victimization, and indicates that a school-based antibullying program reduces the role of non-shared environmental factors in bullying and victimization. The results indicate that prevention and intervention efforts need to target both environmental and (heritable) individual level factors to maximize effectiveness.
Discussion
The present study aimed at examining the magnitude of genetic and environmental contributions to bullying and victimization, in a Finnish sample aged 7–15. In addition, the aim was to test whether the KiVa antibullying program would moderate these contributions. As expected, we found significant and substantial genetic contribution for both bullying and victimization in general, as well as a moderation through the antibullying intervention program of the ratio between genetic and non-shared environmental factors.
Broad sense heritability (H2) for bullying was estimated at 62% (A = 23%, D = 39%) for pre-intervention, with non-shared environmental factors accounting for the rest of the variance. The dominance component has not been reported previously, however, twin analyses have limited statistical power to distinguish between additive and non-additive genetic effects, and estimates of the broad sense heritability are more stable (Eaves, 1972). In that sense, the broad sense heritability is very similar to the previous findings by Ball et al. (2008; 61% heritability for bullying), Veldkamp et al. (2019; ~70%), and Dunbar (2018; 55%), even though different informants were used (i.e. combined parents’ and teachers’ ratings in Ball et al., 2008, and teachers’ ratings in Veldkamp et al., 2019). The present study is the first one to have estimated the heritability of bullying using peer nominations. The heritability estimates for bullying are in line with those found for antisocial behavior and aggression more generally (Brendgen et al., 2008; Burt, 2009; Polderman et al., 2015; Porsch et al., 2016).
Broad sense heritability for pre-intervention victimization scores was quite substantial at 77% (A = 50%, D = 27%), with non-shared environmental influences accounting for the rest of the variance. Again, this estimate is in the range of several previous studies (73% for Ball et al., 2008; 71–77% for Bowes et al., 2013; 70% for Connolly & Beaver, 2016; 67% for Törn et al., 2015; ~65% for Veldkamp et al., 2019). Other studies have reported lower heritability estimates: Brendgen et al. (2008) estimated the heritability of victimization to 0% in one study, and 26% in a later study (Brendgen et al., 2011), and Shakoor et al. (2015), Silberg et al. (2016), and Dunbar (2018) reported heritabilities of 35%, 45% and 48%, respectively. However, these variations could be accounted for by a variety of reasons. For instance, both Eastman et al. (2018) and Veldkamp et al. (2019) found that heritability estimates vary depending on the type of victimization; they found higher heritabilities for physical (42% and 70%, respectively) versus social/relational (0% and 55%) and property-related victimization (0%, only in Eastman et al., 2018). In addition, the twin studies also seem to differ, for example, with regards to informants used and age of participants. The use of single informants may give rise to unstable or low estimates (due to larger measurement error), especially for self-ratings in younger children. When multiple informants were used in a latent model of peer victimization and rejection, a significant and substantial contribution of genes was seen for twins from Kindergarten to grade 4 (H2 = 73–94%; Boivin et al., 2013a), which is in line with our results. Eastman et al. (2018) compared the genetic and environmental estimates derived from children (ages 9–14), versus adolescents (ages 15–20), and found differences in the magnitude and structure of genetic influences. The genetic contribution to victimization indicate that heritable characteristics in the child could evoke a negative reaction from peers and thus play a role in the likelihood of being bullied by others, a form of evocative rGE (Boivin et al., 2013a). Such characteristics could include reactive-impulsive aggression (Boivin et al., 2013b), but also depression, ADHD, risk taking, high BMI or low intelligence, as Schoeler et al. (2019) recently showed, through a polygenic risk score approach, that genetic risks for these characteristics were related to victimization. These modest associations need replications, as well as confirming evidence that they work through the mediating role of the putative child characteristics. Further evidence for the rGE hypothesis is also found in twin studies indicating partial overlap between the genes influencing victimization and those for social anxiety (Silberg et al., 2016), as well as depression/anxiety (Connolly & Beaver, 2016).
Significant genetic contributions to victimization and bullying were also found post-intervention, after pre-intervention levels were accounted for. This was true for both the intervention and the control group. Since pre-intervention levels were regressed out from the post-intervention levels, direct comparisons between heritability estimates between the time-points cannot be made. Crucial to the objective of the present study were the findings regarding the moderating role of the KiVa intervention. Moderation by the KiVa program would be indicated if the post-intervention estimates differed between the control and the intervention group after controlling for pre-intervention levels. This was true for both bullying and victimization. For both phenotypes, the general ADE pattern of estimates differed across groups (intervention vs control), essentially reflected by lower E-estimates in the intervention (bullying 42%, victimization 28%) compared to the control group (bullying 64%, victimization 63%), and thus leaving more room for genes to account for the remaining variance. The significantly smaller E-estimates in the intervention group could reflect a possible leveling out of environmental risk due to the KiVa intervention (e.g. through changing the behavior of bystanders and making bullying behavior less acceptable in the school setting), leaving a higher role to genetically influenced individual characteristics in that context. This pattern of findings is in line with the push hypothesis (Raine, 2002), which posits that genes will play a larger role in an environment freer of environmental risk factors. Evidence for such GxE findings have been found, for example, with respect to the moderating role of socioeconomic status on the genetic and environmental etiology of antisocial behavior (Tuvblad, Grann, & Lichtenstein, 2006) and the role of early adversity in physiological stress (Ouellet-Morin et al., 2008).
