Cyberbullying and cybervictimization patterns among left-behind and non-left-behind adolescents in China: a latent profile analysis

Abstract

Background

Left-behind adolescents in China may face heightened risks of involvement in cyberbullying due to their psychological vulnerability and complex social circumstances. Considering the potential heterogeneity within this population, this study aimed to identify distinct patterns of cyberbullying and cybervictimization among left-behind adolescents and to explore how reactive anger, left-behind patterns, gender, and grade level predict membership in these subgroups.

Methods

A total of 1,351 junior high school students (752 left-behind, 599 non-left-behind) were recruited from five schools. Latent profile analysis (LPA) was used to identify distinct patterns, and multinomial logistic regression was used to examine the relationships between predictors and various profiles.

Results

(1) Three distinct profiles of cyberbullying and cybervictimization were identified among left-behind adolescents. (2) Left-behind adolescents were more likely to experience cybervictimization compared to their non-left-behind peers. (3) Reactive anger, left-behind patterns, gender, and grade level significantly predicted subgroup membership.

Conclusion

These findings underscore the importance of developing targeted interventions and considering the specific psychosocial vulnerabilities of left-behind youth.

Introduction

Cyberbullying and cybervictimization have emerged as a global public health issue that poses significant risks to the physical and mental well-being of both perpetrators and victims [1]. Individuals subjected to repeated online harassment are more likely to experience depression, suicidal ideation, and various psychological difficulties [2, 3], whereas perpetrators face heightened risks of antisocial behavior and broader involvement in delinquency [4, 5]. Early adolescence represents a particularly critical developmental period, as heightened emotional sensitivity and increased susceptibility to peer influence render youth more vulnerable to both perpetrating and experiencing cyberbullying [6–8]. Understanding the risk factors associated with cyberbullying involvement during this stage is therefore crucial for implementing timely and effective intervention strategies.

Recent research has increasingly focused on subgroups of adolescents who may be especially vulnerable to cyberbullying. A prominent example is left-behind adolescents, defined as youth who grow up without the daily presence of one or both parents due to labor migration [9]. Prolonged parental absence often results in reduced supervision, emotional neglect, and weaker parent–child bonds, which may elevate the likelihood of involvement in online aggression, either as victims or perpetrators [10, 11]. Consistent with the susceptibility hypothesis, the lack of stable caregiving heightens psychological distress and fosters maladaptive coping strategies, such as online retaliation or social withdrawal [12, 13]. Supporting this view, Feng and colleagues found that left-behind adolescents reported significantly higher levels of cyberbullying involvement than their non-left-behind peers, with loneliness and poor parent–child relationships partly mediating this association [12]. However, not all evidence supports the notion that left-behind status necessarily increases the risk of cyberbullying. The no-difference hypothesis suggests that observed disparities between left-behind and non-left-behind adolescents may be overstated or largely shaped by contextual factors. For instance, Hu and colleagues found that initial differences in psychological and behavioral outcomes among left-behind, migrant, and local children disappeared once family and school-related variables were controlled, indicating that environmental influences, rather than parental absence alone, accounted for the disparities [14].

These discrepancies may reflect a tendency to treat left-behind adolescents as a uniform group, despite substantial heterogeneity likely existing within this population [6]. In fact, some left-behind adolescents report high levels of cyberbullying involvement, while others remain relatively unaffected [6, 15]. However, most existing studies have emphasized group-level mean differences, overlooking meaningful within-group variation [16]. This limits our ability to identify high-risk subgroups and to determine whether certain behavioral profiles are unique to left-behind adolescents or shared with their non-left-behind peers [6, 15].

Recently, latent profile analysis (LPA) offers a promising approach to address this gap. LPA classifies individuals into distinct subgroups based on shared behavioral patterns [17]. Recent studies using LPA have successfully identified subtypes of cyberbullying and cybervictimization among Chinese adolescents. For example, three subgroups, including “non-involved,” “cyberbully-victim,” and “cyber-victim,” have been identified in previous studies conducted within the Chinese population [6].

