The relationship between peer victimisation, self-esteem, and internalizing symptoms in adolescents: A systematic review and meta-analysis

Background Peer victimisation is common and predicts increased internalizing symptoms. Low self-esteem, which is associated with both greater peer victimisation and higher internalizing problems, may help explain why victimised adolescents experience greater internalizing symptoms. Objectives of the present research were to establish the relationships between peer victimisation, internalizing problems, and self-esteem, and to test whether self-esteem mediates the effect of victimisation on internalizing symptoms. Methods We conducted a systematic literature search in Psychinfo, ERIC, Web of science, and Pubmed, following PRISMA guidelines. Inclusion criteria were: age 10–18 years; empirical studies that measured a) internalizing symptoms, b) self-esteem, and c) peer victimisation or bullying; design was either longitudinal or cross-sectional with a comparison group. Quality assessment were conducted using the Newcastle–Ottawa Quality Assessment Scale. We conducted random effects models and a meta-mediation analysis, with self-esteem acting as a mediator between peer victimization and internalizing symptoms. Results Sixteen papers with a total of N = 35,032 (53% female) participants met the criteria for inclusion in the meta-analysis. The meta-analysis demonstrated an association between peer victimisation and both high internalizing problems (r = .31, CI 95 = .26 to.36) and low self-esteem (r = -.25, CI 95 = -.29; to -.22), and between low self-esteem and high internalizing problems ((r = -.38, CI 95 = -.42 to -.33), as well as an indirect effect of peer victimization on internalizing symptoms via self-esteem (ß = .10, CI lower = .07, CI upper = .13). Conclusions Peer victimization, high internalizing symptoms and low self-esteem are all mutually related. Peer victimization partially mediates internalizing symptoms via self-esteem. Anti-bullying programmes may consider incorporating self-esteem building exercises in bully-victims. Limitations include high heterogeneity of results.


Introduction
Globally, the prevalence of anxiety and depressive symptoms among adolescents is high. More than a third show elevated symptoms of depression [1] and 12% show elevated anxiety symptoms [2], which puts them at risk to develop a clinical presentation. It is therefore vital to identify modifiable risk factors and mechanisms for the development of these symptoms to identify targets for prevention.
Peer victimisation, defined as experiences of repeated maltreatment from one or more peers over time [3], constitutes one of these factors and is experienced by approximately 17% of children and adolecents [4], and this can have detrimental effects on mental health as well as affect a young persons' self-esteem. Meta-analytic evidence points towards a causal relationship between experiences of peer victimisation and the development of internalising symptoms [5]. This relationship seems to be bi-directional however, with internalising symptoms also putting children at risk of victimisation [6,7].
The link between experiences of victimisation and the development of internalising symptoms could be explained by its impact on self-esteem. The development of self-esteem in children is influenced by the level of available social support through family and peers. Changes in the level of support are directly related to intra-individual changes of a child's self-esteem [8]. This bi-directional relationship has been confirmed in a recent meta-analysis [9]. Bullying victimisation could indicate the absence of a good peer support network at school. Indeed, it has been shown that children who are continuously bullied have fewer friends at school [10].
While meta-analytic evidence confirms the negative association between bullying victimisation and self-esteem [11], a large longitudinal study in Chinese school children showed that lower self-esteem mediated the relationship between bullying victimation and internalising symptoms [12]. This would be consistent with the vulnerability model which assumes that low self-esteem contributes to depressive symptomatology [13].
Overall, current evidence suggests an interplay of victimisation experiences, lower self esteem and internalising symptoms. While meta-analyses have examined links between individual variables, no meta-analysis has been conducted which integrates all three variables of interest; bullying victimisation, self-esteem and internalising symptoms.
Our first aim was therefore to conduct a systematic review and meta-analysis to establish the relationship between peer victimisation, self-esteem and internalizing problems. Specifically, we aimed to investigate whether self-esteem is related to internalizing symptoms of adolescents aged between 10 and 18 who have experienced victimisation, and to gain an understanding of the relationship between victimisation and self-esteem and victimisation and internalizing symptoms independently. Our second aim was to assess whether self-esteem mediates the relationship between peer victimization and internalizing symptoms across studies.
Our first aim was to conduct a systematic review and meta-analysis to establish the relationship between peer victimisation, self-esteem and internalizing problems. Specifically, we aimed to investigate whether self-esteem is related to internalizing symptoms of adolescents aged between 10 and 18 who have experienced victimisation, and to gain an understanding of the relationship between victimisation and self-esteem and victimisation and internalizing symptoms independently. Our second aim was to assess whether self-esteem mediates the relationship between peer victimization and internalizing symptoms across studies.

