According to previous research, adolescents who are victims of online and offline violence show greater levels of PSNSU than non-victimized adolescents (Longobardi et al., 2020; Martínez-Ferrer et al., 2018). The explanation for this lies mainly in the fact that victimized adolescents spend more time and effort than their non-victimized peers on social networking sites (SNS) because they find in them an outlet to escape from unwanted moods and distress associated with particularly painful and stressful interpersonal experiences (Boniel-Nissim & Sasson, 2018). Specifically, victimized students find in SNSs a virtual support space beyond their geographical boundaries (Echeburúa & De Corral, 2010) where they can build their self-confidence (Boursier & Manna, 2018), develop close and intimate relationships (Tzavela et al., 2015), and feel connected to peers by belonging to a virtual community (Etgar & Amichai-Hamburger, 2017). The problem is that this coping strategy leads the victims to become overly preoccupied with SNSs and to devote a lot of time and energy, causing them to withdraw from other life cycle activities, and thus creating attitudes of dependence and a lack of control over their behaviors that negatively impact the normal development of their daily lives (Schou Andreassen & Pallesen, 2014; Zimmer-Gembeck et al., 2021).

In addition, it has been found that adolescents who spend more time connected to SNSs are more likely to engage in violent online behaviors (Baldry et al., 2019), such as cyberbullying (Craig et al., 2020; Kırcaburun et al., 2019). This relationship could be explained by several theoretical frameworks, in particular the problem-behavior theory (Jessor, 1991). According to this theory, adolescents who engage in a risky or problematic activity are also likely to participate in other risky behaviors. In this regard, SNSs are linked to greater Internet use and, therefore, greater exposure to cyberspace risks (Machimbarrena et al., 2021). More specifically, the relationship between PSNSU and cyberbullying could be attributed to the general strain theory of Internet addiction and deviant behavior in social networking sites (SNS) (Mubarak & Quinn, 2019). This theory suggests that aggressive behavior in the virtual environment is associated with the stress and anxiety felt by adolescents caused by their excessive desire to connect to social networks and their inability to emotionally self-regulate, which could explain this relationship.

Consequently, victims would be more likely to develop a problematic use of SNSs than a non-victimized adolescent and engage in aggressive behavior towards their peers in the online environment. Nonetheless, little is known about the contribution of adolescents’ problematic use of SNS (PSNSU) when examining the overlap and continuity of roles in both harassment dynamics in adolescents. Further research on this issue could provide useful information about transitions from offline to online contexts.

Previous studies have observed a greater use of SNSs as spaces for social interaction among girls (Park et al., 2019; Yang et al., 2018), whereas boys are more likely to use SNSs for entertainment, such as viewing videos or photos, and playing online games (Su et al., 2020). Moreover, studies have revealed that girls, compared to boys, are more vulnerable to the negative consequences (psychological and emotional problems) of different expressions of aggressive behavior both online and offline (Baldry et al., 2017; Zhang et al., 2020). These particularly stressful interpersonal experiences may, in turn, prompt girls to turn to these virtual spaces for emotional relief and support (Mariën et al., 2021). Thus, the association between peer victimization and problematic use of SNSs may be stronger in girls than in boys (Liu et al., 2020). Regarding cyberbullying, girls have been found to be more vulnerable to cyberbullying through social networks by denigrating comments about their appearance and by private or compromising photos within their peer group, while boys are more likely to be cyberbullied through online games (Gianesini & Brighi, 2015).

In view of the negative consequences of cyberbullying, we believe it is of theoretical and practical importance to explore those factors that may contribute to its emergence and increase among adolescents. Prior studies regarding PSNSU and its relations to offline and online victimization and cyberbullying are scarce and primarily focused on the direct relationship between PSNSU and cyberbullying or victimization. The mediating mechanism (i.e., how PSNSU connects offline and online victimization with cyberbullying) remains largely unknown. Therefore, the current study extends previous research by investigating whether offline and online victimization are associated with cyberbullying and whether the problematic use of social networking sites mediates this relationship. In addition, this study explored whether this mediation process was moderated by gender.