In addition, building on these classification studies, prior research further indicates that several individual and contextual factors may shape adolescents’ involvement in cyberbullying, as reflected in Guo et al.’s multinomial logistic regressions linking social-ecological factors with cyberbullying roles [15]. From the perspective of emotional dysregulation theory, adolescents high in reactive anger are more prone to retaliatory aggression and consequently more likely to become both perpetrators and victims [18, 19]. Attachment theory and empirical findings further show that prolonged parental absence undermines emotional regulation and social competence, elevating vulnerability to online aggression [9, 12]. Gender differences also shape involvement, as girls are more likely to engage in verbal and relational aggression in digital contexts, whereas boys tend toward physical confrontation [20, 21]. Finally, developmental studies highlight middle adolescence, particularly the second year of junior high school, as a peak stage for peer stability and sensitivity to social evaluation; both of which intensify risks of cyberbullying and cybervictimization [6]. Taken together, these findings provide a strong rationale for including reactive anger, left-behind patterns, gender, and grade level as predictors in the present study.

The current study

The present study aimed to examine the heterogeneity of cyberbullying and cybervictimization and to compare their distribution between left-behind and non-left-behind adolescents. First, we investigated whether multiple latent profiles of cyberbullying and cybervictimization could be identified among left-behind adolescents. Second, we compared the distribution of these profiles across left-behind and non-left-behind groups. Third, we tested whether reactive anger, parental migration patterns, gender, and grade level predicted profile membership within the left-behind sample. Finally, we examined whether these predictors contributed to group differences in profile membership between left-behind and non-left-behind adolescents. Based on prior research and theory, we proposed the following hypotheses: (H1) cyberbullying and cybervictimization would show heterogeneous profiles among left-behind adolescents [13, 22]; (H2) left-behind and non-left-behind adolescents would differ in the distribution of these profiles [16]; (H3) reactive anger, parental migration patterns, gender, and grade level would predict profile membership within left-behind adolescents; and (H4) these predictors would partly account for intergroup differences between left-behind and non-left-behind youth [23, 24].

Methods

Participants and procedure

The sample comprised 1,351 adolescents from five middle schools in Shandong and Henan provinces, China. Among them, 752 were identified as left-behind students (417 boys, 335 girls; M_age = 14.35, SD = 1.66), with 37.5% in Grade 7, 31.3% in Grade 8, and 31.3% in Grade 9. The remaining 599 were non-left-behind students (241 boys, 358 girls; M_age = 13.94, SD = 1.27), including 52.9% in Grade 7, 22.0% in Grade 8, and 25.0% in Grade 9.

Ethical approval was obtained from the Human Subjects Review Committee at Guangzhou University. Prior to the formal survey, the objectives and procedures were communicated to school principals, teachers, and parents. After approval was secured, the researchers and investigators entered the designated classrooms to provide a detailed introduction to the study. Written informed consent was obtained from parents or legal guardians, as well as from all participating students, before the questionnaire was administered. Participants were informed that their involvement was voluntary and that they had the right to withdraw at any time without consequence. Data collection took place in classrooms during regular school hours, with each session lasting approximately 25 min.

Measures

Cyberbullying questionnaire (CBQ)

The Chinese version of the CBQ consists of seven items assessing the frequency of cyberbullying behaviors (e.g., “Do you often play malicious jokes on someone online or via text messages?”) [5]. Responses are rated on a 5-point scale (1 = never, 5 = always), with higher scores reflecting greater involvement in cyberbullying. In the present study, the single-factor model showed acceptable fit for both left-behind (CFI = 0.95, TLI = 0.92, RMSEA = 0.08, SRMR = 0.04) and non-left-behind groups (CFI = 0.99, TLI = 0.99, RMSEA = 0.04, SRMR = 0.02). Cronbach’s αs for the total scores were 0.89 and 0.95, respectively. Multi-group confirmatory factor analyses supported measurement invariance, indicating that the CBQ could be validly compared across left-behind and non-left-behind adolescents.

Cyber victimization questionnaire (CVQ)

The Chinese version of the CVQ consists of eight items assessing the frequency of online victimization (e.g., “Do your peers often make fun of you online or via text messages?”) [5]. Items are rated on a 5-point scale (1 = never, 5 = always), with higher scores indicating more frequent experiences of cybervictimization. In the present study, the single-factor model showed excellent fit for both left-behind (CFI = 0.99, TLI = 0.99, RMSEA = 0.03, SRMR = 0.02) and non-left-behind adolescents (CFI = 0.99, TLI = 0.99, RMSEA = 0.04, SRMR = 0.02). Cronbach’s αs for the total scores were 0.93 and 0.96, respectively. Multi-group confirmatory factor analyses supported metric invariance, suggesting that the CVQ is psychometrically comparable across left-behind and non-left-behind groups.