Eligibility criteria
We conducted a systematic review following the preferred reporting items in systematic review and meta-analysis (PRISMA) guidelines (Fig 1) [14]. We screened all studies using pre-defined inclusion and exclusion criteria. Inclusion criteria were: age 10-18 years; empirical studies that measured a) internalizing symptoms (depression, anxiety), b) self-esteem, and c) peer victimisation or bullying; design was a) longitudinal or b) cross-sectional with a comparison group. We included studies conducted in languages other than English. Exclusion criteria were age > 18, no control group or longitudinal design, no standardised measures of internalizing symptoms and self-esteem, did not assess school bullying (e.g. cyberbullying).

Study selection
After removing duplicates, 982 articles remained, which were screened independently by V. M. and S. S. Of these, 139 full-texts were screened. Finally, N = 22 papers were included in the qualitative synthesis. Two of these studies included the same sample [15,16]. Only the baseline measures have been published in a peer-reviewed journal and thus we only included baseline measures in the meta-analysis [16]. Five studies reported values that could not be transformed into r [17][18][19][20][21], and were therefore excluded from the meta-analysis. Therefore, 16 studies were included in the meta-analysis (Fig 1).

Quality assessment
We conducted a quality assessment of the remaining 21 papers using an adapted version of the Newcastle-Ottawa Quality Assessment Scale for Cohort studies [22]. VM and SS conducted quality ratings independently, conlficts were later resolved by discussion with VB. This assessment was originally developed to measure the quality of non-randomised studies. We assessed quality with regard to a) the selection of the participants, b) the comparability of the groups, and c) the measurements used, i.e., whether the scales have been validated (e.g., standardised self-report and teacher reports, such as the Beck Depression Inventory [23], the Hospital Anxiety and Depression Scale [24], and the Social Anxiety Scale for Adolescents [25]). A rating of at least 4 out of 7 stars indicated acceptable quality. All papers had at least 4 stars (median = 5), therefore, no papers were excluded due to poor quality (see S1 Table for ratings of each paper).

