Abstract
Purpose
An emerging body of literature suggests that victims of bullying report detrimental mental health outcomes, such as depression and anxiety. The purpose of this study was to explore the relationship between cybervictimization, depression, and anxiety among school-aged youth over a three-year timeframe.
Methods
Students in grades six through 12 at the initial wave of the study responded to survey items designed to assess their online experiences, including cybervictimization, and self-reported depression and anxiety at three separate time points, over a three year period. In total, 559 school-aged youth participated in the study.
Results
Results suggest a reciprocal relationship between cybervictimization and depression, and cybervictimization and anxiety. More specifically, depression at time 1 predicted cybervictimization at time 2, depression at time 2 predicted cybervictimization at time 3, and cybervictimization at time 1 predicted depression at time 3. Additionally, cybervictimization at time 1 predicted anxiety at time 2, cybervictimization at time 2 predicted anxiety at time 3, and anxiety at time 1 predicted cybervictimization at time 2.
Conclusions
Based on the findings from this study, cybervictimization, depression, and anxiety seem to have a reciprocal relationship. Therefore, educational and mental health professionals should consider interventions that address adolescents’ online experiences, while supporting mental health and social and emotional learning.
Keywords: Cyberbullying, Victimization, Anxiety, Bullying, Harassment, Longitudinal, Internet, Depression
In an era of evolving technology, where youth have always experienced a world with internet access, students are engaging in electronic activities for communication, socialization, leisure, and knowledge acquisition. While these advancements in technology have transformed the social milieu, affording students immediate access to information and potentially increasing intellectual capacity [1], they have also established a precedent for perpetual “connectedness.” For example, 95% of adolescents have access to the internet [2], with electronic media exposure eclipsing ten hours per day [3]. Although the benefits to these advancements cannot be disputed [4], perpetual “connectedness” may set the stage for increased risk of cybervictimization.
We draw on Hinduja and Patchin’s (2009) definition of cyberbullying victimization as “willful and repeated harm inflicted through the use of computers, cell phones, and other electronic devices” [5 (pp 5)], as well as recognize experiences that may occur only once (i.e., online harassment). Drawing parallels with traditional bullying literature, cyberbullying may include an imbalance of power, where individuals cannot easily defend themselves [6], which may be attributed to the technological aptitude of the aggressor [6]. Cyberbullying can include several topographies of behaviors [7] and manifest through diverse contexts [6]. Given the breadth of avenues for cybervictimization, and high rates of electronic media consumption, it is conceivable that long-term exposure to cyberbullying is related to negative psychosocial outcomes.
Traditional versus Cyberbullying and Mental Health
Given the advancing landscape of technology and media consumption among youth, we initially examine the relationship between depression and cybervictimization through a traditional bullying lens. Kaltiala-Heino, Rimpelä, Marttunen, Rimpelä, and Rantanen [8] suggested that students who experienced frequent victimization were 4.2 times more likely to report depression than students who are not victimized. These findings suggest that students who experience moderate to high levels of traditional victimization report higher levels of depression [9–11]. Ttofi, Farrington, Lösel, and Loeber [12] conducted a meta-analysis examining victimization and depression, and determined that victims of school bullying were 1.74 times more likely to report depression during subsequent time points. However, in a 4-year longitudinal study, Klomek, Kleinman, Altschuler, Marrocco, Amakawa, and Gould [13] determined that students who are directly involved in bullying without comorbid serious depression or current suicidal ideation were significantly less likely to report depression and suicide ideation over time than students who began the study as perpetrators, victims, or bully-victims with comorbid depression or current suicidal ideation.
Cross-sectional cyberbullying studies have revealed that excessive internet use is related to detrimental psychosocial symptoms [14,15], and students who experience cybervictimization report higher levels of depression than students who have not experienced cyberbullying [16,17]. Therefore, an overlap between traditional victimization and cybervictimization exists; yet an emerging body of literature also suggests that these forms of victimization may diverge. Bonanno and Hymel [18] suggest that involvement in cyberbullying, either as a perpetrator or victim, uniquely predicted depression and suicidal ideation above traditional bullying involvement. Schneider, O’Donnell, Stueve, and Coulter [19] reported that students who only experience traditional victimization, cybervictimization, and both forms of victimization are 2.31, 3.26, and 5.64 times more likely to report depression than students who are not victimized respectively. However, Gini and Espelage [20] reported that cybervictimization was more strongly associated with suicidal ideation than traditional forms of bullying.
Similar to depression, an empirical link between anxiety and victimization has been established. Depression and anxiety are common outcomes related to traditional victimization [21], and comorbidity is common [22]. Theoretical models suggest that comorbidity may be attributed to overlap in definitions, may represent a single construct, reciprocity between the constructs, population stratification, or a fundamental link between predictive factors [23]. Cummings and colleagues [23] “maintain that anxiety and depression are separate but meaningfully related constructs” (pp 823), where they propose a multiple pathways model that suggests comorbidity may exist in a subset of individuals that maintain parallel antecedents for both depression and anxiety, but in other subsets of individuals comorbidity may develop due to the reciprocity of depression and anxiety. This development may be attributed to a combination of social/environmental, biological, and behavioral risk factors. Anti-social or adverse interactions with peers, including victimization, may serve as a social/environmental risk factor for anxiety [11,24].