Reactive anger scale (RAS)

The Chinese version of the RAS consists of six items assessing the tendency to experience anger in frustrating or threatening situations (e.g., “When I am frustrated, I want to hit someone.”) [25]. Items are rated on a 4-point scale (1 = never, 4 = always), with higher scores reflecting greater reactive anger. In the present study, the single-factor model showed acceptable fit for both left-behind (CFI = 0.94, TLI = 0.90, RMSEA = 0.10, SRMR = 0.05) and non-left-behind adolescents (CFI = 0.99, TLI = 0.99, RMSEA = 0.02, SRMR = 0.01). Cronbach’s αs for the total scores were 0.89 and 0.95, respectively. Multi-group confirmatory factor analyses supported metric invariance, suggesting that the RAS can be validly compared across left-behind and non-left-behind groups.

Statistical analysis

All analyses were conducted using SPSS 26.0 and Mplus 8.3. First, descriptive statistics and internal consistency were computed. To assess potential common method bias, Harman’s single-factor test was performed. In addition, confirmatory factor analyses (CFA) were conducted separately for left-behind and non-left-behind adolescents to evaluate the factor structures of all scales. Model fit was assessed using the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR). Model fit was considered acceptable when RMSEA ≤ 0.08, CFI and TLI ≥ 0.90, and SRMR ≤ 0.08 [26, 27].

Second, latent profile analysis (LPA) was conducted separately for left-behind and non-left-behind students to identify homogeneous subgroups based on their responses to cyberbullying and cybervictimization. A series of LPA models were performed from 1 to 4 classes. To avoid local maxima, each model was estimated with multiple sets of random starts, and convergence was considered satisfactory when the best log-likelihood value was replicated across start sets. Model selection was guided by several criteria: (a) statistical fit indices, including AIC, BIC, and adjusted BIC, with lower values indicating better model fit; (b) likelihood ratio tests (LMR and BLRT), with significant p-values suggesting that the k-class solution was superior to the k-1 class solution [28]; (c) classification quality, as indicated by entropy values, with values above 0.80 considered acceptable; and (d) substantive interpretability and minimum class size, with proportions below 5% regarded as unstable.

Finally, multinomial logistic regression analyses were performed to examine the effects of reactive anger, left-behind patterns, gender, and grade level on latent profile membership, and to test intergroup differences between left-behind and non-left-behind adolescents.

Results

Common method bias test

Harman’s single-factor test indicated that four factors with eigenvalues greater than 1 emerged, with the first factor accounting for 38.38% of the variance, below the critical threshold of 40% [29]. These results suggest that common method bias was unlikely to be a serious concern in the present study.

Latent profile of cyber bullying/cybervictimization among left-behind and non-left-behind students

All estimated models converged to stable log-likelihood solutions with replicated values across multiple sets of random starts, indicating that the solutions were not due to local maxima. The three-class solutions demonstrated excellent classification quality, with entropy values exceeding 0.95 in both the left-behind and non-left-behind groups. The consistently high entropy values, together with adequate minimum class proportions (all > 5%), support the reliability of the identified profiles. Detailed fit indices and class distributions are reported in Table 1.

Table 1.

Fit indices for the latent class models on cyberbullying/cybervictimization for the total sample