Thus, the present study had two aims. The first objective was to analyze a moderated mediation model in which PSNSU mediated the relationship between offline and cybervictimization and cyberbullying. The second aim of our study was to analyze the moderate effect of gender in direct and indirect relations between offline and cybervictimization, PSNSU, and cyberbullying. Based on the theoretical model (Figure 1a and 1b) we proposed the following hypotheses:

H1: Both cybervictimization and offline victimization will relate positively to cyberbullying.

H2: Both cybervictimization and offline victimization will be indirectly associated with cyberbullying, through involvement with PSNSU (mediating effect).

H3: Gender will moderate the mediating effect of PSNSU on the association between both cybervictimization and offline victimization, and cyberbullying. In particular, the indirect association will be greater for girls than for boys.

H4: Boys victimized both online and offline will report greater involvement in cyberbullying.

H5: Girls victimized both online and offline will report higher PSNSU.

The universe of the present study consisted of adolescents of both sexes between 12 and 18 years old. Data in this research were obtained from a sample of 2,011 adolescent students (50.67% boys and 49.32% girls) selected from 4 secondary schools (2 public and 2 state-subsidized) in Andalusia, Spain. Participants were aged between 12 and 18 years (M = 14.17, SD = 1.47) and were selected randomly. A sampling error of ± 2.5% and a confidence level of 95% were established with an expected population variance of .50.

Cybervictimization Scale (CYB-VIC) was used (Buelga et al., 2012). This Likert-type scale is composed of 18 items with a response range from 1 (never) to 4 (always) that rates how frequently an adolescent has been subjected to aggression in the online environment (e.g., “They have hacked into my private account without my being able to do anything”). The Cronbach’s alpha obtained in the present sample was acceptable (α = .82).

Peer Victimization Scale (Mynard & Joseph, 2000), adapted to Spanish by Martínez-Ferrer et al. (2018). It was used to assess offline victimization. It consists of 22 Likert-type items with a response range from 1 (never) to 4 (always) that evaluates how frequently the adolescent has been subjected to aggression in the offline environment. The scale consists of three dimensions: relational victimization (e.g., “A peer has told my secrets to others”), overt physical victimization (e.g., “A peer has beaten me up”), and overt verbal victimization (e.g., “A peer has insulted me”). The Cronbach’s alpha obtained in the present sample was acceptable (a = .92).

Problematic Use of Virtual Social Networks Scale (Martínez-Ferrer et al., 2018). This Likert-type scale is composed of 13 items with a response range from 1 (never) to 4 (always) that evaluates problematic use of SNSs in adolescents (e.g., “I need to be connected to my social networks continuously”). The Cronbach’s alpha obtained in the present sample was acceptable (a = .87).

Cyberbullying Scale (CYB-VIC) (Buelga & Pons, 2012). This Likert-type scale consists of 18 items with a response range from 1 (never) to 4 (always) that measures participation in aggressive behavior through the use of digital media (e.g., “I have sent or manipulated photos or videos of someone without their permission”). The Cronbach’s alpha obtained in the present sample was acceptable (a = .81).

First, we contacted school administrators of the educational centers, explained the objectives and scope of the study, and asked for their participation on a voluntary basis. In addition, an informative meeting was held with the teaching staff in order to inform the entire educational community about the study. At the same time, a letter explaining the study was sent to families to obtain their consent for their children’s participation in the study. Once the authorizations from the educational centers to carry out the research and the active informed consent from families had been obtained, the battery of instruments was administered in two different sessions of approximately 45 minutes during school hours. Furthermore, this study complied with the ethical values required in research with human beings, respecting the fundamental principles included in the Declaration of Helsinki (1964) in its updates and in regulations in force: informed consent and right to information, protection of personal data and guarantees of confidentiality, non-discrimination, gratuity, and the possibility of abandoning the study at any stage.

