As mentioned above, cyberbullying is defined as an aggressive and intentional act or behavior reiterated and/or maintained over time in electronic forms of contact by one or several individuals against a victim who cannot easily defend her/himself (Smith et al., 2008). There are different forms of cyberbullying. Written-verbal cyberbullying (insults or threats in messages, wounding comments through Internet, or intimidating anonymous calls), visual cyberbullying (recording and spreading compromising images or “happy slapping”, that is, aggression or humiliation that is recorded and spread), exclusion (deliberate omission from an online group), and impersonation are types of aggressive behavior presented on digital devices. And among these, verbal cyberbullying and exclusion seem to be the most common (Álvarez-García, Barreiro-Collazo, et al., 2017). However, other studies have shown verbal and visual cyberbullying with humiliating images as the most prevalent (Cuadrado-Gordillo & Fernández-Antelo, 2016).

The psychoemotional adjustment of adolescents is intimately related to victimization by internet (Evangelio et al., 2022; Quintana-Orts et al., 2021) and other types of exposure to violence (Vilariño et al., 2022). Some emotional characteristics have been related to higher risk of cybervictimization of adolescents, such as low esteem, feeling afraid or shame of revealing to others what they feel or what happens to them, high sensation-seeking (Save the Children, 2019), along with feeling lonely (Sahín, 2012; Save the Children, 2019), seem to be factors present in cyberbullying victims. Low emotional control (Hempehill & Heerde, 2014) has also been related to cybervictimization. In this regard, specific moods, such as anger, confusion and tension/anxiety are related to higher risk of being a cyberbullying victim or aggressor, which highlights the importance of emotional management in this issue (De Pasquale et al., 2021).

Moreover, cybervictimization has been related to a series of negative results in terms of the mental wellbeing of youths (González-Cabrera et al., 2020). Specifically, cybervictimization has been pointed out as a factor highly linked to psychological harm among adolescents (Montes et al., 2022). Victims of cyberbullying show a significantly higher prevalence of major depressive disorder than those who are not (Mallik & Radwan, 2020; Thai et al., 2022). Previous studies have shown cyberbullying to be a risk factor in increased depressive and anxiety symptoms in adolescents (Evangelio et al., 2022; Molero et al., 2022). This could in turn lead to feelings of isolation and loneliness, and less wellbeing. Similarly, a longitudinal study found cybervictimization to be a risk factor for future development of depressive and social anxiety symptoms, and diminishing wellbeing in adolescence (Fahy et al., 2016). Emotional self-efficacy and maladaptive emotional regulation, such as rumination, have an important role in this as mediators between cybervictimization and wellbeing. Thus, the capacity of youths to manage their negative emotions is in a way responsible for the adverse psychological consequences of cyberbullying (Schunk et al., 2022). The search for social support has also been shown to be a relevant variable in results derived from cyberbullying, as it moderates depressive symptoms. This coincides with the results found in other groups exposed to violence (Pérez-Fuentes et al., 2021), while other strategies, such as assertiveness or helplessness, are less effective (Machmutow et al, 2012). In this regard, adolescents who are more likely to employ unproductive strategies in facing cybervictimization are also more likely to develop depression, anxiety, and stress (McLoughlin, 2021).

McLoughlin et al. (2019) showed that face-to-face social connections protect against the negative results of mental health linked to cybervictimization. This could be because the connection with others through digital media (for example, social networks) seems to increase feelings of loss and loneliness, both when they do not use them (preventing communication) and when they do (since they feel excluded when they see what others are doing) (Twenge et al., 2021).

Cybervictimization also leads to other negative results such as the appearance of feelings of isolation and self-guilt, and even suicidal thinking and behavior (Dorol-Beauroy-Eustache & Mishara, 2021; Save the Children, 2019).

Research has shown differences in cybervictimization between boys and girls. Cybervictimization by girls seems to be more influenced by online gossip, that is, evaluative comments about third persons. Meanwhile, boys seem to be especially influenced by a problematic use of the internet, which is in turn linked to low self-esteem (Cebollero-Salinas et al., 2022). However, studies such as the one by Álvarez-García, Barreiro-Collazo, et al. (2017) on Spanish adolescents showed that there are no significant differences in the type of cyberbullying received between boys and girls, although there is a generalized trend in which girls are victims more than their male classmates, which coincides with other studies (Carvalho et al., 2021; Evangelio et al., 2022; González-Cabrera et al., 2020). This could be especially problematic considering that women who experience cyberbullying and have unstable family relationships and a problematic Internet use pattern are at a higher risk of becoming future cyberbullies. These adolescents, who show a dual pattern (online victims and bullies), are especially disposed to psychological alterations, such as depression, anxiety, and emotional problems (Lozano-Blasco et al., 2020). Women who are victims of cyberbullying and have a problematic use of internet also have emotional regulation problems (Arrivillaga et al., 2022).

