This 9-month prospective cohort study was conducted in Guangzhou, South China. The methods have previously been described [6]. Briefly, 9 public secondary schools, 3 from each of the city’s 3 regions (ie, core, suburb, and outer suburb regions), were selected by convenience sampling, and all students of the seventh and eighth grade (ie, 7 and 8 years of formal education) were invited to participate in the study. Grade 9 students were not included into the study given that they were too busy preparing for the public examination for entrance to senior high schools and would leave their schools before the end of the 9-month follow-up period; it would thus be practically difficult to follow up such students. The anonymous structured questionnaire was self-administered in classroom settings without the presence of teachers. We announced that the return of the completed questionnaire implied consent to participate in the study; this information was also printed on the cover page of questionnaire. No incentive was given to the participants. Permission to conduct the survey was granted by the school principals. The Survey and Behavioral Research Ethics Committee of the Chinese University of Hong Kong approved the study.

A total of 5365 students (response rate 98.0%) completed the baseline questionnaire. The baseline and follow-up questionnaires completed by each student were matched based on the last 4 digits of the student’s home telephone number, the last 4 digits of the parents’ mobile phone number, the last 4 digits of the student’s identity card number, the student’s birthday, and the last letter of the student’s and parents’ name. A total of 4871 (follow-up rate 90.8%) of the 5365 students provided matched completed questionnaires at the follow-up survey. Students who were lost to follow-up were more likely to be females, be in their junior year of school, to not live with both parents, to have a perceived poor/very poor family financial situation, and to have lower self-reported academic performance (Multimedia Appendix 1). Students who did not use online social networking (n=643) were excluded from the data analysis. The effective sample size was 4237 in this study.

The associations between changes (indicated as △) in online social networking use intensity (SFUI and EFUI) and depressive symptoms were tested. Potential mediators of the associations included changes in father–adolescent conflict, mother–adolescent conflict, and social nonconfidence; potential suppressors included changes in friendship quality, perceived family support, and perceived friends support. Changes were determined by subtracting the scale scores at baseline from those at the 9-month follow-up.

Potential mediation or suppression effects were tested using hierarchical modeling based on the methods proposed by Baron and Kenny [56], which required the presence of significant associations between (1) △SFUI (or △EFUI) and △CES-D; (2) △SFUI (or △EFUI) and changes in potential mediators or suppressors; and (3) changes in potential mediators or suppressors and △CES-D after controlling for △SFUI (or △EFUI). The strength of the indirect effects (mediation or suppression) was estimated by the product a × b, where a is the regression coefficient that relates the independent variable to the potential mediator (or suppressors) and b is the regression coefficient that relates the mediators (or suppressors) to the dependent variable, with significance determined using Sobel Z [57,58]; a × b / (a × b + c’) reflects the magnitude of mediation effect size [57,59], where c’ is the regression coefficients relating the mediators (or suppressors) to the dependent variable and the independent variable to the dependent variable in the same model. As the total effect of the suppressors could not be decomposed into indirect and direct proportions because it comprised associations in opposite directions, the proportion of the indirect effect could not be calculated; thus, only the suppressing effects (a × b) were reported. The interaction effects of gender for the associations between △SFUI (or △EFUI), changes in potential mediators (or suppressors) and △CES-D were further tested by including the interaction terms (gender multiplied by △SFUI, △EFUI, changes in potential mediators, or changes in potential suppressors).

Multilevel linear regression models (level 1: students, level 2: schools) with random intercepts were fitted to take into account the potential clustering effects from the schools. All models were adjusted for baseline background factors that were significantly associated with △CES-D in the univariate models (P<.05), and the unstandardized regression coefficients with standard errors are presented. Analyses were conducted using SAS software (version 9.4, SAS Institute). A 2-sided P<.05 was considered statistically significant.

In the baseline sample, 49.7% (2105/4237) of the participants were male, and 47.5% (2011/4237) were in the seventh grade. Approximately 48.3% (2047/4237) of students self-reported a good or very good family financial situation. The majority of students (88.4%, 3747/4237) lived with both parents. Approximately one-fifth (20.1%, 852/4237) reported lower academic performance, and 23.4% (993/4237) perceived heavy or very heavy academic pressure (Table 1).

