How can adolescent aggression be reduced? A multi-level meta-analysis

Highlights

• Psychosocial interventions are effective in reducing adolescent aggression.

• Interventions targeted at adolescents at risk are more effective.

• Shorter interventions are more effective than longer interventions.

• Problem-solving and behavioural practice are the most effective techniques.

• Including more techniques does not make interventions more effective.

Abstract

Aggressive behaviour among adolescents has significant social and economic costs. Numerous attempts have been made to intervene to reduce aggression in adolescents. However, little is known about what factors enhance or diminish intervention effectiveness. The present systematic review and meta-analysis, therefore, seeks to quantify the effectiveness of interventions to reduce aggressive behaviour in adolescents and to identify when and for whom such interventions work best. Sixteen databases were searched for randomised controlled trials that assessed interventions to reduce aggression among adolescents. After screening 9795 records, 95 studies were included. A multi-level meta-analysis found a significant overall small-to-medium effect size (d = 0.28; 95% CI [0.17, 0.39]). More effective interventions were of shorter duration, were conducted in the Middle East, were targeted at adolescents with higher levels of risk, and were facilitated by intervention professionals. Potentially active ingredients were classified using the Behaviour Change Technique Taxonomy. Behavioural practice and problem solving were components of more effective interventions targeted at the general population. Overall the findings indicate that psychosocial interventions are effective in reducing adolescent aggression. Future trials need to assess the effect of individual techniques and their combination to identify the key components that can reduce aggression in adolescents.

Introduction

Aggression among adolescents is a worldwide problem. For example, Craig et al. (2009) found that bullying is common in 40 countries across Europe and America. In the United States, Lynne-Landsman, Graber, Nichols, and Botvin (2011) found aggression increases through adolescence, with 51% of their sample presenting high levels of aggressive behaviour at the end of middle school. Aggressive behaviours during adolescence are associated with negative immediate and long-term outcomes for both victims and perpetrators. For example, victims of bullying have more mental health problems during adulthood than those who have not been bullied (Arsenault, 2017) and adolescents who display aggression are more likely to have drug problems, present depressive symptoms and be arrested as an adult (Hyde, Burt, Shaw, Donnellan, & Forbes, 2015; Rhoades, Leve, Eddy, & Chamberlain, 2016). In the US alone, the annual cost of serious aggression among adolescents is over $21 billion (National Center for Injury Prevention and Control (U.S.). Division of Violence Prevention, 2019).

Many interventions have been developed to prevent and reduce aggressive behaviour in adolescents. Previous reviews have concluded that these interventions are effective (see Appendix A for an overview of 38 previous systematic reviews and meta-analyses including adolescents). However, it is still not clear which specific characteristics and techniques included in these interventions are most effective among adolescents. Only six of these reviews have investigated what works specifically in adolescents: two meta-analyses and four systematic reviews. The meta-analyses have focused on specific interventions involving sports participation (Spruit, Assink, van Vugt, van der Put, and Stams, 2016) and positive youth development (Melendez-Torres et al., 2016), and although the systematic reviews included a broader range of interventions, they did not quantify their effects (Cox et al., 2016; Gavine, Donnelly, & Williams, 2016; Kelly, 2017; Limbos et al., 2007). Therefore, a meta-analysis that identifies what works to reduce aggression in adolescents is currently unavailable.

Despite the limitations of the previous systematic reviews, some have suggested that interventions aimed at those at greater risk of perpetrating aggression (i.e., targeted interventions) may be more effective than interventions aimed at the general adolescent population (i.e., universal interventions; Gavine et al., 2016; Limbos et al., 2007). The mechanisms underlying effectiveness in universal and targeted interventions might also be different and many reviews focused solely on universal or targeted interventions (see Appendix A). This is important because numerous reviews have concluded that behaviour training and social skills training are the most effective components of interventions that are targeted rather than universal (Fossum, Handegård, Martinussen, & Mørch, 2008; Molina, Dulmus, & Sowers, 2005; Mytton, DiGuiseppi, Gough, Taylor, & Logan, 2006; Özabacı, 2011; Wilson & Lipsey, 2007). Identifying which intervention components are effective in improving behaviour is valuable in guiding intervention optimisation. However, without a common language, it is difficult to describe and compare intervention components. For example, Wilson and Lipsey (2007) defined “behavioural strategies” as giving rewards and incentives, whereas Özabacı, 2011 characterised learning and practising behavioural responses as “behavioural strategies”. This limitation can be overcome by using a common framework or taxonomy. The present meta-analysis will use the Behaviour Change Technique (BCT) taxonomy version 1 (Michie et al., 2013) to identify the BCTs included in interventions and test which are effective in reducing aggression among adolescents. We will also test whether BCT effectiveness differs between universal and targeted interventions. The BCT taxonomy has been widely used to analyse interventions addressing many health behaviours such as diabetes care (Presseau et al., 2015) and physical activity (Cradock et al., 2017). The taxonomy includes 93 techniques such as feedback on behaviour, problem solving and adding objects to the environment that aim to change behaviour.