Moreover, girls have been found to have a stronger tendency to choose coping strategies for dealing with cybervictimization focused on emotions than their male counterparts. And in turn, this type of strategy is related to higher levels of depression, anxiety, and stress (McLoughlin, 2021). Girls’ problems with emotional management have also been found to be related to victimization, but not boys’ (Rodríguez-Álvarez et al., 2021). It has also been shown that depression, and especially anxiety, may be caused by cybervictimization, particularly in women (Molero et al., 2022). All the above suggests that gender may be related to possible negative results of cyberbullying.

Cybervictimization has been related to poor mental health and internalizing symptoms (Beaudequin et al., 2020; Fahy et al., 2016; Mallik & Radwan, 2020). This study attempts to take one step further by relating specific emotional states to the concrete type of cybervictimization. This will provide us with more precise knowledge about whether certain internalizing behaviors are more closely related to types of cyberbullying.

Therefore, this study posed two objectives. First, it was intended to examine the network structure of items and dimensions in the Depression, Anxiety and Stress Scale (DASS-21) in a sample of Spanish adolescents. A network graph was created, where the components of the dimensions and their associations (edge weight and centrality), as well as the robustness and stability of the network, were evaluated.

The second objective was to test how the different modalities of cybervictimization (CYVIC) relate to the presence of emotional states (DASS-21), in this case, by means of network analysis of the dimensions (depression, anxiety, and stress).

The original sample had 2,241 participants who had agreed to participate and gave their consent, but 70 cases were removed from the database because they did not complete the whole battery of questionnaires. The final study sample was therefore comprised of 2,171 adolescents aged 11 to 18, with a mean age of 13.84 (SD = 1.46). Gender distribution revealed equitable representation with 50.53% (n = 1,097) males and 49.47% (n = 1,074) females with a mean age of 13.88 (SD = 1.45) and 13.81 (SD = 1.46), respectively. The courses they were studying at the time data were collected were distributed as follows: 25.51% (n = 554) 1^st year of compulsory education (ESO), 22.98% (n = 499) 2^nd year of compulsory education, 23.35% (n = 507) 3^rd year of compulsory education, 19.94 (n = 433) 4^th year of compulsory education, 4.33% (n = 94) 1^st year high school, 3.27% (n = 71) 2^nd year high school, and 0.59% (n = 13) vocational training.

This was a descriptive cross-sectional study. Before collecting the information, the school directors were contacted to inform them of the objectives of the study and guarantee confidential data processing. Two members of the research team went to the schools to administer the questionnaires. First, they gave the students the pertinent instructions and guaranteed them anonymity of their answers. The students filled in the tests individually in 25-30 minutes. In all cases, ethical research standards were met and all the participants voluntarily agreed to participate and had the written consent of their parents/guardians for their participation. The study was approved by the Bioethics Committee of the University of Almería (Ref: UALBIO2021/022).

First, a descriptive correlational analysis was carried out with the study variables. Mean scores and standard deviations were calculated, and the bivariate correlation matrix was estimated using Pearson’s correlation coefficient. Then any statistically significant gender differences were examined with a comparative analysis of the mean scores of all the study variables. Welch’s t-test (1947) and d (pooled, weighted, and unbiased formula) were performed to estimate mean differences and the effect size (0.20, 0.50, and 0.80 were interpreted as small, medium, or large effect sizes, respectively). Complementarily, the statistical model error, i.e., the probability of error of the statistical model in cases’ classification was estimated with the probability of inferiority score (Gancedo et al., 2021).

Instrument reliability was estimated using Cronbach’s (1951) alpha coefficient and McDonald’s (1999) omega (Ventura-León & Caycho, 2017).

Then, in view of the study objectives, two network models were estimated: Network 1 of emotional symptoms with all the DASS-21 items, and Network 2 combining the emotional symptoms (scales/dimensions) and the different types of cybervictimization (CYVIC). Networks were also estimated separately by gender for this model.

In network analysis, the variables correspond to nodes and connections between nodes are called edges. The strength of the relationship between the nodes was measured in terms of edge weights (the minimum is .30; Isvoranu et al. 2017). In this case, networks were estimated using JASP ver. 0.16.3 (JASP Team, 2022). Missing data were managed using pairwise deletion with the extended Bayesian information criterion (EBICglasso), with hyperparameter set at 0.5 (Epskamp & Fried, 2018). In the graphic representation of the network, edge weights are expressed in terms of edge thickness and color density, so thicker, brighter lines suggest stronger connections. Another aspect to consider is the distance between nodes, such that the nodes closest to each other would be the most closely related. Fruchterman and Reingold’s (1991) algorithm, which places the nodes with the strongest correlations in the center of the network, leaving those with weaker correlations farther out, was used for this.

