Links Between Externalizing And Internalizing Symptoms And Peer Victimization: The Role Of Rumination And Self-Efficacy

Abstract

Objective:

Bullying is a serious problem among school-age children, contributing to both internalizing and externalizing difficulties. The existing literature does not place much emphasis on the factors explaining the link between victimization and psychological difficulties. This study looked into the roles of rumination and self-efficacy in this relationship.

Method:

Data was collected from 362 children (202 girls; Mage = 13.1, SD = 1.16) over two points in time, separated by 6 months.

Results:

Findings revealed a main effect of bullying status (being a victim of bullying vs. being non-involved) on rumination, self-efficacy, and internalizing/externalizing symptoms. Additional analyses explored different existing profiles (non-involved who stayed non-involved, new victims, escaped victims, and continuous victims) in terms of their psychological difficulties, ruminative responses, and self-efficacy. Mediation analyses show that victimization at Time 1 predicts internalizing difficulties at Time 2, but this effect is not mediated by rumination or self-efficacy. Conversely, victimization indirectly predicts externalizing symptoms at Time 2 via rumination. Moreover, internalizing symptoms at Time 1 predict victimization at Time 2 through both rumination and self-efficacy, while externalizing symptoms at Time 1 predict victimization at Time 2 via self-efficacy.

Conclusions:

The findings provide practical insights for those working with victims of bullying and also lay the foundation for future research.

Introduction

Bullying is a pervasive issue among school-age children, characterized by a power imbalance where perpetrators intentionally harm victims repeatedly (Olweus, 1993). Extensive research shows that victimization is bidirectionally linked with both internalizing (e.g., anxiety, depression) and externalizing symptoms (e.g., aggression, conduct problems), suggesting a cycle of vulnerability (Reijntjes et al., 2010, 2011). Despite the well-documented consequences of bullying, prevention programs have only modest effects on reducing victimization (Ttofi & Farrington, 2011), and it remains unclear which psychological processes explain why some victims develop lasting difficulties while others do not (Juvonen & Graham, 2014). Moreover, internalizing and externalizing symptoms themselves increase risk for subsequent victimization, creating developmental cascades (Beeson et al., 2020; Kochel et al., 2012; Vaillancourt et al., 2013). Yet the literature has paid limited attention to explanatory mediators of these cascading effects, which could help identify targets for more effective interventions. To address this gap, we adopt a process-based perspective (Kinderman, 2005, 2009; Philippot et al., 2018, 2024), which highlights a range of transdiagnostic processes involved in mental health difficulties. Among these, rumination and self-efficacy stand out as particularly relevant for understanding the bidirectional victimization–symptoms cycle. Although both have been individually studied, their combined role remains underexplored.

The role of rumination

Rumination, or the repetitive focus on distressing experiences (Nolen-Hoeksema, 1991), may initially serve as an avoidance strategy but ultimately exacerbates affective, cognitive, motivational, and social difficulties (Baeyens, 2016; Kirkegaard Thomsen, 2006). Numerous studies link rumination to both internalizing and externalizing symptoms in adolescence (Aldao et al., 2016)..

Victimization, as a stressful life event, is consistently associated with higher rumination (Arató et al., 2020; Camacho et al., 2021; Garnefski & Kraaij, 2014; Potard et al., 2022; Spyropoulou & Giovazolias, 2023; Stange et al., 2014), which may not be surprising given that victims often rely on self-blame strategies (Tenenbaum et al., 2011). Longitudinal work shows that rumination mediates links between victimization and internalizing symptoms (Barchia & Bussey, 2010; Chu et al., 2019; Feinstein et al., 2014; Labella et al., 2024; Malamut & Salmivalli, 2023; Mathieson et al., 2014; Monti et al., 2017; Peets et al., 2021) and may also explain externalizing outcomes such as aggression (Li et al., 2021; Malamut & Salmivalli, 2021, 2023). However, other externalizing problems remain less studied. In addition, rumination itself may predict later victimization (Camacho et al., 2021; McLaughlin & Nolen-Hoeksema, 2012; Shapero et al., 2013), raising the possibility of a reciprocal process in which symptoms can predict later victimization through rumination.

The role of self-efficacy

Self-efficacy, or the belief in one’s ability to perform desired actions (Bandura, 1997), is another key process. Bandura et al. (1996) and Muris (2001, 2002) showed that self-efficacy across academic, emotional, and social domains predicts both internalizing and externalizing symptoms (see also Suldo & Shaffer, 2007).