The social and environmental factors associated with physical environments may deviate from the purview of online environments. For example, there is a perception of anonymity in online environments [25], yet the perpetual “connectedness” transcends physical and electronic boundaries, where cyberbullying may result in immediate repetition through electronic distribution to large subsets of peers [26]. To illustrate this point, Dempsey and colleagues [26] determined that cybervictimization was a distinctly different construct from traditional and relational victimization, and increased levels of cybervictimization was associated with escalated rates of social anxiety, but not depression.
Are Mental Health Problems Antecedents or Consequences of Cybervictimization?
A small number of longitudinal studies of cyberbullying show associations with poor mental health, substance use, and problem behavior [27,28]. A central question these studies attempt to answer is whether youth that exhibit depressive symptoms go into online spaces or behave in ways that make them more susceptible to experiencing cybervictimization or is depression a consequence of these experiences [29]. For example, Gamez-Guadix, Orue, Smith, & Calvete’s [27] longitudinal study explores associations between cybervictimization, depression, substance use and problematic internet use. They found that victimization at time 1 predicted depression and problematic internet use at time 2. In addition increased levels of depressed mood and substance use at time 1 predicted more victimization at time 2. However, additional time points may be needed to fully understand the temporal sequence of cybervictimization and mental health outcomes.
The Current Study
The current study was designed to directly address these gaps in the literature by exploring the longitudinal relationship between cybervictimization and depression and anxiety, where anxiety and depression were examined as separate constructs, guided by the following research questions. What is the predictive relationship between cybervictimization and depression, and cybervictimization and anxiety over time?
Method
Population
The sample for the current study was derived from a federally funded, longitudinal study designed to explore the online experiences adolescents, with an a priori emphasis on racial and ethnic diversity. The sample includes 559 students that participated in the study over three waves of data collection, with 251 males (44.9%) and 308 females (55.1%) from 12 Midwestern schools in grades 6 (17.5%), 7 (15.0%), 8 (17.2%), 9 (13.1%), 10 (27.7%), 11 (10.7%), and 12 (3.8%) during the initial administration of the study. The racial breakdown includes 183 (32.7%) African American, 180 (32.2%) Caucasian, 96 (17.2%) Latino/a, 44 (7.9%) Biracial, 36 (6.4%) Asian or Asian American, 12 (2.1%) Native American, and 8 (1.4%) “other.” The participation rate averaged 49.8%, with a retention rate of 71% from time 1 to time 2, and 74% from time 2 to time 3 (see Figure 1).
Figure 1.
Total sampling distribution over three time points.
Procedure
Respondents were recruited in collaboration with school administrators, where research assistants gave a 10-minute presentation to selected classes, and distributed approximately 150 parental consent forms per school, which were available in English and Spanish. Care was taken to ensure students were representative of the general population of each school, including a range of SES backgrounds, technical and cognitive abilities, and racial-ethnic groups. On a prearranged date, researchers returned to each school to administer surveys via web link to students who returned the affirmative parental consent forms. During administration, research assistants were present to inform students of confidentiality, explain terms, and troubleshoot any technical difficulties. Surveys were completed at three separate time points over the duration of the study, and each subsequent survey administration was completed between 10 to 12 months following the previous administration. Upon completion or termination of the survey, all students were provided with online counseling and safety resources. The Institutional Review Board of the principal investigator’s institution approved all procedures, and all necessary approvals were acquired from partner schools and school boards.
Measures
Cybervictimization
The 6-item Online Bullying and Harassment subscale from the Online Victimization Scale [30] was used to measure respondents’ experiences with cybervictimization. To assess the longitudinal experiences of cybervictimization, respondents were asked to “choose how many times these things happened to you online in the LAST YEAR.” The cybervictimization items included “People have said mean or rude things about the way that I talk (write) online,” “People have posted mean or rude things about me on the Internet,” “I have been harassed or bothered online because of something that happened at school,” “I have been embarrassed or humiliated online,” “I have been bullied (repeated name calling or harassment) online,” and “People have said mean or rude things about how I look, act, or dress online.” Response options included “0 = Never,” “1 = Once,” “2 = A Few Times a Year,” “3 = A Few Times a Month,” and “4 = Almost Daily.”
The Online Victimization Scale [30] was validated in a two-phase confirmatory factor analysis with a diverse sample of adolescents in grades 9 through 12. In phase 1, the four factor model (i.e., online bullying and harassment, online sexual victimization, individual online racial discrimination, vicarious online racial discrimination) demonstrated acceptable model fit (Χ2(186) = 556.709, RMSEA = .096, TLI = .916, CFI = .927, IFI = .927). In the second phase, an additional confirmatory factor analysis was conducted with a separate diverse sample to demonstrate consistency across samples. In this phase, the four factor model mirrored the previous phase and demonstrated acceptable model fit (Χ2(183) = 570.303, RMSEA = .087, TLI = .930, CFI = .939, IFI = .939). To demonstrate generalizability to the current sample, a confirmatory factor analysis was conducted for the current sample, where the Bullying and Harassment subscale of the Online Victimization Scale [30] maintained acceptable model fit (Χ2(15) = 36.596, RMSEA = .051 (.030; .072), TLI = .987, CFI = .995, IFI = .995). Cronbach’s α for the current study was .84, .90, .92 for Time points 1 through 3 respectively.