In addition, centrality indices (betweenness, closeness, degree, and expected influence) were calculated, although in this study we focused on degrees or strength, which reflect the direct influence that a certain node has in the network.

Finally, for Network 1, edge accuracy was also examined using bootstrap 95% non-parametric confidence intervals (CIs), and stability of centrality estimates using the case-dropping bootstrap procedure. In both cases, 1,000 bootstraps were estimated.

Table 1 shows mean scores and standard deviations of the study variables, and correlation matrix. As shown, positive associations were found (in all cases with a significance of p < .001) between the modalities of cybervictimization (impersonation, visual sexual cybervictimization, written-verbal cybervictimization, and online exclusion) and the dimensions of the DASS-21 (depression, anxiety, and stress) as emotional symptoms.

In addition, a comparative analysis was conducted to test for the existence of statistically significant gender differences in the study variables. Levene’s test was significant (p < .05), suggesting a violation of the assumption of homogeneity of variance, and therefore, Welch’s t-test was applied to estimate gender differences. As shown in Table 2, mean scores were significantly higher in impersonation, t = 2.32, p < .05, d = 0.10, and visual sexual cybervictimization, t = 2.76, p < .01, d = 0.12, for boys. On the contrary, 46.0% (PIS = .460) of the males had scores under the mean of the females’ group (error of the statistical model) in impersonation and in visual sexual cybervictimization, respectively.

Concerning the presence of emotional symptoms (see Table 2), statistically significant differences were observed in the three DASS-21 dimensions: depression, t = -9.93, p < .001, d = -0.43, anxiety, t = -10.16, p < .001, d = -0.44, and stress, t = -7.77, p < .001, d =- 0.33, where girls had higher mean scores in all cases. Conversely, the probability of girls with a score under the mean of boys (error of the statistical model) was 33.4%, 32.9%, and 37.1% (PIS = .334, .320, and .371) in depression, anxiety, and stress, respectively.

With 21 nodes, the maximum number of edges in this network was 210. However, EBICglasso estimation used in the analysis reduced the estimated number of edges to 147. Sparsity is a value between 0 and 1, where the higher the sparsity, the more weakly connected the network. In this case, the sparsity index was .30. As shown in Figure 1, all the nodes were positively associated with each other. Furthermore, the stronger the association between nodes, the thicker and brighter the network edge.

Table 3 shows the indices of centrality (betweenness, closeness, strength, and expected influence) for the items on the DASS-21 (nodes). Graphs were created of these measures to facilitate data interpretation (Figure 2).

As observed in Table 3 and Figure 2, the strongest nodes with a relatively high number of connections were: “close to panic” (DASS15), “close to panic” (DASS15), “difficult to relax” (DASS12) and “meaninglessness of life” (DASS21). Furthermore, the nodes with the lowest centrality of strength were “dryness of mouth” (DASS2), “intolerance of interruptions” (DASS14), and “nervousness” (DASS8). We focused on strength as the main indicator of centrality because it shows reasonably accurate estimations for network analysis in psychology (Santos et al., 2018).

Finally, the accuracy and stability of the network were evaluated. Confidence intervals around some of the edges estimated were large, which suggests that many of them probably did differ significantly from each other, and this could prevent interpretation of the order of the network edges.

Figure 2 shows graphically the stability of the centrality indices examined. Case-dropping bootstrap, which resamples with different data subsets, was used. The correlation stability coefficient of the data subset represents different percentages of the general sample. Associations between estimations decreased from the original sample to the subsample from 95% to 25%. The correlations for all the indices of centrality showed a certain stability, as they were above .50 (Epskamp et al., 2018), with strength above .70 (traditionally interpreted as an indicator of very large effect in behavior sciences; Cohen, 1988) (Figure 3).

Figure 4 shows the combined network of the DASS-21 emotional symptom subscales (depression, anxiety, and stress) and the CYVIC dimensions, which represent different types of cybervictimization. This network had seven nodes, and the maximum number of edges was 21, with 16 remaining after EBICglasso estimation. In this case, the sparsity index was .23. Connectivity between cybervictimization nodes was stronger than with the emotional symptom scales.

Of the three DASS-21 scales, anxiety (2) had the most visible associations with the written-verbal (6) and online exclusion (7) cybervictimization modalities, with edge weights of .080 and .058, respectively. There was no association at all with other types of cybervictimization, such as impersonation or visual sexual.

All the connections with the different types of cybervictimization were positive, but with stronger edge weights between written-verbal (6) and online exclusion (7) with .350, followed by connections between impersonation (4) and written-verbal (6) and visual sexual cybervictimization (5), with weights of .316 and .308, respectively.