Victimization is linked to lower global self-efficacy (Fredstrom et al., 2011; Haraldstad et al., 2019; Kokkinos & Kipritsi, 2012; Xie et al., 2023), and in particular to reduced efficacy in academic (Andreou & Metallidou, 2004; Galand & Hospel, 2013; Kokkinos & Kipritsi, 2012; Thijs & Verkuyten, 2008; Xie et al., 2023), social (Heiman et al., 2018; Olenik-Shemesh & Heiman, 2017; Purcell et al., 2021; Raskauskas et al., 2015; Romera et al., 2016; Xie et al., 2023), and emotional domains (Eden et al., 2014; Xie et al., 2023). Coping self-efficacy specifically mediates victimization effects on social anxiety, depression, and externalizing symptoms (Singh & Bussey, 2011; Trompeter et al., 2018), with similar findings for support-seeking efficacy (Barchia & Bussey, 2010) and social self-efficacy (Purcell et al., 2021). In addition, low confidence in one’s abilities may promote isolation (Bandura et al., 1999) and maladaptive coping (Bandura, 1997), further reinforcing vulnerability to bullying. Despite this evidence, it remains unclear whether general self-efficacy ‒ which integrates all domains ‒ serves as a mediator of victimization effects. It also remains uncertain whether symptoms themselves predict later victimization through self-efficacy.

Integrating the two processes

Together, rumination and self-efficacy may form complementary pathways: rumination sustains distress after bullying, while low self-efficacy undermines coping and increases exposure to further victimization. Importantly, these processes may not only explain how victimization leads to later psychological difficulties, but also how existing difficulties increase the likelihood of subsequent victimization ‒ pointing to a bidirectional cycle. Despite evidence that psychological difficulties can diminish over time (Schoeler et al., 2018), little is known about whether rumination and self-efficacy follow similar trajectories, improving when bullying stops. Smith et al. (2004), for example, identified distinct bullying trajectories with different psychosocial adaptations, yet how rumination and self-efficacy evolve across these pathways remains unclear.

Taken as a whole, prior research suggests that rumination and self-efficacy may be key mechanisms linking bullying victimization with both internalizing and externalizing symptoms, and that they may also contribute to the recurrence of victimization; however, their longitudinal and bidirectional roles have not yet been directly tested.

The current study

This longitudinal study addresses these gaps by investigating the effects of bullying status (victim vs. non-involved) and time on internalizing and externalizing symptoms, rumination, and self-efficacy. The hypotheses are as follows : (1) Hypothesis 1: There will be an effect of status (victim vs. non-involved) on internalizing and externalizing psychological difficulties, as well as rumination. Victims' psychological difficulties and rumination scores are expected to increase from Time 1 to Time 2. (2) Hypothesis 2: There will be an effect of status on overall self-efficacy and its three dimensions. Victims' overall self-efficacy is anticipated to decline from Time 1 to Time 2. (3) Hypothesis 3: Victimization at Time 1 will indirectly predict internalizing and externalizing psychological difficulties at Time 2 via rumination and global self-efficacy. (4) Hypothesis 4: Internalizing and externalizing symptoms at Time 1 will predict victimization at Time 2 via rumination and global self-efficacy. Finally, we examine distinct participant profiles (persistent victims, escaped victims, stable non-involved, and new victims) to explore longitudinal changes in psychological difficulties, rumination, and self-efficacy.

Method

Procedure

The study has been validated by the Research Ethics Committee of the Savoie Mont Blanc University (submitted on 01/06/21, Reference number: 2021_27_ INSTA). Hypotheses were pre-registered on the Open Science Framework at https://osf.io/7jcmy. The dataset and analysis scripts are also publicly available at https://osf.io/3qapc/?view_only=88e31aad6aef4aa5bf775cac9973c14c.

Four private schools in the Savoie department in France (one of which is situated in a rural setting) provided access to the participants. First, parents were notified about the study and asked to return a non-consent form if they did not agree to the participation of their child. Following that, participants were given the information letter in the presence of the principal investigator or research assistants, who were able to explain the study's objectives and answer any questions they had. Consent was given freely and voluntarily. An Olweus definition of bullying was also given orally. The candidates then completed the questionnaires online via the LimeSurvey platform or on paper, depending on the school's capabilities. The overall participation rate was 60%. The procedure was repeated six months later.

Participants

A total of 886 students participated in the first phase (Time 1), compared to 362 participants who completed all questionnaires (Time 1 and Time 2). It should be noted that some 9^th graders were unable to take part in phase 2, as they were preparing for their national examination. In addition, some candidates probably entered different anonymization codes at times 1 and 2, which prevented the two measurement times from being matched. The final sample is thus made up of 362 participants (41% of the initial sample) who responded exhaustively to both measurement times. The average age of the sample was 13.1 years(SD = 1.16), with 202 girls (55.8%) and 160 boys (44.2%). The distribution by school is as follows: 31% of participants are from school 1, 20.4% from school 2, 13.5% from school 3, and 35.1% from school 4. Next, 21% of participants are in 6^th grade, 30.1% in 7^th grade, 30.9% in 8^th, and 18% in 9^th grade. We have checked if participants who completed all questionnaires (n=362) differ from those who only completed questionnaires at Time 1 (n=524). We note that the two groups differ in terms of gender, χ²(1) =7.31, p =.01, and participating school χ²(3) =99.08; p<.001. Nevertheless, the two groups did not differ in terms of their bullying status, χ²(3) = 3.83, p = .28. A weak effect of age is also observed in both groups, t(655.42) = -2.3, p=.02, d=-0.17: It does indeed appear that participants who completed questionnaires at times 1 and 2 are younger overall(M = 12.78, SD = 1.15) than those who completed questionnaires at Time 1(M= 12.99, SD = 1.3). In addition, there were significantly more girls in the first group of n = 362 (55.8%) than in group 2 of n = 524 (46.6%). The two groups thus differed in their anxiety (STAI-C) scores, with a small effect size; t(884) = 2.13, p=.03, d=0.15. In other words, participants in the first group(n=352; M=33.72, SD=9.08) appear more anxious than those in the second group(n=524; M=32.42, SD=8.82). However, there was no significant difference between the two groups in terms of total self-efficacy score, academic self-efficacy score, emotional self-efficacy score, social self-efficacy score, rumination, depression (CDI) score, and total psychological difficulties score.