During a 9-month follow-up period, the mean △CES-D was 0.48 (95% CI 0.22, 0.74; P<.001). There were slight significant reductions in the SFUI (P=.01) and EFUI scores (P<.001) and the scores of 3 potential suppressors (friendship quality: P=.001; perceived family support: P=.009; and perceived friend support: P<.001) and 2 potential mediators (father–adolescent conflict: P=.02; social nonconfidence: P<.001), whereas △mother–adolescent conflict from baseline to follow-up was not statistically significant (P=.21) (Table 2).

After adjustment for grade, academic performance, and perceived academic pressure, we found that (1) both △SFUI and △EFUI were positively associated with △CES-D (Table 3); (2) all 3 favorable interpersonal variables (potential suppressors) were negatively associated with △CES-D, while all 3 unfavorable interpersonal variables (potential mediators) were positively associated with △CES-D (Table 4); (3) △SFUI was positively associated with all interpersonal variables, except △father–adolescent conflict (P=.12); and (4) △EFUI was positively associated with 2 favorable interpersonal variables (△perceived family support and △perceived friend support) and one unfavorable interpersonal variables (△social nonconfidence), but not with △friendship quality, △father–adolescent conflict, and △mother–adolescent conflict (Table 3).

Interaction effects of gender on (1) the association between △SFUI (or △EFUI) and △CES-D (P=.60; P=.41), (2) the association between △SFUI (or △EFUI) and change in psychosocial factors (P=.09-.94; P=.18-.93), and (3) the associations between psychosocial factors and △CES-D (P range: .30–.68) were nonsignificant, except for the interaction between gender and △SFUI on △father–adolescent conflict (interaction P=.006), between gender and △friendship quality on △CES-D (P=.03), and between gender and △perceived family support on △CES-D (interaction P=.04) (Multimedia Appendix 2 and Multimedia Appendix 3).

When a potential mediator or suppressor was entered into the model that contained only △SFUI as the independent variable and had △CES-D as the dependent variable, a significant mediator resulted in a significant decrease in the regression coefficient, while a significant suppressor resulted in a significant increase in the regression coefficient between △SFUI and △CES-D. Accordingly, we found 2 significant mediators of the association between △SFUI and △CES-D: △mother–adolescent conflict (mediation effect size 5.11%, P=.02; model m1, Table 5) and △social nonconfidence (mediation effect size 20.97%, P<.001; model m2, Table 5). The combined mediation effect was approximately 25.33% (model m3).

In addition, we found that all 3 favorable interpersonal relationship variables were significant suppressors: △friendship quality (suppression effect: –0.011, P=.003; model s1, Table 6), △perceived family support (suppression effect: –0.009, P=.003; model s2), and △perceived friend support (suppression effect: –0.022, P<.001; model s3).

△Social nonconfidence (mediation effect size: 30.65%, P<.001; model m4) was the only significant mediator for the association between △EFUI and △CES-D. Significant suppressors included △perceived family support (suppression effect –0.036, P<.001; model s4) and △perceived friend support (suppression effect –0.039, P<.001; model s5).

In this study, we found a significant increase in depressive symptoms (P<.001) among young adolescents during the 9-month period and positive associations between △SFUI (or △EFUI) and △CES-D. Moreover, associations were simultaneously partially mediated by some unfavorable psychosocial factors (ie, social nonconfidence and mother–adolescent conflict) and partially suppressed by some favorable psychosocial factors (ie, perceived social support and quality of friendship).

Consistent with our findings, previous studies conducted in the United States [60] and Iceland [61] had also reported a gradual increase in depressive symptoms among adolescents. The findings suggest that adolescents are vulnerable to gradual deterioration of their mental health and even chronic depression. Attention is required, given that depression may have lasting negative consequences for adolescents.