Model	K	Log	AIC	BIC	aBIC	Entropy	LMR	BLRT	Class Probability
Cyberbullying	14	-6570.00	13168.00	13232.72	13188.26	—	—	—	—
Left-behind	22	-5138.33	10320.67	10422.37	10352.51	0.98	< 0.01	< 0.01	0.80 / 0.20
	30	-4838.21	9736.42	9875.11	9779.84	0.99	< 0.01	< 0.01	0.79 / 0.06 / 0.15
	38	-4650.45	9376.91	9552.57	9431.91	0.97	0.22	< 0.01	0.74 / 0.12 / 0.06 / 0.07
Non-left-behind	14	-5551.43	11130.86	11192.39	11147.95
	22	-3733.50	7511.00	7607.69	7537.85	0.99	< 0.01	< 0.01	0.83 / 0.17
	30	-3412.37	6884.74	7016.60	6921.35	0.98	0.02	< 0.01	0.79 / 0.09 / 0.12
	38	-3291.03	6658.05	6825.07	6704.43	0.98	0.63	< 0.01	0.76 / 0.06 / 0.10 / 0.08
Cybervictimization	16	-8297.46	16626.93	16700.89	16650.09	—	—	—	—
Left-behind	25	-6427.16	12904.32	13019.89	12940.51	0.98	< 0.01	< 0.01	0.76 / 0.24
	34	-6074.86	12217.72	12374.89	12266.93	0.95	0.01	< 0.01	0.69 / 0.19 / 0.13
	43	-5922.18	11930.36	12129.13	11982.59	0.97	0.08	< 0.01	0.68 / 0.16 / 0.09 / 0.07
Non-left-behind	16	-6579.70	13191.41	13261.73	13210.94	—	—	—	—
	25	-4576.60	9203.20	9313.08	9233.71	0.99	< 0.01	< 0.01	0.82 / 0.18
	34	-4130.12	8328.23	8477.67	8369.73	0.99	< 0.01	< 0.01	0.77 / 0.15 / 0.08
	43	-3922.62	7931.25	8120.24	7983.73	0.97	0.73	< 0.01	0.73 / 0.11 / 0.10 / 0.07

Note. AIC = Akaike information criterion; BIC = Bayesian information criterion; aBIC = adjusted BIC; LMR = the Lo–Mendell–Rubin adjusted likelihood ratio test; BLRT = bootstrapped likelihood ratio test

For left-behind students, the three-class model provided the best overall fit, as indicated by lower AIC, BIC, and adjusted BIC values compared with the two-class solution and a nonsignificant LMR test for the four-class model (p > 0.05). The largest subgroup (n = 591; 78.6%) reported uniformly low levels of cyberbullying across all items and was labeled the balanced-low-incidence class. A second subgroup (n = 116; 15.4%) displayed moderate levels across all items and was labeled the balanced-marginal class. A smaller subgroup (n = 45; 6.0%) showed moderate scores on items 1–4 but distinctly higher scores on items 5–7, reflecting a pattern of overt online aggression; this subgroup was labeled the direct-high-incidence class. A comparable three-class solution also emerged for non-left-behind students, with similar proportions (78.7%, 11.7%, and 9.6%). However, unlike the left-behind group, students in the high-incidence class demonstrated consistently elevated scores across all seven items rather than a concentration on direct aggression. Accordingly, this subgroup was labeled the balanced-high-incidence class (see Fig. 1).

Fig. 1.

Latent profiles of cyberbullying (above) and cybervictimization (below) in left-behind (left) and no-left-behind (right) group

For cybervictimization, the three-class model also provided the best overall fit in both the left-behind and non-left-behind groups. The latent profiles were highly similar in structure and proportion across groups. The largest subgroup, labeled the low-victimization class, was characterized by uniformly low scores on all items and included 68.7% of left-behind students (n = 517) and 76.9% of non-left-behind students (n = 461). The medium-victimization class showed moderate levels across items and comprised 18.8% of left-behind students (n = 141) and 15.1% of non-left-behind students (n = 91). The smallest subgroup, labeled the high-victimization class, was defined by consistently high scores across all items, indicating frequent experiences of cybervictimization; this profile included 12.6% of left-behind students (n = 94) and 8.0% of non-left-behind students (n = 48).

Frequency distribution of cyberbullying/cybervictimization among left-behind and non-left-behind students

Table 2 presents the cross-distribution of latent classes for cyberbullying and cybervictimization among left-behind and non-left-behind students. Among left-behind students, 63.8% (n = 480) were simultaneously classified into the balanced-low-incidence cyberbullying class and the low-victimization cybervictimization class. The remaining 36.2% were distributed across higher-risk profiles, reflecting involvement in elevated levels of cyberbullying, cybervictimization, or both.

Table 2.