Measures

Bullying behavior

We measured bullying behavior with the French adaptation of the revised Olweus Bully/Victim Questionnaire (Fr-rBVQ; Kubiszewski et al., 2014) We followed Potard et al.'s (2021) approach, which adopts both continuous and categorical perspectives when using this instrument to evaluate bullying behavior (Potard et al., 2021). First, a definition of bullying is read out to each of the participants. This self-report questionnaire assesses experiences of being victimized (7 items) and experiences of bullying others (7 items) “in the past couple of months.” Various forms of bullying are assessed (verbal, physical, social, etc.). In the present study, we added one other type (i.e., cyberbullying) to each part (victimization/perpetration). Items were rated on a 5-point Likert scale ranging from 1 (Never) to 5 (Several times a week). Two versions of the global measures were used for analyses: (1) a continuous approach, which yielded two mean scores (one for the victimization items and one for the bullying perpetration items), but only the victimization score was used for this study; and (2) a categorical approach, in which participants were classified as victims, bullies, bully-victims, or non-involved, based on Solberg’s criteria (e.g., participants who had been bullied/bullied others “2 or 3 times a month” or more) were categorized as being involved in bullying (Solberg & Olweus, 2003). In our study, internal consistency was found to be acceptable for the perpetration scale (α =.72) and for the victimization scale (α =.82).

Psychological difficulties

Psychological difficulties were assessed using the Strengths and Difficulties Questionnaire (SDQ; R. Goodman, 1997). which has been widely used to measure behavioral and emotional difficulties in 11 to 17-year-olds. This 25-item Likert-type questionnaire measures internalizing and externalizing psychological difficulties in children and adolescents. These are grouped into four subscales (emotional, relational, behavioral, and hyperactivity problems). Scores on these different subscales are added together to obtain a total difficulty score. The fifth scale concerns prosocial attitude, considered as a strength. This dimension was not used in this study. The child chooses the indicator that best corresponds to the statement read ("not true", "somewhat true", "very true") over the last 6 months. A. Goodman et al. (2010) recommend the use of the two subscales "internalizing difficulties" (including relational and emotional difficulties) and "externalizing difficulties" (including behavioral difficulties and hyperactivity) in the general population (A. Goodman et al., 2010). Capron and colleagues carried out the French validation of the questionnaire in 2007 (Capron et al., 2007). The internal consistency found in our sample was satisfactory for the total difficulties score (α = .81), the internalizing symptoms score (α = .81), and the externalizing symptoms score (α = .74).

Anxiety

The State-Trait Anxiety Scale for Children (STAIC - trait) was used to measure anxiety. The scale is derived from the State-Trait Anxiety Scale for Children (STAI-C; Spielberger et al., 1973). The anxiety-trait scale consists of 20 items. The scale ranges from 1 (almost never) to 3 (quite frequently). It reflects the child's usual emotional state and assesses their level of anxiety. The tool was validated in French by Turgeon & Chartrand, 2003. In our sample, the internal consistency of the STAIC- trait scale is excellent (α = .93).

Depression

To measure depressive symptoms, the Child Depression Inventory (CDI; Kovacs & Beck, 1977) was used. This 27-item questionnaire assesses depression severity in children and adolescents aged 7 to 17. This inventory is based on Beck's adult depression inventory and was adapted for children by Beck and Kovacs. For each item, the child must select one of three statements that best describes his or her thoughts and feelings over the previous two weeks. Saint-Laurent (1990) validated the French version, which was translated by Mack and Moor (1982) (Mack & Moor, 1982; Saint-Laurent, 1990). In our sample, the internal consistency of the CDI scale is excellent (α = .92).

Self-efficacy

Self-efficacy was assessed using the French version of the Self-Efficacy Questionnaire for Children (SEQ-C; Muris, 2001), which measures academic, social, and emotional self-efficacy. We chose to assess global self-efficacy rather than coping-specific self-efficacy. This decision reflects a process-based perspective: global self-efficacy captures beliefs about one’s overall capacity to act effectively, which may generalize across situations, including bullying. The French validation of the questionnaire was carried out by the authors and was submitted for publication. The French version of the SEQ-C comprises 18 Likert-type items ranging from (1) Not at all to (5) Very well. It represents three domains of self-efficacy: (1) social self-efficacy, which refers to the perceived ability to maintain relationships with peers and assert oneself (4 items); (2) academic self-efficacy, which concerns the perceived ability to manage one's learning behavior, master school subjects and meet academic expectations (7 items); and (3) emotional self-efficacy, which refers to the perceived ability to cope with negative emotions (7 items). The scale also provides a total score corresponding to the overall sense of self-efficacy. Internal consistency was found to be satisfactory with for overall self-efficacy (α =.88), for academic self-efficacy (α =.86), for emotional self-efficacy (α =.82), and for social self-efficacy (α =.73).

