Effects of sports intervention on aggression in children and adolescents: a systematic review and meta-analysis

Search strategy

We performed a systematic search of Pubmed, Web of Science, Cochrane library, Embase and Scopus databases from database inception to 12 October 2022 for all studies written in English. The search strategy was designed by Yahui Yang and Fengshu Zhu by an initial scoping review of the literature. Studies were identified by using all possible combinations of the following groups of search terms: (a) “adolescent” OR “teens” OR “youth” OR “teenager” OR “juvenile” OR “young” OR “minor”; (b) “physical training” OR “sport” OR “exercise” OR “athletics”; (c) “intervention” OR “behaviour change” OR “prevention” OR “experiment” OR “program” OR “reduction” OR “evaluation” OR “strategy” OR “effect”; (d) “aggression” OR “bullying” OR “violence” OR “assaultive behavior” OR “atrocity” OR “physical assault” OR “fighting”. The specific search was amended as necessary for each database to account for different search functionalities. The reference lists of retrieved articles and grey literature were searched to detect studies potentially eligible for inclusion.

Inclusion and exclusion criteria

Studies were included if they met the following PICO criteria: (1) included typically developing children and/or adolescents (Population); (2) examined different sports including school physical education programs (Intervention); (3) included anactive/inactive comparator (Comparison) and (4) examined associations with aggression (Outcomes). Studies were excluded if they focused on populations with develop-mental disorders (e.g., Down syndrome).

Study selection

Search results were imported into Endnote to remove duplicates. Yahui Yang and Hao Zhu screened the titles and abstracts of the retrieved articles independently to remove irrelevant articles. Then the same reviewers independently screened remaining articles in full to determine the final included studies. Disagreements were resolved by consensus or consultation with Fengshu Zhu.

Data extraction

One reviewer extracted specific characteristics from included studies, including country, study design, paticipants characteristics (age, gender), sample size, intervention programme characteristics (name, type, duration, frequency), comparison programme and outcome variables. Shachar et al. (2016) reported mean and standard deviation (SD) of baseline and change-from-baseline, the reviewer calculated the final mean and standard deviation according to Cochrane Handbook version 5.1.0 (Higgins & Green, 2011). Another reviewer confirmed the content.

Outcomes

The primary outcome was aggression scores. The secondary outcomes were other externalizing behaviors of aggression, including hostility, anger, delinquent acts (including crimes of varying severity levels such as gang fights and extortion, but also minor theft and nuisances), attitude towards violence (ATV) and provocation/bullying scores. If outcomes were reported for more than one time point, we extracted results closest to post-intervention (Fung & Lee, 2018). If two or more measurement tools were used, we referred to a previously described hierarchy of outcome measures (Fung & Lee, 2018). If physical aggression and verbal aggression scores were reported concurrently, we extracted the physical aggression scores (Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002).

Risk of bias assessment

Yahui Yang and Hao Zhu assessed risk of bias of randomised controlled trials (RCTs) using the Cochrane collaboration tool 2.0 (Sterne et al., 2019) and assessed risk of bias of quasi-experimental studies using the JBI Critical Appraisal Checklists for Quasi-Experimental Studies (The Joanna Briggs Institute, 2016) independently. Discrepancies were resolved by consensus or deliberation with Fengshu Zhu.

Data analysis

All analyses were carried out using the Review Manager 5.3 Software. We summarised aggression, hostility and anger scores using SMDs. Summary estimates with 95% confidence intervals were pooled using DerSimonian-Laird random effects model or fixed effects model according to between-study heterogeneity (DerSimonian & Laird, 1986). The heterogeneity was estimated using I², considering I² values of <25%, 25–50, and >50% as small, medium, and large amounts of heterogeneity, respectively (Higgins & Thompson, 2002). Subgroup moderator analyses were conducted to determine whether results differed according to intervention duration and sport type. Sensitivity analyses were used to explore the impact of individual studies. A narrative synthesis of the results was carried out using descriptive statistics in order to summarize characteristics of the studies where data cannot be extracted (Ioannidis, Patsopoulos & Rothstein, 2008).

Study characteristics and risk of bias

Following the screening process, 15 studies (Fung & Lee, 2018; Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002; Mehralian et al., 2022; Rosa et al., 2021; Harwood-Gross et al., 2021; Blomqvist, 2020; Wade et al., 2018; Setty, Subramanya & Mahadevan, 2017; Hortiguela, Gutierrez-Garcia & Hernando-Garijo, 2017; Park et al., 2017; Pels & Kleinert, 2016) were deemed eligible for inclusion in this review. The detailed screening flow is shown in Fig. 1. Included studies were published between 2002 and 2022. Six studies were RCTs and nine studies were quasi-experimental studies. Twelve studies reported aggression outcomes, five studies reported hostility and anger outcomes, two studies reported delinquent acts outcomes and attitude towards violence, and one study reported provocation/bullying outcome. Three studies were comparison between two sports events with no control group. Pels & Kleinert (2016) carried out a single experiment and interventions of other studies varied in duration from 4 weeks to 1 year (see Table 1 for details).

Six RCTs and nine quasi-experimental studies were all identified as “moderate quality”. The assessment results are shown in Tables 2 and 3.