We found that, with increases in depressive symptoms, adolescents may have less favorable relationships (ie, reduced friendship quality and perceived social support) over time. Given that adolescents need to establish an autonomous self-identity, they may spend less time with their families, resulting in reduced perceived family support [62]. Our findings indicate that adolescents seem to face an additional challenge of having their relationships with their peers deteriorate over time, possibly due to the low quality of friendships formed online, the difference between online and general offline social connectedness [63], and important developmental changes during adolescence (eg, decrease in parental attachment, learning to form attachment relationships with peers, exploration of new environments) [64]. Fortunately, adolescents may gain social confidence over time, possibly because they have more practice in social interactions with other people. We observed that changes in interpersonal relationships were associated with changes in depressive symptoms. As social relationships are important determinants of mental, physical, and behavioral health among adolescents, longitudinal studies are warranted to understand other consequences of changes in such interpersonal relationships. In addition, we found a significant and slight decline in online social networking use intensity (both SFUI with P=.01 and EFUI with P<.001) over the 9-month period. To the best of our knowledge, our study is the first to document a longitudinal trend of online social networking use intensity among adolescents. We show that such changes were associated with changes in interpersonal relationships and depressive symptoms.

This study identified a number of important partial mediators and suppressors of the associations between △SFUI (or △EFUI) and △CES-D. First, △social nonconfidence partially mediated the associations of both △SFUI and △EFUI with △CES-D. This is a novel finding. An increase in online social networking use may thus increase, rather than reduce, social nonconfidence, which in turn increases depressive symptoms. While online social networking sites provide convenient platforms for establishing communication, contacts, and relationships among adolescents [65], they might simultaneously deprive adolescents of skills, time allocation, and experiences related to effective direct offline social interactions, resulting in social nonconfidence. Online social networking use does not involve the use of social and nonverbal cues; it may also weaken individuals’ desire for involvement in offline social activities and pleasure seeking [66]. Furthermore, social nonconfidence is potentially associated with isolation and other risk factors for depression, which might mediate the association between △SFUI (or △EFUI) and depression. Future research should distinguish between social nonconfidence in face-to-face versus online social interactions and reveal the relationship of social nonconfidence with online and offline social relationships.

Another important finding was that △father–adolescent conflict mediated the association of neither △SFUI nor △EFUI with △CES-D, while △mother–adolescent conflict partially mediated the association between △SFUI and △CES-D (a relatively small effect of 5.11%) but not between △EFUI and △CES-D. Thus, overall, conflict with parents is not a major mechanism explaining the associations between △SFUI (or △EFUI) and depressive symptoms. Furthermore, mother–adolescent conflict, but not father–adolescent conflict, played a role as a mediator. The difference between the 2 types of parental conflict was evident in the correlation results, which showed that △SFUI was positively associated with △mother–adolescent conflict, but not with △father–adolescent conflict, while △EFUI was not associated with either △father–adolescent conflict or △mother–adolescent conflict. We speculate that mothers tend to be more engaged in parenting than fathers; they also interact more frequently with adolescents concerning personal issues (eg, intensive online social networking use and poor time management) [67,68]. Such parenting may increase adolescent–mother conflict [69]. Additionally, heavy online social networking use may negatively impact adolescents’ academic performance [70,71] and result in adolescent conflicts with their mothers rather than with their fathers, given that mothers may be more involved and concerned about their children’s personal development (eg, academic performance) than fathers [72,73]. Thus, an increase in SFUI might result in adolescents’ more conflict with their mothers than with fathers.

△Friendship quality showed a suppression effect between △SFUI and △CES-D, but not between △EFUI and △CES-D, which is understandable, as SFUI reflects the social functions of online social networking use, while EFUI reflects the entertainment functions. Positive friendship is a protective factor against depressive symptoms among adolescents [74]. Online social networking is doubtlessly one of the most preferred platforms for communication among adolescents, through which they can be socially connected with peers regardless of time and place [75]. For instance, online social networking use can supplement offline interactions and communications with friends to improve friendship quality among adolescents. Mutual and interactive updating of information (eg, messages and photos) online increases perceived closeness with friends [76] and hence supports the maintenance of friendships, especially between friends that are geographically separated.