Rumination

The Children’s Response Styles Questionnaire (CRSQ; Abela et al., 2002) assesses children's response styles (rumination and distraction). The scale is made up of 20 items: 10 items for the "rumination" subscale and another 10 for the "distraction" subscale. The rating scale ranges from 0 (never) to 10 (always). The French version was validated by (Le Van et al., 2021). In our sample, the internal consistency is excellent for the rumination subscale (α = 0.9) and for the distraction subscale (α = .93).

Data Analysis Plan

Data analysis was performed using R software (version 4.1.0). For each analysis carried out, statistical outliers were identified through the examination of studentized residuals and Cook's distance. Outliers exceeding thresholds (greater than 4 or less than -4 for studentized residuals, and exceeding 0.5 and at least twice as high as the highest previous value for Cook's distance) were flagged and removed from the analysis. To test our hypotheses, mixed ANOVAs were performed on the scores obtained on the different measures of interest at Time 1 and Time 2 in both groups (non-involved vs. victims at Time 1). Gender was added as a predictor to control for its effect. The effect size was obtained by calculating the eta squared, considering 0.01 as a "weak" effect, 0.06 as a "medium effect" and 0.14 as a "strong effect" (Cohen, 1988). To address multiple comparisons, the False Discovery Rate (FDR) method was applied to systematically correct p-values and minimize Type I errors. Distinct participant profiles were analyzed concerning their psychological symptoms, rumination, and self-efficacy scores over time using paired-samples t-tests and independent samples t-tests. Structural Equation Modeling (SEM) utilizing the lavaan package in R was employed to investigate the direct and indirect effects of victimization on internalizing/externalizing symptoms, and vice versa, through rumination and self-efficacy pathways.

Results

Preliminary analyses

As our study exclusively recruited participants from private schools, we conducted one-sample t-tests to ensure that our main variables of interest aligned with theoretical samples. Specifically, we found that the mean total score for psychological difficulties (encompassing internalizing and externalizing symptoms) did not significantly differ from that reported in Capron et al.'s sample (2007) at both time points (p > .05). Additionally, the mean victimization score from our sample differed from that documented in Potard et al.’s sample (2021) at Time 1 (p < .001) (at the beginning of the academic year), but not at Time 2 (p = .54) (toward the end of the academic year). However, our sample exhibited significantly lower levels of ruminative responses compared to Le Van et al.'s sample (2021) (p < .01). Unfortunately, we were unable to directly compare self-efficacy scores, as the same population was utilized for the validation study of the French version of the Self-Efficacy Questionnaire for Children.

Description of the selected sample

Table 1 shows the status distribution of participants for measurement times 1 and 2. It should be noted that the participant status differs significantly between times 1 and 2; χ²(3) = 20.3; p<.001. Post-hoc comparisons show that the percentage of victims (14.1% versus 22.6%) and bully-victims (1.6% versus 6.1%) increases significantly (p=.02), whereas the percentage of non-involved decreases significantly (from 81.8% to 69.1%) with p<.001. The percentage of bullies appears to remain constant (p = 1).

Table 1.

Participants' status according to their answers to the Olweus questionnaire at times 1 and 2

	Time 1		Time 2		χ2	p
	n	%	n	%
Status					20.3	<.001
Non-involved	296	81.8	250	69.1		<.001*
			82			0.02*
Victim	51	14.1		22.6
Bully	9	2.5	8	2.2		1
Bully-victim	6	1.6	22	6.1		0.02*

^*Post-hoc analysis with Bonferroni method (correction already applied)

Changes in participants' status from Time 1 to Time 2 were also explored: Three-quarters (76%) of non-involved participants at Time 1 remain non-involved at Time 2, while 17.6% become victims, 4.4% bully-victims, and 2% bullies. Concerning victims at Time 1, half of them (51%) remain victims at Time 2; 37.3% become non-involved, 9.7% become bully-victims, and 2% become bullies.

The scores obtained for our main variables at times 1 and 2 were then compared (table 2). Overall, the means obtained for the different scores (total difficulties score, STAI-C score, rumination score, and overall self-efficacy score) did not differ significantly between Time 1 and Time 2, especially following the p-value adjustment. Nevertheless, mean depression (CDI) and victimization scores increased significantly with time (p < .001).

Table 2.