Depression
A12-item version of the Center for Epidemiologic Studies Depression Scale (CESD-12) was used to measure adolescents’ depressive symptoms [31]. Sample items include “I felt I was just as good as other people” and “I had crying spells”. Response options ranged from “0=Rarely or none of the time” to “3=Most or all of the time.” To establish generalizability to the current sample, a confirmatory factor analysis was conducted for the current sample, where the CESD-12 [31] demonstrated acceptable model fit (Χ2(15) = 32.570, RMSEA = .046 (.024; .067), TLI = .978, CFI = .991, IFI = .991). Cronbach’s α for the current study was .71, .73, .77 for Time points 1 through 3 respectively.
Anxiety
Anxiety was assessed using four items from the Tension subscale of the Profile of Mood States-Adolescents [32]. Participants were asked to describe the extent that they felt each of the following: “panicky,” “anxious,” “worried,” and “nervous.” Responses ranged from “0=Not at all” to “4=Extremely.” To establish generalizability to the current sample, a confirmatory factor analysis was conducted for the current sample, where the Tension subscale of the Profile of Mood States-Adolescents [32] demonstrated acceptable model fit (Χ2(15) = 19.684, RMSEA = .024 (.000; .049), TLI = .996, CFI = .998, IFI = .998). Cronbach’s α for the current study was .84, .90, .92 for Time points 1 through 3 respectively.
Missing Data
To address the issue of missing data, a multiple imputation procedure (MI) was executed using SPSS version 21 through the fully conditional specification MCMC maximum likelihood algorithm. Overall, 10 fully imputed datasets were computed at the item level by time of administration [33], and aggregated to create one parsimonious dataset [34]. Respondents were retained if they completed at least two time points of administration on each of the three constructs, and missing data were treated as Missing Completely at Random (MCAR). A missing values analysis was conducted prior to imputation and it was determined that the mean level of missingness by item was 3.85% in Time 1, 1.61% in Time 2, and 3.37% in Time 3. While Luengo, García, and Herrera [35] reported that missingness below 5% is manageable; the MI procedure was used to create a parsimonious dataset that was the best possible representation of the data.
Analytic Procedures
Longitudinal structural equation modeling (SEM) was used to explore the longitudinal effects of cybervictimization on depression and anxiety. The approach was appropriate for the data structure because SEM allows the exploration of similarities and differences of latent constructs over time, while simultaneously controlling for measurement error [34]. To ensure appropriate data fit, a measurement model was established with freely estimated parameters. Since the scales have been subject to several factor analytic procedures, and it was the a priori decision to test the structural models, the freely estimated measurement model served as the foundation for all subsequent structural models. Once the measurement model was fit, separate structural models were constructed for depression and anxiety by cybervictimization over time.
To establish the measurement and structural models, and item-to-construct balancing procedure was conducted to create parcels for cybervictimization, depression, and anxiety [34]. Parcels represent aggregate-level indicators to establish a just-identified model (i.e., a model with one unique solution), and this method was used because the focus of the study was on the construct, not item-level indicators. To establish the parcels, separate exploratory factor analytic procedures were conducted by construct, and the three highest loadings were used to anchor the parcels, and the next highest loadings were added to the anchors in inverse order [34].
After the data were imputed and the parcels were created, both the freely estimated and longitudinal structural models were constructed using SPSS AMOS version 22. Once all of the nonsignificant paths were trimmed from the original structural model to establish the final structural model, the χ2 difference test was used to determine if model fit was impacted by the removal of the paths. Model fit was evaluated through χ2, Root Mean Square Error of Approximation (RMSEA), Tucker Lewis Index (TLI), and Comparative Fit Index (CFI). RMSEA scores below .08 and TLI and CFI scores greater than .90 are considered acceptable.
Results
Associations between the constructs revealed that each time point of cybervictimization was positively associated with each time point of depression, ranging from r =.20** to r =.38**, and anxiety, ranging from r =.12** to r =.27** (see Table 1). In addition to the associations, basic descriptives were assessed. Percentages of respondents who endorsed each scale (i.e., affirmed at least one item on the scale) were calculated, where 53.5% endorsed at least one item on the cybervictimization scale at time 1, compared to 59.4% at time 3; 98.6% endorsed at least one item on the depression scale at time 1, compared to 98.0% at time 3; and 67.3% endorsed at least one item on the anxiety scale at time 1, compared to 62.6% at time 3. High endorsement for each scale was calculated by establishing a pooled mean score (i.e., total mean score for all time points) and standard deviation, and defining high endorsement by one standard deviation above the pooled mean score by time point. For cybervictimization, 7.2% reported high cybervictimization at time 1, 18.4% at time 2, and 24.3% at time 3. Depression and anxiety followed a similar pattern with 17.0% at time 1, 18.6% at time 2, and 22.2% at time 3 for depression, and 18.6% at time 1, 19.0% at time 2, and 22.7% at time 3 for anxiety (see Table 1).
Table 1.