These results indicate the conditional nature of these environmental and genetic sources of individual differences, but they also point to the importance of providing contexts, through policies (e.g. early education) or intervention (i.e. KiVa), to create a more equitable social and learning environment for all children. When these interventions are successful in leveling out the environmental playing field, they may paradoxically identify individual factors, here genetic factors in the child, as more important for various outcomes. In doing so, they provide useful information in that they point to where the effort for change should be oriented. Specifically, even though the KiVa intervention is amongst the most effective antibullying interventions (Gaffney, Farrington, & Ttofi, 2019), it fails to stop all bullying. Our results indicate that genes play a significant role in accounting for post-intervention variance in bullying and victimization, and that the efficacy of the KiVa program might be enhanced by incorporating components targeting individual heritable characteristics. Previous research indicates that heritable individual characteristics that could evoke victimization from peers include mental health problems such as depression and anxiety (Connolly & Beaver, 2016; Schoeler et al., 2019; Silberg et al., 2016), ADHD, high BMI and low intelligence (Schoeler et al., 2019). Schoeler et al. (2019) suggested that one way to target such individual heritable characteristics with regards to victimization, could be to include components trying to reduce the stigma of mental health problems or other vulnerabilities such as high BMI, or to offer more support to children displaying internalizing or externalizing symptoms. Intervention components could either be universal (targeted at the entire school) such as in components aiming to change the environment to be less discriminatory (e.g., toward people with mental health problems, neuropsychological difficulties, or high BMI), or individual (e.g. support and/or interventions for students at risk). In addition, one should also keep in mind that interventions not specifically aimed at reducing bullying victimization or perpetration, but rather aimed at reducing characteristics that increase the risk for victimization or perpetration could, in turn, also reduce bullying.
With respect to bullying perpetration, less is known about the specific underlying heritable characteristics, but such could include, for example, callous-unemotional traits and/or conduct problems (Viding, Simmonds, Petrides, & Frederickson, 2009; Zych, Ttofi, & Farrington, 2019). It is important to identify at-risk children early, and intervene not only in the school context but also by means of parenting programs (Waller, Hyde, Klump, & Burt, 2018). Further research is needed to identify specific heritable characteristics related to both the risk of victimization and bullying, especially victimization and bullying that persists after intervention efforts. However, as we know that standard interventions are not helpful in all cases (see also Kaufman, Kretschmer, Huitsing, & Veenstra, 2018), school personnel should always follow up after taking action to stop bullying, check whether their intervention was helpful, and to take further action when needed.
A strength of this study is that bullying and victimization were measured with proportion scores derived from peer nominations. In addition to there being multiple reporters, classmates can be considered to be more up to date on what is happening in the school setting (Boivin et al., 2013a; Stassen Berger, 2007), than for instance teachers or parents. Another clear strength of our study is the RCT-design, which effectively controls for potential confounds such as those that might arise from unaccounted rGE. The participating schools can be considered representative of Finnish schools at the time of data collection. In addition, twins did not differ from non-twin individuals on bullying or victimization, suggesting that the results are generalizable to non-twin individuals.
A limitation is the relatively small sample size, especially with regards to the post-intervention comparison between the intervention and the control group. The pre-intervention estimates are likely more robust, as these used data from all twins. We decided to do separate analyses for the pre- and post-intervention data for two reasons. First, even though a multi-group approach is warranted for the post-intervention data, it was preferable to analyze the pre-intervention data in a single group for increased statistical power. Second, an extension of the Cholesky decomposition to a multivariate case (i.e. analyzing both pre- and post-intervention) is problematic in multi-group approaches (Neale, Røysamb, & Jacobson, 2006). A multivariate approach would, however, allow a comparison of the variance components between the time-points, and therefore, replication efforts with larger sample sizes finding a solution to the multivariate approach would be welcome. A possible limitation of the twin method is whether the equal environments assumption holds. It has, however, been tested in a number of studies and appears to be valid (e.g. Derks, Dolan, & Boomsma, 2006; Kendler, Neale, Kessler, Heath, & Eaves, 1993).