Means and standard deviations of main variables for different measurement times

Variables	Time 1		Time 2		df	t	p	padj
	M	ET	M	ET
Total psychological difficulties	11.36	5.58	11.75	6.20	361	-1.73	.08	.09
Victimization	9.81	3.06	10.93	4.01	361	-6.23	<.001	<.001
STAI-C	33.72	9.07	34.48	9.96	361	-2.02	.04	.07
CDI score	12.3	8.23	13.61	9.49	361	-3.7	<.001	<.001
Rumination	4.52	2.67	4.6	2.61	361	-0.7	.49	.49
Distraction	3.98	2.63	3.69	2.68	361	2.11	.04	.07
Total self-efficacy	78.71	16.75	77.37	16.46	361	1.74	.08	.09

STAI-C: State-Trait Anxiety Scale for Children

CDI: Child Depression Inventory

p adj: p adjusted (FDR correction)

Main analyses

We excluded the bully-victim and bully groups from our main analyses due to small and uneven sample sizes across statuses. Analyses therefore focused on victims (n = 51) and non-involved participants (n = 296), for a total of 347 adolescents.

Hypotheses 1 and 2 : Effect of victimization status and time on internalizing andexternalizing symptoms, rumination and self-efficay

We first examined whether status at Time 1 (victim vs. non-involved) and time predicted internalizing and externalizing psychological difficulties, controlling for gender (table 3). A significant main effect of status was observed for the total difficulty score, F(1, 344) = 64.69, p < .001, η_p² = .16, indicating that victims reported more difficulties than non-involved peers. This effect was significant for both internalizing, F(1,344) = 83.83, p = <.001, η_p²=0.20, and externalizing symptoms, F(1,344) = 15.65, p = <.001, η_p²=0.04, with a larger effect size for internalizing difficulties. Additional analyses confirmed the effect of status for anxiety (STAI-C), F(1,344) = 37.44, p = <.001, η_p²=0.10, and depression (CDI), F(1,344) = 45.97, p = <.001, η_p²= 0.12. The only significant time effect was found for CDI scores, which increased from Time 1 to Time 2, F(1,344) = 16.20, p < .001, η_p²= 0.04. No status × time interactions were significant.

Table 3.

Results of Mixed ANOVA Analyzing the Effects of Status (at Time 1) and Time on Psychological Difficulties Score

	Time 1		Time 2
	NI	Vic	NI	Vic	Between-subjects effect (Gender)		Between-subjects effect (Status)			Within-subjects effect Time				Status x Time
	M (SD)	M (SD)	M (SD)	M (SD)	F (1,344)	p	F (1, 344)	p	η2p	F (1, 345)	p	padj	η2p	F(1, 345) p
SDQ
Total difficulties	10.2 (5)	16.7 (5.6)	10.7 (5.6)	16.5 (6.8)	22.47	<.001	64.69	<.001	0.16	3.51	.06	.10	0.01	1.06 .30
Internalizing difficulties	4.2 (3.2)	8.8 (3.9)	4.6 (3.5)	8.6 (4.7)	72.19	<.001	83.83	<.001	0.20	2.46	.12	.15	0.01	1.57 .21
Externalizing difficulties	6.0 (3.1)	7.9 (4.1)	6.2 (3.5)	8.0 (3.9)	0.37	.55	15.65	<.001	0.04	1.70	.19	.19	0	0.10 .75
STAI-C	32.4 (8.5)	40.1 (9.3)	33.3 (9.4)	40.9 (11)	59.36	<.001	37.44	<.001	0.10	4.83	.03	.08	0.01	0 .99
CDI	11 (7.5)	18.5 (9.1)	12.4 (8.9)	20.6 (10.6)	27.72	<.001	45.97	<.001	0.12	16.20	<.001	<.001	0.04	0.85 .36

SDQ: Strengths and Difficulties Questionnaire, STAI-C: State-Trait Anxiety Scale for Children, CDI: Child Depression Inventory NI: Non-involved, Vic: Victims, p adj: p adjusted (FDR correction)

We next examined the effects of status and time on rumination and self-efficacy (table 4). A significant main effect of status was observed for rumination, F(1,344) = 17.36, p = <.001, η_p²= 0.05, with victims reporting higher scores than non-involved participants. A significant effect of status was also found for global self-efficacy, F(1,344) = 10.31, p = .002, η_p²= 0.03, with differences emerging particularly for emotional self-efficacy, F(1,344) = 10.81, p = .002, η_p²= 0.03, and social self-efficacy, F(1,344) = 9.24, p = .004, η_p²= 0.03. Academic self-efficacy did not differ significantly by bullying status, F(1,344) = 1.62, p = .20. No significant main effects of time or status × time interactions were observed across these outcomes.

Table 4.

Results of Mixed ANOVA Analyzing the Effects of Status (at Time 1) and Time on Psychological Difficulties Score

	Time 1		Time 2
	NI	Vic	NI	Vic	Between-subjects effect (Gender)		Between-subjects effect (Status)				Within-subjects effect Time			Status x Time
	M (SD)	M (SD)	M (SD)	M (SD)	F (1,344)	p	F (1, 344)	p	paj	η2p	F (1, 345)	p	paj	F (1, 345)	p
Rumination	4.2 (2.6)	6 (2.7)	4.4 (2.6)	5.6 (2.8)	39.37	<.001	17.36	<.001	<.001	0.05	0.36	.55	.55	2.63	.11
SEQ-C Global self-efficacy	59.4 (12.8)	55.4 (12.2)	59 (12.1)	51.9 (12.4)	6.20	.01	10.31	.001	.002	0.03	1.65	.20	0.21	3.42	.07
Academic self-efficacy	24.4 (6.1)	23.8 (5.6)	24 (6.0)	22 (6.1)	1.62	.20	2.70	.10	.20	0.01	4.79	.03	0.12	3.77	.053
Emotional self-efficacy	20.7 (6.1)	18.4 (6.9)	20.4 (6.1)	17.2 (5.9)	27.81	<.001	10.81	.001	.002	0.03	1.75	.19	0.21	1.05	.31
Social self-efficacy	14.3 (3.6)	13.2 (3.7)	14.6 (3.4)	12.8 (14.0)	5.3	.02	9.24	.003	.004	0.03	1.56	.21	0.21	2.02	.16