Correlations, Means, and Standard Deviations Among the Latent Constructs
| Construct | CV1 | CV2 | CV3 | DEP1 | DEP2 | DEP3 | ANX1 | ANX2 | ANX3 |
|---|---|---|---|---|---|---|---|---|---|
| CV1 | 1.00 | ||||||||
| CV2 | .39** | 1.00 | |||||||
| CV3 | .29** | .55** | 1.00 | ||||||
| Dep1 | .35** | .26** | .23** | 1.00 | |||||
| Dep2 | .20** | .35** | .27** | .50** | 1.00 | ||||
| Dep3 | .26** | .27** | .38** | .48** | .52** | 1.00 | |||
| Anx1 | .27** | .20** | .18** | .45** | .31** | .29** | 1.00 | ||
| Anx2 | .15** | .31** | .19** | .26** | .46** | .29** | .30** | 1.00 | |
| Anx3 | .12** | .16** | .27** | .26** | .32** | .54** | .35** | .30** | 1.00 |
| Means | .24 | .47 | .60 | .85 | .88 | .89 | .72 | .74 | .82 |
| SD | .53 | .76 | .85 | .48 | .49 | .51 | .83 | .90 | .99 |
| % End (n) | 53.5 (299) | 53.3 (298) | 59.4 (332) | 98.6 (551) | 99.1 (554) | 98.0 (548) | 67.3 (376) | 64.6 (361) | 62.6 (350) |
| High End (n) | 7.2 (40) | 18.4 (103) | 24.3 (136) | 17.0 (95) | 18.6 (104) | 22.2 (124) | 18.6 (104) | 19.0 (106) | 22.7 (127) |
Note.
Significant at the .05 level
Significant at the .01 level.
CV = Cybervictimization, Dep = Depression, Anx = Anxiety, Number represents the Time Point of Administration. % End represents individuals who affirmed at least one item on the scale, and High End represents the percentage or respondents who were at least one standard deviation above the pooled mean and standard deviation. Pooled means and standard deviations for cybervictimization = .45, .56; depression = .87, .36; and anxiety = .76, .66 respectively.
Measurement and Structural Models
Depression
To establish preliminary model fit, a freely estimated measurement model (i.e., no specified parameter restrictions) was established for cybervictimization and depression. Loadings by parcels ranged from λ =.81 to λ =.92 for cybervictimization and λ =.31 to λ =.92 for depression (see Table 2). In addition to the acceptable parcel loadings, the measurement model represented a close fit of the data, χ2(108) =265.00 = p <.001, RMSEA =.05, TLI =.96, CFI =.97 (see Table 3).
Table 2.
Loadings, Intercepts, Estimated Latent Variance, Unique Residuals, and Squared Multiple Correlations for Estimated Measurement Model for Cybervictimization, Depression, and Anxiety.
| Scale | λ | τ | λa | θ | R2 |
|---|---|---|---|---|---|
| CV Time 1 | |||||
| Parcel 1 | 1.00 | .29 (.03) | .88 | .08 (.01) | .78 |
| Parcel 2 | .82 (.04) | .20 (.02) | .81 | .10 (.01) | .66 |
| Parcel 3 | .90 (.04) | .22 (.02) | .82 | .11 (.01) | .68 |
| CV Time 2 | |||||
| Parcel 1 | 1.00 | .44 (.03) | .90 | .11 (.01) | .81 |
| Parcel 2 | .90 (.03) | .43 (.03) | .87 | .13 (.01) | .76 |
| Parcel 3 | 1.04 (.04) | .49 (.04) | .89 | .15 (.01) | .79 |
| CV Time 3 | |||||
| Parcel 1 | 1.00 | .55 (.04) | .92 | .11 (.01) | .85 |
| Parcel 2 | .93 (.03) | .56 (.03) | .91 | .11 (.01) | .83 |
| Parcel 3 | 1.04 (.03) | .63 (.04) | .91 | .14 (.01) | .83 |
| Dep Time 1 | |||||
| Parcel 1 | 1.00 | .92 (.03) | .38 | .29 (.02) | .15 |
| Parcel 2 | 2.39 (.29) | .89 (.03) | .89 | .08 (.02) | .79 |
| Parcel 3 | 2.07 (.23) | .71 (.03) | .77 | .14 (.02) | .60 |
| Dep Time 2 | |||||
| Parcel 1 | 1.00 | .95 (.03) | .31 | .32 (.02) | .09 |
| Parcel 2 | 2.84 (.40) | .92 (.03) | .87 | .08 (.02) | .76 |
| Parcel 3 | 2.63 (.38) | .73 (.03) | .76 | .16 (.02) | .58 |
| Dep Time 3 | |||||
| Parcel 1 | 1.00 | .97 (.03) | .38 | .33 (.02) | .15 |
| Parcel 2 | 2.28 (.26) | .95 (.03) | .92 | .05 (.02) | .85 |
| Parcel 3 | 2.20 (.24) | .76 (.03) | .81 | .14 (.02) | .66 |
| Anx Time 1 | |||||
| Parcel 1 | 1.00 | .87 (.04) | .86 | .32 (.02) | .74 |
| Parcel 2 | 1.06 (.06) | .73 (.05) | .80 | .41 (.04) | .64 |
| Parcel 3 | .67 (.04) | .42 (.03) | .70 | .32 (.02) | .48 |
| Anx Time 2 | |||||
| Parcel 1 | 1.00 | .89 (.04) | .85 | .31 (.02) | .72 |
| Parcel 2 | 1.08 (.06) | .77 (.05) | .83 | .38 (.04) | .68 |
| Parcel 3 | .73 (.04) | .42 (.04) | .74 | .28 (.03) | .55 |
| Anx Time 3 | |||||
| Parcel 1 | 1.00 | .94 (.05) | .94 | .13 (.03) | .88 |
| Parcel 2 | 1.04 (.04) | .92 (.05) | .85 | .41 (.04) | .72 |
| Parcel 3 | .71 (.03) | .51 (.04) | .77 | .34 (.02) | .60 |
Note. CV = Cybervictimization; Dep = Depression; Anx = Anxiety; λ = loading estimates (SE); τ = intercept estimates λa = standardized loadings-common metric completely standardized; θ = residual (SE), R2 = squared multiple correlation.