Equally important, both △perceived family support and △perceived friend support suppressed the association between changes in online social networking use intensity (both △SFUI and △EFUI) and △CES-D. Increases in SFUI and EFUI may enhance perceived support from family and friends, which in turn decreases the level of depressive symptoms. In this study, △SFUI and △EFUI were positively associated with perceived social support. Conceptually, online social networking use allows adolescents to express their needs for support readily by means of online self-disclosure (eg, communicating personal information, thoughts, and feelings with others online) [77,78], which has been shown to be correlated with perceived social support [29,79]. Simultaneously, online social networking use provides channels for others to provide social support to adolescents and for adolescents to receive such social support. Interestingly, △perceived family support, similar to △perceived friend support, was a suppressor. It is plausible that online social networking use is not limited to interaction with friends but also includes interaction with family members. The relationship of online social networking use with family support and its mechanisms require future research attention.

Importantly, △SFUI (or △EFUI) possessed both mediators (social nonconfidence and mother–adolescent conflict) that were risk factors for △CES-D and suppressors (friendship quality and perceived social support) that were protective factors against △CES-D. Stakeholders should be made aware that online social networking use by adolescents is neither good nor bad; related interventions should maximize protective suppressors and minimize harmful mediators of online social networking use among adolescents to transform harms into benefits. Attention should be paid to reducing harm that arises from social nonconfidence as a mediator, which was the strongest among all the mediators that were studied. Such interventions should provide training on social skills and communication skills. Conflict resolution strategies for adolescents and their mothers (eg, conciliation and positive problem-copying strategies) might also be useful [80]. In addition, given the high popularity of online social networking among adolescents, web-based platforms might be an innovative avenue to identify individuals at high risk for depression and can be a cost-effective tool to improve psychological well-being among adolescents [81].

This study has some limitations. Reporting bias (eg, social desirable bias and recall bias) might exist although most studies [21,22] used self-reported data on online social networking use. The survey may not be nationally representative, due to large geographic variations and its restriction to secondary school students. Our study focused on early adolescence (ie, secondary school students), caution needs to be used in generalizing the results to older adolescents. Future comparisons of this study’s findings with those of older adolescents are warranted. There were differences in some sociodemographic factors (ie, gender, grade, family financial situation, lives with both parents, and academic performance) between those who were followed up and lost to follow-up in our study. Although the influence of some of these factors (ie, grade, academic performance) have been considered in the multivariable analysis, interpretation of the findings should be made cautiously by considering the potential influence of these differences. Online and offline psychosocial relationships were not differentiated, whereas a study conducted in Hong Kong reported that perceived online social support failed to buffer stress for adolescents [82]. We investigated only perceived social support but no other forms of social support. The psychosocial factors represent self-perceptions, which were measured by widely used and validated scales. The measurements of parent–adolescent conflict could not identify whether the parent–adolescent conflict arose from online social networking use; further refinements in future studies were warranted. The relatively short follow-up time and involvement of only one follow-up (in the same school year) is a limitation; the interpretation of the findings should hence take the design into account (ie, that findings reflect short-term but not long term changes). We also did not include other types of interpersonal and personal mediators (or suppressors). The causal mediational sequence could not be derived from the 2-level study, and the follow-up period was relatively short. Moreover, the mediation and suppression effects found in this study were relatively small. Future multilevel research considering more comprehensive factors is warranted. Despite these limitations, our study has the strengths of being novel and using a longitudinal study design and well-validated instruments.

Consistent with previous findings, an increase in online social networking use was associated with an increase in adolescents’ depressive symptoms. The risk effect was, however, partially a balance between the combination of mediation and suppression effects of psychosocial factors. The increase in online social networking use was associated with both unfavorable interpersonal situations (eg, increase in social nonconfidence) that increased depressive symptoms (mediators) and improvement in favorable interpersonal relationships (eg, perceived social support, friendship quality) that reduced depressive symptoms (suppressors). Future studies should clarify the relative contributions of the various interpersonal effects, and the findings of this study should be considered when designing interventions. Furthermore, we found that an increase in online social networking use also improved perceived family support. However, the mechanism, as well as the degree and role of the involvement of parents in adolescents’ online social networking use, have not been well studied in the literature; this involvement may be a potential moderator of the relationship between online social networking use and depressive symptoms, but the direction is unclear. Future studies are warranted. Finally, our findings provide a reminder for various stakeholders to eliminate the stigma against adolescents who are heavy online social networking users, given that online social networking use can generate both harmful and protective effects against depressive symptoms. The understanding or perception that online social networking use is associated with depressive symptoms needs to be reviewed and elaborated.