Percentage of co-occurrence of potential classes of cyberbullying/cybervictimization

Cyberbullying Class	Cybervictimization Class
	High	Medium	Low
Left-Behind (n = 752)
1. Direct-High-Incidence	28(3.7%)	12(1.6%)	5(0.7%)
2. Balanced-Marginal	45(6.0%)	37(4.9%)	33(4.4%)
3. Balanced-Low-Incidence	21(2.8%)	91(12.1%)	480(63.8%)
Non-Left-Behind (n = 599)
1. Balanced-High-Incidence	36(6.0%)	17(2.8%)	4(0.7%)
2. Balanced-Marginal	9(1.5%)	49(8.2%)	12(2.0%)
3. Balanced-Low-Incidence	3(0.5%)	25(4.2%)	444(74.1%)

In contrast, the majority of non-left-behind students (74.1%, n = 444) also fell into the balanced-low-incidence and low-perception co-occurring class, with only 25.9% involved in more active patterns. Notably, only 6.0% (n = 36) of non-left-behind students were classified in the balanced-high-incidence cyberbullying and high-perception cybervictimization classes. Overall, the proportion of students involved in more severe cyberbullying/cybervictimization patterns was notably higher among left-behind adolescents than among their non-left-behind peers.

Multiple logistic regression

Table 3 presents the relationship between influencing factors (reactive anger, left-behind patterns, gender, and grade level) and profile membership within left-behind adolescents. For cyberbullying, higher levels of reactive anger significantly increased the likelihood of being classified into both the balanced-marginal (OR = 2.20, 95%CI [1.58, 3.05]) and direct-high-incidence (OR = 2.68, 95%CI [1.65, 4.37]) classes relative to the balanced-low-incidence class. Males were less likely than females to be in the direct-high-incidence class (OR = 0.44, 95%CI [0.22, 0.89]). Grade 1 students were less likely to be in higher-incidence classes, while Grade 2 students were more likely to be in the balanced-marginal class (OR = 1.83, 95%CI [1.13, 2.96]). Students whose mothers or both parents migrated for work were more likely to be in the balanced-marginal class compared to those with only a migrant father. For cybervictimization, reactive anger was a significant predictor of both the medium-perception (OR = 1.64, 95%CI [1.20, 2.23]) and high-perception (OR = 2.99, 95%CI [2.08, 4.29]) classes; Grade 2 students were more likely to fall into both elevated cybervictimization classes while Grade 1 students were less likely to be in the high-perception class (OR = 0.37, 95%CI [0.19, 0.71]).

Table 3.

Association of influencing factors with different latent classes of CBQ/CVQ within left-behind adolescents

		Cyberbullying Class				Cybervictimization Class
Factors		Balanced-Marginal		Direct-High-Incidence		Medium-perception		High-perception
		OR	CI (95%)	OR	CI (95%)	OR	CI (95%)	OR	CI (95%)
Gender	female	1.00		1.00		1.00		1.00
	male	0.79	0.50 ~ 1.24	0.44*	0.22 ~ 0.89	1.04	0.68 ~ 1.59	0.64	0.39 ~ 1.06
Grade	grade 3	1.00		1.00		1.00		1.00
	grade 1	0.29**	0.16 ~ 0.53	0.38*	0.15 ~ 0.94	0.67	1.36 ~ 3.41	0.37**	0.19 ~ 0.71
	grade 2	1.83*	1.13 ~ 2.96	2.03	0.98 ~ 4.20	2.12**	1.32 ~ 3.41	1.99*	1.16 ~ 3.42
Left-Behind Patterns	father-migrant	1.00		1.00		1.00		1.00
	mother-migrant	2.93**	1.66 ~ 5.16	2.49*	1.11 ~ 5.63	0.66	0.38 ~ 1.13	1.24	0.68 ~ 2.27
	two-parent-migrant	1.75*	1.05 ~ 2.94	1.25	0.58 ~ 2.71	0.33**	0.20 ~ 0.55	0.54	0.31 ~ 0.95
Reactive Anger		2.20**	1.58 ~ 3.05	2.68**	1.65 ~ 4.37	1.64**	1.20 ~ 2.23	2.99**	2.08 ~ 4.29

Note. *p < 0.05, **p < 0.01. For cyberbullying, the balanced-low-incidence class was treated as the reference group; for cybervictimization, low-perception class was treated as the reference group

To examine whether predictors of class membership differed by left-behind status, logistic regression analyses were conducted (see Table 4). For cyberbullying, boys were higher odds of being in the balanced-low-incidence class (OR = 2.45, 95% CI [1.89, 3.16]), whereas no significant gender differences were found for the other classes. Grade 1 students were less likely than Grade 3 students to be in the balanced-low-incidence class (OR = 0.43, 95% CI [0.31, 0.59]), and reactive anger was negatively associated with membership in the high-incidence class (OR = 0.42, 95% CI [0.19, 0.96]).