Table 3.
Fit indices for Final Measurement Model, Comparative Independence Model, Fully Estimated Structural Model, and Final Trimmed Structural Model for Cybervictimization and Depression and Cybervictimization and Anxiety.
| Model | χ2 | df | p | RMSEA | 90% RMSEA CI | CFI | TLI |
|---|---|---|---|---|---|---|---|
| Measurement Model for Cybervictimization and Depression | |||||||
| Final CFA | 265.00 | 108 | .000 | .05 | .04 – .06 | .97 | .96 |
| Independence Model | 6174.75 | 153 | .000 | .27 | .26 – .27 | .00 | .00 |
| Fully Estimated Model | 510.66 | 125 | .000 | .07 | .07 – .08 | .94 | .92 |
| Final Model | 514.08 | 128 | .000 | .07 | .07 – .08 | .94 | .92 |
| Measurement Model for Cybervictimization and Anxiety | |||||||
| Strong CFA | 197.61 | 108 | .000 | .04 | .03 – .05 | .99 | .98 |
| Independence Model | 6763.78 | 153 | .000 | .28 | .27 – .28 | .00 | .00 |
| Fully Estimated Model | 361.06 | 125 | .000 | .06 | .05 – .07 | .96 | .96 |
| Final Model | 362.96 | 128 | .000 | .06 | .05 – .06 | .96 | .96 |
Note. RMSEA = root mean square error of approximation, 90% RMSEA CI = 90% confidence intervals for root mean square error of approximation, CFI = comparative fit index, TLI = Tucker-Lewis Index.
The full structural model, including all autoregressive (i.e., path from one construct to the same construct at a subsequent time point) and cross-lagged (i.e., path from one construct to a different construct at a subsequent time point) paths represented an acceptable fit of the data (χ2(125) =510.66, RMSEA =.07, TLI =.92, CFI =.94). Following the initial fit of the structural model, all nonsignificant paths were sequentially removed resulting in an acceptable fitting model (χ2(128) =514.08, RMSEA =.07, TLI =.92, CFI =.94). To establish invariance between the models, the χ2 difference test was employed and it was determined that there were nonsignificant differences between the models (Δ χ2(3) =3.42, p >.05), suggesting that the removal of the nonsignificant paths was tenable (see Table 3).
The final structural model included both autoregressive and cross-lagged paths (see Table 4). The autoregressive paths for cybervictimization were retained for time 1 to time 2 (β =.45) and time 2 to time 3 (β =.53), and for depression for time 1 to time 2 (β =.63) and time 2 to time 3 (β =.63). Significant cross-lagged paths included depression at time 1 to cybervictimization at time 2 (β =.21), depression at time 2 to cybervictimization at time 3 (β =.14), and cybervictimization at time 1 to depression at time 3 (β =.15). Figure 2 represents the final structural model for cybervictimization and depression.
Table 4.
Beta Weights and z Scores for the Final Structural Model for Cybervictimization and Depression and Cybervictimization and Anxiety.
| Construct | B (SE) | z Score | β |
|---|---|---|---|
| Structural Model for Cybervictimization and Depression | |||
| Auto-Regressive Paths: Cybervictimization | |||
| CV1 – CV2 | .60 (.06) | 10.06** | .45 |
| CV2 – CV3 | .60 (.05) | 12.89** | .53 |
| Auto-Regressive Paths: Depression | |||
| Dep1 – Dep2 | .65 (.06) | 11.66** | .63 |
| Dep2 – Dep3 | .67 (.06) | 11.97** | .63 |
| Cross-Lagged Paths: Cybervictimization and Depression | |||
| CV1 – Dep3 | .15 (.04) | 3.56** | .15 |
| Dep1 – CV2 | .30 (.06) | 4.75** | .21 |
| Dep2 – CV3 | .21 (.06) | 3.34** | .14 |
| Structural Model for Cybervictimization and Anxiety | |||
| Auto-Regressive Paths: Cybervictimization | |||
| CV1 – CV2 | .63 (.06) | 10.50** | .47 |
| CV2 – CV3 | .66 (.05) | 14.09** | .58 |
| Auto-Regressive Paths: Anxiety | |||
| Anx1 – Anx2 | .38 (.06) | 6.74** | .35 |
| Anx2 – Anx3 | .37 (.06) | 6.57** | .32 |
| Cross-Lagged Paths: Cybervictimization and Anxiety | |||
| Anx1 – CV2 | .20 (.06) | 3.51** | .15 |
| CV1 – Anx2 | .11 (.05) | 2.12* | .10 |
| CV2 – Anx3 | .10 (.04) | 2.19* | .10 |
Note.
Significant at the .05 level
Significant at the .01 level.
CV = Cybervictimization; Dep = Depression; Anx = Anxiety; Number represents the Time Point of Administration.
Figure 2.
Structural models for Cybervictimization and Depression (top) and Cybervictimization and Anxiety (bottom).