Duration is another characteristic that previous reviews focusing on adolescents have found as a significant moderator of effectiveness. Limbos et al. (2007) suggested in their systematic review that interventions that lasted 12 months or more were more likely to be effective than shorter interventions. However, other reviews have found that longer interventions are not more effective than shorter ones (Fagan & Catalano, 2013).

Systematic reviews and meta-analyses of interventions to reduce aggressive behaviour across children, adolescents and adults have suggested that the existence of other factors in addition to targeting and duration that might moderate intervention effectiveness. However, they have not considered adolescents as a group separate from children and/or adults. Therefore, there is a need to investigate whether these factors moderate the effect of interventions in adolescents as it is possible that interventions need to be specifically tailored for this target group (Yeager, Dahl, & Dweck, 2018).

One of the factors that might moderate the effectiveness of interventions to reduce aggression in adolescents is whether intervention is delivered individually or to a group. Group interventions have been found to be less effective with samples containing high proportions of boys (Sawyer, Borduin, & Dopp, 2015) and targeted interventions to be more effective when delivered individually than to a group (Smedler, Hjern, Wiklund, Anttila, & Pettersson, 2015; Wilson & Lipsey, 2007). The person delivering the intervention has also been found to moderate intervention effectiveness. For example, interventions delivered by a member of the research team were more effective than those delivered by mental health professionals in Sawyer et al.'s (2015) meta-analysis, and interventions delivered by intervention specialists were more effective than those delivered by teachers in Park-Higgerson, Perumean-Chaney, Bartolucci, Grimley, and Singh (2008)'s quantitative review. School-based interventions have been found to be more effective in high school than in middle school (Hahn et al., 2007) and to be associated with the amount of training the teachers received (Ttofi & Farrington, 2011). The size of the effect of interventions to reduce aggressive behaviours has also been found to vary depending on how the outcome is assessed. For example, both Grove, Evans, Pastor, and Mack's (2008) and Sawyer et al.'s (2015) meta-analyses found that the reduction in aggressive behaviour was more pronounced when the outcome was measured via official records. In the former, that effect was significantly larger than the effect found for self-reports, and in the latter, it was significantly larger than the effect when the outcome was assessed via parent reports. Finally, Ttofi and Farrington (2011) found that anti-bullying school-based interventions evaluated before 2003 were more effective than those evaluated more recently, and interventions implemented in Norway were more effective than those implemented elsewhere. These findings suggest that children from different cultures may vary in the extent to which they engage with interventions to reduce aggressive behaviour and that they may have become less receptive of such interventions over time. In terms of informing our study, these findings indicate that it is important to test whether date of publication and geographical location also moderate the effectiveness of interventions.

Previous studies of the effectiveness of interventions for aggression have used traditional meta-analysis, which is limited by the assumption of independence of effect sizes that prevents more than one effect size from being included from each study. The present study applies a multi-level meta-analysis, which relaxes that assumption. Multi-level meta-analysis allows all effect sizes from studies that report multiple comparisons to be included as the modelling accounts for the dependence of effect sizes nested within studies (Assink & Wibbelink, 2016). Thus, information is maximized and analysis power improved.

In the present study, we aim to identify what works for whom in the reduction of aggressive behaviour. In order to do this, we classify components of the interventions using the BCT taxonomy and test which BCTs are most effective for universal and for targeted interventions separately. In addition, we examine the moderators of intervention effectiveness highlighted in this introduction with the objective of confirming their moderation effect in interventions with adolescents. The moderators that will be examined include characteristics of the intervention such as duration, characteristics of the participants such as gender and aspects of the study design such as outcome informant.