The average of missing data was 2.1% and never above 5% for an individual measure. The low level of missingness meant that it was not likely to bias the results, thus estimations were accurate to the expected values on the population (Graham, 2009). Missing values by scales or subscales were processed using the regression imputation method. In this method, rows in the data matrix are presumed to constitute a random sample of a normal multivariate population. Univariate outliers were detected via the exploration of standardized scores. Following the criteria provided by Hair et al. (2016), atypical values were those whose standardized scores had an absolute value above 4. For multivariate detection, Mahalanobis distance was computed. A multivariate outlier is identified if the associated probability at a Mahalanobis distance is .001 or less (Tabachnick & Fidell, 2007).

First, descriptive analyses (means and standard deviations) and bivariate correlations were calculated to explore all the variables included in the analysis. Furthermore, a t-test was performed to examine the differences of the study variables as a function of gender. Next, PROCESS macro (Model 59) (Hayes, 2013) was used to test the mediating effect of PSNSU on cybervictimization, offline peer victimization, and cyberbullying. Mediation was deemed to be statistically significant if confidence intervals did not include zero. In addition, the analysis of moderated mediation models was completed using Hayes’s PROCESS macro (Model 59). All analyses were conducted using SPSS 25.

Table 1 outlines means, standard deviations, and zero-order correlations among all study variables separately for boys and girls. Findings revealed that cyberbullying, cyber victimization, offline victimization, and PSNSU were positively related for boys and girls. Next, an independent sample t-test was conducted to test differences among study variables according to gender. As shown in Table 1, boys reported higher levels in cyberbullying, whereas girls reported higher levels in cybervictimization and PSNSU.

The first aim of the study was to examine the mediating role of PSNSU in the relationship between both cybervictimization and offline peer victimization and cyberbullying. Results are presented in Table 2.

Regarding cybervictimization, Model 1 of Table 2 showed a positive and significant effect of both cybervictimization and PSNSU on cyberbullying (ß = .48, t = 24.31, p < .001 and ß = .19, t = 9.38, p < .001 respectively). Model 2 indicated that cybervictimization had a significant positive effect on PSNSU (ß = .29, t = 13.61, p < .001). In Model 3, the mediating effect was examined. Results found that the direct effect of cybervictimization on cyberbullying was still significant (ß = .42, t = 21.03, p < .001). Moreover, results of the bias-corrected percentile bootstrap method revealed that PSNS mediated the indirect association between cybervictimization and cyberbullying (effect = .05, SE = .01, 95% CI [.04, .07]). The mediation effect of PSNSU accounted for the 13.12% of the total effect of cybervictimization on cyberbullying (see Figure 2a).

Concerning offline peer victimization, Model 1 of Table 2 showed a positive and significant effect of both offline peer victimization and PSNSU on cyberbullying (ß = .27, t = 12.73, p < .001 and ß = 26, t = 12.35, p < .001 respectively). Model 2 revealed that offline peer victimization had a significant positive effect on PSNSU (ß = .20, t = 9.24, p < .001). In Model 3, the mediating effect was examined. Results showed that the direct effect of cybervictimization on cyberbullying was still significant (ß = .32, t = 14.99, p < .001). Furthermore, results of the bias-corrected percentile bootstrap method indicated that PSNS mediated the indirect association between cybervictimization and cyberbullying, (effect = .05, SE = .01, 95% CI [.04, .07]). The mediation effect of PSNSU accounted for the 16.25% of the total effect of cybervictimization on cyberbullying (see Figure 2b).

The second aim was to examine whether gender would moderate direct and indirect relations between cybervictimization/offline victimization and cyberbullying through PSNSU. Results are reported in Table 3. To analyze the moderated mediation, we estimated parameters for two regression models with Hayes’ (2013) PROCESS macro (Model 59) and 95% bias-corrected bootstrapped confidence interval (CI) based on 10,000 bootstrap samples at values of one standard deviation (SD) below the mean, at the mean, and one SD above the mean.

We tested whether gender moderated (a) the relation between cybervictimization/offline victimization and cyberbullying, (b) the relation between cybervictimization/offline victimization and PSNSU, and (c) the relation between PSNSU and cyberbullying.