Table 4.

Multiple logistic regression analysis of cyberbullying/cybervictimization class of left-behind and non-left-behind junior high school students

		Comparison between Cyberbullying Class						Comparison between Cybervictimization Class
Factors		Balanced-low-incidence		Balanced-marginal-incidence		High-incidence		Low-perception		Medium-perception		High-perception
		OR	95%CI	OR	95%CI	OR	95%CI	OR	95%CI	OR	95%CI	OR	95%CI
Gender	female	1.00		1.00		1.00		1.00		1.00		1.00
	male	2.45**	1.89 ~ 3.16	1.04	0.55 ~ 1.96	0.48	0.19 ~ 1.23	2.13**	1.64 ~ 2.77	2.10*	1.13 ~ 3.88	0.51	0.24 ~ 1.10
Grade	grade 3	1.00		1.00		1.00		1.00		1.00		1.00
	grade 1	0.43**	0.31 ~ 0.59	0.81	0.34 ~ 1.94	2.90	0.77 ~ 10.99	0.42**	0.30 ~ 0.58	1.12	0.52 ~ 2.34	4.58*	1.26 ~ 16.67
	grade 2	1.00	0.69 ~ 1.46	1.30	0.66 ~ 2.55	1.99	0.81 ~ 4.90	0.91	0.62 ~ 1.34	1.30	0.67 ~ 2.53	2.25*	1.04 ~ 4.89
Reactive Anger		0.90	0.76 ~ 1.07	0.68	0.41 ~ 1.15	0.42*	0.19 ~ 0.96	0.90	0.76 ~ 1.07	0.47**	0.29 ~ 0.75	0.69	0.36 ~ 1.33

Note. *p < 0.05, **p < 0.01. The non-left-behind adolescents were treated as the reference group in the multiple logistic regression analysis. Finally, reactive anger was negatively associated with membership in the medium-victimization class (OR = 0.47, 95% CI [0.29, 0.75])

For cybervictimization, boys were more likely than girls to be in the low-victimization (OR = 2.13, 95% CI [1.64, 2.77]) and medium-victimization (OR = 2.10, 95% CI [1.13, 3.88]) classes. Grade 1 students were less likely to be in the low-victimization class (OR = 0.42, 95% CI [0.30, 0.58]) but more likely to be in the high-victimization class (OR = 4.58, 95% CI [1.26, 16.67]), and Grade 2 students also showed greater odds of being in the high-victimization class (OR = 2.25, 95% CI [1.04, 4.89]).

Discussion

This study investigated the heterogeneity of cyberbullying and cybervictimization among left-behind and non-left-behind adolescents. The results revealed distinct latent subgroups within both groups, along with notable differences in their distribution. In addition, reactive anger, gender, grade level, and left-behind patterns emerged as significant predictors of subgroup membership.

For cyberbullying, three subgroups consistently emerged in both left-behind and non-left-behind groups: balanced-low-incidence, balanced-marginal, and high-incidence classes. However, the structure of the high-incidence class differed across groups. Among left-behind adolescents, this subgroup reflected direct-high-incidence patterns, characterized by elevated involvement in overt forms of online aggression (items 5–7). In contrast, among non-left-behind adolescents, the high-incidence subgroup showed balanced-high-incidence patterns, with consistently high scores across all items. This distinction suggests that left-behind youth may be more inclined toward overt and confrontational cyber aggression, potentially shaped by the cumulative stress of parental separation and reduced emotional support, which could foster more direct coping strategies [30, 31].

For cybervictimization, three subgroups were consistently identified: low-victimization, medium-victimization, and high-victimization classes. While the overall distribution of cyberbullying classes was broadly similar across groups, left-behind adolescents were more likely to be classified into high-victimization profiles than their non-left-behind peers. This finding suggests heightened susceptibility to online victimization among left-behind youth. Such vulnerability may be linked to emotional fragility and increased sensitivity to rejection, potentially rooted in early parental absence, which can hinder the development of secure relationships and adaptive coping strategies [31–33]. As a result, even comparable levels of online conflict may be perceived as more hostile or threatening by left-behind adolescents [13]. In addition, limited access to protective resources, such as parental guidance, peer support, and effective emotion regulation strategies—may further intensify their vulnerability to online harm [10].