Impact of interventions on aggression

The overall mean effect size of nine included studies (Fung & Lee, 2018; Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002; Mehralian et al., 2022; Wade et al., 2018; Park et al., 2017) was SMD = −0.37, 95% Confidence Interval (CI) −0.69, −0.06, indicating that sport interventions reduced aggression compared to a control group (P = 0.020). There was significant heterogeneity between effect sizes between studies (I² = 88%, P < 0.001) (Fig. 2).

According to the types of sport, non-contact sports (Trajković et al., 2020a; Shachar et al., 2016; Mehralian et al., 2022; Wade et al., 2018; Park et al., 2017) were associated with lower aggression (SMD = −0.65, 95% CI [−1.17 to −0.13], P = 0.020; I² = 92%). High-contact sports (Fung & Lee, 2018; Trajković et al., 2020b; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002) were not associated with lower aggression (SMD = −0.15, 95% CI [−0.55 to 0.25], P = 0.470; I² = 79%) (Table 4).

According to the intervention duration, when intervention duration <6 months (Fung & Lee, 2018; Carraro, Gobbi & Moè, 2014; Mehralian et al., 2022; Park et al., 2017), sport interventions was associated with lower aggression (SMD = −0.99, 95% CI [−1.73 to −0.26], P = 0.008; I² = 90%). When intervention duration ≥6 months (Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Reynes & Lorant, 2002; Wade et al., 2018), sport interventions was not associated with lower aggression (SMD = −0.08, 95% CI [−0.44 to −0.28], P = 0.660; I² = 87%) (Table 4).

When the impact of individual studies was examined by removing studies from the analysis one at a time, we observed that when Mehralian et al. (2022), Park et al. (2017), Shachar et al. (2016) and Carraro, Gobbi & Moè (2014) been removed, the pooled results became insignificant (P ≥ 0.05) (Table 5). However, these studies did not share any specific characteristics.

Harwood-Gross et al. (2021) only provided the mean change-score comparison between martial arts training and controls on aggression so that it was not included in the meta-analysis. The aggression scores in both groups were increased, and the difference was not significant (P = 0.85).

Blomqvist (2020) compared effects of Mixed Martial Arts (MMA) intervention and Brazilian Jiu-Jitsu (BJJ) intervention on aggression. The results showed that there was no significant main effect of aggression as a result of training (P = 0.100). However, the interaction between aggression and sport was significant (P < 0.001). Whereas MMA practitioners slightly increased their levels of aggression, BJJ practitioners reduced theirs. Pels & Kleinert (2016) reported a significant reduction of aggressive feelings was found for participants exercising individually in the rowing condition compared with the individual combat exercise condition.

Impact of interventions on hostility

The overall mean effect size of five included studies (Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002) indicated that sport interventions was associated with lower hostility (SMD = −0.29, 95% CI [−0.41 to −0.17], P < 0.001; I² = 0%) (Fig. 3).

When the impact of individual studies was examined by removing studies from the analysis one at a time, we observed that the pooled results estimate remained consistent.

Impact of interventions on anger

The overall mean effect size of five included studies (Trajković et al., 2020a; Trajković et al., 2020b; Shachar et al., 2016; Carraro, Gobbi & Moè, 2014; Reynes & Lorant, 2002) indicated that sport interventions was not associated with lower anger (SMD = −0.26, 95% CI [−0.56 to 0.03], P = 0.08; I² = 78%) (Fig. 4).

Sensitivity analysis showed that when Reynes & Lorant (2002) been removed, the heterogeneity became small (I² = 0%) and the pooled result became significant (P < 0.001) (Table 5).

Impact of interventions on delinquent acts

Blomqvist (2020) indicated that both MMA and BJJ intervention groups reduced criminal behaviour moderately (P = 0.030). Fung & Lee (2018) found that Chinese martial arts group had light decrease in delinquent behavior than did the physical fitness training group, but there was no significant fixed effects of training were found in delinquent behavior (P = 0.760).

Impact of interventions on attitude towards violence

Setty, Subramanya & Mahadevan (2017) and Hortiguela, Gutierrez-Garcia & Hernando-Garijo (2017) reported effects of sport intervention on adolescents’ attitude towards violence. Setty, Subramanya & Mahadevan (2017) showed a significant change in both yoga and control groups in self-reported ATV, pre- and post-intervention ( p < 0.05). But the mean change in the yoga group is 39.59%, compared to 7.51% in the control group, indicating significant improvement. Hortiguela, Gutierrez-Garcia & Hernando-Garijo (2017) reported the results of two dimensions of ATV—unjustified violence and violence linked to self protection. The unjustified violence and the violence linked to self protection fell from high to medium in the judo and capoeira teaching units, significant difference with large effect sizes were found between the pre-test and the pos t-test in unjustified violence (P = 0.021) while there was no difference in the control group.

Impact of interventions on provocation/bullying

Rosa et al. (2021) carried out judo and ball games among children and adolencents. A significant improvement in the domain of provocation/bullying was observed after the interventions, with judo increasing 18.1% and ball games increasing 4.1%. In other words, the participants felt safer and more confident about other people’s negative attitudes.