Anxiety
To establish preliminary model fit, a freely estimated measurement model was established for cybervictimization and anxiety. Loadings by parcels ranged from λ =.81 to λ =.92 for cybervictimization and λ =.70 to λ =.94 for anxiety (see Table 2). In addition to the acceptable parcel loadings, the measurement model represented a close fit of the data, χ2(108) =197.61 = p <.001, RMSEA =.04, TLI =.98, CFI =.99 (see Table 3).
The full structural model, including all autoregressive and cross-lagged paths represented an acceptable fit of the data (χ2(125) =361.06, RMSEA =.06, TLI =.96, CFI =.96). Following the initial fit of the structural model, all nonsignificant paths were sequentially removed resulting in an acceptable fitting model (χ2(128) =362.96, RMSEA =.06, TLI =.96, CFI =.96). To establish invariance between the models, the χ2 difference test was employed and it was determined that there were nonsignificant differences between the models (Δ χ2(3) =1.90, p >.05), suggesting that the removal of the nonsignificant paths was tenable (see Table 3).
The final structural model included both autoregressive and cross-lagged paths (see Table 4). The autoregressive paths for cybervictimization were retained for time 1 to time 2 (β =.47) and time 2 to time 3 (β =.58), and for anxiety for time 1 to time 2 (β =.35) and time 2 to time 3 (β =.32). Significant cross-lagged paths included cybervictimization at time1 to anxiety at time 2 (β =.10), cybervictimization at time 2 to anxiety at time 3 (β =.10), and anxiety at time 1 to cybervictimization at time 2 (β =.15). Figure 2 represents the final structural model for cybervictimization and anxiety.
Discussion
The present study examined the predictive relationship between cybervictimization and depression and anxiety over time. This study was designed to directly evaluate depression and anxiety as independent constructs and their unique relationship to cyberbullying. As anticipated, associations between cybervictimization and depression and anxiety were significant for each time point of administration [18,20]. Additionally, the measurement model demonstrated that cybervictimization, depression, and anxiety were stable over time.
Autoregressive Paths
Positive autoregressive paths were found for cybervictimization, where the level of cybervictimization at one time point was predictive of subsequent time points, after controlling for the levels at previous time points, suggesting that cybervictimization increased over time. While few studies have examined cybervictimization over time, in a cross-sectional analysis, Rose, Simpson, and Moss [36] used the Bullying and Harassment subscale of the Online Victimization Scale [30] to evaluate prevalence rates of cybervictimization among a diverse sample (n = 14,508) of adolescents in grades six through 12, and determined that approximately 47% of adolescents endorse at least one item on the subscale, with 12% reporting high levels of cybervictimization. Although these findings are similar to the current study, Tokunga [37] argued that extant cyberbullying research demonstrates a curvilinear relationship, where prevalence levels peak during seventh and eighth grade. While this postulation supports traditional victimization literature [38], it also demonstrates the necessity of evaluating cybervictimization over time. However, it should be noted that the current study was situated by time of administration, not grade level, which suggests that the number of students who experienced cybervictimization increased over each subsequent time point (one year apart).
Similar positive autoregressive paths were found for depression and anxiety, with levels of depression and anxiety at one time point predicting increased levels of depression and anxiety at subsequent time points. Since the scales used to assess depression and anxiety for the current study are psychometric proxies for a construct of interest, and not a diagnostic tool for depression or anxiety diagnosis, we cannot postulate on psychiatric diagnoses. However, based on Cummings and colleagues [23] multiple pathways model, it is conceivable that if depression and/or anxiety go undiagnosed and without intervention, adolescence are unlikely to independently resolve these complications.
Depression, Anxiety, and Cybervictimization
Cross-sectional studies suggest that increased levels of traditional victimization [9–11] and cybervictimization [16,17] are associated with higher levels of depression. Viewing this relationship through a cross-sectional lens suggests this relationship is directly linear, where victimization is predicting depression [37]. Tokunaga and colleagues [37] argued that cyberbullying should be viewed through a non-recursive (i.e., bi-directional) lens, where the ongoing process of cyberbullying is cyclical.
The current study adds credibility to the non-recursive model, where findings suggest that cybervictimization and depression represent a reciprocal relationship, and students who reported high depression at times 1 and 2 reported higher levels of cybervictimization at times 2 and 3 respectively, and respondents who reported higher levels of cybervictimization at time 1 reported higher levels of depression at time 3. These findings parallel traditional bullying literature that suggest students who report depressive symptoms are more likely to experience subsequent victimization [10,39] because they may appear more vulnerable [21,40]. However, this study suggests that vulnerability may extend to online contexts.
In contrast to the findings with depression, we found students reporting higher levels of cybervictimization at times 1 and 2 reported higher levels of anxiety at times 2 and 3, and participants who reported more anxiety at time 1 reported higher levels of cybervictimization at time 2. It may be the case that students developed a toolkit for managing symptoms of anxiety over time and thus no relationship was found between time 1 and 3, as was the case with depression. For example, Tynes, Umaña-Taylor, Rose, Lin, and Anderson [41] found that ethnic identity moderated the association between race-based cybervictimization and anxiety, but not for depression.
Limitations and Future Research
While this study addressed notable gaps in the literature, it is not without limitation. First, data were examined by wave of administration, so implications for grade and/or age could not be established. Second, demographic variables were not included in the models, limiting the generalizability of the findings. Third, this study did not account for internet access, time online, or online activities, which may further the understanding of the relationship between cyberbullying, depression, and anxiety. Given these limitations, future research should examine these relationships, while accounting for potential mitigating factors, such as age, demographics, and frequency of internet use.