Regarding cybervictimization, we analyzed gender as a moderator on the associations (a) between cybervictimization and cyberbullying (Model 1), (b) between cybervictimization and PSNSU (Model 2), and (c) between cybervictimization and cyberbullying after controlling for PSNSU (Model 3). Specifications of these three models can be seen in Table 3. The association (a) between cybervictimization and cyberbullying, and/or (b) between cybervictimization and PSNSU was moderated by gender.

As shown in Table 3, results of Model 1 indicated that cybervictimization and PSNSU were positively related to cyberbullying (ß = .80, p < .001, ß =.19, p <.01, respectively), whereas gender was negatively related to cyberbullying (ß = -.26, p < .01). It was also observed that there was a moderation effect of cybervictimization and gender (ß = -.24, p < .001). Model 2 revealed that the association between cybervictimization and PSNSU and between gender and PSNSU were significant (ß = .15, p < .05, ß =.13, p <.05, respectively). In addition, the relationship between cybervictimization and PSNSU was moderated by gender (ß = .09, p < .05). However, the association between PSNSU and cyberbullying was not moderated by gender (ß = .01, p = .81).

With regard to the moderation effects, first we plotted predicted cyberbullying and cybervictimization separately for boys and for girls (Figure 3a) as well as predicted PSNSU and cybervictimization separately for boys and girls (Figure 3b). As presented in Figure 3a, simple slope tests indicated that for boys higher levels of cybervictimization were related to higher levels of cyberbullying than for girls (ß_boys = .56, t = 19.57, p < .001, 95% CI [.50, .61], ß_girls = .32, t = 11.57, p < .001, 95% CI [.26, .37]).

As presented in Figure 3b, by contrast, higher levels of cybervictimization were related to higher levels of PSNSU for girls than for boys (ß_boys = .24, t = 7.56, p < .001, 95% CI [.18, .30, ß_girls = .33, t = 11.19, p < .001, 95% CI [.27, .38]). The bias-corrected percentile bootstrap found that the indirect effect of cybervictimization on cyberbullying through PSNSU was not moderated by gender (effect = .02, SE = .02, 95% CI [-.01, .05]). Next, we examined the conditional indirect effects of cybervictimization on cyberbullying by gender.

Concerning offline peer victimization, we assessed gender as a moderator on the associations (a) between offline victimization and cyberbullying (Model 1), (b) between offline victimization and PSNSU (Model 2), and (c) between offline victimization and cyberbullying after controlling for PSNSU (Model 3). If the association (a) between offline victimization and cyberbullying, and/or (b) between offline victimization and PSSNS was moderated by gender, the moderated mediation model was established (Hayes, 2013).

Results of Model 1 indicated that offline victimization and PSNSU were positively related to cyberbullying (ß = .47, p < .001, ß = .27, p < .01, respectively), whereas gender was negatively related to cyberbullying (ß = -.18, p < .01). A moderation effect of offline peer victimization and gender (ß = -.15, p < .001) was also found. Model 2 showed that gender was positively related to PSNSU (ß = .19, p < .01), whereas offline peer victimization was not a significant predictor of PSNSU (ß = -.02, p = .77). However, the relationship between offline peer victimization and PSNSU was moderated by gender (ß = -.13, p < .001).

With respect to the moderation effects, for descriptive purposes, we plotted predicted cyberbullying and offline peer victimization separately for boys and for girls (Figure 4a). Simple slope tests indicated that for boys higher levels of offline peer victimization were related to higher levels of cyberbullying than for girls (ß_boys = .33, t = 11.75, p < .001, 95% CI [.27, .38], ß_girls = .18, t = 5.78, p < .001, 95% CI [.12, .24]).

Next, we plotted predicted PSNSU and offline victimization separately for boys and girls (Figure 4b). Simple slope tests revealed that for girls higher levels of offline peer victimization were related to higher levels of PSNSU than for boys (ß_boys = .15, t = 4.99, p < .001, 95% CI [.09, .20, ß_girls = .27, t = 8.52, p < .001, 95% CI [.21, .34]).

Results of the bias-corrected percentile bootstrap found that the indirect effect of cybervictimization on cyberbullying through PSNSU was not moderated by gender (effect = .03, SE = .02, 95% CI [.00, .07]).