Other connections between CYVIC nodes showed associations between online exclusion (7) and impersonation (4) and visual sexual cybervictimization (5), and this one with written verbal (6), with edge weights of .165, .229, and .183, respectively.

Centrality indices of strength for impersonation (4), visual sexual cybervictimization (5), written-verbal (6), and online exclusion (7) were -0.684, -1.152, 1.196, and -0.299, respectively. Thus, written-verbal victimization may be considered the center node of the cybervictimization modalities in the combined network of emotional symptoms-cybervictimization.

With the estimation of Network 1 above, it was confirmed that the strongest nodes were DASS15, DASS12, and DASS21, represented by an element of each subscale: anxiety, stress, and depression, respectively. In this case, with the estimation of the combined network with different types of cybervictimization, the anxiety node was the strongest (1.552).

Finally, the same combined network was estimated by gender. Figure 5a shows the network for males and Figure 5b for females. In both cases, the network has seven nodes, with 14 edges other than 0 after estimating with EBICglasso, and a dispersion index of 0.33.

Of the DASS-21 subscales, anxiety (2) had associations with the written-verbal (6) and online exclusion (7) cybervictimization modalities, with edge weights of .077 and .057 for males and .080 and .074 for females, respectively. Depression (1) also was associated, although weaker, with written-verbal (6) and online exclusion (7), with weights of .044 and .033 for males, and .053 and .020 for females, respectively. Differences on the Stress (3) scale were in the edge weights of both networks, with no association with type of cybervictimization, except in the network estimated for females, where stress was associated with online exclusion (7) with a weight of .031.

In the different types of cybervictimization in the network estimated for males, the edge weights were stronger (.364) between impersonation (4) and written-verbal (6), followed by the connection between written-verbal (6) and online exclusion (7) with .346. Moreover, looking at the network estimated for females, the connection between the written-verbal (6) and online exclusion (7) stands out, followed by impersonation (4) with visual sexual cybervictimization (5) with .285.

Table 4 shows the centrality indices in the network estimated for both genders. Both networks show stronger centrality indices for written-verbal cybervictimization, and the DASS-21 anxiety subscale, with values of 1.150 and 1.235 for males, and 1.191 and 1.419 for females, respectively.

Gender differences observed between the networks emphasized the impersonation and stress nodes. Even though the strength of the impersonation node has low-strength centrality indices in both networks, it is weaker for females (-1.079) than males (-0.092). In the network estimated for males, the stress node also has a weaker centrality index (-1.214) than the same node in the female network (0.169). The contrary occurs with the depression node, where it is the network estimated for males which has a stronger centrality index (0.108) than females (-0.142).

Thus, observing the data presented in Table 4, it may be said that in the gender networks estimated, anxiety was the strongest node in both cases. However, differences were observed in the centrality of depression and stress nodes: the first one (depression) was stronger in the male network estimated and the second one (stress) was stronger in the female network.

This study enabled the most important nodes in the structure of emotional states to be identified and, therefore, assisted in elucidating the specific symptoms of greatest importance to the emotional wellbeing of adolescents. This could be used to set up specific programs to curtail the psychoemotional symptoms that could develop into mental health problems. In particular, feelings of being about to have a panic attack, difficulties relaxing, or thinking that life is meaningless are major symptoms affecting their psychoemotional wellbeing and could trigger negative circular structural emotional thinking. Therefore, schools could work on emotional intelligence to palliate these symptoms in adolescents. This in turn would enable work with youths who show comorbid symptoms affecting their psychoemotional wellbeing, but not meeting diagnostic criteria for a specific category of affective disorders. Another possible application of the findings of this study would be in promoting intervention addressing anxiety in adolescents who are cyberbullying victims.

Future lines of research could find out whether the emotional symptom network structure differs between populational groups (e.g., at-risk adolescents, preadolescents, and emerging adults). Secondly, in line with the limitations mentioned, measures enabling the identification of specific manifestations of difficulties experienced by adolescents who have been the object of cybervictimization should be continued. Thus, finding out whether the emotional symptom network of youths who experience cyberbullying is similar to the rest of adolescents could help understand its effects.

The new technologies have meant unprecedented progress for society, but they also pose new problems. Cyberbullying is highly prevalent among youths and, therefore, attention must necessarily be given to its consequences. The results of this study show that negative emotional states and types of cybervictimization are related in a complex network where the efforts to approach reduction of emotional symptoms associated with it could be of lasting benefit to the emotional wellbeing of youths. Thus, focusing attention on the central nodes (such as anxiety) in the network could be essential to preventing a cascade of negative emotional symptoms in young people who suffer from cyberbullying. This could put an end to early forms of cybervictimization or provide adolescents with strategies for coping with them to slow down or head off the appearance of other forms of cyberbullying.