SEQ-C: Self-Efficacy Questionnaire for Children

NI: Non-involved, Vic: Victims, p adj: p adjusted (FDR correction

Hypothesis symptoms via 3: rumination Victimization and predicting self-efficacy later

We tested whether victimization at Time 1 predicted internalizing and externalizing symptoms at Time 2 indirectly through rumination and self-efficacy. Correlational analyses (table 5) indicated that victimization at Time 1 was positively associated with internalizing and externalizing symptoms at Time 2, positively associated with rumination, and negatively associated with self-efficacy. Our data (figures 1 and 2) showed that victimization at Time 1 directly predicted internalizing symptoms at Time 2, but indirect effects through rumination (β = -0.01, z = -0.8, p = .42) or self-efficacy (β = -0.004, z = -0.37, p = .71) were not significant. In contrast, victimization at Time 1 predicted externalizing symptoms at Time 2 indirectly through rumination (β = 0.02, 95% CI = [0.001, 0.03], z = 1.9, p = .05). The direct path from victimization to externalizing symptoms was no longer significant, indicating full mediation. No indirect effect was observed through self-efficacy.

Table 5.

Correlations between variables of interest

	1	2	3	4	5	6	7
1. Victimization (T1)	‒
2. Internalizing symptoms (T1)	0.52*	‒
3. Externalizing symptoms (T1)	0.26*	0.27*	‒
4. Rumination (T2)	0.19*	0.44*	0.09ns	‒
5. Global self-efficacy (T2)	-0.21*	-0.42*	-0.38*	-0.23*	‒
6. Victimization (T2)	0.57*	0.43*	0.28*	0.27*	-0.35*	‒
7. Internalizing symptoms (T2)	0.45*	0.74*	0.23*	0.53*	-0.48*	0.60*	‒
8. Externalizing symptoms (T2)	0.22*	0.26*	0.69*	0.21*	-0.47*	0.36*	0.32*

*p<.001, ^ns : non-significant, T1 : Time 1, T2 : Time 2

Figure 1.

Direct and indirect effect of victimization on internalizing symptoms

Standardized regression coefficients are presented with standard errors in parentheses. ** p <.001, *p <.05 Analyses are adjusted for gender and internalizing symptoms at Time 1.

Figure 2.

Direct and indirect effect of victimization on externalizing symptoms

Standardized regression coefficients are presented with standard errors in parentheses. ** p <.001, *p <.05 Analyses are adjusted for gender and externalizing symptoms at Time 1.

Hypothesis 4: Symptoms predicting later victimization via rumination and self-efficacy

We next examined whether internalizing and externalizing symptoms at Time 1 predicted victimization at Time 2 via rumination and self-efficacy. Correlational analyses (table 5) indicated that internalizing symptoms at Time 1 were positively associated with rumination and negatively associated with self-efficacy at Time 2. In contrast, externalizing symptoms at Time 1 were significantly negatively correlated with self-efficacy but not with rumination at Time 2. On this basis, rumination was not retained in the path model for externalizing symptoms.

Results (figures 3 and 4) showed that internalizing symptoms at Time 1 indirectly predicted later victimization through both rumination (β = 0.04, 95% CI = [0.003, 0.08], z = 2.09, p = .04) and self-efficacy (β = 0.09, 95% CI = [0.05, 0.14], z = 3.87, p < .001 for self-efficacy). Externalizing symptoms at Time 1 indirectly predicted victimization at Time 2 through self-efficacy (β = 0.07, 95% CI = [0.04, 0.11], z = 3.76, p < .001). In both cases, direct effects were no longer significant after including the mediators, indicating full mediation.

Figure 3.

Direct and indirect effect of internalizing symptoms on victimization

Standardized regression coefficients are presented with standard errors in parentheses. ** p <.001, *p <.05 Analyses are adjusted for gender and victimization at Time 1.

Figure 4.

Direct and indirect effect of externalizing symptoms on victimization

Standardized regression coefficients are presented with standard errors in parentheses. ** p <.001, *p <.05 Analyses are adjusted for gender and victimization at Time 1.

Additional analyses: victimization profiles overtime

To capture heterogeneity in trajectories, we examined four profiles based on participants’ status across the two time points: continuing victims (V–V, n = 26), escaped victims (V–NI, n = 19), new victims (NI–V, n = 52), and stable non-involved (NI–NI, randomly selected subsample of n = 50). Means and standard deviations for each group are reported in table 6.