Statistical analysis
From each study, we extracted: characteristics of the sample (i.e., country where the study was conducted, sample size, age and sex ratio [where available]) and the measure victimisation, self-esteem, and internalizing symptoms (Table 1). Data was extracted by VM and VB independently and compared. Conflicts were resolved by discussion. We report r as a measure of effect size (see Table 1). If r was not originally reported, it was calculated it from means and    standard deviations where possible. Where papers reported correlations at more than one time point, we averaged across these correlation for the purpose of the meta-analysis. Not all studies assessed the association between self-esteem and internalising symptoms or only reported cross-sectional correlations for those two variables, leading to fewer studies veing included regarding the relationship between self-esteem and internalizing symptoms, and therefore the meta-mediation was restricted to 7 studies (N = 17380).
We computed weighted summary measures for the effects of victimisation on self-esteem and internalizing symptoms, and for the association between self-esteem and internalizing symptoms, using Comprehensive Meta-Analysis software [46]. Some studies reported effects sperately for males and females, and where this was the case, we also included male and female samples separately. Synthesised effects are reported as r with 95% Confidence intervals (CI). The Eggers test was used to explore possible publication bias.
Sensitivity analyses were conducted excluding studies reporting statistic other than r or mean and SD, and one study excluding non-victims from their final analysis. Heteroneheity due to mean age (N = 11 studies) was explored using meta-regression in the relationship between peer victimization and self-esteem and peer victimization and internalizing symptoms. Not enough studies could be included to conduct a meta-regression for the relationship between self-esteem and internalizing symptoms. Not enough studies reported results by gender, so that no meta-regression was possible using gender.
A meta-mediation model was conducted in R, using the metaSEM package [47], where selfesteem mediated the relationship between peer victimization and internalizing symptoms (N = 7 studies included). All meta-analyses were conducted using random-effect models to account for heterogeneity in the effect sizes between studies.
Seven studies assessed depression only, one study assessed anxiety only, five studies assessed depression and additional constructs, such as anxiety, social anxiety or loneliness. Four studies assessed internalizing symptoms (Table 1). Eight studies made group comparisons, nine studies assessed peer victimisation continuously, one of these studies assessed cumulative peer victimisation over time. One study encompassed a non-victimised group but excluded the group from their analysis; this study was removed for the sensitivity analysis. Eight studies reported correlations, six studies reported means and standard deviations, one study reported odds ration for the relationship between victimisation and self-esteem and standardised means for the relationship between internalizing symptoms and victimisation.

Victimisation, self-esteem, and internalizing symptoms
The meta-analysis showed that victimization was significantly associated with higher internalizing symptoms across studies (r = .31, CI 95 = .26 to.36, p < .001; Fig 2). The results were highly heterogeneous (Q = 381.63, p < .001, I 2 = 95). The Egger's test was not significant (p = .343), indicating that the funnel plot is symmetrical and publication bias is unlikely to account for the results (S1 Fig). A meta-regression entering mean age as a covariate for those studies that reported mean age showed a non-significant result (ß = .003, CI95 = -.04 to.05, p = .880).
A meta-regression entering mean age as a covariate for those studies that reported mean age showed a non-significant result (ß = .03, CI95 = -.01 to.01, p = .106).
The meta-mediation model showed a significant indirect effect of peer victimization on internalizing symptoms via self-esteem ß = .10, CI lower = .07, CI upper = .13; for direct effects see Fig 5).

Sensitivity analysis
The sensitivity analysis showed that the results were robust (S4-S6 Figs).