Implications and Contributions
This study demonstrated the reciprocal relationship between cybervictimization and mental health outcomes over time. Given the increasing prevalence of cybervictimization over time and the reciprocity between cybervictimization and depression and anxiety, educational stakeholders must attend to the mental health and bully prevention policies. More specifically, bullying prevention legislation should include provisions and guidance for schools to address and reduce cybervictimization among adolescents. Additionally, Educational stakeholders should implement multi-tiered systems of support that include universal and targeted interventions that specifically address online experiences and safety, while supporting the mental health of school-aged youth through social and emotional learning and skill development.
Implications and Contribution.
A reciprocal relationship exists between cybervictimization and mental health outcomes over time. Given the increasing prevalence of cybervictimization and its relationship to depression and anxiety, interventions should target both mental health and bullying prevention.
Acknowledgments
This work was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number R01HD061584. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Contributor Information
Chad A. Rose, University of Missouri-Columbia
Brendesha M. Tynes, University of Southern California
References
- [1].Prensky M Is the human brain still the smartest thing on the planet? When enhanced by technology, it is. Educ Leadersh; 2013;70:22–27. [Google Scholar]
- [2].Lenhart A, Madden M, Smith A, et al. Teens, kindness and cruelty on social network sites. Pew Internet & American Life Project; Retrieved from http://www.pewinternet.org/files/old-media//Files/Reports/2011/PIP_Teens_Kindness_Cruelty_SNS_Report_Nov_2011_FINAL_110711.pdf. Published November 9, 2011. [Google Scholar]
- [3].Rideout VJ, Foehr UG, Roberts DF. Generation M2: Media in the lives of 8- to 18-year-olds. Menlo Park, CA: Henry J. Kaiser Family Foundation; Retrieved from http://kaiserfamilyfoundation.files.wordpress.com/2013/04/8010.pdf. Published January 20, 2010. [Google Scholar]
- [4].Tynes BM. Internet safety gone wild?: Sacrificing the educational and psychosocial benefits of online social environments. J Adolesc Res 2007;22:575–584. DOI: 10.1177/0743558407303979. [DOI] [Google Scholar]
- [5].Hinduja S, Patchin JW. Bullying beyond the schoolyard: Preventing and responding to cyberbullying. Thousand Oaks, CA: Corwin Press; 2009. [Google Scholar]
- [6].Kowalski RM, Giumetti GW, Schroeder AN, Lattanner MR. Bullying in the digital age: A critical review and meta-analysis of cyberbullying research among youth. Psychol Bull 2014;140:1073–1137. DOI: 10.1037//a0035618. [DOI] [PubMed] [Google Scholar]
- [7].Willard NE. Cyberbullying and cyberthreats: Responding to the challenge of online social aggression, threats, and distress. Champaign, IL: Research Press; 2007. [Google Scholar]
- [8].Kaltiala-Heino R, Rimpelä M, Marttunen M, et al. Bullying, depression, and suicidal ideation in Finnish adolescents: School survey. BMJ 1999;319:348–351. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Cook CR, Williams KR, Guerra NG, et al. Predictors of bullying and victimization in childhood and adolescence: A meta-analytic investigation. Sch Psychol Q 2010;25:65–83. DOI: 10.1037/a00200149. [DOI] [Google Scholar]
- [10].Klomek AB, Marrocco F, Kleinman M, et al. Bullying, depression, and suicidality in adolescents. J Am Acad Child Adolesc Psychiatry 2007;46:40–49. [DOI] [PubMed] [Google Scholar]
- [11].Seals D, Young J. Bullying and victimization: Prevalence and relationship to gender, grade levels, ethnicity, self-esteem, and depression. Adolescence 2003;38:735–747. [PubMed] [Google Scholar]
- [12].Ttofi MM, Farrington DP, Lösel F, Loeber R. Do the victims of school bullies tend to become depressed later in life? A systematic review and meta-analysis of longitudinal studies. J Aggress Confl Peace Res 2011;3(2):63–73. [Google Scholar]
- [13].Klomek AB, Kleinman M, Altschuler E, et al. High school bullying as a risk for later depression and suicidality. Suicide Life Threat Behav 2011;41:501–516. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Ayas T, Horzum MB. Relation between depression, loneliness, self-esteem and internet addiction. Education 2013;133:283–290. [Google Scholar]
- [15].Kelleci M, İnal S. Psychiatric symptoms in adolescents with internet use: Comparison without internet use. Cyberpsychol Beh Soc Netw 2010;13:191–194. DOI: 10.1089/cyber.2009.0026 [DOI] [PubMed] [Google Scholar]
- [16].Kowalski RM, Limber SP. Psychological, physical, and academic correlates of cyberbullying and traditional bullying. J Adolesc Health 2013;53:S13–S20. [DOI] [PubMed] [Google Scholar]
- [17].Ybarra ML, Mitchell KJ, Wolak J, Finkelhor D. Examining characteristics and associated distress related to internet harassment: Findings from the Second Youth Internet Survey. Pediatrics 2006;118:e1169–e1177. [DOI] [PubMed] [Google Scholar]