Table 6.

Means and standard deviations of scores on main variables for the four profiles at times 1 and 2

	Time 1				Time 2
	V-NI (N=19)	V-V (N=26)	NI-NI (N=50)	NI-V (N=52)	V-NI (N=19)	V-V (N=26)	NI-NI (N=50)	NI-V (N=52)
	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Total difficulties score	14.8(5.6)	16.7(4.9)	8.74(4.6)	12.6(5.61)	12.2 (5.3)	18.4 (5.8)	8.92(4.5)	14.8(5.1)
Rumination	5.02(2.8)	6.72(2.4)	3.39(2.5)	5.07(2.3)	4.71(6.4)	6.43(2.6)	3.71 (2.8)	5.46(2.4)
Global self-efficacy	60.1(11)	51.3(11)	60.1(14.1)	56.1(12.7)	55.1(14.3)	51.4(11.2)	61.5 (13.1)	54.0 (9.94)

V-NI: Victims who become non-involved, V-V: victims who remain victims, NI-NI: Non-involved who remain non-involved, NI-V: Non-involved who become victims.

For psychological difficulties (figure 5), continuing victims (V–V) maintained high scores, whereas stable non-involved participants (NI–NI) maintained low scores across time. Escaped victims (V–NI) showed a significant decrease in difficulties over time, t(18)=2.55, p = .02, d = 0.58, while new victims (NI–V) showed a significant increase, t(51)=-3.98, p<.001, d = -0.55. At Time 2, continuing victims (V-V) reported significantly higher difficulties than escaped victims (V-NI), t(43)=-3.68, p<.001, d=-1.14, and new victims (NI-V) reported significantly higher difficulties than NI–NI, t(100)=-6.14, p<.001, d=-1.23. Descriptively, new victims (NI-V) already exhibited somewhat elevated difficulties at Time 1 compared to NI–NI, while escaped victims (V-NI) began with lower difficulties than continuing victims (V-V).

Figure 5.

Longitudinal trends in mean total psychological difficulties scores across four profile groups

For rumination (figure 6), no significant change was observed within groups over time. At Time 2, escaped victims (V-NI) reported significantly less rumination than continuing victims (V-V), t(43)= -2.16, p = .04, d= -0.67, and new victims (NI-V) reported significantly more rumination than NI–NI, t(100)=-3.39, p = .001, d = -0.68. Descriptively, new victims (NI-V) already showed elevated rumination at Time 1 compared to NI-NI, and escaped victims (V-NI) started with lower rumination than continuing victims (V-V).

Figure 6.

Longitudinal trends in mean rumination scores across four profile groups

For self-efficacy (figure 7), no significant change was observed within groups over time. At Time 2, NI–NI participants reported significantly higher self-efficacy than NI–V, t(100)=3.25, p = .002, d =0.65. No significant difference was observed between V–V and V–NI, t(43)=0.97, p = .34, although descriptively the latter began with somewhat higher self-efficacy at Time 1. Similarly, new victims (NI-V) already showed lower self-efficacy at Time 1 than NI–NI.

Figure 7.

Longitudinal trends in mean global self-efficacy across four profile groups

Discussion

To our knowledge, this is among the first studies to simultaneously test bidirectional pathways between victimization and psychological symptoms through rumination and self-efficacy, while also examining their longitudinal trajectories. Hypothesis 1 predicted that victims would report higher psychological difficulties and rumination than non-involved peers, and that these difficulties would increase over time. The first part of this hypothesis was confirmed : victims displayed significantly more internalizing and externalizing difficulties and higher rumination, consistent with prior research (Arató et al., 2020; Camacho et al., 2021; Garnefski & Kraaij, 2014; Potard et al., 2022; Reijntjes et al., 2010, 2011; Spyropoulou & Giovazolias, 2023; Stange et al., 2014; Zych et al., 2015). However, contrary to our expectations, these difficulties remained stable over time, except for depression. One explanation may be ceiling effects: victims already reported high levels of difficulties and rumination at baseline, leaving little room for measurable increases. Hypothesis 2 predicted that victims would report lower self-efficacy than non-involved peers, and that their self-efficacy would decline over time. Focusing on global rather than coping-specific self-efficacy also proved informative, as it highlights broader beliefs that may underlie vulnerability to victimization and psychological distress across contexts. Group effects were confirmed: victims scored significantly lower on global, social, and emotional self-efficacy, in line with previous studies (Fredstrom et al., 2011; Haraldstad et al., 2019; Kokkinos & Kipritsi, 2012; Heiman et al., 2018; Olenik-Shemesh & Heiman, 2017; Romera et al., 2016; Xie et al., 2023). However, similarly to psychological difficulties and rumination, no significant decline was observed longitudinally. The profile analyses provided further support and nuance to these hypotheses. First, stable non-involved participants (NI– NI) consistently exhibited the most favorable scores across psychological difficulties, rumination, and self-efficacy, followed by escaped victims (V–NI), new victims (NI–V), and finally continuing victims (V–V). This pattern parallels findings by Smith et al. (2004). Second, escaped victims (V–NI) reported significantly lower rumination and fewer psychological difficulties than continuing victims (V–V), suggesting that mental health can improve once victimization ceases. This aligns with meta-analytical evidence of recovery over time (Schoeler et al., 2018) and supports the view that rumination is a flexible process that adapts to contextual change (Bonanno & Burton, 2013). Third, however, self-efficacy did not show similar improvement: escaped victims did not differ significantly from continuing victims, indicating that feelings of efficacy may remain impaired even after bullying ends. This persistence is consistent with Bandura’s (1997) model, as bullying undermines mastery experiences, vicarious learning, social persuasion, and emotional states. In particular, victims often experience repeated coping failures and perceive their strategies as ineffective (Tenenbaum et al., 2011), while also observing peers succeed where they struggle, receiving negative feedback from aggressors, and enduring chronic stress. Together, these factors may explain the enduring vulnerability in self-efficacy found in our study.