Moreover, among left-behind adolescents, reactive anger, parental migration patterns, gender, and grade level significantly predicted subgroup membership in both cyberbullying and cybervictimization. Higher levels of reactive anger increased the likelihood of belonging to marginal and direct-high-incidence cyberbullying classes, as well as medium- and high-victimization classes. This finding is consistent with emotional dysregulation models, highlighting that emotionally reactive left-behind youth are particularly vulnerable to occupying dual roles as both perpetrators and victims in online contexts [34–36]. Maternal or dual-parent absence heightened the risk of high-incidence cyberbullying, underscoring the protective role of maternal presence [32, 33, 37]. Girls were more likely than boys to enter direct-high-incidence cyberbullying classes, reflecting greater use of verbal and relational aggression [38–40]. Grade effects showed that second-year students were more likely to fall into marginal cyberbullying and higher victimization profiles, while first-year students were less likely to report high victimization. These results suggest that early adolescence, especially the second year of junior high school, represents a particularly sensitive stage for victimization risks, with weaker but still notable associations for cyberbullying [3, 5, 41].

Importantly, comparisons between left-behind and non-left-behind adolescents revealed distinct group differences in cyberbullying and cybervictimization profiles. Left-behind adolescents were disproportionately represented in high-incidence classes, whereas their non-left-behind peers were more often concentrated in low-incidence classes, underscoring the heightened vulnerability of left-behind youth to both perpetration and victimization in online contexts [9]. Interestingly, the role of reactive anger varied depending on the analytic frame of reference. Within the left-behind sample, higher levels of anger significantly increased the likelihood of membership in high-incidence bullying and victimization classes, consistent with prior evidence that emotional reactivity amplifies both aggressive behavior and perceived victimization [42]. However, in cross-group comparisons, reactive anger was negatively associated with high-incidence bullying and medium-level victimization, a pattern likely reflecting methodological differences in reference groups. Specifically, subgroup analyses capture intra-group variability among left-behind adolescents, whereas cross-group analyses emphasize relative risks between left-behind and non-left-behind peers [43, 44]. Taken together, these findings suggest that anger is a robust risk factor for left-behind adolescents, but its expression may depend on the social context.

Several limitations should be noted. First, the cross-sectional design prevents causal conclusions about the associations between predictors and cyberbullying or cybervictimization profiles. Longitudinal studies are needed to clarify developmental pathways and examine how factors such as reactive anger and parental absence shape cyberbullying behavior over time [36, 45]. Second, the reliance on self-reports may introduce biases related to social desirability. Future research should incorporate multi-informant methods to enhance validity, such as peer nominations, teacher ratings, or behavioral observations [46, 47]. Third, although LPA identified meaningful subgroups, it cannot capture dynamic peer influence or real-time interaction processes. Approaches such as social network analysis or ecological momentary assessment (EMA) may provide deeper insights into how cyber behaviors develop in everyday contexts [48].

Conclusion

This study identified three latent profiles of cyberbullying and cybervictimization among left-behind and non-left-behind adolescents, revealing shared structures but also meaningful group differences. Left-behind adolescents were overrepresented in higher cybervictimization profiles, whereas differences in high-incidence cyberbullying profiles across groups were less consistent. Reactive anger, parental migration patterns, gender, and grade level significantly predicted subgroup membership, with reactive anger emerging as the most robust correlate. These findings highlight the need for targeted interventions, especially for left-behind adolescents.

Author contributions

L.L. was responsible for the conceptualization and methodology of this study, analyzed the data and drafted the manuscript. B.X.Z. contributed to analyzed the data and drafted the manuscript. M.L.S. and Y.X.F helped investigate and collect the data. M.M. and B.P. helped review and revise the manuscript, and provided financial support. All authors have read and agreed to the published version of the manuscript.

Funding

This study was funded by Hunan Provincial Natural Science Foundation of China (No.2022JJ60003).

Data availability

The data supporting the conclusions of this study are available upon request to the corresponding author, Biao Peng.

Declarations

Ethics approval and consent to participate

The study was approved by the Institutional Review Board of the Guangzhou University. Prior to completing the questionnaire, all participant parents or legal guardians provided written informed consent. All the participants provided written informed consent prior to completing the questionnaire. The participants were informed that the study was voluntary, and they could discontinue at any time. All methods in the study were performed in accordance with the ethical standards of the institutional and/or national research committee and the 1964 Helsinki declaration and its later amendments.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.