Participants and recruitment
Using convenience sampling, six junior high schools from two districts of Shanghai, China, were selected from April to May 2018. All first-grade students (a total of 1,329) from the selected schools were recruited. After removing observations with over 5% of items missing and low credibility, 1,243 observations remained. Data from 1,066 (85.8%) students who self-reported themselves as Internet gamers in the past 12 months were used for statistical analysis.
Permission for this in-school survey was obtained before the investigation from schools, legal guardians, and students. As school principals are responsible for students, these were first informed about the study and their consent obtained. Then students and their legal guardians were informed about the study aims and procedure. Verbal informed consent was obtained from legal guardians and students themselves as they participated in the investigation during school time. Participants were informed that all data collected would be used only for research purposes and would be strictly confidential. The background, aim, procedure, and confidentiality of the study were explained at the top of the questionnaire. Participants were free to terminate their participation at any time with no adverse consequences. All eligible participants were asked to complete an anonymous structured questionnaire in class.
The following background characteristics were analyzed: sex, age, mother’s educational level, father’s educational level, perceived family financial condition, residence identity (local or migrant residents), family type (single-parent family or not), and living arrangements (lives with parents or not). These background characteristics were selected by referring to the literature.
Participants were asked how often over the last 12 months they had considered suicide. The three possible response options reflected the frequency of emerging suicidal ideation: “0” (never), “1” (once or twice), and “2” (more than twice). We categorized respondents into two groups for descriptive statistical analysis and logistic regression. Participants who chose “1” or “2” were deemed as exhibiting suicidal ideation and those who chose “0” were considered to show no suicidal ideation. Suicidal ideation category scale scores were used for path analysis.
Internet gaming disorder (IGD)
IGD was assessed using the diagnostic criteria in the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (DSM-V)(30). The measure comprised nine items that assessed IGD symptoms. Participants rated how often they had experienced the symptoms in the previous 12 months on a yes/no scale; “0” indicated absence of the symptom and “1” indicated its presence. Positive responses on ≥5 criteria were considered to indicate IGD (Cronbach’s α = 0.746). IGD continuous scale scores were used for path analysis.
Insomnia was assessed using the Insomnia Severity Index (ISI)(31). The ISI is a 7-item self-report instrument that measures symptoms and insomnia-related problems. The scale has been validated and is widely used in insomnia studies (32-34). The total ISI score ranges from 0 to 28; higher scores indicate more severe insomnia. Scores of 0–7 indicate no insomnia, 8–14 indicate subclinical insomnia, 15–21 indicate moderate clinical insomnia, and 22–28 indicate severe clinical insomnia (Cronbach’s α = 0.838). Participants with a total score >14 are deemed to have clinical insomnia (35). ISI continuous scale scores were used for path analysis.
The 9-item Patient Health Questionnaire (PHQ-9)(36) was used to evaluate depression, as many previous studies indicate its effectiveness and superiority for assessing depression(37, 38). Total PHQ-9 scores range from 0 to 27; higher scores indicate more severe depression. Scores of 0–4 indicate no depression, 5–9 indicate mild depression, 10–14 indicate moderate depression, 15–19 indicate moderately severe depression, and 20–27 indicate severe depression (Cronbach’s α = 0.870). Total scores ≥10 are considered to indicate depression(39). PHQ-9 continuous scale scores were used for path analysis.
Descriptive analyses were first conducted of background characteristics and the prevalence of suicidal ideation, IGD, insomnia, and depression. As the distribution of age was skewed, this continuous variable was described using the median (interquartile range [IQR]), and the median was used to divide this variable into two categories for the subsequent logistic regression. Categorical variables (suicidal ideation, IGD, insomnia level, depression level, sex, father’s educational level, mother’s educational level, perceived family financial condition, residence identity, family type, and living arrangements) were described using frequencies (percentages).
Univariate logistic regression was then performed to examine the association between background characteristics and suicidal ideation, and the association between psychological variables (IGD, insomnia, and depression) and suicidal ideation. After controlling statistically significant background characteristics, we included IGD, insomnia, and depression into a logistic regression model to obtain adjusted ORs (AORs) and the corresponding CIs. Moreover, pairwise correlation analysis of measurements (DSM-V for IGD, ISI for insomnia, and PHQ-9 for depression, questionnaire for suicidal ideation) was used to test the relationships among the variables.
The serial multiple mediation hypothesis for IGD, insomnia, depression, and suicidal ideation was tested using Preacher and Hayes’s method(40). Bootstrapping analysis with 5,000 resamples was conducted to test the significance of the mediation effects(41). The weighted least squares and mean and variance estimator was used as the outcome was categorical. The significant background variable of suicidal ideation reported in the regression analysis was controlled. Model fit indices (root mean square error of approximation [RMSEA], comparative fit index [CFI], Tucker–Lewis index [TLI], standardized root mean square residual [SRMR]) were calculated to assess the model goodness of fit. RMSEA and SRMR values <0.08, and CFI and TLI values >0.90, indicated acceptable goodness of fit(42).
We used IBM SPSS Statistics 24.0 (IBM Corp., Armonk, NY, USA) to conduct the descriptive analysis, logistic regression, and pairwise correlation analysis, and used Mplus Version 8.3 (Muthen & Muthen, Los Angeles, CA, USA) to conduct the path analysis. P values <0.05 were considered statistically significant.