Next, hypothesis 3 predicted that victimization at Time 1 would indirectly predict later psychological difficulties through rumination and self-efficacy. This hypothesis was only partially supported: an indirect effect emerged for externalizing difficulties via rumination, but no mediation was observed for internalizing once baseline difficulties were controlled. This contrasts with several cross-sectional studies in youth that have reported the expected indirect effects through rumination and/or self-efficacy ‒ often alongside a direct link from victimization to symptoms (Chu et al., 2019; Gini et al., 2019; Mathieson et al., 2014; Monti et al., 2017; Purcell et al., 2021; Singh & Bussey, 2011; Trompeter et al., 2018). In longitudinal work that adjusts for prior symptoms, indirect effects have been observed for internalizing via rumination (Barchia & Bussey, 2010; Peets et al., 2021) and, specifically for depression, via self-efficacy to enlist support (Barchia & Bussey, 2010). At the same time, our findings align with recent evidence that victimization relates to externalizing outcomes through rumination (Li et al., 2021; Malamut & Salmivalli, 2021, 2023). Our results may therefore reflect both methodological differences and contextual ones. Given that our sample came exclusively from private schools, it is possible that greater resources and support attenuated the mediating role of these processes for internalizing symptoms.

Hypothesis 4 was supported. Internalizing symptoms at Time 1 predicted later victimization, with the effects fully mediated by both rumination and self-efficacy. Externalizing symptoms also predicted later victimization, but this pathway was fully mediated by self-efficacy alone. These findings extend developmental cascade perspectives and align with prior evidence that psychological vulnerabilities can heighten exposure to peer adversity through maladaptive coping processes (Potard et al., 2022; Troop-Gordon et al., 2015; Undheim et al., 2016).

Limitations

This study is subject to several limitations that warrant consideration. First, we observed a significant dropout and non-completion rate among participants. Retained participants were younger, more often female, and differed in school composition and anxiety. However, they did not differ in victimization, rumination, self-efficacy, depression, or total difficulties, suggesting that attrition, while a limitation, is unlikely to have systematically biased the main results.

Second, the sample was drawn exclusively from private schools. This composition may have influenced our results: rumination levels were lower than expected, and self-efficacy is known to vary with socioeconomic status (Bandura et al., 1996). Nonetheless, psychological difficulties and victimization scores were similar to theoretical norms, suggesting that bullying itself may be less sensitive to socioeconomic context (Debarbieux & Montoya, 2011).

Finally, our design involved only two measurement waves due to logistical constraints imposed by the schools, which limited our ability to separate temporal sequencing of predictors, mediators, and outcomes. Future work should use at least three time points to establish causal pathways more firmly.

Implications and future directions

Ultimately, and in line with a process-based approach (Kinderman, 2005, 2009; Philippot et al., 2018, 2024), our findings emphasize the critical role of addressing psychological processes such as rumination and self-efficacy to prevent psychological distress among vulnerable youth, both among those who are chronically victimized (tertiary prevention) and those with pre-existing psychological difficulties (secondary prevention). This perspective highlights the mechanisms through which symptoms can be alleviated and victimization reduced.

Future research should replicate these findings in more diverse populations, including public schools, and extend longitudinal designs beyond two waves. Additional mediating processes, such as experiential avoidance, also warrant investigation.

At the practical level, interventions that reduce rumination (e.g., metacognitive or mindfulness-based strategies) and strengthen self-efficacy may be particularly effective. Because our findings suggest that self-efficacy remains impaired even when bullying ceases, programs explicitly designed to restore mastery and social confidence could be of particular interest. In this regard, Rigby’s (2010) typology of bullying interventions is highly relevant: one category, reinforcement of the victim, seeks to empower victims by developing their social, emotional, and behavioral competencies (Rigby, 2010). In theory, this approach can enhance victims’ self-efficacy ‒ a process directly consistent with our findings ‒ but empirical evidence for its effectiveness remains scarce. Testing whether reinforcement-based interventions can restore self-efficacy and reduce recurrent victimization represents an important next step for both research and practice.

Taken together, these insights suggest that tackling bullying requires more than reducing victimization itself: targeting the psychological processes that sustain vulnerability can both improve victims’ adjustment and lower their risk of continued peer victimization.