- [18].Bonanno RA, Hymel S. Cyber bullying and internalizing difficulties: Above and beyond the impact of traditional forms of bullying. J Youth Adolesc 2013;42:685–697. DOI: 10.1007/s10964-013-9937-1. [DOI] [PubMed] [Google Scholar]
- [19].Schneider SK, O’Donnell L, Stueve A, Coulter RWS. Cyberbullying, school bullying, and psychological distress: A regional census of high school students. Am J Public Health 2012;102:171–177. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Gini G, Espelage DL. Peer victimization, cyberbullying, and suicide risk in children and adolescents. JAMA Pediatr 2014;185:435–442. DOI: 10.1001/jamapediatrics.2013.4143. [DOI] [PubMed] [Google Scholar]
- [21].Hong JS, Espelage DL. A review of research on bullying and peer victimization in school: An ecological system analysis. Aggress Violent Behav 2012;17:311–322. DOI: 10.1016/j.avb.2012.03.003. [DOI] [Google Scholar]
- [22].Garber J, Weersing VR. Comorbidity of anxiety and depression in youth: Implications for treatment and prevention. Clin Psychol (New York) 2010;17:293–306. DOI: 10.1111/j.1468-2850.2010.01221.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Cummings CM, Caporino NE, Kendall PC. Comorbidity of anxiety and depression in children and adolescents: 20 years after. Psychol Bull 2014;140:816–845. DOI: 10.1037/a0034733. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Espelage DL, Holt MK. Dating violence & sexual harassment across the bully-victim continuum among middle and high school students. J Youth Adolesc 2007;36:799–811. DOI: 10.07/s10964-006-9109-7. [DOI] [Google Scholar]
- [25].Patchin JW, Hinduja S. Bullies move beyond the schoolyard: A preliminary look at cyberbullying. Youth Violence Juv Justice 2006;4(2):148–169. [Google Scholar]
- [26].Dempsey AG, Sulkowski ML, Nichols R, Storch EA. Differences between peer victimization in cyber and physical settings and associated psychosocial adjustment in early adolescence. Psychol Sch 2009;46:962–972. DOI: 10.1002/pits.20437. [DOI] [Google Scholar]
- [27].Gamez-Guadix M, Orue IO, Smith PK, Calvete E. Longitudinal and reciprocal relations of cyberbullying with depression, substance use, and problematic internet use among adolescents. J Adolesc Health 2013;53:446–452. [DOI] [PubMed] [Google Scholar]
- [28].Lester L, Cross D, Dooley J, Shaw T. Internalising symptoms: An antecedent or precedent in adolescent peer victimization. Australian Journal of Guidance and Counseling 2012;22(2):173–189. [Google Scholar]
- [29].Patchin JW, Hinduja S. Cyberbullying among adolescents: Implications for empirical research. J Adolesc Health 2013;53:431–432. [DOI] [PubMed] [Google Scholar]
- [30].Tynes BM, Rose CA, Williams DR. The development and validation of the online victimization scale for adolescents. Cyberpsychology: Journal of Psychosocial Research on Cyberspace 2010;4(2), article 2. [Google Scholar]
- [31].Radloff LS. The CES-D Scale: A self-report depression scale for research in the general population. App Psychol Meas 1977;1(3):385–401. [Google Scholar]
- [32].Terry PC, Lane AM, Lane HJ, Keohane L. Development and validation of a mood measure for adolescents. J Sports Sci 1999;17(11):861–872. [DOI] [PubMed] [Google Scholar]
- [33].Enders CK. Applied missing data analysis. New York, NY: Guilford; 2010. [Google Scholar]
- [34].Little TD. Longitudinal structural equation modeling. New York, NY: The Gilford Press; 2013. [Google Scholar]
- [35].Luengo J, García S, Herrera F. A study on the use of imputation methods for experimentation with radia basis function network classifiers handling missing attribute values: The good synergy between RBFNs and event covering method. Neural Netw 2010;23:406–418. [DOI] [PubMed] [Google Scholar]
- [36].Rose CA, Simpson CG, Moss A The bullying dynamic: Prevalence of involvement among a large-scale sample of middle and high school youth with and without disabilities. PSYCHOL SCHOOLS 2015; 52, 515–531. DOI: 10.1002/pits.21840 [DOI] [Google Scholar]
- [37].Tokunga RS. Following you home from school: A critical review and synthesis of research on cyberbullying victimization. Comput Human Behav 2010;26:277–287. [Google Scholar]
- [38].Nansel TR, Overpeck M, Pilla RS, et al. Bullying behaviors among US youth: Prevalence and association with psychosocial adjustment. JAMA 2001;285:2094–2100. DOI: 10.1001/jama.285.16.2094. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [39].Turner I, Reynolds KJ, Lee E, et al. Well-being, school climate, and the social identity process: A latent growth model study of bullying perpetration and peer victimization. Sch Psychol Q 2014;29(3):320–335. DOI: 10.1037/spq0000074. [DOI] [PubMed] [Google Scholar]
- [40].Fekkes M, Pijpers FIM, Fredriks AM, et al. Do bullied children get ill, or do ill children get bullied? A prospective cohort study on the relationship between bullying and health-related symptoms. Pediatrics 2006;117:1568–1574. [DOI] [PubMed] [Google Scholar]
- [41].Tynes BM, Umaña-Taylor AJ, Rose CA, et al. Online racial discrimination and the protective function of ethnic identity and self-esteem for African American adolescents. Dev Psychol 2012;48:342–355. DOI: 10.1037/a0027032 [DOI] [PubMed] [Google Scholar]