Discussion
The results of this meta-analysis show that individuals who have been victimised by their peers have higher internalizing problems and lower self-esteem, while lower self-esteem is also associated with higher internalizing problems. Additionally, a small but significant indicrect effect  showed that self-esteem mediates the relationship between peer victimization and internalizing symptoms. The results clarify that one way in which bullying can impact internalizing symptoms is by affecting self-esteem, although the size of the effect indicates that other factors might play a role as well. The findings confirm original research pointing to self-esteem as a mediator of the effect of peer victimization on depression and anxiety [12,[15][16][17]53,54].
The finding that peer victimization and internalizing symptoms are closely linked has been recognized in a wealth of literature [15,52,55,56] and is in line with findings from other metaanalyses showing that peer victimization has a small effect on later internalizing symptoms and that young people with internalizing symptoms are at high risk to experience peer victimization, irrespective of age and gender [6,56]. School programms addressing peer victimization have been shown to have a small to moderate positive effect [57]. The most promising programmes involved the establishment of an anti-bullying school policy and elements of peer counselling and emotional control [57].
Our results are congruent with previous meta-analyses that showed a small effect of peer victimization on internalizing symptoms [6], however, the results showed that the direct effect of peer victimization on self-esteem was larger than on internalizing symptoms, with a medium effect size. This is important to considers. Self-esteem, in turn, has an effect on social relationships [9], mental, and physical health [58]. A medium relationship between self-esteem and internalizing symptoms was also reflected in the results of this meta-analysis.
The direct effect between self-esteem and dinternalizing symptoms is consistent with the vulnerability model of depression, which highlights low-self-esteem as a risk factor for the development of depression [13]. However, the vulnerability model might be too simplistic as it does not consider the influence of shared risk factors or mediators. Our results are also consistent with the Identity Disruption Model [59], which assumes that adverse experiences are associated with negative mental health outcomes through disruptions of identity development (i.e., an unclear/ unstable sense of self). While this meta-analysis did not focus on the concept of self-concept clarity, this variable is highly correlated with self-esteem [60] and might be affected by peer victimization.
Our results highlight bullying victimization and low self-esteem as possible targets for intervention or prevention approaches. Most anti-bullying programmes focus on reducing aggressive acts by installing consequences, removing the reinforcing factors on the environment (e.g. peers as spectators), and to intervene in bullying acts immediately. Some programmes involve talks with bullies, victims, and their parents, and assertiveness traiuninfgwith the bully-victim  [61][62][63]. Overall, the focus seems to be on perpetration. Our results suggest that it might be useful to include interventions targeting the bully-victims as well as the perpetrators. While it might be difficult to treat internalizing symptoms directly, bully-victims may benefit from programme elements that help them recognize the effects bullying can have on self-esteem and help bully-victims build or re-build self-esteem, not only assertiveness. Our results indicate that the relationship between low self-esteem and internalizing symptoms became weaker with higher age, suggesting that interventions would be particularly important for younger children.
Evidence for the effectiveness for this type of interventions stems from research in adults: A cognitive-behavioral group intervention that specifically targeted self-esteem, the Overcoming Low Self-Esteem Intervention, demonstrated an increase in self-esteem, and a decrease in internalizing symptoms at the three months follow-up in adults [64]. Future research might assess whether building self-esteem in bully-victims might reduce both internalizing symptoms and the risk of future peer victimization. Implementation of similar programmes for the school context might be promising. The UK government recently established the role of Education Mental Health Practitioners, who deliver low-intensity cognitive-behavioral therapy within a school context [65]. While the role focuses on the treatment of mental health problems, the evidence from this meta-analysis would also support the establishment of self-esteem building groups in schools that could be led by people in this specialized role.

Strengths and limitations
Our research had a number of strengths. This was the first comprehensive meta-analysis to simultaneously explore the associations between peer victimisation, self-esteem and internalizing problems. Further, only longitudinal studies or studies with a control group were included in the analyses and the majority of studies investigated large groups of participants.
Cyberbullying was not included into the analysis as a focus was put on bullying that occurred in schools. However, cyberbullying is related to self-esteem and internalizing symptoms in a similar way. A cross-sectional study on Vietnamese students demonstrated that selfesteem mediated the relationship between experiencing cyberbullying and depression symptoms [66], and a study on Italian adolescents found that lower self-esteem was related to a greater risk of being cyberbullied [67]. Moreover, a longitudinal study on Chinese adolescents found that cyberbullying positively predicted internalizing problems [68].
A further weakness of this study is that is was not possible to pinpoint the source of heterogeneity in the results. Only three studies reported results by gender, therefore, even though the results were not significant, it would be difficult to interpret the absence of a significant effect for so few studies. Gender can affect the relationship between peer victimization and internalizing symptoms [69], however, we could not show that the mediation via self-esteem was affected by gender. Given the high heterogeneity, results should be interpreted with caution. It should however be pointed out that almost all results across all outcomes were significant and change occurred in the same direction. While the high heterogeneity makes it harder to determine the true effect size, there is no doubt that there is a significant relationship between the variables. Lastly, this review was not registered and we did not write a protocol.

Conclusions
Our meta-analysis showed that peer victimisation poses a risk factor for developing low selfesteem and internalizing problems, and that peer victimization increases internalizing problems partially via lower self-esteem, although with a small effect size.

S1 Fig. Funnel plot for the relationship between victimization and internalizing symptoms.
The eggers test was non-significant, indicating that a publication bias is unlikely to account for the results.